研究条目库

完整代码与数据状态

official fulltext; data on request; no verified code/models/manifest; VOR CC BY4 Figure2 mirrored

完整论文与来源状态

PureApplGeophys10.1007/s00024-026-04066-0 onlineSep12 created07:26UTC

下一触发器

actual issue/future-label manifest;1216vs633 counts; fold-internal oversampling/feature selection; blocked-year test and code

上次判断

Manaus2012-16 sounding24to16indices; daily lightning target; QDA lowthreshold POD.99 FAR.20 author; nominal12UTC input available13:30-14 and16-20valid window do not establish actual4h future skill; day-shuffle not chronological holdout

完整代码与数据状态

publisher abstract only; fulltext/XML403; NRELWINDToolkit and synthetic data; no verified code/weights/splits

完整论文与来源状态

WindEngineering10.1177/0309524X261488169 onlineSep11 created20:27UTC buffer

下一触发器

causal change detection/update; exact leads/units/threshold/year split; fulltext/code

上次判断

regime segmentation plus lightweight predictor; RMSE1.86 TSS.81 FAR.15 BS.131 author; Transformer better nRMSE/MAE/CSI/ETS; do not infer uniform superiority or causal operational update

完整代码与数据状态

no retrieved abstract/fulltext/code/weights or reusable figure license

完整论文与来源状态

AppliedSoftComputing10.1016/j.asoc.2026.116434 createdSep12 06:54UTC; September cover only; first-online unknown

下一触发器

actual online date; fulltext; dataset/lead/rare-event threshold/split; artifacts

上次判断

title-only rarewind candidate; September1 TDMlicense start is not verified publication time

完整代码与数据状态

IEEE unavailable; Crossref no abstract; data/method/leads/scores/code unverified

完整论文与来源状态

ICCCBDA10.1109/ICCCBDA69573.2026.11676482 April2026 conferenceApr24-26; createdSep11 19:04UTC lateindex

下一触发器

abstract/fulltext and actual multimodal forecast contract; code/data

上次判断

late indexing not fresh September scientific result; title alone insufficient to infer model or skill

完整代码与数据状态

full arXivHTML read; anonymous mlwp-evaluation-28AF linked but payload/license not verified; no new weights; nonexclusive license no image

完整论文与来源状态

arXiv2609.10564v1 explicitAug25 13:32UTC conflicts with SeptemberID/listing; preprint only no journal

下一触发器

date clarification; frozen code/predictions; real initialization and nativeHRES comparison; station/block bootstrap

上次判断

12stations2016-22 00UTC+6to72h shared pairs; HRES2.89 FCN32.96 GC2.94m/s RMSE; highwinds all underestimate; ERA5initialization and0.25degree HRES not fully operational equivalence

完整代码与数据状态

Sep13 01:10UTC9f1a2f07 changes forecasts.json only; no rootlicense; README historical backtest not new result

完整论文与来源状态

existing July2026 app; no scientific paper verified

下一触发器

substantive model/validation change; independent station truth; actual forecast met and frozen splits/license

上次判断

5Philippinecities HGB1-24h OpenMeteo; latest automatic products not new skill; historical24hMAE3.76vs4.45ug/m3 remains author-only

完整代码与数据状态

publisher abstract only; fulltext403; no verified code/data

完整论文与来源状态

WAF10.1175/WAF-D-26-0041.1 Sep11 formal journal record

下一触发器

fulltext; years/truth/member count; joint probability and tail validation; code

上次判断

Quebec32catchments ECMWF2/5/9days; median absolute deviation linear variant most consistent; nonlinear maps not uniformly better

完整代码与数据状态

ERA5/HWind/IBTrACS identified; no verified code/weights/splits; VOR CC BY-NC-ND no image

完整论文与来源状态

JGRML10.1029/2025JH001170 onlineSep11

下一触发器

storm-isolated splits; wind averaging period and exact grids; forecast coarse inputs; artifacts

上次判断

fivefold historical ERA5 toHWind1998-2013; MSW/R17 correlations.81/.92; IBTrACS1990-2020 does not itself ensure time/event independence

完整代码与数据状态

publisher abstract; fulltext/XML403; no frozen errors or scripts

完整论文与来源状态

Earth10.3390/earth7050149 Sep11

下一触发器

original model split/years/leads; log-error transformation; error tables/code; confounding controls

上次判断

345180model-grid-month records11models CHIRPS; conditional error diagnosis not new nowcasting; LaNina/negativeIOD confounded; MAE54.10mm aggregate does not establish reliability

完整代码与数据状态

Sep11 d654ad243d Apache2 wrapper; HF Artamta/FuXi-S2S-ONNX pinned5d7a6b1 lists1588847byte graph and2078823204byte weights; checkpoint CC BY-NC-ND; no payload rerun

完整论文与来源状态

new integration of2024 model, not new paper or trained weights

下一触发器

official sample equivalence; hourly aggregates/availability; common issue and members; mirror-original payload comparison

上次判断

two daily means76variables1.5degree; 00UTC labels day aggregate; tp hourly-accumulation mean times24 for daily total; member00 stochastic; four6h samples approximate not formal preprocessing

完整代码与数据状态

ECMWF anemoi-core Sep11 89f26fadd frozen sparse CrossGridProjector and distributed tests; Apache2; no model weights

完整论文与来源状态

new input-grid preprocessing module, complete diffusion example deferred

下一触发器

full configuration/weights/manifest; link toSep7 institutional demo; conservation and real forecast drivers

上次判断

projection after sharding before normalization is interpolation infrastructure, not trained spatial generator or new skill

完整代码与数据状态

official abstracts; data on request; code promised but no verified repo/weights; no reusable figure license

完整论文与来源状态

AAS10.1007/s00376-026-5840-z onlineSep10 created08:15UTC; buffer

下一触发器

real vs synthetic gap mask; grid/height/year splits; FSSscale; public code; downstream forecast impact

上次判断

attention3DU-Net improvedCGAN SWAN contemporaneous gap filling;6562test samples >35dBZ POD.71 FAR.13 TS.67 FSS.81 author scores not future forecast

完整代码与数据状态

official abstract; no complete data/code/splits verified

完整论文与来源状态

WES10.5194/wes-2026-154 discussionSep10 buffer; inreview preprint

下一触发器

fulltext/splits; fitted dynamics uncertainty; richer weather inputs and artifacts

上次判断

AR3Gaussian wind to empirical powercurve GaussHermite conditional Bayes risk;2UKsites9leads1-48h; not universal physical bound beyond wind-only information

完整代码与数据状态

SSRN publisher abstract; exact variables/site/leads/split/code unverified

完整论文与来源状态

10.2139/ssrn.7446076 createdSep11 13:18UTC; first-public date unresolved

下一触发器

fulltext and actual minute-scale contract;30day vs6year controlled split; artifacts

上次判断

ABL turbulence LSTM/GRU/TCN/Transformer/ESN withMFDFA/Langevin interpretation; do not generalize30day sufficiency

完整代码与数据状态

official2page extended abstract; no blind-year/code/weights; no PDF mirrored

完整论文与来源状态

10.11159/icert26.113 createdSep11 19:47UTC; conferenceAug20-22 publicationSep only; first-public unresolved

下一触发器

first-public date; independent-year contract; metric definitions/code

上次判断

Donghae2011-20 hourly WS WD SSR;TransUNet1D DDPM8760hour synthetic year; annual scenarios not futureS2S forecasts

完整代码与数据状态

publisher abstract; fulltext403; no verified artifacts; VOR CC BY-NC-ND

完整论文与来源状态

10.1002/met.70238 onlineSep9 createdSep10 02:18UTC outside48h

下一触发器

train/test years; stations/TFA; cycles; strong baseline/code

上次判断

2022-23 Beijing6-72h GraphCast/Pangu drivenIFS/GFS;XGB RMSE24.1percent better than simplemean author claim; actual independent chronology unverified

完整代码与数据状态

Zenodo22687889 v0.1.0 CC BY4.0 lists241458byte archive; downloadTLS; no weights/full inputs verified; complete v1 Figure4 mirrored

完整论文与来源状态

arXiv2609.11696v1 Sep10 15:21:40UTC; preprint only no journal; manuscript JAMES label is not acceptance

下一触发器

matched member/input budgets; independent scale tuning and years/events; code payload/weights

上次判断

Swiss2018/2021 48h events at1/2/3weeks; IFS50 vs selected dry/middle/wet WRF3; DDPM-ens largely retains CRPS while improving spatial scores; 2018-23 training removes event dates but includes later years; SSR remains low

完整代码与数据状态

official abstract; PDF403; no verified code/weights/data; specific figure unavailable

完整论文与来源状态

NSR DOI10.1093/nsr/nwag570 accepted manuscript onlineSep10, not typeset final

下一触发器

full manuscript; CTM/met/lead/split contract; process budgets and artifacts

上次判断

2022-23 China338cities; neural correctors inside CTM equations; author accuracy gains6.1percentPM2.5/3.0percentO3 vs forecaster-adjusted forecasts; do not relabel as RMSE reductions

完整代码与数据状态

PDF read; WRF/process data on request, stations restricted; code promised upon publication; no weights; nonexclusive arXiv license

完整论文与来源状态

arXiv2609.11230v1 Sep10 08:28:44UTC; preprint only no journal

下一触发器

blind region/seasons; WRF nudging provenance; jointUV and extreme/probability validation; public code

上次判断

3km Shanxi240x176 WRF2024 prior, JFM2025 contemporaneous stations; separateUV1000stepDDPM10members;238station region chosen using high ERA5 errors;19percent median RMSE gain is selected-region reconstruction not forecast

完整代码与数据状态

BSD3 github.com/blackjjh21-afk/r2p-nowcasting; Zenodo22684438 v1.1.0 lists6.55MB; code plus anonymous CSV; raw data/weights/full predictions not bundled

完整论文与来源状态

Sep10 substantive adapted exPreCast training-selection-export and figure reproduction; no formal paper verified

下一触发器

raw-to-tuples; training years/RN60 intervals; checkpoint/full arrays; licensed manifest and independent reproduction

上次判断

4km10min seven radar frames toRN60;128heldout/514fit;35088common issues2024-25JJA+Sep;2023selection;60minCSI-M .4798vs.3883exPreCast but180minRMSE1.733vs1.724mm; same-checkpoint masked context is sensitivity not retrained radar-only

完整代码与数据状态

publisher Crossref abstract only; sourceTLS; no verified code/weights/data or specific figure

完整论文与来源状态

ResearchSquare10.21203/rs.3.rs-10600756/v1 postedSep9; buffer; in-review preprint not journal acceptance

下一触发器

full text; actual coarse source; temporal split; physical and probability tests; code

上次判断

Germany2020-24 described; supervised100to10to1km; author Tmaxr.99/NSE.98 and W.37 unitunknown; precipitationP95/terrain weak; not demonstrated real forecast skill

完整代码与数据状态

official abstract/license read; full archive/checkpoints/file-level licenses not verified

完整论文与来源状态

ScientificData10.1038/s41597-026-08043-z Sep9 date-only; buffer

下一触发器

archive manifest and data licenses; training years/code; joint-variable extremes

上次判断

10CMIP6 models2SSPs1980-2059 daily4kmP/Tmin/Tmax SRCNN/SRGAN; Daymet comparison without retrieved numeric tables; climate projection ensemble not S2S probability ensemble

完整代码与数据状态

publisher abstract accessible; no full evaluation/code/weights verified

完整论文与来源状态

TAC10.1007/s00704-026-06545-9 onlineSep10

下一触发器

lead/lag and time split; AQI/pollutant target overlap; metric definitions

上次判断

about10years Kmeans+XGBoost meteorology+pollutants; claimed98percent accuracy undefined; do not upgrade contemporaneous AQI estimation to core future forecasting

完整代码与数据状态

publisherTLS; no abstract/code/data/weights verified

完整论文与来源状态

bookchapter10.1201/9781042015863-26 Sep10

下一触发器

official abstract/fulltext; region/data/model/lead/split/metrics

上次判断

title-only air-quality candidate, no skill inference

完整代码与数据状态

official Dergipark abstract; NASA POWER identified; no verified code or forecast timing

完整论文与来源状态

10.47495/okufbed.1799905 Sep9 online; buffer

下一触发器

exact location, lag/lead, time split, reproducible code

上次判断

2001-24 single-site daily12models; ANN/XGB/RF R2.680/.674/.665; LSTM/GRU.063/.056 not general evidence against sequence models

完整代码与数据状态

official full text identifies monthly0.1degree aggregation

完整论文与来源状态

DiscoverEnvironment10.1007/s44274-026-01047-x Sep11

下一触发器

genuine future forecast contract, reference and temporal validation

上次判断

MODIS/Sentinel2/MERRA2 monthly estimation; not0-72h rain forecast

完整代码与数据状态

Publisher abstract only; MDPI full text/XML 403; no verified code, weights or frozen data

完整论文与来源状态

Applied Sciences 16(18)8964; DOI 10.3390/app16188964 online Sep9, created 15:01:56 UTC

下一触发器

full text; per-origin causal VMD and imputation; folds; persistence; true forecast meteorology; extremes

上次判断

Lima SENAMHI 2015-24 hourly PM2.5/PM10/NO2, 24-72h; five-fold time split; reported RMSE 9.04/19.93/9.41 at24h to9.79/24.56/11.02 ug/m3 at72h; training-only fitting does not prove causal test decomposition

完整代码与数据状态

Zenodo 20699195 Jun15 lists 91.8kB scripts, not full data or weights; VOR CC BY-NC-ND; accepted-version figure not checked

完整论文与来源状态

JGR Machine Learning and Computation DOI 10.1029/2026JH001275 online Sep9; accepted Jul16

下一触发器

full text; training years; variables; independent truth; real forecast forcing; weights/data

上次判断

25km ERA5 to4km WRF surface fields using U-Net; AP2014 extreme cases; do not infer1km or operational forecast skill from older title

完整代码与数据状态

Zenodo22693613 v1.0.1 MIT createdSep10 16:14UTC lists126538byte codeZIP; Sep11 ed51a1e citation; no data/weights

完整论文与来源状态

software archive verified buffer; JGRsubmitted not acceptance

下一触发器

actual weights/paper; restricted data reproduction; archive payload

上次判断

upgrade metadata-only to fixed software archive identity; 6to36min lightning target; source-only not complete reproduction

完整代码与数据状态

code/configs/author scorecards at github.com/VDT37/Dissertation; no verified weights, full dataset or root license

完整论文与来源状态

Sep9 12:39:28 UTC commit07f88b4d9; thesis artifacts, no formal paper verified

下一触发器

blind test; years/region/leads and manifest; weights/license; matched FSS/PSD subsets

上次判断

UK precipitation project;153487train/13281val,8members25Heun; val RMSE mean0.899 vs advection1.140 and member1.151mm/h; fairCRPS0.1735; no independent skill or calibrated probability claim

完整代码与数据状态

Sep12 6c8e009c JSON source labels and patch verifier; MIT; LFS sampledtrain pointer106881532bytes not fetched; storm39.91796875 stillmm/5min

完整论文与来源状态

material sampling/manifest verification; no new model skill

下一触发器

raw-to-LFS manifest; counts before/after recomputation; threshold units/consumer; natural-distribution calibration

上次判断

2015-22train2023val2024testJJAS;27.2M/3.38M prepatch counts author recompute vs externally supplied12.85M/1.59M after and32021test;top p99>0 keep.6171 rest.3829; val resampled testnatural; not independent reproduction

完整代码与数据状态

Sep11 22625c00 four conditionalSFNO checkpoint calls nonreentrant; gradient equality tests

完整论文与来源状态

material training fix; no new checkpoints/results

下一触发器

affected run/config inventory; retraining; other checkpoint paths; real forecast scores

上次判断

encoder grads could beNone with checkpoint>=1; fix only conditionalSFNO; does not invalidate all ACE weights

完整代码与数据状态

github.com/carminecoppola/Hi-WeFAI_tutorial now MIT; RainPredRNN18to6 frames512square;249MB weights require administrator contact, no public link

完整论文与来源状态

Sep9 14:50:26 UTC commit237a7749d adds LICENSE

下一触发器

actual weights URL; frame cadence; data/model licenses; training/test and metrics

上次判断

license-only substantive state;7.7GB external geospatial data not assumed covered by code MIT; cannot claim runnable complete inference

完整代码与数据状态

AORC/ERA5 identified; no study-specific code/weights/32case manifest verified; complete CC BY4.0 v1 Figure1 mirrored

完整论文与来源状态

arXiv2609.09247v1 Sep8 11:13:32UTC; Sep10 ao-ph listing; buffer; preprint only no journal

下一触发器

205k/165k checkpoints and manifest; blind year/stations; density-normalized guidance; calibrated uncertainty

上次判断

CONUS2020-21train/2022validation-selection;32patches800m,4members32steps;1percent RMSE0.318 vs nearest0.431K, sparser loses; exact-time reconstruction not forecast; different sweep setup cannot pool scores

完整代码与数据状态

abstract and data references verified; full text403; no code/weights/frozen data verified

完整论文与来源状态

Forecasting8(5)80 DOI10.3390/forecast8050080 onlineSep8; created11:23:51UTC buffer

下一触发器

years/provinces/full scores; historical met latency; frozen independent test; extreme reliability

上次判断

four provinces daily historical pollution/met to nextday; RNN/LSTM/GRU no universal winner; peaks underestimated; not NWP-driven/global validation

完整代码与数据状态

github.com/zeaccepted/PCSDiff-Medium-term-precipitation-forecast empty; API409; no code/weights; full arXiv text read

完整论文与来源状态

arXiv2609.06942v1 Sep7 02:24:57UTC; preprint only no journal; outside48h despite Sep9 listing

下一触发器

actual repo files; exact years/member/threshold/unit contract; global labels; calibrated probability and operational trial

上次判断

10day1.5degECMWF to0.25degCRA reanalysis; PIMD then regression-DDIM; authors14.8-16.1percentRMSE gains, no independent verification

完整代码与数据状态

abstract, Oxford accepted record and Crossref dates verified; full new experiment/code not checked

完整论文与来源状态

QJRMS DOI10.1002/qj.70309 onlineSep6, acceptedAug26, createdSep7; old arXiv2601.04701

下一触发器

data correction status; independent observations; complete code/test manifest

上次判断

6h GraphCast exposes Ethiopia06UTC ERA5 warm artifacts from temporally displaced assimilation; approx7percent dates; reanalysis mismatch not automatically model error

完整代码与数据状态

arXiv2609.10135v1 uses LightGBM and prompt self-optimization; no verified forecast contract or reproducible artifacts

完整论文与来源状态

Sep9 preprint only, no journal verified

下一触发器

independent forecast leads/samples, event scores and released artifacts

上次判断

tourism alert text-quality4.2to8.9 is not precipitation skill; monthly AQ drivers, indoor monitoring and routine app refreshes excluded

完整代码与数据状态

BSD-3-Clause code; v1.0.0 release verified this run as published Sep2 00:24:38UTC, not fresh; containers/scripts/cached intermediates; no full forecast.zarr archive

完整论文与来源状态

arXiv 2609.08412v1 submitted Sep 8 08:18:40 UTC; preprint only, no journal information; arXiv nonexclusive license, no figure mirrored

下一触发器

frozen input/mask manifest and full-year independent blocks; precipitation, budgets and equal-member tests

上次判断

retain 1-15 day training-free ensembles from Aurora/GraphCast/SFNO/AIFS; 112 initialisations in eight weekly blocks and ten members; 240h CRPSS deficits vs AIFS-ENS 0.036-0.131; architecture-specific tuning, mean-field offsets, IFS missingness and absence of precipitation limit claims

完整代码与数据状态

public CEDA/TU Wien soil moisture and ROA/ERA5 sources; NetNCC notebook linked at github.com/JawairiaAA/NetNCC but study-specific training code, weights and code license not verified

完整论文与来源状态

EGUsphere DOI 10.5194/egusphere-2026-4687 discussion Sep 8; HESS under review, not accepted; CC BY 4.0

下一触发器

soil-input training code/weights; test-independent probability thresholds; calibration; other months and actual rainfall/radar truth; real ASCAT latency

上次判断

retain 4h West Africa convective-core experiment: three hourly IR fields plus ASCAT wetness/mask, June 2007-20 train and June 2021-24 test; 0.1 degree input often helps evening/night but loses to IR-only in morning; proxy target and tuned CSI thresholds prevent universal rain-skill claim

完整代码与数据状态

ORAS5 and ERA5 identified; no dedicated public code, weights or frozen forcing manifest verified; complete CC BY 4.0 v1 Figure 1 mirrored

完整论文与来源状态

arXiv 2609.08606v1 submitted Sep 8 11:42:25 UTC; preprint only, no journal information

下一触发器

real forecast atmospheric forcing; released code/weights; high-resolution ensembles; matched dynamical/AI baselines; OHC/EKE and ice probability definitions

上次判断

retain prescribed-atmosphere 60-day ocean/sea-ice emulator, not coupled atmospheric S2S forecasting: 13M CNN/SFNO residual model, 1993-2016 train/2017-18 val/2019-21 test; eight members at 1 degree but only deterministic 0.25 degree; currents, mixed layer and mesoscale energy degrade

完整代码与数据状态

github.com/yyalexlee/EastAsiaClimateExtremes README lists external data downloads and analysis scripts; Zenodo 22139821 API returned 504, archive completeness/license and full payload not verified

完整论文与来源状态

arXiv 2609.08241v1 Sep 8 04:36:38 UTC; buffer recovery; preprint only, no journal information; article CC BY 4.0

下一触发器

archive/file license; valid-week coordinates and full manifest; shared years and model samples; train-only thresholds and overlapping-event control

上次判断

retain 1.5 degree East Asia weekly extremes from ERA5 1940-2024/OISST 1982-2024 and ECMWF week3 V2016/V2024; separate model climatologies and 1996-2015 vs 2004-2023 periods do not establish model-upgrade skill or AI forecast gain

完整代码与数据状态

TIGGE/DWD sources identified; no dedicated model code, weights or frozen station split verified; nonexclusive arXiv license, no mirrored figure

完整论文与来源状态

arXiv 2609.07512v1 Sep 7 14:00:29 UTC; buffer recovery; preprint only, no journal information

下一触发器

code and frozen station/tuning manifests; cross-region tests; precipitation and multivariate extension; calibration versus sharpness

上次判断

retain 48h 50-member ECMWF T2M/wind postprocessing over Germany with 30 percent stations held out; altitude-aware pools beat ordinary pools but DRN ALP CRPS 0.8812 is worse than standalone 0.8743; no universal GNN or combination advantage

完整代码与数据状态

IEEE document 11683859 returned empty HTTP 202; no abstract, dataset, model, leads, split, scores, code, weights or image reuse license verified

完整论文与来源状态

TGRS DOI 10.1109/TGRS.2026.3731815 journal-article record created Sep 8 19:03:22 UTC, pages 1-1 and year 2026; exact online date unverified

下一触发器

official abstract/full text and online date; input/target/event split; skill and public artifacts

上次判断

retain title-level water-vapor/radar/terrain fusion metadata only, not a new verified core-nowcasting result

完整代码与数据状态

Elsevier page blocked; no abstract, model details, dataset, lead times, scores, code or weights verified; VOR license CC BY-NC-ND 4.0 is outside figure whitelist

完整论文与来源状态

Science of Remote Sensing DOI 10.1016/j.srs.2026.100500 created Sep 7 23:33:31 UTC in buffer; September cover date and Aug 29 license start do not verify exact online date

下一触发器

full text and true first-online date; flow supervision, data/splits, probability and baseline contract; code/license

上次判断

retain buffer indexing metadata; possible older publication, not established fresh scientific event; do not infer equations or skill from title

完整代码与数据状态

OneScience-Group/PrecipitationSRCNN explicitly uses synthetic mm/day samples on 216x488 for workflow testing under CC BY-NC-ND; enhansome commit 3f2fea00 changes stars/issues/commit-time fields only

完整论文与来源状态

no new real-data forecast result; OneScience HF deposit Sep 7 09:56 UTC is buffer-window artifact tied to older paper

下一触发器

real data, frozen evaluation and substantive algorithm/data/license change

上次判断

exclude synthetic workflow validation and repository-metadata refresh from scientific progress

完整代码与数据状态

MeteoSwiss c4dl-multi supplies the earlier architecture; gpacey/C4-UNetGRU-sens-exps contains result CSVs, reliability data and two plotting notebooks but no paper-specific training entry point, checkpoint, complete sample manifest or root license; POH is available through MeteoSwiss OGD

完整论文与来源状态

arXiv 2609.04427v1 entered the Sep 7 recent listing; deposit timestamp is Sep 3 19:48 UTC; arXiv preprint only and the author note under review for GRL is not acceptance; v1 is CC BY 4.0 and complete Figure 1 is mirrored

下一触发器

paper-specific training code, weights, frozen sample manifest and license; independent ground-hail or impact truth; cross-country radar tests; motion-direction and reliability/Brier/FSS stratification

上次判断

retain as direct 0-60 minute radar-proxy hail probability evidence: fixed UNet-GRU experiments show six convective seasons extend equal-BCE skill from about 35 to 60 minutes compared with one-year training, while rotation or mirroring helps large datasets but degrades small ones; POH remains a radar-derived proxy and later-lead probabilities become overconfident

完整代码与数据状态

public ccores algorithm identifies MSG convective cores; EUMETSAT source data have access conditions and the core database or NFLICS fields are request-only; the accepted paper says NCAST model and evaluation code are request-available and will be public after publication, but no model repository or checkpoint was found; complete Figure 3 is mirrored from the CC BY accepted version

完整论文与来源状态

Quarterly Journal of the Royal Meteorological Society DOI 10.1002/qj.70303, formally online Sep 6 under CC BY 4.0; Crossref record created Sep 7 05:01 UTC and recovered in the current buffer window

下一触发器

public NCAST/PANCAST code and checkpoints; final VOR bibliographic details; real-time latency and failure logs; independent rain-gauge or radar verification; calibration in low-base-rate northern areas

上次判断

retain direct satellite convective-core nowcasting evidence: 2004-2018 training, 2019 validation and 2020-2024 testing support 1/3/6-hour AUC 0.96/0.91/0.89 versus NFLICS about 0.82/0.80/0.78, but the target is a CTT-wavelet core proxy rather than rainfall, and the five random-initialisation members do not represent meteorological uncertainty

完整代码与数据状态

Apache-2.0 Anemoi documentation supports EDM diffusion, stochastic interpolants and a flow-matching-like ODE setup; no experiment-specific config, variable list, checkpoint, training manifest or evaluation package is linked

完整论文与来源状态

ECMWF AIFS Blog dated Sep 7 2026, DOI 10.21957/f59cb63b99; technical institutional prototype rather than a peer-reviewed paper

下一触发器

dedicated config and weights; training years and variables; ERA5-perfect versus AIFS/physics-driven tests; member CRPS, spectra, extremes, conservation and temporal-coherence results

上次判断

retain as a high-value modularity signal: one flow-matching model maps O96 roughly 100-km six-hourly states to N320 roughly 30-km hourly five-step chunks and can attach to low-resolution AIFS ENS without fine-tuning, but animations and interface compatibility do not establish forecast or S2S skill

完整代码与数据状态

repository created Sep 7 11:30 UTC contains a 4.2 kB README and three 2.6-3.0 MB figures only; code is said to be under organisation and no weights, data, split manifest, full result table or license is present

完整论文与来源状态

no public paper, DOI or arXiv record; submitted to IEEE Transactions on Artificial Intelligence is an author statement only

下一触发器

public manuscript; source code and checkpoints; license; Wannan dataset provenance; threshold and lead-wise metrics, baselines, ensemble and sampling cost

上次判断

retain framework-only metadata: optical-flow priors, multiscale truncated residual diffusion and lead-aware correction are relevant, but three headline CSI-M/HSS/SSIM rows cannot be compared without definitions, common cases and a citable paper

完整代码与数据状态

no public code, weights, dataset, model card, abstract or reusable figure license verified; Crossref exposes only a Springer PDF TDM link

完整论文与来源状态

Lecture Notes in Computer Science book chapter DOI 10.1007/978-3-032-37657-2_28, online Sep 8; authors Yanan Guo, Xiaoqun Cao and Mengge Zhou

下一触发器

accessible abstract or full text; forecast horizon, grid, variables and splits; exact physical-consistency tests; code, weights and license

上次判断

retain title-level metadata only; do not infer S2S or short-range skill from physically consistent and multi-scale wording

完整代码与数据状态

no accessible IEEE abstract or PDF, event labels, predictors, lead-time definitions, calibration output, code or data verified

完整论文与来源状态

IEEE eIT 2026 proceedings article DOI 10.1109/EIT68936.2026.11670437; print date May 2026, Crossref created Sep 7 19:11 UTC, so this is late indexing rather than fresh research

下一触发器

full text or official abstract; tornado labels and event split; lead-wise probabilistic metrics and calibration; code, data and operational evidence

上次判断

retain as high-relevance historical metadata only; a fresh DOI index does not change the May publication date or establish a new nowcasting result

完整代码与数据状态

IDSNR Community-Air-Quality-Intelligence is a 12 kB MIT synthetic-demo framework with causal lags, persistence and ridge but no real observations or results; THULab WiDS 2023 HF data only consolidates existing fragments into single tsfiles; new weather and AQ dashboards, agriculture or energy tools and a bilingual thesis brief add no research artifact

完整论文与来源状态

Navi Mumbai Sadhana article was published Sep 7 but its public abstract gives no forecast lead or temporal split and reports near-perfect station-category tree classification; an Aug 1 monthly global AQI paper is only late-indexed

下一触发器

reconsider for frozen real sensor and meteorological data, explicit lead and valid-time contract, temporal or spatial holdout, independent baseline, uncertainty and a material scientific artifact

上次判断

exclude monitoring classification, synthetic demos, file-layout-only updates, applications and proposal material from the four core research lines

完整代码与数据状态

no paper-matched code, weights, data archive or frozen evaluation package was found; the accessible evidence is the publisher abstract plus an earlier MetMed 2025 conference abstract

完整论文与来源状态

Meteorological Applications DOI 10.1002/met.70237, accepted Aug 12 and formally published Sep 6 under CC BY-NC 4.0; Crossref record created Sep 7 02:05 UTC

下一触发器

full text; training and test years, forecast leads, station count and absolute marginal or multivariate scores; code, weights and extreme heat-humidity verification

上次判断

retain as transferable joint-distribution postprocessing evidence, not a precipitation or radar result: a bivariate DRN jointly models air and dew-point temperature from a 10-member Meteocat multimodel ensemble and outperforms univariate DRN plus ECC and bivariate EMOS in reported multivariate skill, while rank-histogram calibration remains similar to EMOS

完整代码与数据状态

GPL-3 mixnhreg R package version 0.1.0 implements mixture regression, non-cyclic boosting and LogS or CRPS fitting, but its last commit is Dec 17 2024 and no paper-specific frozen runner or data snapshot was released in this window

完整论文与来源状态

Environmetrics 37:e70145, DOI 10.1002/env.70145, formally published Sep 6 under CC BY 4.0; upgrades the Dec 2024 CC BY arXiv 2412.09583v1 to a journal article

下一触发器

VOR-to-arXiv content check; frozen paper runner and data manifest; spatial or unobserved-station extension; multi-lead, precipitation and multimodel-group validation

上次判断

retain formal-status update as high-value probabilistic postprocessing method: 2015-2019 ECMWF and DWD data at 280 German stations train local plus-24-hour temperature models and 2020 is held out; MIXSAMOS-GB improves aggregate CRPS to 0.69 and median station CRPSS by about 4.5% versus SAMOS, but intervals are slightly wider than SAMOS-GB and the method does not model space

完整代码与数据状态

Sep12 781b746b LAYER_A_ALIGN_WITH forwarded to harness;09a866f testname; fixed diff read; no new scores/corpus/weights

完整论文与来源状态

material common-case evaluation script fix

下一触发器

matched case list/count and coverage; rerun paired bootstrap; distinguish oracle futureinputs

上次判断

oracle futurepair and median3sixframes need stricter cases than confidence; prior batch omitted alignment; fix does not itself revalidate old results

完整代码与数据状态

MIT Python and Home Assistant code reads CWA O-A0059-001 1.25-km reflectivity, combines 1, 3 and 10 km maxima or coverage with Marshall-Palmer QPE, and uses two-frame phase correlation for 10-60 minute ETA; commit 6c2b2a69c67a on Sep 6 14:07 +08 adds the coverage and QPE scoring path

完整论文与来源状态

personal GitHub tool, no paper, formal release, trained model, frozen event benchmark or independent score

下一触发器

co-located gauge events and arrival-time truth; pre-registered thresholds and weights; CSI, FSS, ETA error and latency by bright band, blockage and regime

上次判断

retain as a transparent deterministic baseline and current-rain infrastructure, not an AI result or calibrated probability system; the 18-dBZ threshold, 35/40/25 percent weights and score threshold 50 are rules without training or held-out validation

完整代码与数据状态

HF repository created Sep 5 17:54 UTC contains DDPM-V and DLPM-Eps configs, normalizations and two 44,918,712-byte safetensors; 80/10/10 random split and validation or test losses are declared, but data years, region, forecast cycle, precipitation accumulation, units, source code and license are absent

完整论文与来源状态

model artifact only, no paper or formal release; buffer-window catch

下一触发器

complete HRRR provenance and redistribution license; temporal or event split; conditioning variables and forecast lead; code; CRPS, spectrum, extremes and conditional consistency against baselines

上次判断

retain as low-evidence generative precipitation-field artifact: the card is tagged unconditional-generation and provides no initialization or lead time, so it is neither precipitation nowcasting nor verified forecast downscaling; losses from different diffusion objectives are not ranked directly

完整代码与数据状态

no accessible abstract, region and period, fusion inputs, model description, split, metrics, code, data package or reusable figure license was verified

完整论文与来源状态

Water Resources Management DOI 10.1007/s11269-026-04881-z, accepted Aug 25 and first online Sep 5; subscription record with Springer TDM terms only; buffer-window catch

下一触发器

accessible abstract or full text; precipitation and gauge inputs; chronological and spatial validation; rainfall-to-flood propagation metrics; code, data and license

上次判断

retain title-level metadata only: the title indicates same-time IMERG fusion used for flash-flood simulation, not future precipitation-field prediction, and no technical effect should be inferred from the DOI record alone

完整代码与数据状态

WP-MIP was delivered Sep 7 15:30 CEST and the official session now labels an online presentation, but supporting material is restricted to registered attendees; no public slides, recording, central database schema, forecast archive, code or weights are available; EMS2026-484, 302 and 124 remain future sessions at this cutoff

完整论文与来源状态

meeting runs Sep 6-11 in Utrecht; all scientific abstracts were published or updated in June, and a talk taking place does not refresh the underlying abstract

下一触发器

public WP-MIP contribution protocol, model list, common and centre-specific initialisation archive and verification code; open slides or recordings; public artifacts from EMS2026-484, 302 and 124

上次判断

update the event state from scheduled to delivered-but-not-public for WP-MIP; retain as an institution signal rather than a result, because attendee-only material cannot support new technical claims

完整代码与数据状态

main-window GitHub and HF hits include an empty thunderstorm-nowcasting repository, automatic GOES or forecast JSON image refreshes, SIH urban-flood templates, weather applications and repeated three-day AQI coursework; no frozen scientific model, dataset, independent result or license change was verified

完整论文与来源状态

riverine and one-day-ahead streamflow papers published Sep 6 forecast hydrological discharge rather than atmospheric or radar fields; EdgeAQ-Net carries a future Sep 26 publication date and monitoring-style accuracy claims without a forecast contract

下一触发器

reconsider only for original atmospheric prediction data, event-isolated evaluation, a real model or checkpoint, a formal in-scope publication, or a material license or result change

上次判断

exclude routine output commits, empty or template repositories, non-atmospheric flow prediction, image restoration, traffic or energy forecasting and monitoring-only AQ systems from research updates

完整代码与数据状态

Zenodo deposit contains one 464646-byte handout only; no code, checkpoints, labels, frozen bootstrap stream or license is included

完整论文与来源状态

Zenodo DOI 10.5281/zenodo.22358588, deposited Sep 5 and created Sep 6 00:38 +08; supplementary handout self-identifies as companion to a Sep 9 S2S2D poster, not a peer-reviewed paper

下一触发器

public claims ledger, code, checkpoints and prospective holdout; independent conference-program record; Phase 2 conditional precipitation model and joint trajectory verification

上次判断

retain direct 1-42 day regime-probability evidence: leave-one-meteorological-year-out TCN experiments beat climatology and a Markov floor at leads 10-20, but the five slow boundary channels add no detectable skill, all results are cross-validated, the 42 leads are independent marginals, and conditional precipitation generation is deferred

完整代码与数据状态

author repository provides a public CSV, 7.65 MB data archive and figures but no executable model code, notebook, weights, environment freeze or license despite the paper claiming a complete workflow

完整论文与来源状态

Urban Climate DOI 10.1016/j.uclim.2026.103120 received a formal Crossref record Sep 5 12:14 UTC; Elsevier exposes only a September 2026 cover date, so exact online publication time is unverified; earlier arXiv 2602.19654v2 is CC BY 4.0

下一触发器

exact VOR online date; runnable code and weights; physical-unit correction; independent station observations; future forecast meteorology; spatial and external-year holdouts

上次判断

retain as a 3-hour CAMS-reanalysis target experiment, not a high-resolution operational AQ forecast: 21 days of past CAMS and ERA5 predict the next 3-hour step at four pollutant locations; reported R2 is high, but no real station truth or future NWP driver is used and reported concentration/RMSE units conflict with the public CSV magnitude

完整代码与数据状态

Apache-2.0 GitHub code, environment files and a 3.2 GB Apache-2.0 Hugging Face checkpoint are public; latest code commit remains Aug 13 and the checkpoint is about five months old, so no new artifact event accompanies publication

完整论文与来源状态

Journal of Geophysical Research: Machine Learning and Computation DOI 10.1029/2026JH001313, formally published Sep 5 under CC BY 4.0; earlier arXiv 2602.18646v2

下一触发器

frozen start and boundary manifests; station or radar validation; stochastic small-scale and precipitation calibration; independent reproduction across forcings and climates

上次判断

update from February preprint to formal publication: Swin-based 4-km hourly LAM plus U-Net diagnostics is stable across multiyear and 30-year forced runs and preserves CESM ensemble spread, but systematically loses 10-100 km variance and warm-season convective precipitation objects; this is dynamical downscaling rather than an initialized 0-72 h forecast

完整代码与数据状态

Zenodo v1.0.0 provides 4.51 MB of BSD-3-Clause Python and CC BY 4.0 derived results, ten rebuildable figures and verification expectations for 204 table values; raw forecast fields, per-case predictors, competition labels and one extended ERA5 generator are absent

完整论文与来源状态

code-and-results deposit DOI 10.5281/zenodo.22342203 created Sep 5 19:53 +08 and synchronized to an undistributed Sep 5 manuscript snapshot; the deposit explicitly is not publication of the manuscript

下一触发器

public manuscript or preprint; raw/frozen FuXi-S2S and CFSv2 input manifests; independent rerun from source; explicit figure reuse scope

上次判断

retain as high-value S2S verification infrastructure: it separates coarse-grid vortex detection, source-side threshold calibration and downstream probabilistic post-processing across four basins and ten-member diagnostics, showing that better source fidelity does not automatically replace or improve downstream calibration

完整代码与数据状态

MIT GitHub tag and Zenodo archive DOI 10.5281/zenodo.22303529 contain VQVAE, diffusion-OT coupling, training and 45-day sampler code; inference still points to checkpoints/xxx.ckpt, no weights are released, and the older HF data package exposes only one sample per month from 2018-2021 under CC BY-NC 4.0

完整论文与来源状态

existing arXiv 2512.12545 preprint remains without formal venue; v1.0 release and 21.63 MB Zenodo archive were created Sep 4 19:06 +08 in the buffer window

下一触发器

trained VQVAE and diffusion checkpoints; complete test starts and preprocessing; locked environment; independent 51-member reproduction and equal-member ECMWF comparison

上次判断

upgrade reproducibility state only: paper reports a 51-member 45-day ERA5-trained ensemble and up to 12% CRPS improvement versus 11-member ECMWF S2S beyond day 10, but the archived code cannot reproduce reported forecasts without absent checkpoints and complete evaluation inputs

完整代码与数据状态

no abstract, assimilation configuration, cases, forecast leads, metrics, code, data package or reusable figure license was accessible

完整论文与来源状态

Atmospheric Research DOI 10.1016/j.atmosres.2026.109303; Elsevier core metadata says available online Sep 5, Crossref created Sep 5 15:22 UTC; subscription article

下一触发器

accessible abstract/full text; exact WRF or other NWP and DA contract; common-case precipitation scores; code/data and downstream AI relevance

上次判断

retain title-level short-range observation-assimilation metadata only; FY-4A water-vapor assimilation may inform upstream initialization, but there is no evidence here for an AI method or quantifiable forecast improvement

完整代码与数据状态

no public code, manually labelled curve data, split manifest, forecast archive or standard reusable figure license verified

完整论文与来源状态

IEEE RadarConf 2026 DOI 10.1109/RADARCONF2663773.2026.11674988, published May 11; Crossref record was created Sep 4 19:03 UTC, so this is late indexing rather than fresh science

下一触发器

full text and case count; automatic NFGDA integration; lead-wise position metrics and reliability; code, labels and independent events

上次判断

recover as high-relevance historical nowcasting context: fitted and vectorized gust-front curves from polarimetric radar feed point-wise and mean-motion models whose fused position forecast is reported skillful to about 30 minutes, but manual labels and abstract-only evidence prevent operational or calibrated-probability claims

完整代码与数据状态

ray3389/fr-radar-rainfall copies the card and 2020-2024 tar layout of the existing 2025 Météo-France 71.2 GB release; a one-page AQI report lacks source, site, split and code; weather-analyzer commits are cloud-audit timestamps; Santiago updates are routine database backups; fresh Spaces are generic weather or non-weather radar apps

完整论文与来源状态

no new peer-reviewed paper, official dataset version, model checkpoint, scientific result or licensing clarification relevant to the four core lines

下一触发器

reconsider only for original provenance, new years or QC, real forecast evaluation, executable scientific code or a formal publication state

上次判断

exclude the duplicate radar mirror, generic AQI report, schema-only multimodal weather notes, routine database or forecast-feed refreshes and application-only Spaces from scientific updates

完整代码与数据状态

public ERA5 and SRTM sources; 5243-station CMA-S source and derived grid restricted; current code and trained weights unavailable; article CC BY 4.0 and complete Figure 2 mirrored

完整论文与来源状态

Remote Sensing 18:3026, DOI 10.3390/rs18173026, published Sep 4; formal open-access article

下一触发器

cleaned code and weights; CMA-S processing manifest or shareable substitute; true low-latency inputs; multi-member CRPS, reliability and spread-skill; out-of-region tests

上次判断

retain 1-6 h retrospective Zhejiang precipitation sequence study: independent ERA5 and CMA-S models use chronological 2024 tests; strongest baseline-relative CSI gains are about 6.5% and 9.5%, but data are reanalysis or IDW station fields rather than radar and stochastic baselines use one realization, so no operational or calibrated-probability claim

完整代码与数据状态

Pha Din and DWD datasets link to Mendeley and Zenodo; github.com/DucTrung-GiaSon/SR2GAN has four 2025 notebooks only, no README, root license, weights, environment freeze or paper-matched manifest; no new repo commit

完整论文与来源状态

Mobile Networks and Applications DOI 10.1007/s11036-026-02532-6, formal article published Sep 4; subscription full text unavailable

下一触发器

full text and per-lead contract; paper-to-notebook mapping; frozen splits and common baselines; weights, requirements and license; independent extreme-rain verification

上次判断

direct radar sequence result retained at abstract evidence level: ConvLSTM plus refined CBAM, PatchGAN and adversarial-MSE-SSIM loss; authors report MSE reductions of 44.4% and 25% and SSIM gains of 12.8% and 4.5% versus CNN-GRU on Pha Din and DWD, but public notebooks cannot reproduce or audit those claims

完整代码与数据状态

github.com/itcallsitcalls/Radar-Echo-Nowcasting-CNN-LSTM has MIT source, config and course report; HF LilaSin/radar-cnn-lstm-nowcasting has a real 3,261,931,698-byte LFS checkpoint with SHA-256 OID; the 943-sample dataset, CSV and provenance are absent

完整论文与来源状态

no paper or formal release; repository and HF model created Sep 4 in the main window

下一触发器

dataset source, station/domain and redistribution terms; event-separated train/validation/test manifest; dBZ mapping and lead contract; independent reproduction and simpler baselines

上次判断

retain as low-evidence direct artifact: 10 five-minute input frames predict 10 future frames on 256-square grayscale images; all headline metrics come from one 80:20 validation split, the 0.184 threshold was selected from the same validation distribution, no independent test exists, and the report inconsistently mentions 981 versus 943 samples

完整代码与数据状态

commit bb12f8938bb4 adds 11 raw five-minute Maly Javornik scans, generate_samples.py and an updated 2.7 MB evaluation notebook; removes 18 precompiled samples; README now claims MIT code and CC BY-NC-ND data, but the linked root LICENSE is still 404 and baseline notebook primarily plots precomputed result files

完整论文与来源状态

same IEEE Access DOI 10.1109/access.2026.3730270; no new publication status, buffer commit Sep 3 17:56 UTC

下一触发器

add actual root license and provenance for adapted EDM/RainNet/LUPIN components; frozen 18-event manifest and executable metric recomputation; common masks and independent 30-member calibration

上次判断

correct prior tracker state that said last pushed Aug 11: the buffer-window commit materially improves raw-to-sample reproducibility for one event and exposes 18 anchors, but full benchmark inputs, other-model generation and independent metric reproduction remain incomplete

完整代码与数据状态

author-associated GitHub repository created Sep 3 10:22 UTC and commit 8f7822d adds MIT Python package, configs and component tests; no KMA HSP, GK2A or AWS data, preprocessing package, pretrained weights or frozen result files

完整论文与来源状态

existing ICLR 2026 conference paper and arXiv 2603.00418; code release is the new buffer event, not a new paper

下一触发器

dataset access recipe and aligned manifests; pretrained proposal and splatting checkpoints; reproduce cross-domain metrics and test resolution transfer; quantify radar-free inference latency

上次判断

retain as same-time precipitation-field reconstruction infrastructure, not future nowcasting: QCGS fuses satellite brightness temperature and sparse AWS anchors to render a continuous radar-like field; useful for observation recovery and initialization, not evidence of forecast lead skill

完整代码与数据状态

github.com/fabioscarparo/Nembo is a new MIT TypeScript implementation using Italian DPC five-minute radar tiles, consensus motion plus dense optical flow, Lagrangian warping and Open-Meteo 850-500 hPa steering fallback; no frozen verification package

完整论文与来源状态

personal software project created Sep 3 06:39 UTC in the buffer; no paper or formal scientific release

下一触发器

event manifest, observation latency and data-use terms; persistence and pySTEPS comparison; CSI, FSS, displacement and growth-decay diagnostics

上次判断

retain as a transparent 30-minute deterministic baseline and application reference; browser runtime is reported around 120 ms, but it advects existing echoes only, cannot initiate cells, and has no published skill evaluation or AI model

完整代码与数据状态

ph-w2000/HARECast is publicly cloneable again at unchanged commit b50a4052d960; code, SEVIR and MeteoNet loaders, evaluation entry point and Google Drive checkpoint link are visible; no root LICENSE, release or checkpoint hash

完整论文与来源状态

arXiv 2605.13181v2 remains a preprint; README says ACM Multimedia 2026, but no official ACM page or registered Crossref DOI was verified; exact repository restoration time is unknown

下一触发器

official ACM record or DOI; root license, archived checkpoint and hash; end-to-end portable configs and independent reproduction

上次判断

reverse the Aug 8 accessibility downgrade only: artifacts are visible again, but unchanged July code and an unverified venue string are not new scientific results

完整代码与数据状态

commits 286c0c9 through 3296f70 merge PRs 158-159, reuse true scoring weights in M0 linearity diagnostics, preserve radar-prior mean and spatial log-standard-deviation JVP or VJP plus fallback derivatives, and add 16-by-16 synthetic theory and learning approval artifacts; no independent suite was run here

完整论文与来源状态

repository engineering and numerical-review update only; no new paper, real-event score, model weights, dataset, standard license or release

下一触发器

event-isolated multi-radar before-after skill with independent observations; complete CPU, CUDA and full MPS suite; compact scientific interface, release and license audit

上次判断

update infrastructure state only: the spatial cotangent and weighting chain is more internally consistent and the synthetic phase-one demonstration passes author checks, but the repository still marks real radar-dependent prior evaluation and confirmatory publication evidence as external or HOLD and full MPS or operational deployment as NO-GO

完整代码与数据状态

public abstract and supporting-information text describe UFS meteorology, emissions, AirNow, VIIRS AOD and TEMPO NO2 inputs; no public paper-matched code, trained weights or frozen forecast package verified; figure reuse permission not established

完整论文与来源状态

Environmental Science & Technology DOI 10.1021/acs.est.6c04600, accepted Aug 26 and published online Sep 1; Crossref record created Sep 3 09:00 UTC, but scientific publication is outside 48 h

下一触发器

code, weights, training and independent-year manifests; real-cycle latency and forecast meteorology; PM2.5 and O3 threshold reliability and extreme-event errors

上次判断

high-value air-quality context kept window-out: observation fusion updates initial PM2.5 and O3 fields before 72 h DeepAQM forecasts and fine-tuning improves reported independent-period R2, but it is not a fresh Sep 4-5 publication event

完整代码与数据状态

Crossref provides title, authors and Elsevier full-text API links; no accessible experiment details, code, weights or dataset package verified; article license is CC BY-NC-ND 4.0 and no figure mirrored

完整论文与来源状态

Energy and AI 100887, DOI 10.1016/j.egyai.2026.100887; version-of-record license starts Aug 26 and print month is September, while Crossref record was only created Sep 4 23:58 UTC

下一触发器

official abstract or full text; coarse-to-fine NWP variables, temporal resolution, probabilistic calibration, code and checkpoints

上次判断

retain as methodologically adjacent window-out metadata: shape-prior flow matching may inform conditional small-scale recovery, but its target is day-ahead solar irradiance and the indexing event is not a fresh weather-forecast paper release

完整代码与数据状态

publisher page and PDF were blocked during verification; no abstract, model, data version, split, effect size, code, weights or reusable figure permission obtained

完整论文与来源状态

Journal of Remote Sensing DOI 10.34133/remotesensing.1078, Crossref created Sep 4 10:33 UTC and reports Sep 4 publication

下一触发器

official abstract or full text; GPM DPR product/version and label definition; held-out geography/time; code and data; retrieval-to-forecast value

上次判断

title supports satellite precipitation-radar classification only; keep upstream metadata separate from ground-radar nowcasting and do not infer forecast skill

完整代码与数据状态

no official abstract, full text, dataset, lead-time contract, code, weights or license verified

完整论文与来源状态

IET Conference Proceedings DOI 10.1049/icp.2026.2709; Crossref record created Sep 4 09:11 UTC but publication date is Sep 1

下一触发器

official abstract or full text; meteorological task and input source; actual forecast horizon, split, metrics and artifacts

上次判断

generic title-level record only; cannot classify as radar nowcasting or report scientific performance

完整代码与数据状态

Hugging Face repository created Sep 4 with about 12.2 GB of LFS NPZ files, but no README, license, provenance, executable code or checksums beyond LFS OIDs; filenames identify WeatherNext 2 and WeatherNext Cyclones, not WeatherNext 3

完整论文与来源状态

no paper-linked release identity or author provenance verified

下一触发器

an official or author-verifiable model card, license, code and exact linkage to a published system

上次判断

exclude from WeatherNext 3 state changes: large files alone do not establish authentic WN3 weights or a usable research release

完整代码与数据状态

experimental forecasts available from official WeatherNext page; no WN3 code, weights, frozen evaluation bundle or full intermediate data verified; PDF says Google all rights reserved, no figure mirrored

完整论文与来源状态

arXiv 2609.03582v1 submitted Sep 3 09:30:21 UTC; preprint only, no journal information

下一触发器

full code/weights and frozen manifests; full-season/year quasi-realtime evaluation; common AIFS/ENS truth/interpolation; extreme precipitation calibration

上次判断

retain 15-day 64-member system with low-latency GEO conditioning and hourly initialization; author reports early precipitation CRPS gains varying from about 60% on IMERG to 10% on gauges and roughly 10% upper-air CRPS versus AIFS ENS in a six-week test; truth sensitivity, training overlap, artifacts and underdispersion constrain claims

完整代码与数据状态

identifiable CERRA/NWCSAF/MEPS sources; github.com/fmidev/cloudcast still exposes old v1 U-Net only, no v2/CFM code, weights, processed NWCSAF data or root license verified; no figure mirrored

完整论文与来源状态

arXiv 2609.03763v1 submitted Sep 3 12:30:10 UTC; preprint only, no journal information

下一触发器

v2 code/weights/data manifest; equal-member probabilistic verification; displacement and latency; 1-3 h advection baseline

上次判断

retain 5-km Scandinavian 1-12 h observation-adapted cloud forecast; 2025 test has 2855 starts; CFM improves class FSS but degrades MAE/ACC/MSESS versus no-CFM, and no calibration/ensemble test supports probabilistic claim

完整代码与数据状态

official abstract accessible; no matched code, weights, complete data description or reusable figure permission verified

完整论文与来源状态

AIES DOI 10.1175/AIES-D-26-0044.1 created Sep 3 16:47:57 UTC; formal journal record, publisher full text not obtained

下一触发器

full text and experiment contract; code/weights; conditional/noise attribution tied to calibration and extremes

上次判断

retain 16x precipitation downscaling diagnostic: mean mainly low-resolution-conditioned, maxima more stochastic, only cWGAN meaningfully uses Gaussian noise with differences at bottleneck/early decoder; method insight, not forecast skill

完整代码与数据状态

public ARPA/SIAS/OSM sources identifiable; derived data and Python code only on request; article CC BY 4.0 and complete Figure 9 mirrored

完整论文与来源状态

Atmosphere 17(9):865 DOI 10.3390/atmos17090865 published Sep 3, Crossref created Sep 3 17:13:25 UTC

下一触发器

runnable code/data; multiple independent stations or mobile sensors; frozen forecast-meteorology provenance; emissions uncertainty

上次判断

retain 1-12 h XGBoost/RF corridor forecast, 2020-25 hourly data and chronological test; XGB R2 .7906 to .6692, but only one independent NO2 station and SIRANE-generated spatial targets mean 11-m maps are not independent spatial validation

完整代码与数据状态

MERRA-2 HAQAST CNN national daily data stated but not packaged; code unavailable and data by request; article CC BY-SA 4.0, no figure mirrored

完整论文与来源状态

DOI 10.57041/6x6rrz53 created Sep 4 02:23:55 UTC; Crossref title says hourly while PDF title/task say daily

下一触发器

corrected metadata; station/spatial and forecast-meteorology inputs; unclipped extreme-event tests; multiseed public artifacts

上次判断

retain as low-evidence one-day baseline only: 2193 daily averages, 7-day window, no weather predictors; linear regression RMSE 3.866/R2 .905 slightly beats complex ensembles; train-only IQR capping may suppress smog extremes

完整代码与数据状态

Anemoi and WeatherBenchX ecosystem identified; no paper-matched Laxmi code, weights, preprocessing manifest or 94-start bundle verified; no explicit reusable figure license

完整论文与来源状态

arXiv 2609.03210v1 submitted Sep 2 22:50:10 UTC, buffer recovery; preprint only, author note submitted to Science is not acceptance

下一触发器

code/checkpoint and start manifest; independent radar/gauge and land-ocean validation; same-cycle IFS; >100/300 mm/day calibration

上次判断

retain 1-15 day global precipitation ensemble trained to IMERG; day-1 CRPS gains 19.3% vs AIFS-CRPS and 20.1% vs IFS with reported 50/100 mm/day capture gains, but >300 mm/day misses, retrieval-target bias and post-day6 IFS cycle mismatch limit claims

完整代码与数据状态

MIT WindFlow pretrained code/weights at github.com/tjvandal/windflow and study stereo/fine-tune/distillation code at github.com/myzeus/stereo-winds; public GOES, IGRA, ERA5 and EarthCARE sources

完整论文与来源状态

arXiv 2609.03100v1 submitted Sep 2 19:22 UTC, buffer recovery; preprint only, no journal information

下一触发器

frozen release; cross-satellite geography; multiple active-sensor cases; data-assimilation/downstream precipitation impact

上次判断

retain upstream wind retrieval: fine-tuning cuts aggregate RMSVD 18%; triple collocation favors stereo on three water-vapor bands but operational AMV on 11.2 um; single-satellite student adds 1.34 m/s RMSVD and global transfer remains untested

完整代码与数据状态

github.com/Hecthorio/AgsAir-PM25-beta has Streamlit source, real 3.12 MB wind and 1.13 MB emission H5 models and scaler; no LICENSE, training manifest or validation package

完整论文与来源状态

GitHub repository created Sep 3 00:27:02 UTC in buffer; no paper or formal release

下一触发器

license and model cards; real observed initialization; RUOA comparison; frozen events and skill; refactored solver

上次判断

retain executable 24-h Aguascalientes advection-diffusion prototype only; manually uniform C0 and substituted Open-Meteo drivers, no independent skill, so maps are not validated air-quality forecasts

完整代码与数据状态

anonymous-code statement only; exact repository, weights and evaluated artifacts not verified

完整论文与来源状态

OpenReview yTeXmoeZ79 created May 25 and modified Sep 3 11:16:56 UTC; TMLR decision pending; exact diff inaccessible

下一触发器

visible revision diff, decision or identifiable code/results

上次判断

modified-record signal only; abstract describes FNO plus Transformer-parameterized Neural ODE on WeatherBench2 1.5 degrees, but timestamp is not a new scientific result

完整代码与数据状态

no newly verified code, checkpoint or evaluation bundle from arXiv mirror

完整论文与来源状态

arXiv 2609.03364v1 submitted Sep 3 04:47 UTC, but Atmospheric Research formal DOI was created Apr 6 and article published Aug 2026

下一触发器

substantive new version/data/code or corrected evaluation

上次判断

old formal study newly mirrored on arXiv; do not count as a fresh S2S result

完整代码与数据状态

no paper-matched code, weights, California test manifest or boundary-condition package verified

完整论文与来源状态

OpenReview NMtCIMSi8i created Jul 5, modified Sep 2 05:06 UTC in buffer; GAIA 2026 extended abstract

下一触发器

substantive revision diff, full paper, artifacts and common-sample HRRR evaluation

上次判断

older abstract metadata change only; 36-h California multi-reanalysis pretraining and one-month few-shot claims are not new-window science

完整代码与数据状态

no matched public code, checkpoint or processed dataset verified; official deposited abstract read, publisher full text inaccessible

完整论文与来源状态

AIES DOI 10.1175/aies-d-25-0117.1 registered Sep 2 15:54 UTC; journal date Sep 2; same-title EGU26-2056 abstract predates this event

下一触发器

full text; Kyushu years/events/grid; observation arrival times; matched JMA C4 evaluation and code

上次判断

journal/abstract-level state: radar-gauge, station and satellite attention plus focal/ordinal loss; authors report nonzero CSI for >=50 mm/h to 6 h and 4-6 h JMA comparison; no exact metrics or robust operational-skill claim

完整代码与数据状态

no dedicated public coupled-code/checkpoint package verified; UM institutional license; article CC BY-NC-SA 4.0, no figure mirrored

完整论文与来源状态

arXiv 2609.02566v1 submitted Sep 2 13:17:10 UTC; preprint only, no journal information

下一触发器

unseen initializations/seasons; multivariable errors and long integration; code/checkpoints and random-seed replication

上次判断

UM N320 ~40 km/70 levels/720 s with rank-local DDPG; ten nudged training episodes and frozen non-nudged inference share Dec 12 2021 00 UTC initialization; Z500 tropics improves but T1.5m/T500 worsen in all six bands; feasibility, not generalization

完整代码与数据状态

NASA TROPICS_CNN_2026 SOFT and L2 directories accessible; scripts dated July, no new release today; checkpoint and code/product license not verified; article CC BY 4.0

完整论文与来源状态

EGUsphere DOI 10.5194/egusphere-2026-4156 discussion Sep 2 date-only; AMT under review, not accepted

下一触发器

frozen years/events; independent ground validation; weights/product licenses; upstream-error propagation

上次判断

satellite rainfall/LH retrieval: 14 features, 7x7 patches, GPM 5-to-30 km aggregation and 15/30 min matching; random 80/20 split and hyperparameter ensemble selection do not establish heldout-weather probability skill

完整代码与数据状态

no dedicated code/checkpoint/frozen-match bundle verified; identifiable ERA5/GridSat/IBTrACS and weather-forecast sources; article CC BY 4.0

完整论文与来源状态

arXiv 2609.02085v1 submitted Sep 2 04:17:22 UTC, buffer recovery; preprint only, no journal information

下一触发器

code/weights/matched manifests; strict unseen-backbone selection; consistent wind definition and probabilistic ensemble evaluation

上次判断

3.6M model, GraphCast training with 24 h history and real future fields; 1990-2017 train, 2018 val, 2019-20 test, WN-C 2025 transfer; +24h GraphCast track 48.68 vs 57.03 km, intensity definition and changing TE coverage constrain comparison; WN-C members individually scored, not probability verification

完整代码与数据状态

github.com/yingkaisha/RAL-GWC-TC Apache-2.0 code and HF yingkaisha/CONUS404-AI-TC Apache-2.0 card with real 4.94 GB LAM/674 MB diffusion weights; releases predate window; 2023-24 CONUS404 extension by request; article CC BY 4.0 Figure 1 mirrored

完整论文与来源状态

arXiv 2609.02034v1 submitted Sep 2 03:06:19 UTC, buffer recovery; preprint only, no journal information

下一触发器

true forecast-initialized boundaries; full reference availability; forcing/initial/generative uncertainty and extreme calibration

上次判断

8 km LAM plus residual EDM, 1980-2018 train/2019 val/2020-24 twenty-TC test, 20 members; continuous ERA5 or GDAS analysis boundaries establish downscaling rather than forecast lead skill; FSS 9x9 ~72 km, reliability underforecasting remains

完整代码与数据状态

ERA5 source accessible; study code promised on Zenodo upon publication, not yet verified; article CC BY 4.0

完整论文与来源状态

Research Square DOI 10.21203/rs.3.rs-10739679/v1 created Sep 2 01:37 UTC, buffer; under-review preprint, no verified journal acceptance

下一触发器

fix Table 1 NE Africa Apr 30 initialization versus Jun 6-7 peak and Table 3 ten-day integration; code; physical perturbation feasibility/probability

上次判断

2.8-degree deterministic NeuralGCM/37 levels/10 days, four selected 2024 heat events; optimize initial state with variable-wise regularization, reported HI +1.2-3.0 C; process stress test at 1000 hPa, not probabilistic or general S2S skill

完整代码与数据状态

no abstract/full text, matched code, weights or dataset verified; specific figure permission unknown

完整论文与来源状态

IEEE JSTARS DOI 10.1109/jstars.2026.3729834, document 11676193, Crossref created Sep 2 19:10 UTC

下一触发器

official abstract/full text; actual lead/grid/data/split; latency and reproducible artifacts

上次判断

title-level multimodal nowcasting candidate; do not infer architecture performance, extreme skill or efficiency from title

完整代码与数据状态

publisher inaccessible; no official abstract, dedicated code, weights or data verified

完整论文与来源状态

Resources Conservation and Recycling Advances DOI 10.1016/j.rcradv.2026.200381 created Sep 1 15:40 UTC, buffer; issue September, exact first publication unverified

下一触发器

official full text and actual task/lead; dataset, splits, metrics, venue context and artifacts

上次判断

spatial-frequency nonlocal architecture is title-only evidence; no direct forecast-skill claim

完整代码与数据状态

no official abstract/full text, matched public code, weights or processed data verified

完整论文与来源状态

Atmospheric Pollution Research DOI 10.1016/j.apr.2026.103191 created Sep 1 06:45 UTC, buffer; first-online date unverified

下一触发器

full text; retrospective versus future forecast contract, pollutants/years/stations/drivers and heldout validation

上次判断

CMAQ retrospective CONUS correction from title; do not promote to core future air-quality forecasting without timing evidence

完整代码与数据状态

official full text inaccessible; code/data/checkpoint/license and calibrated probabilities unverified

完整论文与来源状态

International Journal of Environmental Studies DOI 10.1080/00207233.2026.2723829 published Sep 2, created Sep 2 17:45 UTC

下一触发器

official abstract/full text; pollutants, geography, meteorological driver, lead, split and calibration

上次判断

probabilistic architecture title does not establish forecast probability skill; retain candidate only

完整代码与数据状态

primary full text unavailable; no code, model, initial conditions, reference or evaluation package verified

完整论文与来源状态

Weather DOI 10.1002/wea.70123 created Sep 2 08:23 UTC

下一触发器

official full text and model/ensemble/init/reference definitions; historical hindcast data

上次判断

historical forecast reconstruction by title; do not import secondary-page AIFS claims as primary scientific evidence

完整代码与数据状态

restricted ground-lightning/HRES data; no dedicated code/weights verified; article CC BY 4.0, no figure selected

完整论文与来源状态

Atmosphere 17:856 DOI 10.3390/atmos17090856 published Aug 31 outside 48 h; DOI index Sep 1 10:32 UTC in buffer

下一触发器

public artifacts; latest-cycle and observation-delay contract; equal-budget alert thresholds; more independent HRES events

上次判断

adjacent lightning method recovered by delayed index, not new-window science: 2014-18 train/2019 val/2020 test; July 2025 frozen-model ERA5-to-HRES CSI .60 to .57, POD .85 to .80; retains historical lightning and direct litota1 nonzero baseline is not equal-tuned

完整代码与数据状态

HRES restricted and processed data by request; no dedicated code/checkpoint verified; article CC BY 4.0

完整论文与来源状态

Atmosphere 17:845 DOI 10.3390/atmos17090845 published Aug 28 outside 48 h; DOI index Sep 1 10:32 UTC in buffer

下一触发器

per-lead and city/episode holdouts; real meteorological-error propagation; threshold probability and public artifacts

上次判断

BTH 157 stations/13 cities, 2019-22 train/2023 test, genuine HRES 12 UTC 0-240 h forecasts; pooled 0-24h RMSE17.03 and pooled0-240h24.57 ug/m3 are not day10 point skill; retain old-paper design reference

完整代码与数据状态

no open code/checkpoint/evaluation bundle verified; primary Springer abstract available

完整论文与来源状态

Springer DOI 10.3103/S1068373926130029 published Sep 2; original 10.37162/2618-9631-2025-2-28-43 is 2025

下一触发器

new experiments/data; common-member budget, years and hindcast reference from full text

上次判断

translation state only: SLAV072L96 S1 breeding61 versus S2 LEnKF41 to46 days, both underdispersive; useful dynamic S2S baseline, not a new AI result

完整代码与数据状态

no dedicated code, weights or test dataset verified; official Springer abstracts available

完整论文与来源状态

DOIs 10.3103/S1068373926130030 and 10.3103/S1068373926130042 published Sep 2; originals 10.37162/2618-9631-2025-3-8-31 and 10.37162/2618-9631-2025-4-7-32

下一触发器

substantive new model or event data, full verification contract, reproducible artifacts

上次判断

RF/pySTEPS/COSMO-Ru2.2, central Russia AWS May-Sep2024, >20 m/s useful FSS to30min per abstract; old adjacent wind-hazard studies, not new radar precipitation skill

完整代码与数据状态

KNMI/ERA5 sources identified; no paper-matched public code, weights or preprocessing manifest verified; arXiv non-exclusive license, no mirrored figure

完整论文与来源状态

arXiv 2609.00544v1, submitted 2026-09-01 01:30:33 UTC; buffer item recovered from Sep 2 listing; preprint only with no journal information

下一触发器

code/weights and unfiltered sample manifest; separate 5 min and 30 min physics ablations; stronger-rain calibration, same-case baselines and multi-network reproduction

上次判断

retain direct 0-180 min radar nowcasting: operator/adversarial architecture and 10-member MC dropout are relevant, but primary physics gains concern the secondary 30 min setup; coarse ERA5 interpolation and CRPS alone do not establish resolved physics or reliable extreme forecasts

完整代码与数据状态

MIT github.com/G-Jha-G/generative-precipitation-downscaling, pushed Aug 10; CC BY 4.0 Zenodo 21903246 published Aug 12 contains 2.04 GB weights and 2.63 GB outputs, existence verified but not executed

完整论文与来源状态

Research Square v1 DOI 10.21203/rs.3.rs-10772032/v1 posted Aug 31, DOI created 11:47 UTC in buffer; npj Climate and Atmospheric Science in-review page, not accepted; CC BY 4.0 Figure 1 mirrored

下一触发器

frozen release; real NWP/ensemble drivers; speed-extreme-calibration scores for one identical sampler configuration; longer independent extreme validation

上次判断

retain common-U-Net generative downscaling benchmark: MSWX 0.1-degree targets are average-pooled to 1 degree, not actual forecasts; 100-NFE Heun extreme evaluation differs from 16-NFE Euler ensembles; both DDPM and FM remain underdispersive

完整代码与数据状态

code on reasonable request, Weatherbit subscription data, no public weights; CC BY 4.0 VOR Figure 1 mirrored with full caption/attribution

完整论文与来源状态

formal Atmosphere 17(9), 858, DOI 10.3390/atmos17090858 published Aug 31; Crossref-created Sep 1 10:32 UTC is the main-window event, publication hour unknown

下一触发器

reconstructable code/data; independent station targets; archived forecast meteorology; cross-city/year and pollution/dust FAR plus probabilistic calibration

上次判断

retain direct 1 h AQI evidence: eight cities, 24 h input, 2022-23 training and full-2024 holdout, six seeds; MAE improves persistence but R2/RMSE are close; Weatherbit CAMS/SILAM-observation fusion is not independent station truth, and single-seed threshold alarms are not calibrated probabilities

完整代码与数据状态

MIT github.com/afmonsalves/SAPAD; six actual checkpoint assets in v1.0.0-models and CC BY 4.0 Zenodo 21793955 labels, all released Aug 6, not new today

完整论文与来源状态

arXiv 2609.01514v1 submitted Sep 1 16:43:27 UTC; preprint only with no journal information; arXiv non-exclusive license

下一触发器

more held-out abnormal orbits/sensors; false removal of true extremes; QC-to-QPE/forecast impact; turnkey inference/data manifest

上次判断

retain upstream GPROF satellite precipitation artifact QC: ImageNet MobileNetV3-Large and incremental scarce-label sensor adaptation help early PR-AUC; later joint training is not uniformly better and few abnormal holdout orbits limit claims

完整代码与数据状态

radar/gauge data restricted and require provider approval; no matched public code or weights verified; article CC BY 4.0 does not license raw data

完整论文与来源状态

formal Remote Sensing 18, 2937, DOI 10.3390/rs18172937 published Sep 1; Crossref-created 11:33 UTC

下一触发器

code/data and more independent years/radars/events; reconcile abstract versus holdout metrics; heavy-rain bias and downstream forecast propagation

上次判断

retain 13-radar/5489-gauge hourly QPE: 2021-22 summer training uses 70/30 stratified split plus independent June 2024 validation, with RMSE 3.62-3.66 versus traditional 4.51-4.63 mm/h; ten scans ending at t estimate the preceding accumulation, not future precipitation

完整代码与数据状态

data on request with Sun Yat-sen University/corresponding-author approval; no public code or weights verified; article CC BY 4.0

完整论文与来源状态

formal Remote Sensing 18, 2941, DOI 10.3390/rs18172941 published Sep 1; Crossref-created 12:41 UTC

下一触发器

event-wise holdout and realtime-available extrema; explicit initialization/lead time and FAR; independent hydrometeor observations; convection/multisite tests

上次判断

retain Zhuhai single-site microwave/MRR retrieval and retrospective precipitation indication: 17280 pairs, Mar-Apr 2022 training/testing versus Dec-Feb validation, model-derived hydrometeor references; 97% presence accuracy and 60 min onset-time MAE do not establish a 60 min operational lead

完整代码与数据状态

CC BY 4.0 checklist and literature-based metrics; no new forecast model, training data or checkpoint release

完整论文与来源状态

arXiv 2609.00847v1 submitted Sep 1 07:44:38 UTC; preprint only with no journal information; conference-related author note is not acceptance evidence

下一触发器

measured lifecycle benchmarks at fixed forecast quality, solver NFE, ensemble size, hardware and output resolution

上次判断

retain cross-line evaluation guidance only: account for data processing, experiments, training and inference; quoted literature energy estimates are not a new measurement or weather-skill result

完整代码与数据状态

WMO page now exposes detailed team abstracts but no fifth-webinar PDF, recording, code, weights or frozen scored sample archive; exact page-update timestamp is not shown

完整论文与来源状态

official WMO page remains dated Jun 22; webinar is Sep 24 13:00-14:00 UTC; retrieved Sep 6 with Fengshun, MicroEnsemble PBC and CLINT method summaries

下一触发器

Sep14 JJA2026 actual file release per official ECMWF timetable; Sep 24 presentations, recording and overall rankings; exact common samples and RPSS; Fengshun, PBC and CLINT artifacts

上次判断

update evidence without assigning it to the 24-hour window: WMO says FengshunHybrid won season three and FengshunAdjust leads season four, while team abstracts emphasize probabilistic bias correction, lagged ensembles and hybrid calibration; these are organizer-hosted claims pending full materials

完整代码与数据状态

no full text, code, data, weights or reusable figure verified

完整论文与来源状态

formal Journal of Hydrology: Regional Studies metadata, DOI 10.1016/j.ejrh.2026.103914, Crossref-created Sep 1 06:52:59 UTC; issue dated Oct 2026; first-online hour unverified

下一触发器

official abstract/full text, actual network type, lead/input/target contract, data splits, results and artifacts

上次判断

retain new index only; network in the title is insufficient to infer deep learning or a verified nowcasting gain

完整代码与数据状态

full experimental content, code/data/weights and license not verified

完整论文与来源状态

IEEE ISCID 2025 proceedings DOI 10.1109/iscid68789.2025.00028; Crossref-created Aug 31 19:12:14 UTC in buffer

下一触发器

accessible official text and forecast/data/split/baseline contract with reproducible artifacts

上次判断

retain delayed index of historical conference item, not new scientific publication in the current window

完整代码与数据状态

ACM full text inaccessible; no sufficient abstract, experiments, code/data/weights or figure license verified

完整论文与来源状态

DOI 10.1145/3817124.3817354, conference dated Mar 2026; Crossref-created Aug 31 05:20:33 UTC in buffer

下一触发器

full text, dual-polarization inputs, lead times, event split, heavy-rain scores and code/weights

上次判断

retain metadata-only late indexing; the relevant title alone does not justify promotion to direct core-nowcasting evidence

完整代码与数据状态

official IAHR abstract available; no model/data split, numerical results, code, weights or reuse license verified

完整论文与来源状态

existing HIC 2026 proceedings DOI 10.64697/iahr.proc.hic2026.258; new DOI registration Sep 1 18:53 UTC is index delay

下一触发器

complete experiment/results and actual archived QPF drivers rather than perfect future rainfall

上次判断

retain historical adjacent upper-bound design for 344 km2 Geul: observed rainfall versus perfect future rainfall does not demonstrate actual radar-QPF-to-flood skill

完整代码与数据状态

Nj-1111/weathergpt-m2-bias-correction contains only .gitattributes and a short README, no weights; unchanged Fed-Cast and routine UI/card refreshes not promoted

完整论文与来源状态

main/buffer discovery checks; no new retained scientific model/artifact result in these candidates

下一触发器

actual weights/code plus forecast contract and validation; otherwise no repeated tracker refresh

上次判断

exclude empty model branding, wind-resource siting, same-time ozone mapping and material/financial keyword collisions from direct atmospheric forecast counts

完整代码与数据状态

exPreCast backbone code/weights are public, but paper-specific residual-diffusion code and weights are only available on request/promised after publication; KMA release remains subject to KMA terms and MeteoNet is public

完整论文与来源状态

arXiv 2608.30205v1, submitted 2026-08-31 03:35:35 UTC in the buffer/previous-cutoff overlap and recovered after list delay; preprint only, submitted-to-npj is an author note; CC BY 4.0; Figure 1 mirrored with complete metadata

下一触发器

paper-specific code/weights and formal status; more radar networks/backbones/events; threshold-independent tail calibration; equal-resolution 1 km baseline; independent reproduction

上次判断

retain as high-value direct radar probabilistic nowcasting: 30-member residual diffusion improves full-2023 KMA deterministic/probabilistic scores and recovers missed heavy rain, but +60 min heavy-rain rank histograms remain strongly underdispersive, the P>=0.3 alarm was validation-selected and 1 km native comparison uses an interpolated backbone

完整代码与数据状态

public MIT GitLab code and MIT HF dataset bundle at SHA 36ba69cbedcd845e2c0923b88af9cec01981e9f5; HF API shows 359 normalization/checkpoint/result files; no frozen release/tag verified

完整论文与来源状态

arXiv 2608.30795v1, submitted 2026-08-31 13:49 UTC; preprint only under default arXiv non-exclusive license; 2007-2017 train, 2019 validation, 2018/337 starts test, 1-10 d, 1.5-degree ERA5 plus HadISD station verification

下一触发器

release/DOI; more years/resolution/observations; decoder uncertainty; IFS ENS common contract; station calibration; sensitivity to decomposition order and dropout rate

上次判断

retain as direct end-to-end probabilistic weather evidence across 1-10 d: 7x7 nested encoder/dropout ensemble improves deterministic Aardvark and fair CRPS, but trails IFS ENS, station output is underdispersive and component-aligned variance attribution is order-dependent rather than unique causality

完整代码与数据状态

MIT Zenodo 19564334 published 2026-04-23 contains model.pth and training/inference scripts; RIKEN runs a research display currently showing up to 8 h; no new weight event in this window

完整论文与来源状态

ESSOAr v1 DOI 10.22541/essoar.15008115/v1, published-date 2026-08-31 and Crossref-created 2026-09-01 02:50 UTC; preprint only with no journal acceptance/license verified; abstract reports 2024 full-year 0-12 h evaluation

下一触发器

accessible full text/formal venue and version license; exact 2024 tables; 12 h live-product parity; event/regime reliability, extremes, latency and independent reproduction

上次判断

retain as direct global satellite precipitation short-range evidence, not radar nowcasting: stabilized adversarial and progressive multi-resolution ConvLSTM targets GSMaP discontinuity/blur and author reports 12 h gains over tracking/persistence, while evidence is not yet sufficient for calibrated extreme or operational claims

完整代码与数据状态

no paper-matched code, observation-operator weights, frozen FY-3G/PMR cases or cycling package verified

完整论文与来源状态

formal Remote Sensing article DOI 10.3390/rs18172912, online and Crossref-created 2026-08-31 09:21 UTC, CC BY 4.0; abstract reports 1D-Var plus 3D-Var and up to 19.9% TS gain at severe-torrential threshold

下一触发器

public implementation/data and supplement; traditional-operator/no-assimilation common cases; increments and balance; threshold sample counts; lead-wise multi-event/season verification

上次判断

retain as satellite-to-regional-model assimilation infrastructure, not a standalone ML forecast: the observation operator and humidity retrieval directly affect short-term precipitation, but one case plus one-month cycling and an abstract maximum do not establish general skill

完整代码与数据状态

MeteoNet NW-2016 radar/gauge archives are public under Etalab Open Licence 2.0; manual annotations, analysis outputs and code are only available on reasonable request

完整论文与来源状态

Preprints.org v1 DOI 10.20944/preprints202608.2170.v1, submitted 2026-08-28 and posted 2026-08-31; explicitly not peer reviewed, CC BY 4.0

下一触发器

public labels/code and formal review; multi-network/event tests; weak-rain false-removal; full-field QPE; QC-to-assimilation/nowcast propagation

上次判断

retain as upstream reflectivity-only radar QC: complete-event RF F1 0.833 versus window-level 0.975 exposes leakage, and targeted station FAR/CSI improve while RMSE is unchanged, so do not convert clutter suppression into a forecast-skill claim

完整代码与数据状态

CC BY 4.0 Zenodo 15574458 exposes four Python model files and Reims input/output NetCDF samples but no frozen training weights; operational Atmo Grand Est deployment is author-reported

完整论文与来源状态

formal JGR: Machine Learning and Computation article DOI 10.1029/2025JH000915, online 2026-08-31; formal upgrade from 2025 ESSOAr preprint; 25 m U-Net emulator for PM10/PM2.5/NO2/O3 over Grand Est

下一触发器

weights/training manifest; more dates/cities/stations; archived forecast meteorology/emissions; hotspot extremes, probabilistic calibration and independent operational audit

上次判断

retain as core AQ emulator status change with a hard boundary: near-real-time 25 m fields and strong SIRANE agreement support scalability, but seven 2021 test days/three stations are narrow, traffic hotspots are smoothed and the neural emulator inherits SIRANE and driver biases

完整代码与数据状态

BSD-3-Clause Zenodo 22201757 created 2026-08-31 17:01 +08 contains 73.8 MB uode.zip; GitLab project is linked; no LES/NWP integration package verified

完整论文与来源状态

arXiv 2608.30610v2, replacement submitted 2026-09-01 15:58:39 UTC; v1 submitted Aug 31 11:23 UTC, preprint only, CC BY 4.0; 24 KCE/superdroplet train trajectories and five held-out tests in an idealized box

下一触发器

LES/NWP/climate integration; more kernels/spectral shapes/out-of-domain states; conservation and long-run stability; independent reproduction

上次判断

Sep 2 metadata-only version check: normalized HTML changes concern four citation presentation/punctuation spans, no new scientific content found; retain as open scientific-ML microphysics method, not forecast skill: UODE recovers in-domain process rates without rate targets and often improves analytic closure loss, but extrapolation weakens and idealized warm-rain boxes do not establish precipitation forecasts

完整代码与数据状态

no paper-matched public model, weights, training manifest or frozen MPAS-A/WINTRE-MIX package verified from the accessible record

完整论文与来源状态

EGUsphere preprint DOI 10.5194/egusphere-2026-4026, discussion started 2026-08-31 and under review for GMD; not accepted; CC BY 4.0

下一触发器

code/data/model; more IOPs/seasons; spin-up, phase, water conservation and climate drift; independent regional forecast evaluation

上次判断

retain as NWP-embedded ML microphysics evidence: two WINTRE-MIX IOPs preserve 24 h total precipitation and help early spin-up, but ML bias reduces drizzle and slightly worsens phase, so few-case component tests cannot become a general short-range skill claim

完整代码与数据状态

commit201020545 removes duplicate parent history stage-out while preserving best-model dependency; author FABRIC pilot68jobs3DAGs success and Tier2R1/L1/R4 pass

完整论文与来源状态

third-party reproduction execution update Sep9 15:14UTC; no formal paper change

下一触发器

complete E1 metrics/artifacts; license and frozen data; independent repeat

上次判断

execution validation improved; full scientific reproduction and weights still unverified

完整代码与数据状态

paper/supplement are public; no paper-matched code, frozen 36-city station manifest or weather-normalization package verified

完整论文与来源状态

EGUsphere preprint DOI 10.5194/egusphere-2026-4127, ACP discussion started 2026-08-31; not peer-reviewed final; CC BY 4.0; 2015-2025 observations plus ERA5/LightGBM/SHAP

下一触发器

code/data mapping and formal review; normalization sensitivity; emission proxies and causal limits; future forecast meteorology/event calibration if extended to prediction

上次判断

retain as AQ meteorological-driver attribution, not forecasting: weather-normalized baseline and positive meteorological modulation on all 229 exceedance days motivate error decomposition, while historical SHAP/normalization has no initialization-valid-time forecast contract

完整代码与数据状态

no frozen public implementation/release or independent end-to-end result package verified from the arXiv record; paper describes F2PY-based Python-main-controller integration of Fortran models and Python AI/DA

完整论文与来源状态

arXiv 2608.29532v1, submitted 2026-08-30 03:33 UTC, about 90 minutes before the 48 h buffer; revised version of an EGUsphere discussion paper; CC BY 4.0

下一触发器

public code/config; exact revision diff; performance/robustness; real 1 km DA and forecast impact across events

上次判断

retain as window-out AI-physical coupling infrastructure, not a current-window forecast result: climate SCDA and nested 1 km DA examples show interface potential, but accessible evidence does not establish reusable software or forecast skill

完整代码与数据状态

no paper-matched code, weights, OEM label manifest or train/test split verified; Crossref abstract is accessible; CC BY-NC-ND 4.0 forbids adapted figure mirroring under the project policy

完整论文与来源状态

formal JGR: Machine Learning and Computation article DOI 10.1029/2026JH001562, online 2026-08-30 and Crossref-created 2026-08-31 02:41:45 UTC; ML predicts FY-4B/GIIRS NH3/HCOOH columns, CO profiles and averaging kernels over East Asia, with abstract-reported R2 >0.9 for CO, >0.8 for NH3/HCOOH and ~1000x speedup versus OEM

下一触发器

code/data/splits; OEM-label volume; cloud/night stratification; independent ground/aircraft validation; AK uncertainty; assimilation or operational composition-monitoring use

上次判断

retain as main-window atmospheric-composition observation infrastructure, not core AQ forecast evidence: joint AK prediction improves physical comparability, but the task is retrieval/monitoring and provides no future pollutant, meteorological-error or calibrated-forecast result

完整代码与数据状态

no accessible abstract, full text, author manuscript, code, radar configuration, event list or quantitative effect verified

完整论文与来源状态

formal Remote Sensing Letters article DOI 10.1080/2150704X.2026.2720795, online 2026-08-29 and Crossref-created 2026-08-29 19:06:55 UTC; buffer-window registration; title/author metadata only

下一触发器

official abstract/full text; interference taxonomy and algorithm; precipitation-retention false-removal tests; multi-site/event coverage; real-time latency; downstream QPE/assimilation/nowcast impacts

上次判断

retain metadata only as upstream radar QC; do not infer method, efficacy or forecast benefit from the title

完整代码与数据状态

public GitHub has five 2026-08-23 commits and executable GenCast distillation workflows but no repository license; HF yyiming3/gencast_distillation provides distilled_model.npz under CC BY-NC-SA 4.0; code and weight rights differ

完整论文与来源状态

arXiv v1 submitted 2026-08-27 21:39:50 UTC, surfaced on the 2026-08-31 new list and outside 48 h; CC BY 4.0; 1-degree ERA5, 1979-2018 train, only 12 Jul-Dec 2022 starts and one Nanmadol case; supervised energy-distance student uses 1 versus teacher 39 NFEs per 12 h step

下一触发器

broader seasons/basins/starts; 50-1000-member calibration and tail tests; exact runtime/memory; code license/tag/release; independent reproduction; higher-resolution detail diagnostics

上次判断

retain as high-value indexing-delay method related to core-short-range, not a current-window event: distribution-aware distillation is promising, but 12 starts, one cyclone and 1-degree resolution do not establish robust extreme or local-detail skill

完整代码与数据状态

no public code, weights, frozen configuration or result package verified; default arXiv non-exclusive distribution license, so no figure mirroring

完整论文与来源状态

arXiv v1 submitted 2026-08-27 10:07:40 UTC and surfaced later; ERA5 5.625-degree, 2006-2015 train, 2016 validation, 2017-2018 test; advection prior plus learned velocity/source and variational simulation-free dynamics on 42 h and 5-month tasks

下一触发器

public code/config; higher-resolution common-sample baselines; exact lead/metric tables; ablations of physics prior; probabilistic/extreme evidence; 1-65 d evaluation

上次判断

retain as indexing-delay adjacent method, not core skill: physics anchoring addresses error accumulation, but the grid is coarse, the monthly task exceeds scope and open evidence is incomplete

完整代码与数据状态

HF recently modified results are generic national forecast tables, Sri Lanka documents, space-weather/energy data and a Taiwan AQ compilation; GitHub hits are weather MCP/dashboard or old project UI updates; buitjes commits only change Android charts and location/readout UI

完整论文与来源状态

IJRSI paper DOI 10.51244/IJRSI.2026.1308000020 claims ~87% 1-7 d accuracy without a metric contract; arXiv 2608.27665 is QBWO theory using GPT assistance, 2608.27692 is IASI retrieval, and the 2023 SSW paper has no AI forecast task

下一触发器

reconsider only when a frozen scientific forecast task, substantive artifact, exact model/data version, objective common-sample evaluation or downstream forecast effect appears

上次判断

exclude collectively: web/app activity, generic tables, vague accuracy, retrieval-only and physical-theory work must not fill a quiet core forecast window

完整代码与数据状态

HF API verifies exact lastModified times on 2026-08-25 UTC for all six packages and shows weight/.gitkeep with no pretrained file; model cards expose synthetic-data, train, inference and plotting entry points under Apache-2.0 top-level tags, while upstream code/data/weight rights remain item-specific

完整论文与来源状态

correction to the 2026-08-30 discovery: organisation-page relative timestamps did not prove a 2026-08-30 publication/update event; all packages cite older papers and are not current-window papers, releases or independent reproductions

下一触发器

actual weight files and hashes; tags/releases; precise repository-name fix; frozen real-data splits/preprocessing; end-to-end outputs; common-sample extremes/spectra/calibration; source/licence audit

上次判断

retain as framework-only infrastructure and retract the prior event-time inference: CorrDiff/NowcastNet remain directly relevant, but no current-window change or scientific baseline is verified

完整代码与数据状态

CAMS Aerosol Alerts, AirLens, WMO marine bulletins, MOSDAC/radar pages and generic forecast dashboards are operational or webpage refreshes; OpenReview hits are older ICLR 2026 papers; no new method, model version, weight, frozen science dataset or objective evaluation event

完整论文与来源状态

no relevant paper or formal status change after the previous cutoff; CAMS pages explicitly update daily and OneScience GLONET/other off-line adapters add no current-line scientific evidence

下一触发器

reconsider only if a method/version/production configuration, substantive artifact, exact status change or common-sample evaluation appears

上次判断

exclude collectively: routine products and batch packaging cannot fill a quiet scientific window; keep the daily operational-data clock separate from the research-event clock

完整代码与数据状态

no public paper-matched code, weights, frozen train/test manifest or software license verified; official HSAF portal runs NetNCC V1 at original MSG ~3 km over three African domains and V2 pan-African at 0.05 degrees for 60-360 min leads

完整论文与来源状态

formal QJRMS article DOI 10.1002/qj.70284, published online 2026-08-28; VOR CC BY 4.0; specific representative figure/full caption could not be verified and no figure was mirrored

下一触发器

code/weights/samples/supplement; exact train-test-event split; calibrated probability and joint radar/rain-gauge validation; 4-6 h local-scale evidence

上次判断

retain as high-value satellite convective-core probability nowcasting: three successive MSG IR frames support 1-6 h signals and beat persistence, but the target is neither radar reflectivity nor quantitative rainfall and longer-lead skill survives mainly at coarse 180-270 km scales

完整代码与数据状态

no public correction code, HFSS configuration, matched XPAR-SPOL dataset or software/data license verified

完整论文与来源状态

EGUsphere preprint DOI 10.5194/egusphere-2026-3965, posted 2026-08-28; open discussion and under review for AMT, not accepted; CC BY 4.0

下一触发器

code/config/data; multi-site and multi-event validation; downstream hydrometeor, assimilation and lead-time forecast impacts

上次判断

retain as upstream radar QC, not AI prediction: RC-ZDR combines system, attenuation, pointing-dependent isolation and radial-interference corrections; one squall-line case reduces XPAR-SPOL ZDR bias from 0.619 to 0.198 dB

完整代码与数据状态

Apache-2.0 GitHub first release includes executable SEVIR/HKO-7 training/evaluation code and 39.8/20.6 MB checkpoints; no tag/release, numerical result table or original SEVIR HDF5-to-PNG preprocessing script; derived dependencies/data terms require separate compliance

完整论文与来源状态

repository citation claims a 2026 Computers & Geosciences article but explicitly says it will be finalised on acceptance; no DOI, official acceptance or paper version verified; no representative figure mirrored

下一触发器

formal paper/status; frozen result/ablation tables; reproducible raw-data conversion and event split; independent common-sample extremes, spectra and calibration

上次判断

retain as direct radar-nowcasting artifact, not established skill: frequency enhancement, PhyCell, VMamba and high-frequency loss directly target detail preservation, but released weights without public numbers cannot support superiority claims

完整代码与数据状态

first frozen HF bundle contains 14 manifest-bound files/360,142,290 bytes, forward-time and site-OOD roles, baseline metrics and verifier; EPA AQS source is public domain, compilation/docs CC BY 4.0 and original code Apache-2.0, while the HF top-level tag remains license:other

完整论文与来源状态

public dataset first uploaded 2026-08-27; no paper/DOI or operational forecast status verified

下一触发器

versioned tag/DOI/paper; historical as-of availability; forecast-meteorology/CAMS/MLWP drivers; high-pollution and dust-stratified calibration

上次判断

retain as strong AQ evaluation infrastructure, not core forecast evidence: 2010-2025 ozone/PM2.5 next-day examples enforce time/site leakage boundaries and split-conformal intervals but contain no future meteorology or weather-error propagation

完整代码与数据状态

no downloadable screening/extraction matrix, unified benchmark, model code, weights or common-sample outputs verified from the official article

完整论文与来源状态

formal Frontiers in AI review DOI 10.3389/frai.2026.1908848, published 2026-08-27 and recovered from the buffer; CC BY 4.0

下一触发器

screening matrix/extraction data; correction or update; structured task/horizon/reference audit of the 139-study evidence base

上次判断

retain as a multi-horizon evidence map, not a leaderboard: it separates 0-6 h, 1-4 d and S2S predictability sources and highlights UQ/physics/generalization gaps, but performs no fair common-sample re-evaluation

完整代码与数据状态

DataCite/USC record has no downloadable asset, content URL, method, coverage, results, creator, rights, related identifier or license

完整论文与来源状态

dataset DOI 10.25549/usctheses-ouc11399o19r was registered 2026-08-27 but says publicationYear=2021; landing page returned 403 and the date/status conflict is unresolved

下一触发器

author/abstract/files/rights; CAMS version, starts/leads/reference, stacking/calibration design and reported results

上次判断

retain only as highly relevant title metadata; do not infer a new CAMS PM2.5 method or result until the malformed/incomplete record is resolved

完整代码与数据状态

IOP AQ-health review has no future forecast contract; HF weather audio/text/multimodal demos and routine tables have no atmospheric model release; RadarEyes is indoor/autonomous-driving mmWave; GitHub UI/hackathon/API refreshes lack scientific artifacts

完整论文与来源状态

no relevant new forecast paper, model release, weights, frozen scientific dataset, license or objective evaluation event

下一触发器

reconsider only if a future-valid atmospheric forecast, radar/AQ task, substantive artifact or common-sample evaluation appears

上次判断

exclude collectively and keep states separate: trustworthy-AI health governance, generic weather data, space weather, nonmeteorological radar and UI churn cannot fill core-line gaps

完整代码与数据状态

NOAA-PSL/nested-eagle is Apache-2.0 with inference/evaluation configs and a v1.0.0 checkpoint; latest repository commit remains 2026-03-20, and the paper-specific reforecast archive is only promised soon

完整论文与来源状态

arXiv 2608.26822v1, submitted 2026-08-27 09:01 UTC; preprint with no journal information; CC BY 4.0; original Figure 1 mirrored as a complete equal-aspect resize with metadata

下一触发器

frozen 293-variable and 1426-precipitation reforecast archive; ensemble/CRPS/reliability; fixed-intensity FSS, extremes, spectra, balances and independent operational tests

上次判断

retain as high-value 0-15 d nested global-regional evidence: HRRR-trained near-surface gains reassert in 6-30 h and persist for days, but absolute precipitation amplitude remains blurred and percentile-FSS location skill cannot substitute for correct intensity

完整代码与数据状态

github.com/hiepdang-ml/SimCast-S2S now accessible after prior 404; 77 tree files, data/train/inference/evaluation and quantile-mapping code; latest commit 23c5f041a21f remains June 17, public-switch time unknown; no README, standard LICENSE, weights or release

完整论文与来源状态

existing arXiv 2608.26594v1 plus Research Square DOI 10.21203/rs.3.rs-10872625/v1 registered Sep 1 05:23 UTC; both preprints, no formal acceptance verified

下一触发器

license/README/release/checkpoints; frozen 28-member CESM2 settings versus 14 currently enabled; exact calibration and ECMWF initialization/reforecast/postprocessing/member contract; reliability and common-sample reproduction

上次判断

update artifact accessibility and cross-platform metadata, not new skill: code can now be inspected but current default configuration is not the paper freeze; prior CESM2-to-ERA5 days-15-28 result and underdispersion limits remain

完整代码与数据状态

paper-linked mqpena/logistic-error-growth-limits returned 404; no public executable, frozen trajectories or standard license verified

完整论文与来源状态

arXiv 2608.26492v1, submitted 2026-08-27 00:27 UTC and recovered from the 48 h buffer; preprint with no journal information; arXiv non-exclusive license

下一触发器

public code/data; independent trajectory-level shape/clock/residual diagnostics; real AI/dynamical weather ensemble transfer

上次判断

retain as predictability diagnostic, not a forecast model: aggregate logistic fits can arise from heterogeneous clocks and cancellations, so mechanism claims require scale-shape, propagation-clock and budget-residual tests

完整代码与数据状态

no paper-matched public code merge, runnable MPAS-JEDI case, radar sample or frozen experiment package verified

完整论文与来源状态

formal Monthly Weather Review Early Online article DOI 10.1175/MWR-D-25-0276.1, published 2026-08-27; AMS default reuse rights do not authorize a figure mirror here

下一触发器

P-PRO/JEDI code record; multi-case/cycle tests; lead-wise reflectivity/QPF/probability skill and separable KDP/ZDR impact

上次判断

retain as radar-to-model short-range infrastructure, not AI nowcasting: ZH/vr/KDP/ZDR assimilation improves analysis consistency and early directly-assimilated-variable errors, but extra polarimetric forecast gains are not robustly separable in one case

完整代码与数据状态

no versioned 1 km AGCD download, code, frozen station/pseudo-observation manifest or data DOI verified as newly released in this run

完整论文与来源状态

formal Journal of Southern Hemisphere Earth Systems Science article DOI 10.1071/ES25008, published 2026-08-28; CC BY-NC-ND 4.0, so no figure mirror

下一触发器

operational dataset DOI/download/version; station-withholding and extreme diagnostics; license and change log

上次判断

retain as high-resolution precipitation-reference infrastructure, not forecast: 1880-present statistical interpolation is evaluated over 1990-2022, while automated daily 1 km delivery is still described as planned

完整代码与数据状态

no accessible abstract, forecast system identity, starts/leads, hydrologic setup, code, data or reproducible result verified

完整论文与来源状态

formal Digital Water article DOI 10.1080/28375807.2026.2725897, published online 2026-08-27; CC BY 4.0

下一触发器

official abstract/full text; common initialization/reference; basin split, flood metrics, code/data and figure-level source

上次判断

retain only as AI-precipitation-to-flood metadata: title and formal status are verified, but no scientific performance or relationship to the four core lines can be inferred safely

完整代码与数据状态

no accessible abstract, domain/resolutions, transfer protocol, train/test split, code, weights or open artifact license verified

完整论文与来源状态

formal Environmental Modelling & Software DOI 10.1016/j.envsoft.2026.107139; Crossref record created 2026-08-27 16:47 UTC and gives only August 2026, so exact online day is unverified; TDM terms are not open reuse

下一触发器

official VOR/abstract and exact date; forecast versus climate-characterization contract; data, metrics, code and license

上次判断

retain only as transfer-downscaling metadata; do not infer forecast skill, recovered fine scales or publication-day freshness from title and Crossref creation time

完整代码与数据状态

no accessible abstract, official repository, versioned configuration, data examples, metric inventory or standard software license verified

完整论文与来源状态

formal Computers & Geosciences DOI 10.1016/j.cageo.2026.106262; Crossref record created 2026-08-27 16:42 UTC and gives only August 2026, so exact online day is unverified; TDM terms are not open reuse

下一触发器

official VOR/abstract; code/release/license; supported products, ranking rules, uncertainty and reproducible case studies

上次判断

retain only as potentially relevant precipitation-evaluation metadata; do not upgrade to infrastructure until the executable framework and scientific protocol are verified

完整代码与数据状态

arXiv 2608.26767 is climate-scale ENSO forced response beyond 65 d; 2608.27205 is nitrogen management; HF items are weather documents/tables or a city demo; buitjes commit 4d1092b14df3 only repairs basemap switching

完整论文与来源状态

no new forecast paper, model release, weights, scientific dataset version, license or evaluation event relevant to the four core lines

下一触发器

reconsider only if a new initialized 0-72 h or 1-65 d forecast, radar/AQ task, substantive artifact or objective evaluation appears

上次判断

exclude collectively: long-climate response, general environmental management, routine tabular refresh and UI maintenance should not fill a daily scientific update

完整代码与数据状态

ERA5, PlaSim and official Pangu resources are identified; paper-trained models and inference data are only promised upon publication and no frozen code/weights package is public

完整论文与来源状态

arXiv 2608.25835v1, submitted 2026-08-26 14:14 UTC; preprint with no journal information; CC BY 4.0; original Lorenz-96 Figure 3 mirrored with complete metadata

下一触发器

public models/code/config; independent GPU/perturbation sensitivity; spectrum saturation, probability calibration and long-run climate tests

上次判断

retain as high-value predictability mechanism: spatial/temporal coarse-graining suppresses butterfly-like error growth and can extend deterministic large-scale skill, but this advantage can distort irreversibility, ensemble spread and physical realism

完整代码与数据状态

Crossref exposes the ESSOAr PDF link but it returned 403 in this run; no code, weights, full experiment configuration or artifact license verified

完整论文与来源状态

ESSOAr v1 preprint DOI 10.22541/essoar.15007917/v1, posted 2026-08-26; no journal information or reusable figure license verified

下一触发器

accessible full text/code; primitive-equation/emulator/window details; state versus trajectory distances; real-analysis and multimodel common-sample verification

上次判断

retain direct one-month mechanism evidence with a boundary: controlled twin experiments say extended skill is trajectory shadowing, while optimized initial conditions can be no closer to truth and do not measure atmospheric predictability from an analysis

完整代码与数据状态

ERA5 and ERA5-Land sources are identified; no public code, weights, exact date manifest or independent reproduction

完整论文与来源状态

arXiv 2608.25823v1, submitted 2026-08-26 14:04 UTC; preprint with no journal information; arXiv non-exclusive license, so no figure mirror

下一触发器

code/weights and exact chronological split; operational NWP inputs; independent non-integer/high-resolution truth; external years/extremes

上次判断

retain as direct 0-6 h regional temperature prototype: coordinate/lead/resolution queries improve +3 to +6 h aggregate scores, but ERA5-to-ERA5-Land testing and unverified query resolutions do not establish operational or super-resolution skill

完整代码与数据状态

ERA5, ERA5-Land and Prithvi WxC foundations are available; paper-specific code, weights, training config and generated samples are not public

完整论文与来源状态

arXiv 2608.25858v1, submitted 2026-08-26 14:30 UTC; preprint with no journal information; CC BY 4.0; original Figure 3 mirrored with complete metadata

下一触发器

release artifacts; true GCM/NWP forecast drivers; multi-basin/era tests; ensemble reliability, water balance, spectrum and tail confidence intervals

上次判断

retain as generative downscaling evidence, not forecast: Prithvi conditioning improves distribution bias, spectra and rare-event frequency while pixel CRPS is worse than concatenation, exposing a useful accuracy-structure trade-off on only three test years

完整代码与数据状态

SEVIR is available, but southeast-China paired data, split manifests, code, weights and artifact licenses are not public

完整论文与来源状态

arXiv 2608.25818v1, submitted 2026-08-26 14:01 UTC; preprint with no journal information; arXiv non-exclusive license, so no figure mirror

下一触发器

release data/code/weights; event-isolated and cross-sensor/radar tests; seeds/uncertainty; downstream future-nowcast impact

上次判断

retain as same-time satellite-to-radar flow-matching retrieval, not 0-3 h nowcasting: a 2.61M spectral-local-wavelet model improves author-reported MAE/SSIM/heavy CSI over a matched lightweight U-Net, but no future valid time or reproducible artifact exists

完整代码与数据状态

synthetic SDE generators and Beijing PM2.5 source are described; no public implementation, preprocessing, frozen splits or weights were found

完整论文与来源状态

arXiv 2608.25898v1, submitted 2026-08-26 15:18 UTC; preprint with no journal information; CC BY 4.0

下一触发器

code/frozen data; high-dimensional weather/AQ fields; forecast meteorology; cross-lead/spatial joint calibration and independent cities

上次判断

retain as uncertainty method, not core AQ: per-task likelihoods and input-dependent last-layer variance improve NLL in low-dimensional stochastic systems and a 24 h PM2.5 series, but the real arm is small, single-site and lacks meteorological drivers

完整代码与数据状态

FCI, CERRA, ERA5/ERA5-Land and IGRA sources are public; no training code, weights or processed collocation release

完整论文与来源状态

arXiv 2608.25700v1, submitted 2026-08-26 12:16 UTC; preprint with no journal information; CC BY-NC-ND 4.0, so no figure mirror

下一触发器

executable/weights; information-content and correlated-error analysis; independent domains/satellites; DA/nowcast impact

上次判断

retain as forecast-independent observation retrieval infrastructure, not prediction: a spatial Residual U-Net retrieves 15-level all-sky profiles and validates against more than 11k radiosondes, but location climatology, CERRA targets and no DA impact limit forecast claims

完整代码与数据状态

implementation derives from public DoWnGAN; no paper-specific code, weights, frozen GDPS/HRDPS pairs or independent package verified

完整论文与来源状态

formal Artificial Intelligence for the Earth Systems article DOI 10.1175/AIES-D-25-0022.1, published online 2026-08-26; earlier arXiv 2412.06958

下一触发器

public model/artifacts; direct +48 h to 10 d validation; station/extreme/terrain/ensemble tests and independent reproduction

上次判断

update to formal publication: high-resolution static covariates and frequency separation improve +6 to +17 h wind RMSE/spectra over DownGAN, but the claimed 10-day extension remains potential because no longer leads were tested

完整代码与数据状态

conceptual evaluation paper with no new model, code or dataset

完整论文与来源状态

formal Environmental Research Communications article DOI 10.1088/2515-7620/ae9f1f, published 2026-08-26; CC BY 4.0; earlier arXiv 2602.00622

下一触发器

operational falsification suites; hard-constraint/balance/perturbation benchmarks; links to skill and decision value

上次判断

update to formal publication and retain as evaluation infrastructure: functional, structural and physical realism require verification, diagnostics and falsification rather than a single RMSE or visual sharpness test

完整代码与数据状态

Apache-2.0 repository commit 4250d37462b9 adds an STDiT denoiser, chunked autoregressive sampler, UOT/quantile-weighted objectives, CIKM2017/Shanghai2020/MeteoNet loaders and evaluation tools; no weights/results/locked environment

完整论文与来源状态

public derivative code repository created/pushed 2026-08-26; no new paper, tag/release or independent verification

下一触发器

frozen environment/data hashes; public weights; common-sample ablations for calibration, extremes, spectra and efficiency; upstream review

上次判断

retain as substantive code infrastructure without promoting science: implementation breadth increased, but no result table or artifact proves improvement over official FlowCast or other nowcasting baselines

完整代码与数据状态

no accessible abstract, data, starts/leads, split, code, weights or artifact license verified

完整论文与来源状态

formal Journal of Earth System Science article DOI 10.1007/s12040-026-02919-9, published 2026-08-26; publisher page/PDF blocked by a security response in this run

下一触发器

official abstract/full text; NWP source and issuance-valid contract; band thresholds, common samples, extremes/FSS/reliability; artifacts and figure license

上次判断

retain only as direct-topic metadata: title indicates transformer heavy-precipitation bias correction, but no scientific content or skill can be inferred safely and no figure can be mirrored

完整代码与数据状态

paper lists Anemoi package versions and DWD exposes operational forecast files; no AICON-specific configuration, weights, frozen ICON-DREAM training manifest or evaluation package was released

完整论文与来源状态

arXiv 2608.24651v1, submitted 2026-08-25 14:58 UTC; preprint with no journal information; CC BY 4.0; Figure 13 mirrored with complete metadata

下一触发器

public weights/config and training years; independent operational common samples; precipitation extremes, spectra, physical balance, ensemble calibration and 0-72 h regional tests

上次判断

retain as high-value deterministic short-range evidence: single-step 3 h training reduces rollout-induced small-scale damping and supports operational 13 km forecasts, but mesoscale activity, strong wind and tropical-cyclone intensity remain too weak and no dedicated model package is public

完整代码与数据状态

data and code are available only from the corresponding author upon reasonable request; no public frozen station manifest, split or executable package

完整论文与来源状态

formal Meteorology and Atmospheric Physics article DOI 10.1007/s00703-026-01183-2, volume 138 article 89, published 2026-08-24 and recovered from buffer; exclusive Springer publishing rights

下一触发器

public artifacts; external years/regions; true precipitation target and threshold verification; archived commercial/NWP issuance-time common samples

上次判断

retain direct 1-168 h station forecasting with a boundary: five-fold walk-forward tests favor simple RNNs at 1-6 h and show severe wind degradation at 168 h, but only two years/one region are used and precipitation is excluded from model targets

完整代码与数据状态

paper creates a 4.8 GB SEVMOD-CM derivative set from NASA MODIS and EUMETSAT SEVIRI but exposes no public dataset, code, weights, frozen split or software/data release license

完整论文与来源状态

arXiv 2608.24715v1, submitted 2026-08-25 15:34 UTC; preprint, no independently verified conference record; CC BY 4.0

下一触发器

release data/code; split by scene, orbit, date, region and year; boundary/object/calibration metrics; future cloud/precipitation transfer

上次判断

retain as same-time cross-sensor cloud-mask downscaling, not forecast: SpatialGAN improves SSIM/MSE over bicubic while PSNR is slightly worse, SpatialCNN loses on all three metrics, and random 80/10/10 patch splitting leaves scene leakage uncertain

完整代码与数据状态

no public paper-matched code, weights, radar samples, event manifest or frozen split verified

完整论文与来源状态

formal IET Radar, Sonar & Navigation article DOI 10.1049/rsn2.70129, published 2026-08-25; CC BY-NC-ND 4.0

下一触发器

full dataset/split and per-variable errors; cross-radar/year/intensity tests; code/weights/license; downstream QPE and forecast impact

上次判断

retain as same-time upstream polarimetric reconstruction, not radar nowcasting: a U-Net generative autoencoder synthesizes dual-pol fields from reflectivity and reports about 85% classification in selected scenarios, but temporal prediction and generalization evidence are absent; no figure mirror under NC-ND

完整代码与数据状态

code is available only on request; KMA wind-profiler and radiosonde data are restricted by access policy and cannot be redistributed by the authors

完整论文与来源状态

EGUsphere preprint egusphere-2026-3168, discussion started 2026-08-25 and under review for AMT; not accepted/final; CC BY 4.0

下一触发器

public executable/frozen sample; external station and hardware tests; strong-wind weighted or probabilistic loss; operational latency

上次判断

retain as upstream wind-observation/QC infrastructure: raw-spectrum DL benefits depend on hardware error morphology, hybrid QC helps 1290 MHz transient outliers but not 437.5 MHz coherent range-smearing, and MSE underestimates extremes

完整代码与数据状态

no public code, frozen GraphCast cycles, radiosonde/GNSS matchup package or data/software license was verified from accessible material

完整论文与来源状态

formal Earth and Space Science article DOI 10.1029/2025EA004409, published 2026-08-25; CC BY-NC 4.0

下一触发器

exact model version, starts/leads and common samples; analysis versus true forecast drivers; code/data; downstream precipitation or DA impact

上次判断

retain as AI-forecast-driven observation retrieval, not direct weather forecast skill: GraphCast Tm/ZHD yields author-reported PWV RMSE 0.73 mm versus radiosonde and 1.7 mm versus GNSS, but the accessible record cannot separate analysis-near-time quality from lead-time error propagation; no figure mirror under NC

完整代码与数据状态

no accessible abstract, code, weights, data, split, results, artifact license or open figure permission verified

完整论文与来源状态

formal IEEE Geoscience and Remote Sensing Letters article DOI 10.1109/LGRS.2026.3727533; Crossref record created 2026-08-25 19:05 UTC, exact online date unavailable

下一触发器

official abstract/full text; satellite inputs and precipitation reference; 0-3 h issuance-valid contract; common-sample metrics and artifacts

上次判断

retain only as direct-topic metadata candidate: title supports satellite precipitation nowcasting relevance, but no scientific content can be safely inferred and it cannot enter core-nowcasting

完整代码与数据状态

MIT repository ingests KNMI CC BY 4.0 radar and 20-member seamless precipitation ensemble; commit ce97b027091e first max-pools each member on space, then computes p10/p50/p90 at a coarser grid with block-aware radius and expanded tests; author measures 35% less cycle time and 77% fewer spread bytes; no frozen release/evaluation

完整论文与来源状态

public GitHub substantively updated 2026-08-28 13:13 UTC; no paper, DOI, tag/release or independent verification

下一触发器

frozen release/source versions; event/lead coverage, reliability, Brier/CRPS/FSS and threshold utility before/after pooling; independent deployment

上次判断

update infrastructure state: the processing order preserves neighbourhood-versus-member semantics while reducing cost, but implementation tests and one-cycle efficiency do not establish probabilistic forecast skill

完整代码与数据状态

MIT algorithm and immutable prospective result log are public; no frozen release or DOI; current record has 208 usable historical days, 153 walk-forward origins and 7 scored prospective days

完整论文与来源状态

public GitHub result repository created 2026-08-25; no paper or independent verification

下一触发器

one full annual cycle and many prospective origins; lead-wise interval coverage/width; high-pollution and dust thresholds; future-weather-driver comparison and external zones

上次判断

retain as prospective AQ evaluation infrastructure, not established core skill: causal level-plus-anomaly forecasts expose 80% ranges and score later, but weather covariates are excluded from the concentration model and seven scored days cannot calibrate uncertainty

完整代码与数据状态

no forecast artifact tracked because the endpoint is district-level casualty vulnerability/risk mapping from 2019-2025 geography, population, land cover and lightning-stroke history

完整论文与来源状态

formal Weather, Climate, and Society article DOI 10.1175/WCAS-D-25-0057.1, Crossref-created in the main window

下一触发器

reconsider only if a separate initialized lightning/thunderstorm forecast supplies starts, leads, future meteorology and out-of-sample event verification

上次判断

exclude from forecast advances: retrospective risk mapping has no future valid time and cannot be treated as lightning or severe-convection nowcasting

完整代码与数据状态

no atmospheric forecast artifact retained from arXiv 2608.23813v1 or 2608.24428v1

完整论文与来源状态

both are arXiv preprints; the former is generic PDE/EnKF localization without real atmospheric evidence, the latter forecasts Nino-3.4 up to six months and only carries an author submitted-to note

下一触发器

reconsider on a real weather/observation experiment inside 0-72 h or 1-65 d with code and fair verification

上次判断

exclude collectively: generic numerical examples and lead ranges beyond 65 d should not fill a quiet direct S2S/nowcasting window

完整代码与数据状态

no paper-matched code, frozen ECMWF ensemble fields, initialization/member manifest or processed verification data verified

完整论文与来源状态

formal JGR: Atmospheres article DOI 10.1029/2026JD046489, volume 131 issue 16, Crossref-created 2026-08-24 19:28 UTC; Wiley VOR/TDM terms are not an open figure license

下一触发器

full text/supplement; exact starts, members, rainfall reference and filtering endpoints; reproduce across years and ECMWF/NCEP/CMA/UKMO/AI systems with absolute/anomaly separation

上次判断

retain as direct core-s2s mechanism evidence: the 2024 ECMWF annual cycle advances too rapidly after week one and author attributes about one day of East-Asia rainfall skill loss to the drift; accessible metadata does not support finer protocol claims or a figure mirror

完整代码与数据状态

no public paper-matched code, weights, frozen station manifest or split verified; Shanghai hourly source period is 2021-07-01 to 2023-12-01

完整论文与来源状态

formal Scientific Reports article DOI 10.1038/s41598-026-67862-7, published 2026-08-24 as an early citable version; CC BY-NC-ND 4.0

下一触发器

exact horizons, stations and chronological split; issuance-time weather contract; extremes, probabilistic calibration, external-city test and code/weights

上次判断

retain as direct six-pollutant AQ prediction: DenseNet-LSTM with horizon-aware attention reports PM2.5 RMSE 6.35 and MAE 4.12 microgram m-3, but accessible page cannot separate autoregressive memory from deployable future-driver skill; no figure mirror under NC-ND

完整代码与数据状态

Zenodo software 10.5281/zenodo.19673042 and 39.5 GB data 10.5281/zenodo.19029543 are CC BY 4.0; public GitHub zywnuist/Diffwind has no detected standard LICENSE

完整论文与来源状态

formal Geoscientific Model Development article DOI 10.5194/gmd-19-7835-2026, published 2026-08-24; CC BY 4.0; Figure 2 mirrored with complete metadata

下一触发器

cross-season/region/independent-station tests; reconcile reported generation runtime; high-wind tails, spatial dependence and spread-skill; precipitation/radar transfer with conservation

上次判断

retain as uncertainty method, not radar or precipitation nowcasting: a conditional DDPM learns log SIVA wind-error distributions and generates 16-member +1 to +6 h ensembles, improving mean/probabilistic scores while remaining underdispersed and smoothing high-wind tails

完整代码与数据状态

public MIT GitHub has campaign code, but the paper-declared v1.0-campaign release/tag was not visible through the public GitHub release/tag API at cutoff; OpenAQ/Open-Meteo/UCI data are public

完整论文与来源状态

Research Square posted-content DOI 10.21203/rs.3.rs-10515545/v1, posted 2026-08-24; preprint, no journal information; CC BY 4.0

下一触发器

visible frozen release/tag; archived operational weather forecasts; threshold coverage, sensor heterogeneity and contamination audit; independent 29-city reproduction

上次判断

retain as AQ forecast-protocol evidence, not a universal TSFM win: causal last-known versus perfect-future weather changes rankings/equivalence, and zero-shot Chronos is competitive on PM2.5 but not significantly superior by Wilcoxon after the full protocol

完整代码与数据状态

no paper-matched public code or weights verified; GPM/ERA5/OLR source families are identified

完整论文与来源状态

arXiv 2608.22358v1, submitted 2026-08-23 10:50 UTC and recovered from buffer; preprint, no journal information; arXiv default non-exclusive license

下一触发器

remove retrospective zero-phase OLR filtering and future-centroid recentering; fixed-grid end-to-end positions; code/weights, unseen-year pretraining and cross-region tests

上次判断

retain high-value 1-7 d representation/probability prototype with a hard boundary: proxy-variable JEPA plus 16-member rectified flow slightly beats matched IFS fair CRPS, but supplied storm motion makes the patch-frame comparison an upper-bound representation test rather than operational S2S forecasting

完整代码与数据状态

paper claims public curian127/score-pff with code/notebooks/weights, but the repository/API returned 404 at cutoff

完整论文与来源状态

arXiv 2608.22454v1, submitted 2026-08-23 15:08 UTC and recovered from buffer; preprint, no journal information; CC BY-NC-ND 4.0

下一触发器

accessible frozen repository; localization/model-error safeguards; real atmospheric observations and complex model; independent high-dimensional reproduction

上次判断

retain as score-based non-Gaussian DA method only: 40D/1000D Lorenz-96 results are promising, including author-reported 49.5% RMSE reduction over PFF in 1000D, but no real-weather evidence or verifiable artifact yet

完整代码与数据状态

no public code, frozen 24-member initialization sweeps, ERA5 extracts or evaluation package verified from accessible record

完整论文与来源状态

formal Environmental Research Letters article DOI 10.1088/1748-9326/ae9d97, published 2026-08-24; CC BY 4.0

下一触发器

full methods/data; multi-event and multi-region replication; window-position labels in AI/NWP training and lagged-ensemble comparisons

上次判断

retain as initialization/evaluation infrastructure, not AI skill: an ERA5 assimilation-window boundary changes the moisture corridor and precipitation outcome between adjacent hourly starts, so window position belongs in fair high-resolution forecast contracts

完整代码与数据状态

CMA radar-gauge data are restricted and available on request subject to approval; no public implementation or frozen sample verified

完整论文与来源状态

formal SOLA article DOI 10.1007/s44393-026-00037-9, published 2026-08-24; CC BY 4.0

下一触发器

public reference implementation/sample; leave-radar/season/intensity tests; propagate correction uncertainty into nowcast ensembles and thresholds

上次判断

retain as upstream radar-QPE/label-quality evidence, not future nowcasting: Lagrangian log-bias advection nearly removes mean bias under spatial and temporal holdouts, but optical flow omits lifecycle/microphysics and gains decay beyond about 8-10 km

完整代码与数据状态

MIT repository now contains ST-GNN checkpoints, VIIRS and ERA5/Open-Meteo inputs, 2015-2020 evaluation and 2025-2026 recovery replay; heterogeneous source licenses and no frozen release

完整论文与来源状态

no paper/DOI/release or independent verification; substantive checkpoint/data commit 595234274aa3 followed by UI/scaling commits through cutoff

下一触发器

frozen release/data-license manifest; untouched external-year evaluation; archived future-weather drivers; severe-onset calibration/recall and independent reproduction

上次判断

retain as executable 24/48/72 h PM2.5 infrastructure: aggregate RMSE improves over persistence and 2025-2026 replay transfers, but DLinear is competitive, COVID/severe regimes weaken, severe-onset recall is low and no probability is expressed

完整代码与数据状态

no public paper-matched calibration code, frozen 58-case dataset or data/software license verified

完整论文与来源状态

formal Remote Sensing article DOI 10.3390/rs18172854, published 2026-08-23 and registered in the main incremental window; CC BY 4.0

下一触发器

public implementation/data; independent network/season/intensity tests; weak-cloud auxiliary truth and downstream QPE/nowcast impact

上次判断

retain as upstream X-band calibration, not prediction: opposing RHI scans support stable midline/region bias estimation across 58 cases, while weak echoes and cross-network generalization remain uncertain

完整代码与数据状态

no abstract, code, weights, data, split, result package or software/data license verified

完整论文与来源状态

formal Signal, Image and Video Processing article DOI 10.1007/s11760-026-05666-y, volume 20 issue 11 article 605, published 2026-08-25; publisher terms do not establish reusable figure permission

下一触发器

official abstract/full text; variables, radar dataset and 0-3 h contract; persistence/optical-flow/generative common samples, FSS/spectrum/object/calibration and artifacts

上次判断

retain only as direct-topic metadata candidate: title suggests dual-polarization echo sequence prediction, but no method or skill claim can be verified and it cannot yet enter core-nowcasting

完整代码与数据状态

no accessible code, frozen WRF-CMAQ cases, wind-correction model or data/software license verified

完整论文与来源状态

formal Atmospheric Environment article DOI 10.1016/j.atmosenv.2026.122324, available online 2026-08-24; CC BY-NC-ND 4.0

下一触发器

official abstract/full text; starts/leads; training split; pollutant impacts, actual forecast versus analysis drivers; code/data

上次判断

retain as high-priority core-AQ candidate at metadata level only: title and formal record confirm ML wind correction in coupled WRF-CMAQ, but no method, result or causal forecast contract is currently accessible; no figure mirror under NC-ND

完整代码与数据状态

no accessible code, data, model or artifact license verified

完整论文与来源状态

formal npj Climate and Atmospheric Science article DOI 10.1038/s41612-026-01522-5, published 2026-08-25; CC BY-NC-ND 4.0

下一触发器

accessible abstract/full text; AI status, lead, nesting and structure/uncertainty results; direct transfer to precipitation/S2S

上次判断

retain metadata only: soil moisture is outside the four direct endpoints and accessible record cannot establish a transferable AI or 1-65 d result; no figure mirror under NC-ND

完整代码与数据状态

one public commit uploads about 938 files/214 MB with message Add radar AE and latent predictor; no README, LICENSE, paper link, data provenance, environment or frozen metrics

完整论文与来源状态

no paper, DOI, release or independent result verified

下一触发器

reconsider only with traceable data, executable model, frozen evaluation, standard license and a scientific report

上次判断

exclude once and do not reopen on timestamp alone: an opaque bulk code/data upload is not sufficient evidence of a new radar-nowcasting result

完整代码与数据状态

public GitHub has data-generation/training/evaluation scripts, a 50,000-point analytic CSV, baseline/PINN weights and plots; repository has no standard LICENSE or release although the preprint is CC BY 4.0

完整论文与来源状态

engrXiv/Open Engineering posted-content DOI 10.31224/8035, issued 2026-08-23 and Crossref-created 2026-08-23 21:14 UTC; preprint, no journal information

下一触发器

real stations/emissions/forecast meteorology; strictly future-held times and source regimes; PDE/driver uncertainty; code/weight license and independent reproduction

上次判断

retain as physical-constraint interpolation method, not core AQ forecast: 200 random sensors and 2,000 collocation points span the same analytic 0-1 h domain later evaluated, so author MSE 0.031533/R2 0.9274 and 40.8% MSE reduction do not establish causal future or real-world pollution skill; no figure mirror because the concrete repo image provenance/license chain was not closed

完整代码与数据状态

public Beijing Multi-Site AQ dataset is identified, but no paper-matched code, weights, frozen split or environment was found; Figure 1 mirrored from arXiv v1 under CC BY 4.0 with metadata

完整论文与来源状态

arXiv 2608.21277v1, submitted 2026-08-21 16:35 UTC; preprint, no journal information; CC BY 4.0

下一触发器

issuance-time CAMS/MLWP/NWP future covariates; forecasted co-pollutants; causal imputation; code/weights/split; cross-city extremes, CRPS/reliability and residual-bypass audit

上次判断

retain as interpretable AQ time-series method, not core forecast: 96 h future-aware results use actual future meteorology and other pollutants, full-series interpolation precedes the chronological split, look-back-only errors are much larger, and the residual channel exposes a material accuracy-interpretability trade-off

完整代码与数据状态

BSD-3-Clause; merge 7ebb7688d728 adds Serbian radar-IFS linear/saliency/STEPS blending, MLCast RADCLIM extrapolation/FSS and lectures; commit f683246ff84d only fills two CRPS exercise lines and removes one overly technical exercise; no frozen data manifest, locked environment, expected-output hashes, AI weights or independent skill table

完整论文与来源状态

course material for 2026-08-23 ERAD26 Academy, not a paper/model release/frozen benchmark; latest main-window commit is routine exercise maintenance, not a new scientific event

下一触发器

final archive/data-license manifest/locks/hashes; full-member ensemble; AI exercise/weights; common-sample extrapolation/STEPS/NWP/blend FSS/CSI/CRPS/reliability across radar domains

上次判断

retain the post-course infrastructure state without promoting the latest small edit: the 00 UTC IFS-to-15:40 radar contract and cross-European path remain useful, while one-member STEPS and incomplete/frequently edited exercises do not prove probabilistic or operational skill

完整代码与数据状态

no code, data, model, input-time contract or reusable figure license verified

完整论文与来源状态

formal Sustainable Cities and Society article DOI 10.1016/j.scs.2026.107860, available online 2026-08-23; non-open-access Elsevier record exposed title/authors only and was not yet indexed by OpenAlex/Semantic Scholar

下一触发器

official abstract/full text; sensor/CFD inputs; reconstruction versus forecast definition; AI/model details; data/code and quantitative validation

上次判断

retain only as fine-structure candidate: current metadata cannot distinguish ML, data assimilation, reduced-order or physical reconstruction and cannot support core-short-range classification or a figure mirror

完整代码与数据状态

no forecast artifact tracked because the accessible records identify historical/observational analysis or sensor calibration endpoints

完整论文与来源状态

new records include CMIP6 winter-extreme-precipitation variability DOI 10.1016/j.wace.2026.100948, Mentawai diurnal precipitation DOI 10.61310/mjst.v24i1.2606 and cross-city low-cost PM2.5 sensor calibration DOI 10.1002/met.70228

下一触发器

reconsider only if a separate initialized forecast, issuance-time meteorological driver, future target or transferable forecast-evaluation artifact appears

上次判断

exclude collectively: historical CMIP/IMERG/GSMaP diagnosis and sensor calibration are not 0-72 h/S2S/AQ forecasts and should not fill a quiet core window

完整代码与数据状态

no paper-matched code, 65-parameter manifest, hindcast configuration, outputs or data/software license verified; Wiley VOR PDF returned 403

完整论文与来源状态

formal JGR: Atmospheres article DOI 10.1029/2026JD047873, received 2026-06-22, accepted 2026-08-13 and online 2026-08-21; Crossref record surfaced after the prior cutoff; Wiley VOR/TDM terms are not an open figure-reuse license

下一触发器

full text/supplement; hindcast years, initialization, members, observations, absolute/anomaly and exact skill/uncertainty definitions; code/parameter table; AI/dynamical common-sample transfer

上次判断

retain as direct dynamical core-s2s evidence, not AI: two PPE types across 65 parameters and >500 seasonal experiments identify convection entrainment and hydrometeor fall-speed sensitivities, reduce several regional monsoon precipitation biases, but leave Bay of Bengal/western Pacific low skill and uncertainty; no figure mirror

完整代码与数据状态

no code, frozen CMIP6 member list, processed diagnostics or data/software license verified; ESSOAr PDF returned 403

完整论文与来源状态

ESSOAr posted content DOI 10.22541/essoar.15007752/v1, posted 2026-08-22; no journal information and no reusable license in Crossref

下一触发器

accessible full text; model/member/period manifest, detrending/forcing handling, statistical-learning method and validation; code/data; initialized 1-65 d reproduction

上次判断

retain as low-frequency predictable-information mechanism, not core-s2s: CMIP6 underestimates one-year-plus precipitation/SLP autocorrelation and SST conditioning relative to ERA5/20CRv3, but the task is decadal variability rather than initialized forecasting; no figure mirror

完整代码与数据状态

no paper-matched SOM code, frozen in-situ/gridded-PM2.5/ERA5 sample or licenses verified

完整论文与来源状态

earlier SSRN DOI 10.2139/ssrn.6797261 posted 2026-05-19; formal Atmospheric Research DOI 10.1016/j.atmosres.2026.109287 appeared in-window with only August 2026 date; Elsevier copyright/TDM policies are not open figure permission

下一触发器

VOR methods/data; grid/station and SOM stability; code; archived CAMS/MLWP or NWP-driven future forecasts with regime-stratified threshold reliability

上次判断

record formal-status change and retain as PM2.5 meteorological-regime evidence, not AI or forecast: transport/accumulation modes and AO/Rossby-wave modulation can stratify driver-error evaluation, while 2017-2022 diagnosis cannot establish future forecast skill; no figure mirror

完整代码与数据状态

commit c620eb87 adds a 205-line NSW API ingestion script for PM2.5, PM10, O3, NO2 and colocated meteorology with station filters, validation, chunking and retries; no standard LICENSE, committed data, model, results or release

完整论文与来源状态

no paper/DOI; Charles Darwin University 2026 course project; README proposes next-24 h threshold probabilities, Open-Meteo forecast covariates and leave-one-station-out evaluation

下一触发器

archived issuance-time Open-Meteo driver; feature/model/LOSO implementation; chronological tests, Brier/reliability and cost loss; standard license and report

上次判断

retain only as early AQ forecast infrastructure: the probability target, persistence/seasonal baselines and LOSO design are relevant, but the main-window change implements observations only and supplies no forecast evidence

完整代码与数据状态

no code, data, forecast artifact or license verified; Elsevier TDM endpoint returned 400

完整论文与来源状态

formal Journal of Hydrology article DOI 10.1016/j.jhydrol.2026.136293; Crossref created in-window and gives only August 2026; accessible metadata has no abstract or forecast contract

下一触发器

accessible full text shows a genuine future target, initialization, lead, time-out evaluation and transferable artifact for a monitored track

上次判断

exclude from fresh forecast advances: title/metadata indicate retrospective extreme-precipitation and terrain interpretation, not 0-72 h or S2S prediction, and absent methods do not justify stronger claims

完整代码与数据状态

prior empty/scaffold repos remain unchanged; jniimi/weather_forecast_japan continues hourly Upload dataset commits and nuuuwan/lk-dmc-weather-forecasts-{docs,chunks} continues roughly two-hourly forecast-document appends; no schema, license, method or evaluation change was verified

完整论文与来源状态

no matched paper, release or independent result; JMA dataset card is CC BY 4.0 but scheduled data additions alone are not new research events

下一触发器

executable scientific payload, frozen evaluation, model/data/result, schema/method or license change beyond automated append

上次判断

exclude collectively and do not reopen on timestamps alone; descriptions, empty scaffolds and scheduled forecast/observation uploads are not evidence of skill or a substantive research release

完整代码与数据状态

no paper-matched code, frozen 34-station sample or data license verified; accessible publisher abstract identifies 2015-2024 New Jersey station data but not the full split

完整论文与来源状态

formal Discover Applied Sciences article DOI 10.1007/s42452-026-09423-1, online 2026-08-22, CC BY 4.0

下一触发器

full text; seven-model list and chronological split; persistence/climatology and wet/extreme-day skill; code/data and external-region replication

上次判断

retain as +1 d precipitation predictability negative baseline, not core short-range: antecedent-rainfall/time-only models have near-zero or negative R2 despite SVR MAE 0.178 mm, motivating dynamic atmospheric inputs; no figure mirrored because exact figure/caption asset was not verified

完整代码与数据状态

no public scripts, frozen eight-product manifest, station list, AWS sample or data/software license verified from accessible record

完整论文与来源状态

SSRN posted content DOI 10.2139/ssrn.7331055; Crossref created 2026-08-22 02:46 UTC and gives only year 2026; no journal information; copyright link is not a reusable license

下一触发器

accessible full text; full product list/years/station count; disaggregation formula and leakage audit; code/data snapshot and licenses

上次判断

retain as hourly precipitation data-generation/evaluation infrastructure, not forecast: CMA stations plus independent AWS support product diagnosis, while CHMPRE/CMFD author metrics do not establish future valid-time skill; no figure mirror

完整代码与数据状态

no public differential-evolution implementation, frozen 2020-2024 radar-gauge-disdrometer matchups or data/software license verified

完整论文与来源状态

formal JGR: Atmospheres article DOI 10.1029/2026JD046447, publication record dated 2026-08-21; Wiley/AGU terms not verified as reusable figure license

下一触发器

accessible supplement/sample details; cross-event and radar-distance strata; code/matchups/licenses and external-winter validation

上次判断

retain as upstream snowfall radar-QPE and label-uncertainty evidence, not AI or forecast: stable b=1.72 but variable a links Z-S conversion to microphysics and system organization; no figure mirror

完整代码与数据状态

no accessible abstract, code, weights, data, experiment package or software/data license verified; exact GitHub title/DOI search returned no match

完整论文与来源状态

IEEE TCSVT DOI 10.1109/TCSVT.2026.3725698, document 11661771; Crossref created 2026-08-21 19:06 UTC but gives only year 2026 and no exact online day; IEEE general terms are not reusable figure permission

下一触发器

IEEE abstract/VOR; radar/satellite input and 0-3 h contract; datasets/splits; persistence/optical-flow/generative common samples; FSS/spectrum/object/calibration; code/weights/licenses

上次判断

retain only as direct-topic metadata candidate: title suggests vision-language spatiotemporal alignment for precipitation nowcasting, but no technical claim can be verified and no figure may be mirrored

完整代码与数据状态

no accessible full text, code, frozen municipal/SPEI dataset, split or software/data license verified

完整论文与来源状态

SSRN posted content DOI 10.2139/ssrn.7331054; Crossref created 2026-08-22 02:45 UTC and gives only year 2026; no journal information

下一触发器

full text and exact posted day; SPEI timescale and forecast lead; issuance-time inputs; chronological/leave-region tests, baselines, code and formal venue

上次判断

retain only as spatial-transfer candidate: tumbling-window LSTM is reported robust in 71.8% of neighboring districts, but absent lead/input/split contracts prevent core-s2s classification

完整代码与数据状态

public repository includes real-time radar display, replay/export features, 6,000-sample synthetic training script, two pickle files and TypeScript empirical inference; repository tree has no standard LICENSE file although README says MIT; front end does not load pickle models

完整论文与来源状态

no paper, DOI, release, tag or independent verification record; substantive commit b8a522f at 2026-08-21 16:23 UTC followed mainly by automatic radar snapshots

下一触发器

real hail-report/radar common samples; storm/year/region-isolated tests; production loading of frozen trained model; POD/FAR/CSI and reliability/coverage; standard license file

上次判断

exclude once and do not reopen on live snapshots/UI changes: hail labels derive from synthetic MESH-like formulas and random splits test self-consistency, while hand-coded browser trees/empirical blend cannot support claimed calibrated probability or uncertainty

完整代码与数据状态

no accessible abstract, data, code, calibration artifacts or licenses verified

完整论文与来源状态

formal SPIE proceedings article DOI 10.1117/12.3103367; Crossref created 2026-08-21 22:49 UTC and records publication on 2026-08-21

下一触发器

reconsider only if the probabilistic method is transferred to a monitored weather/AQ task with direct calibration evidence

上次判断

exclude from monitored advances: atmospheric seeing for astronomy is outside radar/precipitation, regional weather, S2S and air-quality endpoints, and accessible metadata is insufficient for method transfer

完整代码与数据状态

no AI forecast artifact tracked because the accessible record identifies an index-tracking study rather than a prediction experiment

完整论文与来源状态

formal Journal of Earth System Science article DOI 10.1007/s12040-026-02918-w, online 2026-08-22

下一触发器

future AI/dynamical forecast with explicit initialization, valid time, 1-65 d lead and out-of-sample skill

上次判断

exclude from daily retained items: physical/SPEI flash-drought tracking is not itself an AI forecast or S2S prediction result

完整代码与数据状态

paper-matched GitHub was created in-window with MIT code for ClimaX/ViT finetuning, TopK/GeoTopK SAE, Skope-Rules explainers, ablations and evaluation; external SeasFire/ClimateNet data, ClimaX checkpoint and two git submodules are required; no frozen paper weights, logs or one-command environment

完整论文与来源状态

arXiv 2608.20117v1, submitted 2026-08-20 14:49 UTC; arXiv preprint only, no journal information; author says Under Review and identifies two non-archival oral versions; CC BY-SA 4.0

下一触发器

frozen checkpoints/logs/environment; multi-seed rule stability and causal sanity checks; direct precipitation/radar/AQ forecast transfer and main-task skill

上次判断

retain as location-aware sparse-feature and rule-surrogate method, not direct weather core result: SeasFire is 8-d wildfire forecasting and ClimateNet is same-time TC/AR detection; event reconstruction can worsen for one ViT setting and explanation fidelity is not forecast causality; Figure 2 mirrored under CC BY-SA 4.0

完整代码与数据状态

no paper-matched public code verified; data available only on reasonable request; all experiment data are generated from Hénon, folded-towel and Lorenz-96 systems

完整论文与来源状态

arXiv 2608.20017v1, submitted 2026-08-20 13:27 UTC; arXiv preprint only, no journal information; arXiv perpetual non-exclusive license

下一触发器

public code/frozen experiments; real ECMWF/NCEP/CMA/AI common-initialization error-geometry test; independent weighting, member dependence, CRPS/reliability and extreme coverage

上次判断

retain as mechanistic ensemble-error evidence only: errors partly cancel along unstable manifolds and heavy tails weaken ideal ensemble-size scaling, but shared-data random reservoirs are not real weather multi-model systems; no figure mirror without reusable license

完整代码与数据状态

public author GitHub contains scripts/data entry points but has no detected standard license and no in-window commits; AQCLIM/EEA, CAMS, ERA5/ERA5-Land, emissions and DEM sources are identified

完整论文与来源状态

April arXiv 2604.18823v1 gained formal Spatial Statistics DOI 10.1016/j.spasta.2026.101034; Crossref record was created 2026-08-20 16:07 UTC and gives August 2026 but no exact online day; arXiv is CC BY-NC-SA 4.0

下一触发器

exact VOR date/license; code license; causal historical window and real forecast meteorological/CTM drivers; external-year/region, error propagation and calibration

上次判断

record formal-status change and retain as same-day NO2 spatial fusion/downscaling with conditional uncertainty, not future AQ forecast: 30-d windows include 14 future days and CAMS/ERA5 analyses; controlled CTM reconstruction improves 3.2%, while station CV gain is small; no figure mirror under NC license

完整代码与数据状态

no public algorithm package, frozen radar-gauge evaluation sample, code license or data license verified

完整论文与来源状态

formal AMS Journal of Atmospheric and Oceanic Technology article DOI 10.1175/JTECH-D-25-0092.1, published record dated 2026-08-19 and Crossref-created in buffer; figure/article reuse rights unverified

下一触发器

accessible full text/supplement; network-average and distance/rain-rate/season/hail strata; code/reference implementation, frozen sample and licenses

上次判断

retain as upstream operational C-band radar QC, not AI or future nowcasting: radiometric emission, reflectivity and polarimetric constraints reduce author-reported hourly rainfall RMSE by up to 39%, but sample details and general average are not yet accessible; no figure mirror

完整代码与数据状态

no paper-matched code, frozen monthly-precipitation dataset or software/data license verified; National Tibetan Plateau Data Center source family is identified

完整论文与来源状态

formal Engineering Reports article DOI 10.1002/eng2.71024, online 2026-08-19; Crossref created in buffer; article CC BY 4.0

下一触发器

full text/supplement; initialization-valid-time contract, train/test years, baselines and numeric skill; block-boundary, spectra/extremes, code and frozen data

上次判断

retain only at metadata level: abstract supports one-fifth pooling, block-wise LSTM training and 1-km reassembly under limited hardware, but does not establish a next-month/1-65 d target, time-out split, recovered fine-scale information or numeric advantage; no figure mirror without exact caption/source check

完整代码与数据状态

no accessible full text, code, frozen aircraft-model matchups or data/software license verified

完整论文与来源状态

SSRN posted-content DOI 10.2139/ssrn.7321643; Crossref created 2026-08-20 17:37 UTC but posted date supplies only year 2026, so exact publication day and journal status are unverified

下一触发器

accessible full text and exact posted day; lead bins/common matchups, numeric version-by-version errors, artifacts/licenses and formal venue

上次判断

retain short-range evaluation candidate at metadata level: abstract supports 408 aircraft, 2017-2023, 3-km CMA-MESO V4.2-V5.1 T/RH/LWC error structure, but exact leads, units, magnitudes and reproducibility contract are unavailable; not AI

完整代码与数据状态

no accessible full text, code, frozen Jinan/Taian data split or software/data license verified

完整论文与来源状态

SSRN posted-content DOI 10.2139/ssrn.7291629; Crossref created 2026-08-20 22:24 UTC but posted date supplies only year 2026, with no journal information

下一触发器

accessible full text; PM2.5 units/years, explicit lead times and meteorological inputs; chronological OOD tests, code/license and external-city validation

上次判断

retain AQ candidate only at metadata level: abstract reports ARIMA+BiGRU+attention and city RMSE/MAE, but forecast horizon, driver availability, split and units are absent, so it cannot enter core-air-quality

完整代码与数据状态

no paper-matched code or frozen split verified; CPCB daily AQI source is identified; article is CC BY 4.0

完整论文与来源状态

formal Frontiers in Artificial Intelligence article DOI 10.3389/frai.2026.1892338, published 2026-08-20

下一触发器

reconsider only with interpolation performed strictly inside training folds, original-sample chronological evaluation, explicit future horizon, meteorological drivers and external-year/city tests

上次判断

exclude from core AQ monitoring: univariate daily AQI, undefined operational lead and fractal interpolation from 3,011 points to about one million before an insufficiently documented 80/20 split create leakage/independence risk; reported R2 about 0.9976-0.9998 is not business evidence

完整代码与数据状态

no code, data, abstract or license verified from accessible official metadata

完整论文与来源状态

journal DOI 10.22271/multi.2026.v8.i8a.1166; publication record dated 2026-08-01, while Crossref registration was created 2026-08-20

下一触发器

revisit only on substantive new artifact/status change with accessible methods and strict forecast contract

上次判断

exclude as delayed indexing outside the scientific window; title-only metadata cannot support model, lead, data or skill claims

完整代码与数据状态

no paper-matched code, weights, frozen multimodel-member manifest or software license verified; CMA precipitation, ERA5 and C3S source families are identified

完整论文与来源状态

arXiv 2608.19163v1, submitted 2026-08-19 17:43 UTC; arXiv preprint only, no journal information; CC BY-NC-SA 4.0

下一触发器

2026 JJA observation verification; code/weights; fixed system-version/member list; lead-wise 1-65 d tests, probability calibration, leave-one-system/year and independent reproduction

上次判断

retain as circulation-to-precipitation bridge and seasonal OOD method, not direct core-s2s: March-May to JJA mostly exceeds 65 d, 2026 forecast is unverified, hindcast ACC is modest and deterministic MME does not quantify uncertainty; do not mirror figures under NC-SA license

完整代码与数据状态

Zenodo software DOI 10.5281/zenodo.22000716 is CC BY 4.0; paper-linked GitHub returns 404; cBottle Hugging Face UI and model card expose conflicting Apache-2.0 versus NVIDIA evaluation-license statements

完整论文与来源状态

arXiv 2608.19008v1, submitted 2026-08-19 15:09 UTC; arXiv preprint only, no journal information; arXiv perpetual non-exclusive license

下一触发器

accessible GitHub/frozen cases; resolved checkpoint license; autoregressive training/evaluation; spectra, balances, spread-skill and independent reproduction

上次判断

retain as future-endpoint-conditioned extreme-storyline generation, not initialized forecast probability: 1,000 members expose antecedent ambiguity but systematic event biases, underdispersion and dynamical inconsistency limit inference; do not mirror figures without reusable paper license

完整代码与数据状态

IFS 49R2 experiment is not public; OpenIFS is not the same version; NEMO is traceable; authors state outputs will enter MARS before publication but they are not available now

完整论文与来源状态

arXiv 2608.18712v1, submitted 2026-08-19 09:13 UTC; arXiv preprint only, no journal information; CC BY-NC-ND 4.0

下一触发器

MARS outputs; SEAS6/version documentation; lead-wise anomaly skill, forcing trend, bias and uncertainty separation; formal venue

上次判断

retain as non-AI seven-month seasonal predictability/attribution evidence, not 1-65 d core-s2s: counterfactual forcing changes trends while similar detrended skill is retained, but tropical-Pacific gradient and aerosol effects remain deficient; do not mirror figures

完整代码与数据状态

paper-linked Zjut-MultimediaPlus/Tianmu-TC returns 404; TropiCycloneNet Zenodo 15009527 is CC BY 4.0; processed Google Drive materials and their license are unverified

完整论文与来源状态

arXiv 2608.18500v1, submitted 2026-08-19 03:46 UTC and recovered from buffer; arXiv preprint only, no journal information; CC BY 4.0

下一触发器

public licensed code/weights; clarify future-derived MGTCF variables and 24 versus 48 h history; realtime fields; CRPS/reliability/coverage and 48-72 h common-sample tests

上次判断

retain direct 6-72 h TC track/intensity generation with severe evidence caveat: author short-lead errors are strong but the input contract may contain forecast-time leakage, ensemble calibration is absent and ECMWF is much better by 48-72 h; Figure 2 mirrored under CC BY 4.0

完整代码与数据状态

no paper-matched code, reproducible WoFS experiment bundle, observation package or data/software license verified

完整论文与来源状态

formal Weather and Forecasting article DOI 10.1175/WAF-D-25-0210.1, published record dated 2026-08-18 and Crossref-created in buffer

下一触发器

accessible full methods/supplement; more cases; lead-wise common samples, significance, artifact release and equal-cost observing-network comparison

上次判断

retain as direct non-AI short-range information-preservation evidence: assimilated lower-ABL UAS profiles improve reflectivity/rotation and moisture-advection aspects in three cases, but evidence is not yet generalizable; no figure mirror because figure-level reuse rights were not verified

完整代码与数据状态

code is available only on request; FY-3G source data are public through NSMC but no frozen train/test package or software license is verified

完整论文与来源状态

EGUsphere preprint egusphere-2026-4804, discussion started 2026-08-18; not peer-reviewed final; CC BY 4.0

下一触发器

public code/frozen orbits; real discrete scans and instrument simulator; land/intermittent precipitation, cross-season/region, temporal extrapolation, spectral/object/uncertainty tests

上次判断

retain as same-time sparse-radar 3-D reflectivity reconstruction, not radar nowcasting: two held-out orbits support masked reconstruction but simulated masking, ocean-only continuous rain and no time evolution constrain transfer

完整代码与数据状态

no article code, model package or software license verified; public Beijing dataset identity is recoverable from related author material but article mapping is not frozen

完整论文与来源状态

Crossref DOI 10.1016/j.envsoft.2026.107140 created in buffer; print issue is October 2026, exact formal online date and full text were not verified

下一触发器

official VOR/full text; exact lead, chronological split and bidirectional-imputation audit; code/license; CRPS/reliability/coverage and external station/city tests

上次判断

retain AQ candidate at metadata level only: 1/6/12/24 h are input-history lengths rather than forecast leads, and mean plus standard deviation across four deterministic models is not calibrated uncertainty

完整代码与数据状态

no accessible full text, code, frozen ten-station sample or license verified

完整论文与来源状态

SSRN preprint DOI 10.2139/ssrn.7308276; Crossref metadata/abstract created in buffer; no journal information verified

下一触发器

accessible full text; identify GNN/model version; initialization and common samples; threshold metrics, artifacts and independent regional tests

上次判断

retain short-range precipitation candidate only at metadata level: abstract reports 1-3 d California station errors and LightGBM correction, but model identity, time split and extreme verification cannot be safely reconstructed from the abstract

完整代码与数据状态

no accessible full text, code, frozen ICE-POP sample or software/data license verified

完整论文与来源状态

SSRN preprint DOI 10.2139/ssrn.7309898; Crossref metadata/abstract created in buffer; no journal information verified

下一触发器

accessible full text; initialization/lead and split; per-type common samples; code/data and external winters

上次判断

retain as short-range NWP-profile correction and precipitation-type candidate, not precipitation amount/field skill: abstract reports improved hit rate and type scores but timing and reproducibility contracts are absent

完整代码与数据状态

v3 upgrades training/run code to Zenodo 20784771 and adds rollout initial-condition/forcing data 21952904 plus evaluation/CMOR pipeline 21913147; weights/configs remain promised for Hugging Face and per-archive licenses/version mapping need audit

完整论文与来源状态

arXiv 2605.29976v3, revised 2026-08-18 12:00 UTC under CC BY 4.0; buffer-window version state change, not a new paper/result

下一触发器

public weights/configs; archive-license and version audit; independent 45-year rollout and forced-scenario reproduction

上次判断

record substantive artifact-availability change only: v2-v3 text comparison shows no material scientific-result revision, so do not recycle the climate-simulation findings as new progress

完整代码与数据状态

no paper-matched public code, weights, processed samples, frozen station manifest or software/data license; FY-4A AGRI and gauge inputs are identified but not packaged for reproduction

完整论文与来源状态

arXiv 2608.17753v1, submitted 2026-08-18 13:17 UTC; arXiv preprint only, no journal information; arXiv perpetual non-exclusive license

下一触发器

formal venue; code/weights/data manifest/license; radar-satellite-gauge common samples; cross-year/region/typhoon tests; Brier/reliability/CRPS and field FSS/spectrum/object metrics; processing latency relative to the last input scan

上次判断

retain as direct 0-3 h satellite-to-station neighborhood heavy-rain event prediction, not radar echo extrapolation or satellite-to-radar: matched inputs and event verification support the event-oriented pathway, but endpoint-mismatched training, 10-40 km spatial tolerance and no probability-calibration evidence prevent claims of fine-field recovery or reliable uncertainty

完整代码与数据状态

figure-reproduction code and AIMIP data-access instructions are public; analysis repo has no standard software license; specific AIMIP outputs still require per-dataset license checks

完整论文与来源状态

arXiv 2608.17986v1, submitted 2026-08-18 16:32 UTC; arXiv preprint only, no journal information; CC BY-NC-SA 4.0

下一触发器

formal peer review; code DOI and license; 2015-2024 holdout diagnostics; per-model data licenses; larger forced ensembles, coupled/radiative forcing, extremes and weather/S2S common-initialization transfer

上次判断

retain as OOD physical-consistency diagnostic, not 1-65 d S2S skill: Köppen-zone migration reveals land-boundary propagation failures hidden by training-climatology agreement, while uniform SST forcing, unequal boundary handling, small ensembles and partial holdout overlap limit model ranking; do not mirror figures under NC license

完整代码与数据状态

v2 exposes no public paper-matched code or weights link; ERA5 is traceable, while IFS/MEPS analyses, station observations and derived training manifests are not packaged; no independently runnable experiment bundle verified

完整论文与来源状态

arXiv 2511.23043v2, revised 2026-08-15 20:07 UTC; arXiv preprint, no verified journal information; stable page now confirms 13%/10% average CRPS improvements for 2 m temperature/mean sea-level pressure versus MEPS

下一触发器

code/weights/license; initial-condition ensemble; precipitation extremes, object/spectrum/reliability diagnostics; exact common-sample reproduction and independent operational evaluation

上次判断

status metadata resolved: retain as direct 0-60 h high-resolution probabilistic regional evidence, but do not count stable v2 as a new 2026-08-19 item; deterministic-control initialization remains generally underdispersive and the finest scales can have excess energy

完整代码与数据状态

no public code, weights, raw CSI/rain-reference dataset or explicit data/software license verified

完整论文与来源状态

arXiv 2608.16088v1, submitted 2026-08-17 04:31 UTC; arXiv preprint, no journal information; arXiv perpetual non-exclusive license now verified

下一触发器

event-level real-rain split; co-located gauge/radar validation; cross-site/device/frequency tests; calibration, heavy-rain thresholds, latency and open artifacts

上次判断

metadata status resolved; retain as upstream rainfall-observation infrastructure, not radar/precipitation nowcasting: controlled Wi-Fi/LTE CSI results do not establish future prediction, radar comparison or field generalization

完整代码与数据状态

no forecast artifact state tracked because the item is a physical urban-climate review

完整论文与来源状态

arXiv 2608.17782v1, submitted 2026-08-18 13:45 UTC; author comment says submitted to Reviews of Geophysics, not accepted or published

下一触发器

reconsider only if a separate AI forecast study supplies initialization, lead-time and out-of-sample weather/AQ verification

上次判断

exclude from monitored advances: forcing-response-feedback review has no AI prediction task, forecast lead, radar/S2S/AQ endpoint or operational validation

完整代码与数据状态

no atmospheric forecast artifact state tracked

完整论文与来源状态

arXiv 2608.17971v1, submitted 2026-08-18 16:23 UTC; arXiv preprint only

下一触发器

reconsider only if linked work evaluates future weather/S2S inputs or atmospheric forecast skill rather than crop yield

上次判断

exclude: county corn yield is an impact forecast with meteorological predictors, not weather, radar, S2S or air-quality prediction

完整代码与数据状态

no atmospheric application artifact verified

完整论文与来源状态

arXiv 2608.18004v1, submitted 2026-08-18 16:53 UTC; arXiv preprint only

下一触发器

atmospheric/radar/precipitation/AQ experiment with task-specific baselines, calibration, physics and structure diagnostics

上次判断

exclude from retained methods for now: general PDE-field flow-matching energy and inversion results lack validation on any monitored atmospheric task

完整代码与数据状态

no verified public code, weights, synthetic 0.1-degree data manifest, generation pipeline or artifact license on the accessible abstract record

完整论文与来源状态

arXiv 2608.14652v1, submitted 2026-07-31 03:43 UTC; the 2026-08-18 list exposure is delayed and not a fresh submission; page author comment says accepted by ECCV 2026, but no official proceedings record was verified

下一触发器

official ECCV record; code/weights/data pipeline and license; temporal leakage audit; real high-resolution observations; extremes, spectrum, conservation, probabilistic and independent tests

上次判断

retain as window-out short-range method evidence only: variable-wise super-resolution expands 0.1-degree training data and author abstract reports gains through 120 h, but synthetic-detail bias and evaluation contracts are unresolved; do not count as a 24-48 h event

完整代码与数据状态

no code/data state promoted because the task is outside the monitored forecast lines

完整论文与来源状态

arXiv 2608.14935 new submission; no verified journal information

下一触发器

reconsider only if coupled to an actual NWP forecast experiment with out-of-sample weather skill, uncertainty and operational comparison

上次判断

exclude from daily retained items: neural-network/random-forest surface flux fitting is upstream boundary-layer parameterization, with reported cross-site degradation, but it contains no forecast lead, radar task, S2S evidence or air-quality prediction evaluation

完整代码与数据状态

paper states tcpearce/CAIRN releases the dual-pathway S4D model, 13 weekly checkpoints per species, fold scalers, feature/label code, acceptance tests and a 2023-09-01 to 2024-12-31 principal-site Parquet file; repository could not be independently opened in this run, so repository license/downloadability remain unverified; transfer-entropy, XGBoost and figure scripts are promised on acceptance, and complaint data are restricted

完整论文与来源状态

arXiv 2608.14254v1, submitted 2026-08-14 12:32 UTC under CC BY 4.0; arXiv preprint only, no journal information; submission precedes the current 48 h boundary although it appeared on the 2026-08-17 new-list

下一触发器

independently accessible repository and software license; Zenodo DOI and deferred scripts; formal venue; external source/site tests; future-weather inputs or positive-lead exposure forecasts

上次判断

retain as window-out adjacent-method evidence connected to AQ: measured 1/6 h meteorological timescales anchor a causal S4D nowcaster, but the output estimates present H2S/CH4 exposure rather than future pollution, weather or radar; do not count the new-list appearance as a fresh submission

完整代码与数据状态

paper links JdHondt/MONARCH-downscaling, but the repository currently returns 404; paired MONARCH 10/5 km outputs and ancillaries are promised upon acceptance rather than public now; GHOST observations are public through cited providers

完整论文与来源状态

Research Square v1 DOI 10.21203/rs.3.rs-10627504/v1, posted 2026-08-14 under CC BY 4.0 and not peer reviewed; Figure 1 mirrored with complete caption/version/license metadata

下一触发器

stable code archive and paired dataset; unseen-year and external-CTM/region tests; calibration and joint meteorological-chemical uncertainty; scale factors beyond 2x

上次判断

retain as adjacent diagnostic AQ downscaling evidence, not future AQ forecasting: independent CTM resolutions require probabilistic structural correction before constrained refinement; 2023 observations are unseen by training but fall within the CTM training years

完整代码与数据状态

no public code, weights, frozen CPCB mapping or experiment package verified from accessible records

完整论文与来源状态

formally online 2026-08-14 in Discover Artificial Intelligence, DOI 10.1007/s44163-026-01892-y, CC BY 4.0; abstract reports six cities, 2018-2023 hourly data and TimeGAN plus SMOTE

下一触发器

accessible full text; forecast lead and city/time splits; fold-contained SMOTE/TimeGAN audit; reconcile MAE 3.72 with MSE 4.25; future meteorology, code/data and independent reproduction

上次判断

retain cautiously as direct AQI forecasting evidence recovered from the buffer; author metrics are not accepted as operational skill because lead/input availability and leakage controls are unverified and the abstract metrics may be internally inconsistent

完整代码与数据状态

MIT repository exposes model/training/evaluation code, tests, result CSVs and a final checkpoint link with author-supplied SHA-256; original ERA5 and baseline checkpoints are not redistributed; no local end-to-end reproduction

完整论文与来源状态

public research-code package created 2026-08-15 UTC; no public paper, DOI or formal venue verified

下一触发器

public paper/venue; frozen ERA5 conversion and baseline artifacts; true forecast inputs; probabilistic calibration, FSS/spectrum/object/extreme uncertainty and independent reproduction

上次判断

retain as deterministic one-hour ERA5 long-tail precipitation method, not radar nowcasting or operational short-range skill; small high-threshold CSI gain and low-threshold trade-off motivate tail-aware evaluation, while MoE does not itself express uncertainty

完整代码与数据状态

MIT processing pipeline, synthetic 15-frame/36-station demo, schemas and expected outputs are public; restricted Bangkok BMA observations and event windows are not distributed

完整论文与来源状态

new code for EGUsphere preprint 10.5194/egusphere-2026-1314; preprint remains in public discussion and is not a peer-reviewed final

下一触发器

real or substitutable open event data; station-density/interpolation and threshold sensitivity; radar comparison, author-review resolution and independent execution

上次判断

record substantive code-availability change only: it closes the prior code gap but not data completeness or the physical attribution of optical-flow residuals; not a forecast or AI result

完整代码与数据状态

M1 ingestion and M2 preprocessing cover calibration, 0.04-degree 875 x 875 regridding, masks, gap handling and normalisation manifests; author reports 349 tests; real MOSDAC inputs are not redistributed and no standard LICENSE exists

完整论文与来源状态

hackathon development framework with no paper, forecast model, weights or accuracy claim; M3-M15 remain planned

下一触发器

standard license; real-data temporal contract; labels and model modules; frozen events, lead-wise deterministic/probabilistic verification and independent run

上次判断

retain as satellite-input infrastructure, not satellite-to-radar or completed precipitation nowcasting; explicit absence of a model prevents any skill inference

完整代码与数据状态

no verified code, data, fixed rolling-origin splits or experiment package

完整论文与来源状态

SSRN posted content DOI 10.2139/ssrn.7267818; Crossref record created 2026-08-15 UTC; accessible abstract only, with no journal information verified

下一触发器

accessible full text; initialization, 1-65 d lead and target accumulation; forecast system/data years, common samples, probabilistic baseline, code and data

上次判断

retain as an S2S candidate at metadata level: multiscale causal screening and season-specific reliability are relevant, but the required initialization-valid-time contract is absent

完整代码与数据状态

README-only shell describes seven Delhi sites, satellite/reanalysis/history inputs and XGBoost/LightGBM; no executable code, data, weights, results or standard LICENSE

完整论文与来源状态

no paper, DOI or scientific release verified

下一触发器

implementation, license, frozen data/input availability, chronological 24 h evaluation, absolute metrics and artifacts

上次判断

retain metadata only; a proposed architecture and task description do not establish AQ forecast skill or reproducibility

完整代码与数据状态

no code, weights, frozen data manifest or reproducible experiment package verified

完整论文与来源状态

formally published 2026-08-13 in Advanced Electromagnetics, DOI 10.7716/aem.v15i3.4025, CC BY 4.0; Crossref indexing was delayed into the current window

下一触发器

exact years/sample counts/units; lead-wise common-sample results; model and bias-correction details; code/data and independent regional verification

上次判断

retain as window-out 1-72 h multi-model fusion evidence only; reported aggregate ACC lacks an auditable sample and lead contract, so do not use it as a quantitative short-range baseline

完整代码与数据状态

repository includes scripts/results but no standard LICENSE; data fetches Open-Meteo grids rather than claimed IoT observations; full-series VMD and full-data standardization precede the split, and the main script references an undefined dataset

完整论文与来源状态

no verifiable paper or venue; README metrics conflict with result CSVs and placeholder citation text remains

下一触发器

corrected data provenance; training-only preprocessing, runnable package, baseline/common-sample evaluation, license and independent reproduction

上次判断

exclude from core AQ evidence because provenance, leakage, executability and metric consistency failures invalidate the current forecasting claims

完整代码与数据状态

MIT minimal repository Semllo/imerg_correctability includes a 3 h smoke benchmark with 345,742 rows, three representative cells and M0-M3b; full-domain 1/6 h experiments, CNN M4 and raw AVAMET observations are absent, while IMERG V07 remains public

完整论文与来源状态

formally online 2026-08-14 in Stochastic Environmental Research and Risk Assessment, DOI 10.1007/s00477-026-03336-6; Crossref exposes TDM terms rather than verified figure-reuse rights; methods/results were cross-checked against the March Research Square preprint but VOR differences remain unverified

下一触发器

accessible VOR and version reconciliation; full 1/3/6 h package and M4 reproduction; explicit VOR figure license; future-forecast inputs, independent reference and external-region tests

上次判断

retain as adjacent satellite-QPE correction and uncertainty evidence, not precipitation forecasting: event-block evaluation shows strong continuous-RMSE gains but weak fixed-extreme recovery, motivating direct probability ranking and input-information-limit diagnostics

完整代码与数据状态

MIT GitHub/Zenodo package exposes training, ablation, SHAP, Moran's I, transfer/retraining, curated tables/figures and the timestamped pre-registration record; raw IMERG/ERA5/CN05.1/SRTM inputs and trained model binaries are not redistributed

完整论文与来源状态

arXiv 2608.12988v1, submitted 2026-08-13 09:11 UTC; arXiv preprint only, no journal information; paper license CC BY-NC-ND 4.0, software MIT

下一触发器

formal venue and independent reproduction; cold-season and external-climate validation; finer time/space scales, event/regime labels, real forecast meteorology and calibrated extremes

上次判断

retain as adjacent cross-region satellite-precipitation correction evidence: transfer-versus-retraining and locked coherence proxies are useful applicability tests, but 0.25-degree daily South-China results and Moran's I do not establish unique physical mechanisms or future forecast skill; do not mirror figures

完整代码与数据状态

MIT GitHub CosuleaBianca/eea-pm25 releases the pipeline/config/masks; HF dataset eea-pm25-forecasting and selected model package eea-pm25 are public and tagged CC BY 4.0; README contains alternate HF names, so paper links and current SHAs must be frozen; complete rerun is author-estimated at 6 days/30 GB

完整论文与来源状态

formal CC BY 4.0 article DOI 10.1007/s00477-026-03331-x, online 2026-08-14; full text now accessible: 38 stations/five European countries, 2018-2022 train and 2023-2024 test, direct 1/3/6/12/24 h; matched Protocol B best RMSE 2.18/3.48/4.42/5.37/6.05 with DL leading 1-3 h, tie at 6 h and boosted trees leading 12-24 h; Figure 4 legally mirrored with metadata

下一触发器

versioned release/DOI and HF-name reconciliation; units/mask rerun; forecast-meteorology versus perfect/historical drivers; high-pollution/threshold stratification; calibrated intervals/CRPS; strict cold-start and denser spatial transport tests

上次判断

upgrade from metadata-only to core AQ benchmark: dual coverage/common-mask protocols and horizon-resolved driver shift are useful, but deterministic point scores, no future NWP/MLWP meteorology, 1000-sample DM tests without multiplicity correction and historically calibrated LSTM transfer bound operational claims

完整代码与数据状态

no paper code, weights, frozen training-date manifest or releasable GRAF/MRMS package verified

完整论文与来源状态

arXiv 2608.12685v1, submitted 2026-08-13 01:03 UTC under CC BY 4.0; submission precedes the current 48 h boundary although it appeared on the 2026-08-14 arXiv new-list

下一触发器

public artifacts; independent calibration/test periods; spatially coherent ensemble or joint-field scores; MRMS-quality sensitivity, extreme reliability, spectrum/FSS/object tests and operational reproduction

上次判断

retain as window-out delayed-index direct short-range probability evidence: terrain-conditioned zero-inflated two-Gamma output improves author BSS most over rough terrain, but same-year baseline tuning, marginal-only distributions and unavailable artifacts limit inference; Figure 3 mirrored under CC BY 4.0

完整代码与数据状态

no verified forecast dataset, meteorological-driver contract, code/weights package or scientific result artifact

完整论文与来源状态

formally online 2026-08-14 in SN Computer Science, DOI 10.1007/s42979-026-05273-1; accessible metadata supports sparse-network monitoring, not a future AQ forecast task

下一触发器

explicit forecast target/lead and forecast-time inputs; chronological evaluation, meteorological coupling and reusable artifacts appear

上次判断

exclude from the AQ forecast update until a future-prediction contract is demonstrated; monitoring or contemporaneous sparse-network reconstruction is not enough

完整代码与数据状态

GHCN-hourly, ERA5, SRTM/Mapzen, TESSERA v2 and Aurora dependencies are identified; no paper-matched code, weights, frozen station/date manifest or software license verified

完整论文与来源状态

arXiv 2608.12271v1, submitted 2026-08-12 17:10 UTC; arXiv preprint only, no journal information; specific version CC BY 4.0

下一触发器

public code/weights and exact manifest; operational analysis initialization; precipitation/extremes, joint spatial uncertainty, spectrum/gradient and independent evaluation

上次判断

retain as direct probabilistic station-downscaling evidence: simultaneous station/year holdout and Aurora +6/+24/+72 h tests support transferable local correction, but ERA5-initialized Aurora, marginal-only outputs and two near-surface variables do not establish operational high-resolution weather-field skill; Figure 6 mirrored under CC BY 4.0

完整代码与数据状态

source points to Air Korea/Kaggle daily-aggregated station data; scripts only available on request, with no public weights or frozen rolling-fold package

完整论文与来源状态

formally published 2026-08-13 in Scientific Reports, DOI 10.1038/s41598-026-64590-w, CC BY-NC-ND 4.0

下一触发器

public rolling-origin code/data mapping; reconcile abstract/text versus Table 7 best selector; real CAMS/MLWP/NWP drivers, probabilistic/extreme and multi-station validation

上次判断

retain direct one-step PM2.5 residual-forecast evidence for fold-contained preprocessing; best Table 7 RMSE 4.016 is author-reported and does not establish operational AQ forecasting without future meteorology; do not mirror figures under CC BY-NC-ND

完整代码与数据状态

no paper-specific code, initialization manifest, reforecast package or processed data verified

完整论文与来源状态

formally online 2026-08-12 in Climate Dynamics, DOI 10.1007/s00382-026-08320-y, CC BY-NC-ND 4.0; accessible author materials report atmospheric/land initialization sensitivity

下一触发器

full methods, resolution, initializations, common samples, absolute lead-wise scores, uncertainty and artifact release

上次判断

retain as direct dynamical S2S predictability-source evidence, not an AI advance: reported 35 d MJO and 40 d global initialization optima and soil-moisture sensitivity motivate coupled initial-state audits, but accessible evidence is insufficient for cross-system numerical comparison; do not mirror figures

完整代码与数据状态

supporting data DOI 10.5281/zenodo.20667784 is provided; model developments and evaluation sources are documented, but no local reproduction was performed

完整论文与来源状态

EGUsphere preprint DOI 10.5194/egusphere-2026-4786, discussion started 2026-08-13 under CC BY 4.0; not peer-reviewed final

下一触发器

code/data version mapping; fixed-initialization attribution of meteorology, chemistry cycling and feedbacks; operational and independent verification

上次判断

retain as non-AI AQ forecast infrastructure: daily GFSv16-initialized 2020 wildfire experiments and AirNow/AERONET/ISD evaluation expose two-way physics-chemistry error pathways, but do not count as an AI result

完整代码与数据状态

existing CC BY 4.0 community ports; Sep 8 15:32-15:36 UTC commits rename FiLM/edge-bias/MLP/head keys; cards and open Transformers PR47874 verified, full weight diff and rollout equivalence not verified

完整论文与来源状态

no new scientific publication, training result or WeatherNext3 release; underlying artifact remains an existing infrastructure resource

下一触发器

released upstream support plus full conversion and rollout equivalence; frozen environment and matched benchmarks

上次判断

record compatibility metadata only; do not refresh the old model performance or claim newly trained weights

完整代码与数据状态

OpenAQ raw PM2.5 is accessible; code and retrieval scripts only available on request, no verified live service or public model artifact

完整论文与来源状态

Zenodo preprint DOI 10.5281/zenodo.21894664, created 2026-08-11 21:22 UTC under CC BY 4.0; recovered from 24-48 h buffer

下一触发器

reconcile three-year claim with 2023-09 to 2025-03 data and creator/PDF authorship; public code; real forecast meteorology; event reliability and independent deployment audit

上次判断

retain cautiously as direct high-AQI event evidence: chronological split and hit/false-alarm reporting are useful, but analysis/history-only inputs, metadata inconsistencies and low six-hour onset detection do not establish a mature warning service

完整代码与数据状态

four multi-station datasets are described; processed data, code, weights and reproducible split package are not public

完整论文与来源状态

arXiv 2608.11839v1, submitted 2026-08-12 09:26 UTC; arXiv preprint only, no journal information; recovered from 24-48 h buffer

下一触发器

chronological split and leakage audit; public artifacts; partial-network outages, multi-pollutant/static features and independent tests

上次判断

retain as adjacent station-imputation evidence, not AQ forecasting: SATADL uses contemporaneously online neighbors to reconstruct an offline PM10 station and therefore cannot be cited as future, meteorology-driven prediction

完整代码与数据状态

NeuralGCM/public observations/ERA5 dependencies are identifiable; no paper-specific code, frozen 360-cycle package or software release verified

完整论文与来源状态

arXiv 2608.11515v1, submitted 2026-08-12 00:06 UTC under CC BY-SA 4.0; reanalysis/data-assimilation experiment recovered from 24-48 h buffer

下一触发器

public experiment package; background-term and moist-gradient sensitivity; real observations and forecast-from-analysis evaluation

上次判断

retain as information-preservation infrastructure: four-day windows improve author Z500 reanalysis scores over 20CRv3, while degradation beyond four days, no background term and zero moist-variable gradients limit forecast inference

完整代码与数据状态

initial repository contains 3-D reflectivity NetCDF loading, 128 x 128 centroid crops, a validity channel, SimVP-style code/tests and 13-input/12-output configuration; no license, weights, released data or result package

完整论文与来源状态

no paper, DOI or release; repository created 2026-08-12 and recovered from 24-48 h buffer

下一触发器

explicit timestep and forecast contract; license, paper/data/weights; CSI/FSS/object, vertical and lightning verification on frozen events

上次判断

retain as early 3-D radar pipeline infrastructure, not nowcasting skill; a roughly one-hour output is filename-based inference only and masked-MSE/MAE/RMSE tests do not establish convective lifecycle fidelity

完整代码与数据状态

no accessible code, weights, dataset contract or artifact package verified

完整论文与来源状态

Crossref metadata for DOI 10.1016/j.rsase.2026.102203 was created in-window, but exact online publication event, abstract/full text and version-specific reuse rights were not verified

下一触发器