研究条目库
同一条目可以多次出现，表示它从预印本、代码提交、检查点到正式发表的状态变化。这里不把历史事件误写成新的研究成果。
98 条状态记录
7 类证据分桶
持续更新 下一触发器
direct 直接预测未来雷达/降水场
near-direct 输入或目标接近，但任务不完全相同
related 可迁移的方法、质控、检索或评测证据
最新状态记录#
| item | last_seen | bucket | code_status | paper_status | next_trigger | last_decision |
|---|---|---|---|---|---|---|
Nowcast3D: Reliable precipitation nowcasting via gray-box learning |
2026-07-26 | direct / fully 3D radar reflectivity nowcasting code-state upgrade | GitHub commits through 8eedb5ba9f add 512x512 PhyPredNet, diffusion structure/intensity modules, fill/demo pipeline and nationwide 3D radar BIN-to-NPY conversion; public Drive has 512/1024 sample ZIPs and three checkpoints per variant; 11 Python files pass syntax compile; no full training entry, metric rerun package, release/tag or license; best operational checkpoints remain confidential |
arXiv 2511.04659v4, updated 2026-01-25; no journal/conference information verified; 2024 South-China train, 2025 North-China cross-region test, 2024/2025 Maoming fine-tune/test; up to 3 h; author paper reports neighborhood CSI/PSD/LPIPS/CRPS and 160-meteorologist blind evaluation |
checkpoint checksums, portable inference run, exact public-vs-best checkpoint mapping, training scripts, split manifest, metric reproduction, license, formal venue or independently verified deployment page appears | promote as high-value partial inference release, not complete reproducibility; public samples are not full training data and README deployment claims are not independently verified |
Learning to Refine: Spectral-Decoupled Iterative Refinement Framework for Precipitation Nowcasting / SDIR |
2026-07-26 | direct / deterministic spectral-refinement radar precipitation nowcasting; checkpoint release | commit 5054823279 on 2026-07-25 13:56 UTC adds Drive link; actual folder contains cikm.pth, sevir.pth, shanghai.pth around 132 MB each; training/inference/metrics code present and syntax-compiles; README checkpoint names inconsistent and says MIT but repository has no standalone LICENSE file |
formal ICML 2026 paper, PMLR 306; CIKM, Shanghai and SEVIR; SFG-Former + Fourier residual refiner + PCPSD; author tables report best spatial metrics across listed baselines but DiffCast has lower spectral slope error | checkpoint hashes, exact preprocessing/split reconstruction, independent high-threshold/PSD/latency reproduction, filename/docs fix, license file, or cross-radar validation appears | promote as the strongest paper-code-weight upgrade in this run; treat author results as unreplicated and test spectral sharpness together with localization skill |
Stable Attention Response for Reliable Precipitation Nowcasting / HARECast |
2026-07-26 | direct / radar and radar-satellite diffusion nowcasting code and checkpoint release | official GitHub code uploaded 2026-07-25 through b50a4052d9; datasets, PhyDNet/SimVP backbones, HARECast, run/eval and metrics present; Drive link resolves to ckpt-harecast.pt; 13 Python files pass syntax compile; hard-coded local SEVIR path, README path typo, no license/release/tag |
arXiv 2605.13181 preprint, no journal information; repository claims ACM Multimedia 2026 but no official ACM programme/paper record independently verified; SEVIR/MeteoNet unimodal 5->20 and SEVIR radar-satellite 12->36; author SEVIR ablation CSI-M/HSS 0.344/0.437 vs 0.315/0.408 without HARE |
official venue record, portable data config, license, checkpoint size/hash, end-to-end evaluation reproduction, batch-size/multigpu stability, or additional weights appears | promote fresh code/weight release but retain preprint-only status; HARE gains are a few points and batch-statistics sensitivity needs independent testing |
holys519/solafune_precipitation_nowcasting exp056 causal audit and OOF-LB transfer analysis |
2026-07-26 | 48 h buffer / related satellite-to-current-GPM evaluation governance; non-radar and non-future nowcasting | commits 835ba4d74d to e3ed8daceb add exp059/060/062 and seed configs, OOF-to-LB meta-analysis, submission registry, 40/40 causal replay compliance and static config audit; new experiments have no results yet; code public but raw data/final weights/submissions/license not verified |
no paper/DOI; exp056 public RMSE 0.6839628 beats exp038_sigmafixed 0.6866381; small paired sample suggests architecture/simplification changes transfer better than feature/loss tweaks | final exp059/060/062 OOF and public LB, final leaderboard, geography-held-out validation, weights/data/license, formal writeup, radar-target extension, or independent reproduction appears | retain as leakage/validation-governance evidence; do not generalize the small meta-analysis or satellite-to-same-time GPM score to radar future nowcasting |
photoli93/precipitation_nowcasting_from_space |
2026-07-26 | low-value exclusion / framework-only satellite precipitation repository | created in main window; only environment.yml, .gitignore, and zero-byte config/notebook files; no README, data, model, training, evaluation or license |
no paper/DOI | non-empty data pipeline, model, dataset, results, paper, or license appears | exclude until real scientific payload is uploaded |
A Bio-Inspired Weather-System Sensing Framework for Physically Constrained Precipitation Nowcasting Correction / PCPNet |
2026-07-25 | related method / physically constrained NWP precipitation correction; no radar input and no future echo nowcast | no official code or weights found; operational NWP/station data unavailable publicly and request requires Hunan Meteorological Bureau permission | MDPI Biomimetics 11(8):526, DOI 10.3390/biomimetics11080526, formally published 2026-07-24; Hunan 2021-2023, 98 stations, NWP + terrain + synoptic diagnostics; test MAE 0.148 vs raw NWP 0.190, but MSE 0.360 vs 0.366 and heavy-rain samples only 334 |
official code, data-access package, station/region-held-out validation, multi-event heavy-rain results, or radar-conditioned extension appears | retain as physical-diagnostic/constraint reference; do not call radar nowcasting, and do not generalize MAE improvement to all metrics or spatial generalization |
NOAA-GSL/HRRRCast-live |
2026-07-25 | institution / related regional convection-allowing AI-NWP; not radar-observation nowcasting | commit b1d9b415e64a on 2026-07-24 10:13 UTC removes unused SkipScaleLayer and shortens Slurm forecast/PMM walltimes; no new model, checkpoint, dataset, metric, or scientific result; MIT license; latest release remains v1 from 2026-04-24 |
no paper/status change in this run; README describes HRRR/GFS-conditioned deterministic/diffusion regional forecasts including REFC/APCP | new release, checkpoint, evaluation table, radar-conditioned configuration, HWT/UIFCW evidence, or reproducible convective-scale comparison appears | log as minor operational maintenance only; do not promote as a model or radar-nowcasting advance |
Clustering Pola Intensitas Curah Hujan Berbasis Data Radar NOAA NEXRAD Level II Menggunakan Algoritma K-Means |
2026-07-25 | window-out / index-delay related radar regime clustering; non-nowcasting | no code, exact 30-day sample manifest, or reproducible feature package verified; source NEXRAD Level II is public in principle | Teknik: Jurnal Ilmu Teknik dan Informatika DOI 10.51903/teknik.v6i1.1135; Crossref created 2026-07-25 but official page says published 2026-05-02; KTLX 2023, seven sweep-level features, K=5, silhouette 0.3871, DBI 0.8599 |
code/sample manifest, forecast extension, independent-event validation, or formal source-status change appears | retain only as radar sampling/regime-clustering reference; do not count as today's publication, flood forecast, or precipitation nowcast |
Assessment of Nowcasting Precipitation Schemes Initialized from LAPS Analysis Fields over the Attica Region |
2026-07-24 | related deterministic nowcasting method / complex-terrain LAPS; non-AI | no dedicated code, LAPS configuration, event dataset, or reproducible package verified; MDPI full text/XML returned 403 | MDPI Atmosphere DOI 10.3390/atmos17080714, published 2026-07-23; three Attica high-impact events; Control advection vs AD vs ADLOP; radar verification; abstract reports up to ~33% bias reduction at longer leads but regime-dependent error |
accessible full text, exact events/leads/absolute scores, code/config/data, multi-season validation, or AI/radar-sequence adaptation appears | retain as interpretable complex-terrain baseline and diagnostic evidence; do not label a deep-learning advance or generalize three cases |
Radar Reflectivity Assimilation Method Based on Adaptive Hydrometeor Phase Identification Algorithm Using Dual-Polarization Radar and Its Impacts on Severe Weather Prediction |
2026-07-24 | related radar assimilation / hydrometeor-phase QC; non-AI and non-radar-sequence nowcasting | no official code, WRFDA configuration, HCA parameters, radar case data package, or executable environment verified; AGU full text/XML returned 403 | AGU JGR: Atmospheres DOI 10.1029/2025JD045344, published 2026-07-23; dual-pol HCA-constrained blended retrieval in WRFDA, four convective cases; 3-h precipitation and surface-wind verification |
full text, data/code statement, WRFDA/HCA configs, radar site/case details, more independent cases, or AI initial-condition integration appears | retain as input/initial-state physics reference; never convert NWP precipitation/wind gains into direct future radar-echo skill |
parkermichaeljee3750/accessible-rain-radar-2026 |
2026-07-24 | low-value exclusion / accessible radar-web wrapper; no own AI nowcast | GPL-3.0 single-page HTML app referencing DWD, Bright Sky, Open-Meteo, and Photon; latest main-window commits only append one line to .github/update-log |
no paper, model training, scientific validation, radar extrapolation implementation, weights, or dataset | own forecast-model code, reproducible radar verification, research paper, or substantive data/model release appears | exclude recurring automated log refreshes from radar-AI progress |
| Precipiteau July 2026 gauge validation | 2026-07-24 | metadata-only / date-uncertain operational radar-satellite platform; non-AI | public page describes OPERA radar, H-SAF/JAXA blend and a 1-h PySTEPS optical-flow nowcast; no exact-dated code/model release verified | undated July 2026 page claims validation against 5,274 gauges and 43M matched observations from Mar-Jul 2026; no exact publication day, formal paper, event split, uncertainty table, or DOI |
exact publication timestamp, methods/report, code/data, station-independent validation, formal paper, or learned model appears | do not count by crawl time; retain as optical-flow operational validation lead only |
Convolutional Neural Network post-processing to generate spatially correlated ensemble precipitation forecasts |
2026-07-23 | related method / spatially correlated NWP precipitation ensemble post-processing; not radar echo nowcasting | no official code, weights, or dedicated data package verified; AMS full-text page returned 403 in this run | AMS Monthly Weather Review DOI 10.1175/MWR-D-25-0219.1; Crossref created 2026-07-22 15:55 UTC; Brisbane Drainage Basin, field-wide CNN calibration + spatial extraction + Monte Carlo dropout |
full text, code/data, exact years/split/leads, absolute CRPS/Brier/reliability/extreme-rain metrics, or radar-nowcasting adaptation appears | promote as fresh related probability-calibration method; do not treat NWP post-processing or MC-dropout spread as direct radar-nowcasting skill or automatically calibrated uncertainty |
abc-radar/NJU-CPOL-Nowcasting-Data |
2026-07-23 | window-out edge / direct-topic dual-polarization radar data mirror; not verified as a new or official dataset release | Hugging Face repo is public/ungated and contains ~279 MB kdpRain plus ~5.07 GB update2308 ZIPs; no README, license, checksums, provenance, temporal coverage, QC, or split metadata |
no companion paper or official release relationship stated; repo created 2026-07-21 02:07 UTC and last modified 03:55 UTC, slightly outside strict 48 h at report time | uploader/original-source identity, license, checksums, version mapping, time axis, complete variables/heights, event-isolated split, or official paper/repository link appears | retain as access/provenance lead only; files appear nowcasting-relevant, but do not call them a new scientific dataset or use in publishable experiments before authorization and lineage are verified |
A Deep Learning Approach for Retrieving Cloud Microphysical Parameters with Multi-Frequency Cloud Radar / MiRA-Net |
2026-07-23 | window-out edge / related cloud-radar retrieval; non-nowcasting | no official code verified; some flight-level in-situ and WCR observations have public DOIs, while Ka-band KPR data are available on reasonable request | Advances in Atmospheric Sciences DOI 10.1007/s00376-026-5772-7, online 2026-07-21; simulated X/Ka/W-band inputs, residual attention network + transfer learning; two independent field cases |
official code/configs, full simulated-data recipe, more independent cases, cross-radar/frequency generalization, or forecast/nowcasting extension appears | keep as simulated-pretraining and physical-variable representation reference; never convert retrieval RMSE/correlation into future radar-echo forecast evidence |
ankl00/Sadetutka |
2026-07-23 | low-value exclusion / rain-radar PWA wrapper; no own extrapolation or AI model | public MIT-licensed frontend/backend with tests; README says it consumes MET Norway Nowcast 2.0 optical-flow series and visual radar sources, with simple rain start/stop/gap logic | no paper, training, forecast-model implementation, scientific validation table, or new radar dataset | own learned/extrapolation model, reproducible forecast evaluation, research paper, or new dataset appears | exclude from AI radar-nowcasting progress; application integration and heuristic messaging are not algorithm/model evidence |
Multimodal Deep Learning in Precipitation Nowcasting: A Review from a Geometric Evolution |
2026-07-22 | direct-topic review / multimodal precipitation nowcasting synthesis; no new model | no dedicated code, weights, literature table, or new data package verified; publisher full text blocked in this run | IETE Technical Review DOI 10.1080/02564602.2026.2704967, online 2026-07-21, pages 1-16; Crossref references indicate radar/station/lightning/satellite/NWP fusion, transformer/physics/foundation-language/diffusion coverage, but Crossref has no abstract |
accessible full text, structured dataset/model table, reproducibility checklist, official companion repository, or corrected/expanded version appears | promote as fresh formal review and navigation source; do not treat it as a new forecasting model, unified benchmark, or performance proof |
Severe convective weather in Italy: Current understanding and Research Priorities in the TIM campaign |
2026-07-22 | related observation/research-priority paper / strong-convection monitoring; not direct nowcasting | open-access article; no companion AI code, weights, or newly released TIM dataset verified | Natural Hazards and Earth System Sciences 26:3395-3415, DOI 10.5194/nhess-26-3395-2026, formally published 2026-07-21; final version of EGUsphere 2026-1478 |
TIM rapid-scan/multi-radar dataset, interoperability specification, AI-nowcasting experiment, hydrological coupling benchmark, or campaign data DOI appears | promote formal journal state as related context; do not claim AI model construction, training, or skill improvement |
holys519/solafune_precipitation_nowcasting exp056 / OOF-public-LB audit |
2026-07-25 | related satellite-to-current-GPM engineering and evaluation governance; non-radar and non-future nowcasting | commit d475bd1d4e updates public-score diagnostics; code public, raw data/final weights/predictions/license not verified |
no paper/DOI; exp050_sigmafixed OOF 0.60643 but public LB 0.6870176973, essentially no transfer vs champion 0.6866381029; exp047_sigmafixed OOF 0.60747 but LB degrades to 0.7001420621 and is withdrawn; likely hemisphere x satellite geographic shortcut after math check found no bug |
full 5-fold/final leak-audited leaderboard, geography-held-out validation, weights/data/license, formal writeup, radar-target extension, or independent reproduction appears | promote as OOF-to-public-domain failure evidence, not a new best score; never transfer same-time satellite-to-GPM results to radar echo nowcasting |
EffiRainNet: A Parameter-Efficient Spatiotemporal Network for Precipitation Nowcasting |
2026-07-22 | metadata-only / index-delay / window-out conference paper | no official code, weights, data package, abstract, or accessible full text verified | IEEE CRSIT 2026 DOI 10.1109/CRSIT69981.2026.11605383, Nanjing, pages 197-203; Crossref metadata created 2026-07-20 UTC but OpenAlex/venue metadata give actual publication 2026-06-05 |
full text/abstract, dataset, architecture, parameter/FLOP table, split, metrics, code, or weights appears | do not count as current publication; title-level parameter-efficiency claim is insufficient for technical or performance judgment |
dlimtx/lightning-risk-nowcast |
2026-07-22 | low-value exclusion / rule-based current lightning-risk proxy; non-AI and non-future-field nowcasting | public Python/Flask code fetches MSS Singapore radar image, applies dBR>=3.5 threshold, connected components and 6-km buffers, and compares township labels with 2 h public forecast; no license/release/weights | no paper, training, test split, scientific validation table, or future radar/precipitation prediction | learned model, event-isolated validation metrics, future-field target, paper, license, or reproducible data package appears | exclude from AI radar-nowcasting additions; operational interface engineering is not evidence of a deep-learning forecast model |
NOAA/EPIC Unifying Innovations in Forecasting Capabilities Workshop 2026 / UIFCW26 |
2026-07-25 | institution / conference signal; related AI-NWP and convective-scale operations, not direct radar echo nowcasting | event page now has live-stream links, Event Guide and agenda; slide links visible lower on the page are 2025 materials, not 2026 AI4NWP artifacts; no new radar-nowcasting repo, checkpoint, dataset, 2026 score table, or result file linked | official agenda names Project EAGLE, WoFSCast, HRRRCast, TC-Cast and 2026 HWT evaluation, but no public model metrics or radar-conditioned deliverable; no 2026 recording/slides verified | public 2026 slides/video archive, HWT score tables, WoFSCast/HRRRCast/EAGLE version notes, code/data links, radar-conditioned configuration, or reproducible convective-scale comparison appears | keep as waiting signal; never treat historical 2025 slides or agenda topics as 2026 released results |
Uncertainty-aware precipitation nowcasting with diffusion model simulating precipitation evolution processes / NowcastDiff |
2026-07-20 | direct paper / window-out high-value radar precipitation nowcasting | no official GitHub/HF repo, weights, training logs, or processed Radar Shanghai package verified; CIKM 2017 is an existing dataset lead, not proof of complete reproducibility | Elsevier Engineering Applications of Artificial Intelligence DOI 10.1016/j.engappai.2026.115140, Volume 178 Part 2 article 115140; surfaced in today's crawl but secondary coverage already dated 2026-05-22, so not fresh; evolution autoencoder separates motion/intensity residual + STA-UNet conditional diffusion; Radar Shanghai and Radar CIKM; high-intensity CSI/HSS relative improvements at least 10.9%/9.8% in author summary |
full text, exact split/threshold/lead-time/absolute scores, official code/weights, ensemble calibration, sampling cost, or cross-region reproduction appears | retain as high-value direct retrospective; never promote by search crawl date, and do not call probabilistically calibrated without CRPS/reliability evidence |
Improving precipitation nowcasting via multiphysical parameter fusion in radar echo extrapolation / PIEE |
2026-07-20 | direct paper / window-out multimodal polarimetric radar echo extrapolation | no official code, weights, public dataset, preprocessing scripts, or license verified; full text requires access | Elsevier Journal of Hydrology 668:134947, DOI 10.1016/j.jhydrol.2026.134947, April 2026; ZH/ZDR/KDP/rhoHV/sigmaV, Inter-Intra dual-branch fusion + gated spatiotemporal self-attention; author summary gives up to 24.2% CSI and 20.3% HSS relative improvement for >=40 dBZ, but dataset site/split/absolute scores unavailable |
full text, radar site/years/events/split, variable QC, absolute per-lead scores, code/data/weights, missing-variable and cross-radar tests appear | retain as direct multimodal reference, not fresh; “physics-informed” here means radar-variable priors/architecture, not PDE-constrained PINN, and maximum relative gains are not overall SOTA proof |
SA-DGANet: Precipitation Nowcasting with Attention Mechanism and dual-path residual |
2026-07-18 | direct paper / deterministic KNMI radar precipitation nowcasting; Research Square preprint to formal journal | KNMI/SmaAt-UNet data path public; preprint says code not applicable; no official SA-DGANet repo, weights, release, or license verified | Springer The Journal of Supercomputing DOI 10.1007/s11227-026-08731-y, online 2026-07-17; same Jiani Chen/Xing Wang manuscript as Research Square 10.21203/rs.3.rs-7945830/v1; 12->6 frames, NL-50/NL-20, LWSA+BPRB+GSA; NL-50 0.5 mm/h CSI 0.7427 vs UNet 0.6965 |
official code/weights, event-isolated split, exact preprocessing, independent reproduction, true probabilistic forecast evaluation, longer lead time, or cross-region test appears | promote formal-publication status; keep caveat that validation is random 10% from 2016-2018, forecast is 30 min, and sigmoid-derived CRPS is not ensemble CRPS |
A Radar-Based AI Framework for Nowcasting Severe Hailstorms |
2026-07-18 | direct / high-relevance conference abstract; composite-reflectivity severe-hail nowcasting | no code, weights, event list, data package, presentation, or architecture verified | Plinius Conference Abstracts Vol. 19 DOI 10.5194/egusphere-plinius19-34; abstract updated 2026-07-17 for PLINIUS19 oral scheduled 2026-10-08; Colorado-Wyoming, 15 hail events, 1-3 h, 40 dBZ, POD/FAR/CSI/FSS/SSIM; qualitative comparison with extrapolation and HRRR |
conference slides, full paper, architecture, training/event split, exact scores, code/data, or cross-season validation appears | promote as fresh direct conference signal but not performance breakthrough; evidence is abstract-only and 15 events |
Precipitation Nowcasting over complex terrain in Greece, using weather radar and lighting data assimilation |
2026-07-18 | near-direct radar/lightning assimilation nowcasting; non-AI | no WRF/VAR configs, event data, scripts, or quantitative table verified | Plinius Conference Abstracts Vol. 19 DOI 10.5194/egusphere-plinius19-17; abstract updated 2026-07-17, poster scheduled 2026-10-08; polarimetric reflectivity/radial wind 4DVAR + lightning-derived profiles 3DVAR, 1-km WRF, 3 h, 2024-2026 Greece cases, ICON/GFS initialization |
slides/full paper, event count, no-DA controls, quantitative reflectivity/rain metrics, configs/code, or operational chain appears | keep as non-AI hybrid-baseline/complex-terrain reference; abstract gives only qualitative improvements |
Required Timesteps for Weather Radar Quantitative Precipitation Estimation |
2026-07-18 | related radar QPE/QC / temporal sampling; non-AI and non-nowcasting | high-resolution PAR, MRR, and 2-DVD described; no dedicated code/event data package verified | AMS Journal of Hydrometeorology DOI 10.1175/JHM-D-25-0151.1; Crossref created 2026-07-17 16:44 UTC; two convective cases; reflectivity <30 s, QPE <=5 mm/h <1 min, QPE >5 mm/h <30 s recommendations |
full text, event list, code/data, uncertainty curves, operational-radar comparison, or AI label-uncertainty study appears | promote only as radar input/label QC evidence; do not claim faster input cadence automatically improves nowcast skill |
GPM IMERG V07B and V06B: Evaluation Using Ground-Based Radar Observations and Application in Global Mesoscale Convective System Tracking |
2026-07-18 | related satellite precipitation truth/benchmark and radar evaluation; non-AI and non-nowcasting | IMERG public; dedicated five-region collocations, analysis code, and updated MCS dataset download not verified | AGU JGR: Atmospheres DOI 10.1029/2026JD046673, online 2026-07-17; five radar regions, 1-3 h/0.1-0.25 deg; V07B land heavy-rain low bias larger, ocean intense-rain retrieval improved |
data/code statement, matched radar-IMERG samples, updated MCS data DOI, or use in AI target-bias correction appears | keep as supervision/benchmark caveat; IMERG RMSE is not equivalent to ground-truth extreme-rain skill |
holys519/solafune_precipitation_nowcasting exp039/exp040 |
2026-07-18 | related satellite-to-current-GPM engineering; non-radar and non-future nowcasting | commits add 4-source OOF blend + smoothing/blur/threshold/overlap patch and mean-intensity x normalized-shape U-Net; code public, raw competition data/weights/predictions/submissions absent, no license verified | no paper/DOI; exp039 patched public RMSE 0.6619116739607654 vs prior 0.6661746681900441; raw 0.6789588628265085; exp040 no result yet | exp040 OOF/LB result, weights, final leaderboard, license, formal writeup, radar target, or leakage-free external validation appears | report as substantive related code; most patch benefit is target-overlap leakage and task is same-time satellite-to-IMERG estimation |
The Synoptic, Microphysical, and Radar Characteristics of Upper Great Lakes Snow Events |
2026-07-17 | related radar QPE/QC + SOM regime classification; non-nowcasting | exact-title/DOI searches found no dedicated GitHub or Zenodo artifact; over-decade reanalysis, Marquette microphysics/accumulation and radar-QPE inputs are described, but processed data/code not verified | AGU JGR: Atmospheres DOI 10.1029/2026JD046349; online 2026-07-16, Crossref created 2026-07-17 00:18 UTC; SOM separates lake-effect/synoptic/ambiguous regimes and exposes regime-dependent Z-S bias |
full text/data statement, reanalysis/radar/event details, SOM config, code/data package, operational Z-S parameters, or nowcasting-QC extension appears | promote only as related radar QPE/QC evidence; no future-frame forecast or nowcast skill, and snow-regime findings should not be generalized to warm-season convection |
Joint bias correction and downscaling of subseasonal forecasts via diffusion models |
2026-07-17 | 48 h buffer related generative/downscaling method; temperature-only subseasonal forecast | article open access; exact-title/DOI searches found no dedicated GitHub, Zenodo, weights, or processed training data | Cambridge Environmental Data Science e22, DOI 10.1017/eds.2026.10047, online 2026-07-15; conditional denoising diffusion maps ECMWF IFS 36-km ensemble-mean subseasonal forecasts to 2-km temperature observations with lead-time conditioning; CRPS/extreme discrimination weaker than QM |
code/data statement, dedicated repo, precipitation extension, improved CRPS/extreme discrimination, or radar/high-frequency adaptation appears | keep related-method bucket; useful lead-time-conditioned generative post-processing and negative baseline, not precipitation/radar nowcasting |
Dynamic Ensemble Learning for Rainfall Estimation from Microwave Link Attenuation / duplicate SSRN metadata |
2026-07-17 | metadata duplicate / related CML rainfall estimation; non-radar and non-nowcasting | OpenRainER 2021-2022, 8 links/15 directional sublinks described; no new code/data artifact verified | Crossref created SSRN DOI 10.2139/ssrn.7129062 on 2026-07-16, but title/authors/abstract duplicate 2025 SSRN DOI 10.2139/ssrn.5579223; LSTM/GRU/Transformer/RF/XGBoost dynamic ensemble |
formal version, substantive new version statement, code/data release, or temporal forecasting extension appears | do not count as new paper; deduplicate by title+authors+abstract despite new DOI metadata |
Improving Specific Attenuation-Based Radar Rainfall Estimation |
2026-07-16 | related radar QPE / observation-chain QC; non-AI and non-nowcasting | exact-title/DOI/author searches found no dedicated GitHub or Zenodo artifact; Hangzhou DSD/radar data availability not verified | IEEE TGRS DOI 10.1109/TGRS.2026.3713480; Crossref created 2026-07-15 20:04 UTC; long/event DSD + one severe-rain C-band polarimetric radar case; dynamic alpha in AH-ZH self-consistency, beam blockage/wet-radome correction, AH-R vs KDP-R |
full text, DSD years/sample, gauges, multi-event metrics, code/data, overcorrection flag, or AI-QC extension appears | promote only as related radar QPE/QC input-chain item; abstract reports qualitative AH benefit and overcorrection risk, not AI training or universal nowcast improvement |
Estimating Uncertainty of PMW-Based Convective Fraction Product |
2026-07-16 | related satellite precipitation retrieval uncertainty / KuPR reference; non-AI and non-nowcasting | uses GPM GMI/KuPR 2015-2023 collocations; processed pairs/code not verified; exact-title/DOI searches found no dedicated GitHub/Zenodo artifact | AGU JGR Atmospheres DOI 10.1029/2026JD046377; online 2026-07-15, Crossref created 2026-07-16 00:24 UTC; environment-dependent CVF uncertainty and two independent 4-year stability check |
data/code statement, processed collocations, uncertainty formulation, regional tables, or use as conditional supervision weights in radar/satellite nowcasting appears | keep related retrieval/QC bucket; useful heteroscedastic-label lesson but not satellite-to-radar nowcasting or a new ML model |
Comparative analysis of deep learning models for multi-horizon rainfall forecasting in flood-prone tropical highlands |
2026-07-16 | 48 h buffer low-value exclusion / daily station rainfall forecast | no code/data repository verified; historical rainfall plus meteorological variables only in accessible abstract | JUTI DOI 10.12962/j24068535.v24i2.a1518, online 2026-07-15; LSTM/GRU/TFT/AutoTFT at 1/3/7/9/14-day horizons; no radar or minute-hour nowcasting evidence |
radar/minute-scale extension, reproducible split/code/data, extreme-event metrics, or task reframing appears | exclude from radar/precipitation nowcasting; daily 1-14 day forecasting is outside the strict short-term window despite DL/rainfall title |
TITAN-LSTM: A Weather Radar Nowcasting Tool |
2026-07-15 | direct paper / object-based weather-radar nowcasting | Zenodo DOI 10.5281/zenodo.19022209 has source zip, TITAN config/preprocessing, README PDF, and 76.2 MB 2016-2019 storm-track dataset; no separate GitHub repo found |
AGU JGR: Machine Learning and Computation DOI 10.1029/2026JH001337; online 2026-07-14; TITAN tracking/displacement + 3-layer LSTM area/intensity, 5-30 min; DAEE/SAO-SPOL Sao Paulo radar; 15-min POD >0.70/FAR <0.20 in author evaluation |
inspect Zenodo code/split, independent reproduction, full-field comparison, split/merge/new-cell handling, cross-radar validation, or updated release appears | promote as direct and reproducible object-level hybrid; do not compare its object metrics directly with gridded SEVIR models or call SOTA |
ST-TriMambaUNet |
2026-07-15 | direct paper / deterministic gridded radar echo extrapolation | SEVIR and CIKM datasets public; exact-title/acronym/DOI repo searches found no official code, weights, release, or license | MDPI Sensors DOI 10.3390/s26144461, published 2026-07-14; STFA + tri-branch Mamba/channel-shuffle/directional-conv MSIM; SEVIR 5->20 and CIKM 5->10 frames; strongest gains at high thresholds, but low-threshold/POD not uniformly best |
official repo/weights, exact preprocessing, external reproduction, FAR/FSS/object metrics, calibration, cross-region evaluation, or efficiency benchmark appears | promote as direct paper; phrase as strong author-reimplementation results at high thresholds, not across-the-board SOTA; SEVIR VIP thresholds are not dBZ |
HourGlass: A probabilistic data-driven temporal downscaler for global and regional weather forecasting |
2026-07-15 | 48 h buffer related method / probabilistic temporal downscaling | built on open-source Anemoi; no dedicated HourGlass repository, exact configs, checkpoints, or release verified | arXiv 2607.11457, submitted 2026-07-13; arXiv preprint only, no journal information; AIFS/Bris 6-hour-to-hourly reconstruction with graph transformer, noise, CRPS/window-statistic/difference losses; not radar nowcasting |
dedicated code/config/weights, formal venue, precipitation-extreme diagnostics, or radar/satellite interpolation adaptation appears | keep related-method bucket; useful temporal-consistency/loss reference but never promote as direct radar nowcasting |
U-Net-based spatial synergistic correction of gridded summer precipitation forecasts for CMA-MESO in Central China |
2026-07-15 | metadata-only related / NWP precipitation correction | no public code/data verified; abstract/full text unavailable in this run | Atmospheric Research DOI 10.1016/j.atmosres.2026.109217, available online 2026-07-14; title/core metadata only; not nowcasting |
abstract/PDF, observation truth, years, inputs, split, lead times, metrics, code, or data appears | retain metadata-only; do not infer architecture details or effects from title |
Advancing Spaceborne Radar Precipitation Detection with Deep Learning |
2026-07-15 | metadata-only related / spaceborne radar precipitation detection | no public code/data verified; IEEE abstract/full text unavailable | IEEE TGRS DOI 10.1109/TGRS.2026.3713017; DOI metadata created 2026-07-13 20:04 UTC; task is detection/retrieval, not temporal nowcasting |
IEEE abstract/PDF, radar platform, network, split, metrics, code, or data appears | retain as related radar-AI metadata lead; do not guess CloudSat/GPM or claim performance |
An Incremental Learning Framework Linking Reanalysis and Radar Data for Passive Microwave Precipitation Retrievals / TLP-R2S |
2026-07-15 | related retrieval / reanalysis-radar supervision bridge | NSIDC-0804 DOI 10.5067/L4S3PU4C9XU7 and GitHub Buddha-subedi/PMWPrecip_TLP-R2S, MahyarGarshasbi/TLP-R2S-ATMS exist; data/code predate this run |
IEEE TGRS DOI 10.1109/TGRS.2026.3713037; new DOI metadata in buffer; passive-microwave precipitation retrieval, not nowcasting |
full article, exact incremental-learning stages/metrics, code version changes, or radar-nowcasting transfer study appears | promote as related retrieval with existing artifacts; radar is a supervision bridge, not forecast target |
ECMWF Building shared foundations for machine-learned weather prediction |
2026-07-15 | related institution / MLWP infrastructure | Anemoi ecosystem open source; no new radar code/data/model released in this post | ECMWF AIFS Blog dated 2026-07-14, DOI 10.21957/fd7333c16b; EUMETNET E-AI MLP collaboration on shared MLWP, LAM/stretched-grid/probabilistic Bris |
concrete regional model/config/weights, precipitation benchmark, or radar integration appears | keep as institution engineering signal, not direct radar/precipitation nowcasting evidence |
shobhit1kapoor/solafune-precipitation-nowcasting |
2026-07-15 | related code / satellite-to-GPM precipitation engineering | public repo with ~247 files, tests, manifests, metrics, experiment artifacts and some best.pt; raw competition archives/large caches absent; no license |
no paper/DOI; initial commits 2026-07-14 19:06-19:12 UTC; 3-frame 16-band geostationary imagery to 41x41 GPM-IMERG, location-held-out 5-fold; internal RMSE 1.572460 is not public leaderboard | license, full data/repro script, public leaderboard/final score, formal paper, or radar-target extension appears | promote as substantive related code release; not weather radar, and internal artifact metric is not an external benchmark |
holys519/solafune_precipitation_nowcasting exp028-exp032 |
2026-07-16 | related satellite-to-precipitation engineering / ablation configs and HPC runner | prior commits add target-time/delta-time, sensor-aware IR proxy, dilated bottleneck, focal+Tversky occurrence loss, and satellite-conditional heads; commit fe59b785 adds robust experiment-dir checks, analyze_fold0-4, and Slurm afterok single-fold analysis scheduling; no new weights/data/license |
no paper/DOI and no new fold/OOF/public leaderboard result; 2026-07-16 change is 7 shell scripts, +65/-5, not model/metric evidence | validated fold/OOF metrics, public score, weights, final leaderboard, formal writeup, or radar-target extension appears | report as operational code update only; useful experiment orchestration, but no claimed skill gain and task remains satellite-to-GPM rather than weather radar |
ST-HFNet: A Spatiotemporal Hierarchical Fusion Network for Radar Echo Extrapolation |
2026-07-13 | direct paper / deterministic radar echo extrapolation | no GitHub/HF repo, release, weights, training scripts, data package, or license verified; exact-title/DOI repo search returned 0 | ICIC 2026 / Springer LNCS chapter DOI 10.1007/978-981-92-3384-7_50; Crossref created 2026-07-13; Springer First Online 2026-07-14; Shanghai Dataset; Spatiotemporal Gated Attention + Hierarchical Feature Fusion; abstract-only quantitative evidence |
full text, exact Shanghai dataset identity, split, input/output length, baselines, CSI/SSIM/MSE values, FLOPs/latency, code, or weights appear | promote as direct new DOI/page; do not call SOTA or equate Shanghai Dataset with Shanghai-2020 until verified |
A Localized Fuzzy-Logic Model with a Correction-Aware Pipeline for Bird Echo Identification in China's Operational Dual-Polarization Weather Radars |
2026-07-13 | related radar QC / biological clutter removal | no code, parameter file, public cases, GitHub/HF artifact, or usable PDF verified; SciEngine PDF endpoint returned 0-byte response | Journal of Tropical Meteorology DOI 10.3724/j.1006-8775.2026.015; online 2026-07-13, print issue metadata 2026-06-01; Yingkou Z9417 S-band dual-pol radar; Q=0.82 fuzzy threshold; not nowcasting |
full PDF/HTML, case dates/counts, variable list, membership functions, confusion matrix, cross-site/season validation, or code/data appears | promote as related radar QC; keep single-site/case-evaluation caveat and do not generalize “complete removal” beyond reported cases |
Connected Radar: A New Paradigm for Atmospheric Observation and Experimentation |
2026-07-13 | related AI radar-observation framework / active sensing | no companion code, weights, benchmark data DOI, or release verified | AMS Bulletin of the American Meteorological Society DOI 10.1175/BAMS-D-25-0307.1; published/created 2026-07-13; MAAS-derived cloud multisensor concept using self-supervised learning, uncertainty estimation, active sampling, and information-gain-driven sensing; not nowcasting |
full implementation, adaptive-scan benchmark, open MAAS data/code, quantitative information-gain experiment, or nowcasting closed-loop evaluation appears | promote as strategic observation-method paper; not a radar echo forecast model or skill result |
holys519/solafune_precipitation_nowcasting exp018 / exp026 / exp027 |
2026-07-13 | related satellite-to-precipitation engineering / localization and ensemble | commit 3e1c1ecf0d adds high-res wet-mask/multiscale exp018, overlap-patch exp026, and seed-ensemble exp027; code public; data/checkpoints/prediction TIFF/submission zip not public |
no paper/DOI; exp024 public RMSE 0.6919274860606568 documented; exp018 cloud A/B pending; exp026 ~0.673 is expectation only; exp027 leaderboard pending | exp018 fold/OOF metrics, public score, weights, final leaderboard, formal writeup, or radar-target extension appears | promote as substantive related code update; reuse wet-mask/multiscale ideas, but classify train/eval-overlap GPM copy patch as leakage and non-generalizable |
Spatially Aware Calibration of NWP and AI Precipitation Forecasts |
2026-07-12 | related method / NWP-AIWP precipitation spatial calibration | Zenodo software DOI 10.5281/zenodo.20500534; DataCite confirms Software, issued 2026-06-02, CC BY 4.0; files not inspected because Zenodo endpoint failed in this run; no separate data package verified |
AGU Journal of Geophysical Research: Machine Learning and Computation DOI 10.1029/2026JH001327; accepted 2026-06-26, published 2026-07-11; HRES vs AIFS pointwise/neighbor calibration; not radar or nowcasting |
Zenodo files accessible, observation dataset/region/years, neighborhood size, lead times, exact metrics, or radar calibration extension appears | promote as highest-value related method in 2026-07-12 report; do not call it radar nowcasting or infer that AIFS never needs calibration |
Synoptic-aware explainable artificial intelligence (SA-XAI) model for bias correction and regionalization of satellite precipitation over complex terrain |
2026-07-12 | metadata-only related / satellite precipitation correction and regionalization | no GitHub/Zenodo/HF code, data, or weights verified | Atmospheric Research DOI 10.1016/j.atmosres.2026.109213; available online 2026-07-11; only core metadata verified, no abstract/full text |
abstract/PDF, satellite/reference datasets, synoptic features, XAI method, split, metrics, or code appears | keep metadata-only; related to satellite precipitation/gap filling, not temporal nowcasting |
| ECMWF informal seminar on Earth-2 / StormScope | 2026-07-12 | institution schedule / direct-observation nowcasting carry-forward | StormScope checkpoint already public at nvidia/stormscope-goes-mrms; NVIDIA Open Model License; Earth2Studio/PhysicsNeMo recipes; no new model release in this run |
ECMWF event scheduled 2026-07-15; page abstract says StormScope DiT with neighborhood attention beats WoFS on 0-3 h object-based probabilistic scores including storm initiation; event-page publication date absent; StormScope research publication is 2026-01 | slides/video, WoFS evaluation definition, object metrics, storm-initiation cases, or new checkpoint/version appears | track as fresh institution schedule signal, not a 24-48 h paper/model release |
How to use score-based diffusion in earth system science: A satellite nowcasting example |
2026-07-11 | related method / satellite cloud diffusion nowcasting | Zenodo software DOI 10.5281/zenodo.21075796 with cira-diff-main.zip MIT license; Dryad DOI 10.5061/dryad.xpnvx0kwc with GOES-16 ABI Ch13 train/val/test Zarr tar files and all_models_Chase2025.tar; not downloaded locally |
AMS Artificial Intelligence for the Earth Systems DOI 10.1175/AIES-D-25-0046.1; DOI/VOR created 2026-07-10; arXiv preprint 2505.10432; task is GOES IR 0-3 h cloud nowcasting with Diff/CorrDiff/LDM, not radar or precipitation |
code inspection, exact metrics table, CorrDiff implementation notes, or radar/dBZ adaptation appears | keep as high-value generative-method reference; do not call it radar echo or precipitation nowcasting |
matko-iv/budva-radar / DGMR checkpoint page |
2026-07-11 | direct engineering / single-site radar DGMR fine-tuning | commit 813b41a984f2 on 2026-07-10 17:25 UTC adds docs/dgmr-models.html, models_data.js, DGMR prediction PNGs, and updates verification files; code/data frames public but no .pt/.pth/.ckpt/.onnx weights verified |
no paper/DOI; page documents Uljenje radar 2023-2026, 285,193 products, 10,612 sequences, DGMR grid-cell-only fine-tuning; +60 min DGMR improves some persistence metrics but pySTEPS/ANVIL remains stronger | checkpoint weights, full training logs, broader event verification, objective comparison scripts, license/data-permission notes, or paper appears | promote as meaningful direct engineering update; conclusion is nuanced, not DGMR superiority |
Hail detection and sizing for storm objects: A random forest framework using Swiss C-band radar and crowdsourced reports |
2026-07-11 | related radar AI / hail diagnostics | no official code/data found in GitHub or Zenodo; joshua-wx/pyhail is old related package, not this paper release |
AMS Artificial Intelligence for the Earth Systems DOI 10.1175/AIES-D-26-0001.1; DOI/VOR created 2026-07-10; random forest detection/regression from Swiss C-band polarimetric radar and crowdsourced hail reports; PSS 0.91 vs POH 0.81, size RMSE 7.5 mm vs MESHS 13.9 mm, underpredicts >=32 mm hail |
AMS full text, predictor list, storm-object segmentation, supplement, data/code release, or operational validation appears | track as weather radar AI hazard-diagnosis reference, not nowcasting |
ilianrvd/nowcasting-public |
2026-07-11 | related engineering / public radar nowcasting non-AI | commit f89330f3f1cb on 2026-07-10 05:04 UTC adds ICON 0.1 deg grid cache; code uses pySTEPS S-PROG / optical flow, radar+ICON-EU blending, GitHub Pages maps; no DL weights |
no paper/DOI; Bulgarian/Romanian radar + Blitzortung + ICON-EU operational-style pipeline | objective verification, AI/DL component, data-license notes, release, or paper appears | useful baseline/blending reference; not AI progress |
Integrating IoT-AI weather intelligence into power systems: an SVM-based approach with polarimetric radar data |
2026-07-11 | metadata-only related / polarimetric radar AI application | no code/data/weights found; IET page/PDF 403 | IET Conference Proceedings DOI 10.1049/icp.2026.1797; published 2026-07-01, Crossref created 2026-07-10; only title metadata verified |
abstract/full text, radar variables, SVM target, evaluation, or code appears | keep low-confidence metadata-only; not radar echo or precipitation nowcasting |
The Southern Ontario Lidar (SOLID) Mesonet |
2026-07-11 | related observation infrastructure / non-AI regional nowcasting | no code/data package found | AMS Journal of Atmospheric and Oceanic Technology DOI 10.1175/JTECH-D-25-0102.1; DOI/VOR created 2026-07-10; 8 Doppler lidars + 1 DIAL for aviation/regional nowcasting and NWP verification |
open data, AI assimilation/nowcasting use, or radar-fusion extension appears | background observation source only; non-AI and non-radar |
holys519/solafune_precipitation_nowcasting exp021-exp025 |
2026-07-12 | related satellite-to-precipitation engineering | commit e46497d90d on 2026-07-11 23:25 UTC adds fold completion, feature ablation, serving/calibration diagnostics, prediction blending, and multi-seed runner; code public; data/weights/submissions not public |
no paper/DOI; Solafune competition repo; exp017 folds 0-3 internal tile RMSE 0.6284 vs exp016 0.6320 with larger fold variance; public best still exp014 RMSE 0.6968727727408199; no new public score | complete 5-fold/multi-seed results, exp021-exp025 public score, final leaderboard, weights, formal writeup, or radar-target framing appears | promote as substantive validation-engineering update; internal tile RMSE is not public benchmark; not weather radar |
Enabling Physics-Based Deep-Learning Thunderstorm Nowcasting for Critical Decision Making in Innovative Air Mobility |
2026-07-10 | near-direct application / thunderstorm nowcasting for aviation IAM | no GitHub/HF code, data, weights, or platform demo verified | Materials Research Proceedings / CEAS - AIDAA Conference 2025 DOI 10.21741/9781644904251-5; Crossref created 2026-07-09 16:00 UTC; abstract says remote sensing + physics-based DL thunderstorm nowcasting in IAM flight planning, but radar input/model/metrics not verified |
accessible full text, radar/satellite input details, model architecture, lead time, objective skill, IAM decision metrics, or code appears | promote as 2026-07-10 near-direct application; do not call it radar echo nowcasting until input/task are verified |
Rainfall Field Reconstruction From Commercial Microwave Links via a Conditional Diffusion Model / CML-RainCDM |
2026-07-10 | related method / conditional diffusion rainfall field reconstruction | no GitHub/HF code, data, or weights verified | Water Resources Research DOI 10.1029/2025WR042876; published online 2026-07-09; CML attenuation to gridded rainfall fields via physics-informed conditional diffusion, not radar and not temporal nowcasting |
code/data, CML network details, evaluation dataset, physical constraint implementation, or radar/gauge fusion extension appears | keep related generative-method bucket; useful for sparse-observation rainfall reconstruction, not direct radar nowcasting |
Deep learning-based radar reflectivity retrieval using satellite observations |
2026-07-10 | related method / satellite-to-radar reflectivity retrieval metadata | no code/data/weights verified; IET page/PDF 403 in this run | IET Conference Proceedings DOI 10.1049/icp.2026.1535; published 2026-07-01; Crossref created 2026-07-09; authors include Chengdu University of Information Technology and CMA labs; no abstract/full text verified |
accessible PDF/abstract, satellite/radar data sources, model architecture, metrics, code, or data access appears | track as related satellite-to-radar retrieval; do not infer model/data/effect beyond metadata |
Heat-SwinIR: super-resolution reconstruction method for noisy low-resolution weather radar data based on heat conduction attention |
2026-07-10 | related method / weather radar super-resolution and QC metadata | no code/data/weights verified; IET page/PDF 403 in this run | IET Conference Proceedings DOI 10.1049/icp.2026.1525; published 2026-07-01; Crossref created 2026-07-09; title indicates noisy low-resolution weather radar SR using Heat-SwinIR |
accessible PDF/abstract, radar data source, degradation/noise setup, PSNR/SSIM or meteorological metrics, code, or pretrained model appears | track as radar input-quality method, not nowcasting |
FastNet: Improving the Physical Consistency of Machine Learning Weather Prediction Models through Loss Function Design |
2026-07-10 | related institution/method / global AI weather physical-consistency loss | no code/weights/data verified | AMS Artificial Intelligence for the Earth Systems DOI 10.1175/AIES-D-25-0090.1; Crossref created 2026-07-08; Met Office news 2026-07-08; global MLWP GNN, modified spherical harmonic loss, gradient terms, alternative wind representation |
code/model release, paper details useful for spectral/gradient loss transfer, or high-resolution UK/nowcasting extension appears | keep as related loss-design reference; not radar nowcasting or precipitation nowcasting |
andrewnakas/globalnowcast optional ML REFC correction |
2026-07-10 | related engineering / NWP reflectivity post-processing | public code added commit 0894d1836b19 on 2026-07-09 17:47 UTC; adds residual U-Net, IMERG/MRMS dataset builder, ONNX CPU hook; no weights committed and pipeline remains passthrough without refc_correction.onnx |
no paper/DOI; GitHub project uses NOAA GFS REFC visualization and optional ML correction toward observed precip | ONNX weights, evaluation metrics, corrected live manifest, code release/license clarity, or observed-radar blending appears | track as NWP/precip proxy correction code; not observed radar nowcasting |
holys519/solafune_precipitation_nowcasting |
2026-07-10 | related satellite-to-precipitation engineering | latest commit 2cdc0311 on 2026-07-10 adds exp013/exp015, tile-overlap docs, public scores; code public, data/weights/submissions not public, no release/license |
no paper/DOI; Solafune competition repo; exp014 public RMSE 0.6968727727408199 via tile-overlap GPM copy patch | final leaderboard, released weights, dataset/license clarity, formal writeup, or satellite-to-radar/radar target framing appears | track as satellite-to-GPM engineering and benchmark-leakage lesson; not weather radar |
RadarEchoMamba: A Fast, High-Fidelity Pyramidal Bidirectional Mamba Model for Radar Echo Extrapolation |
2026-07-09 | direct paper / deterministic Mamba radar echo extrapolation | no official repo, release, weights, HF model, or data package verified; GitHub repo search for RadarEchoMamba returned no repo |
Remote Sensing DOI 10.3390/rs18142287; published 2026-07-08; MDPI PDF extracted; South China + Shanghai-2020 datasets; Mamba backbone; strongest evidence is visual fidelity/latency, not across-the-board CSI dominance |
official code, data split/preprocessing, South China data access, Shanghai-2020 scripts, checkpoints, license, or external reproduction appears | promote as 2026-07-09 direct new paper; keep metric caveat that MS-RadarFormer wins some CSI/POD thresholds |
ST-Mamba-LDM: An Attention-Mamba Hybrid Spatio-Temporal Latent Diffusion Model for Radar Echo-Based Precipitation Nowcasting |
2026-07-25 | direct paper / generative radar echo nowcasting; code reproducibility degraded | official GitHub commit 83c23c3e0a on 2026-07-24 09:18 UTC deletes 7406 lines and the entire nowcasting/ directory; current tree retains core model modules but lacks train.py and requirements.txt still referenced by README; no release/tag/license/weights/result table |
IEEE JSTARS DOI 10.1109/jstars.2026.3710785; formal paper status unchanged |
deleted files are restored or replacement pipeline appears; README, license, release, checkpoints, complete SEVIR split, metrics table, or training logs become available | report as a negative code-state change, not a model advance; do not infer deletion reason without maintainer evidence |
Flash flood nowcasting based on deep learning and radar rainfall estimates / BoM MODSIM2025 |
2026-07-08 | near-direct application / radar rainfall to flood nowcasting | no code/data/weights verified; PDF available | MODSIM2025 proceedings DOI 10.36334/modsim2025.c07.hou; Crossref created 2026-07-08; conference paper date 2026-05-01 |
code, technical report, operational BoM material, RAINFIELDS/STEPS experiment details, or reproducible dataset appears | track as impact/hydrology nowcasting using radar rainfall and DL; not radar echo field generation |
SRA-UNet: A Small Residual Attention UNet model for precipitation nowcasting |
2026-07-09 | metadata-only precipitation nowcasting candidate | no code/data/weights verified; ScienceDirect page returned challenge; Elsevier API returned only coredata | Procedia Computer Science DOI 10.1016/j.procs.2026.06.153; Crossref created 2026-07-08 15:38 UTC; abstract/data/model details not verified |
accessible abstract/PDF, dataset type, radar-vs-non-radar input, metrics, code, or venue issue details appear | keep as candidate, not direct radar echo paper until source task/data are verified |
A Case Study on Deep Learning Approach for Multicellular Thunderstorm Classification and Nowcasting Using X-Band Weather Radar Data in East Coast Station of India |
2026-07-09 | index-delay direct/near-direct conference metadata | no code/data/weights verified; IEEE full text/abstract not obtained | InGARSS 2025 DOI 10.1109/ingarss67683.2025.11583856; Crossref created 2026-07-08 19:40 UTC; proceedings date 2025-12-10 |
IEEE abstract/PDF, station/data details, model architecture, nowcast lead time, object-vs-grid task, or metrics appear | track as new-indexed weather-radar DL nowcasting candidate; do not call same-day research publication |
Bridging Satellite Data Latency with Spatio-Temporal Learning for Real-Time Nowcasting Applications |
2026-07-09 | related satellite real-time nowcasting metadata | no code/data/weights verified; IEEE full text/abstract not obtained | InGARSS 2025 DOI 10.1109/ingarss67683.2025.11583866; Crossref created 2026-07-08 19:40 UTC; satellite latency title, not radar |
IEEE abstract/PDF clarifies precipitation/radar relevance, data source, lead time, or code appears | track as related deployment/latency signal, not direct radar echo nowcasting |
CBAM-U-Net-Based Retrieval of Radar Composite Reflectivity from FY-4A Satellite Observations over Complex Terrain in Sichuan, China |
2026-07-09 | related satellite-to-radar retrieval / radar blind-spot compensation | no public code/weights; data available on reasonable request from corresponding author | EGUsphere posted-content DOI 10.5194/egusphere-2026-2395; posted 2026-07-07; FY-4A to CMA SWAN RCRF retrieval over Sichuan, 14,023 samples, CBAM-U-Net, not nowcasting |
formal journal status, code, FY-4A/SWAN sample release, data access, or time-prediction extension appears | keep related method; useful for radar gap filling, not future echo prediction |
A spatiotemporal framework for GNSS-based rainfall nowcasting with FluxFormer and attention-based graph neural networks |
2026-07-09 | related non-radar precipitation nowcasting | no code/weights verified; data sources documented for IMERG, Hong Kong SatRef, and HKO rainfall | GPS Solutions DOI 10.1007/s10291-026-02121-w; published online 2026-07-08; GNSS ST-FluxFormer + attention GNN for 0-3 h station/IMERG rainfall nowcasting |
code, model weights, reproducible scripts, or radar-fusion extension appears | track as external water-vapor prior candidate, not radar echo nowcasting baseline |
A Bin-Aware Deep Learning Framework for Precipitation Nowcasting from Disdrometer Spectra / TSFNet |
2026-07-08 | related method / disdrometer precipitation nowcasting | no code/data/weights verified | SSRN/Crossref posted-content preprint DOI 10.2139/ssrn.7069618; Crossref created 2026-07-07 04:36 UTC; no journal venue verified |
PDF, code, Parsivel2/DSD data, station split, full metrics table, or formal venue appears | keep related; microphysical minute-scale nowcasting, not radar echo nowcasting |
Physics-informed deep learning-based adaptive beamforming for phased array weather radar |
2026-07-08 | related radar observation/QC method | no code/data verified | Scientific Reports DOI 10.1038/s41598-026-60229-y; Crossref created 2026-07-06 15:06 UTC |
accessible full text details, code, datasets, beamforming metrics, or PAWR operational evidence appears | track as radar observation-chain input quality method; not nowcasting |
holys519/solafune_precipitation_nowcasting |
2026-07-09 | related satellite-to-precipitation engineering | public competition code materially updated through commit 5c158699 on 2026-07-09; added exp012 successor-row + satellite-adapter two-head code, data-characteristics docs, and public scores; no data/weights/release/license |
no paper/DOI; Solafune competition repo created 2026-07-07 | final leaderboard, released weights, dataset/license clarity, formal writeup, or satellite-to-radar framing appears | track as satellite-to-GPM/IMERG engineering update, not weather radar |
andrewnakas/globalnowcast |
2026-07-08 | low-value exclusion / NWP visualization | GFS REFC byte-range fetch + render + GitHub Pages; ML listed only future work | no paper | changed files add actual ML post-processing, observed radar/GPM blend, objective metrics, or paper appears | exclude from direct AI/radar progress; GFS diagnostic reflectivity visualization, not observed radar |
kagtgi/Forecasting-Through-Meteorological-Reasoning / ASG-WM / FaithCast |
2026-07-03 | direct engineering | framework/specs/eval scripts updated; quantitative results TBR; neural baselines mostly stub; no verified license/release/checkpoint | no DOI/arXiv verified in the 2026-07-03 report | license, release, real checkpoint, baseline adapters, SEVIR/NEXRAD/MRMS metrics, arXiv/DOI, or official submission appears | track with caveats; do not call it a verified performance advance |
matko-iv/budva-radar / SKALA NOWCAST DGMR adapter |
2026-07-03 | direct engineering system | DGMR nowcast page and local frames cache updated; real-time output visible; objective DGMR skill not verified | not a new paper | DGMR vs LINDA/extrapolation verification, license, data-permission notes, or reproducible experiment docs appear | track as engineering reference; do not generalize Budva/OOD DGMR output |
Hugging Face precipitation-nowcasting-features-v3 empty datasets |
2026-07-03 | low-value exclusion | only .gitattributes; no README/data/license/task definition |
n/a | real dataset files, dataset card, license, task definition, or metrics are uploaded | exclude until real files appear |
sushantakundu/gsmap-live |
2026-07-25 | low-value exclusion | latest checked commit bcf234d on 2026-07-25 00:35 UTC remains Auto-update GSMap_Combined.gif and changes only that GIF; no model/data-processing/training/evaluation evidence |
n/a | changed files expand beyond GIF-only visualization refresh | exclude from direct AI nowcasting progress |
pySTEPS/ERAD-nowcasting-course-2026 / ERAD26 Academy Open Precipitation Nowcasting |
2026-07-07 | direct community/course signal | GitHub repo created 2026-07-06; current tree has README.md and LICENSE only; no notebooks, datasets, environment file, model code, or metrics yet |
course material, not a paper | hands-on notebooks, processed datasets, install instructions, AI nowcasting examples, or benchmark material appear | track as educational/community signal; do not call it a new method or reproducible release yet |
Anchor-Guided Asymmetric Block Sparse Attention for Precipitation Downscaling / CaLoG |
2026-07-07 | related method / precipitation downscaling | no code/data/weights verified | SSRN/Crossref posted-content preprint DOI 10.2139/ssrn.7068285; no journal venue verified |
PDF, code, datasets, full metrics table, or formal venue appears | keep in related-method bucket; precipitation SR/downscaling, not radar echo or precipitation nowcasting |
AI for EW4All AI for early warnings for all: From innovation to impact |
2026-07-07 | related institution signal | ITU/WMO/UNDRR/IFRC workshop/report launch; no radar nowcasting code/data verified | workshop/report launch, not a paper | report/tools/case studies expose radar, nowcasting, QPE/QPF, or model artifacts with technical detail | track as broad operational AI early-warning context; not direct radar nowcasting |
Weather radar nowcasting of rainfall using optical flow with Recurrent All-Pairs Field Transforms (RAFT) |
2026-07-07 | window-out direct paper / high-value review | no code/data verified | Atmospheric Research DOI 10.1016/j.atmosres.2026.109007; Crossref created 2026-04-11; not a 24-48 h new publication |
accessible full text, code/data, objective metrics, or reproducible experiment details become available | keep as window-out review item; direct radar nowcasting but not today's new item |
FusionTempoNet: Dual-Stream Residual Categorical Nowcasting of Short-Duration Heavy Rainfall with Data Augmentation and Ordinal Loss |
2026-07-06 | direct preprint / carry-forward | no code/data/weights verified | SSRN/Crossref posted-content preprint DOI 10.2139/ssrn.7051911; not a journal article |
PDF, code, data, metrics table, formal venue, or source-status change appears | track as direct preprint, but keep source label conservative |
QuarkOther/RadarNowcastingAI |
2026-07-09 | direct engineering / carry-forward | CHMI radar download, ConvGRU/ConvLSTM/CNN/persistence, train/predict/compare/backtest flow; HEAD still unchanged at 20a253d |
no paper/DOI verified | checkpoint, backtest metric table, data-license notes, paper/tech report, or substantial new code appears | track as engineering code signal, not a validated method advance |
EchoCast-3D / Generative machine learning for skilful 3D radar nowcasting |
2026-07-05 | window-out high-value direct paper | code/pretrained weights available by request after publication per Nature page; no direct open repo verified | npj Climate and Atmospheric Science 9, 145 (2026), DOI 10.1038/s41612-026-01407-7; published 2026-04-27, version of record 2026-07-02 |
code/weights become directly accessible, data-access path clarified, or reusable metrics/supplement extracted | keep as near-term review item, not same-day new item |
CMA procurement: 国省短时临近智能数字预报算法协同开发-中国气象局气象干部培训学院开发组件 |
2026-07-06 | window-out related institution signal | public announcement only; requirements say see bid document; no public algorithm/data/code details verified | procurement notice, not a paper; published 2026-07-03 08:11 +08, outside the 2026-07-06 48 h buffer | tender document, award notice, contract, acceptance note, or public technical spec mentions radar echo / precipitation nowcasting / AI components | track as institutional demand clue; do not count as 24-48 h direct progress |
MvReNowcast: Disentangling radar product features for improved nowcasting of raincloud evolution |
2026-07-03 | window-out / index-delay direct paper | code/data not verified | Journal of Hydrology DOI confirmed; Crossref created 2026-06-27 and indexed 2026-07-02; outside main window | abstract/PDF becomes accessible, code/data appears, or metrics can be verified from source | mention only as window-out/index-delay supplement unless status changes |
| GK2A proxy radar gap-compensation paper | 2026-07-03 | related method | code/data not verified | Korean Journal of Remote Sensing; satellite-to-radar / proxy radar observational gap compensation, not direct nowcasting | PDF/model details/code become accessible | keep in related-method bucket, not direct radar nowcasting |
| S2R-GAN++ | 2026-07-02 | window-out direct paper | code/weights not verified | Springer Discover Computing, published 2026-06-29 | code, weights, PhaDin/DWD data-permission details, CSI/FSS/POD/FAR details, or reproducible metrics appear | follow as window-out supplement, not same-day new item |
| IMPA-Net | 2026-07-02 | retrospective direct / high-value review | GitHub mentioned in report; completeness not verified here | arXiv 2026-04 in prior report context | formal page, license, weights, complete code, or reproducible metrics appear | keep as retrospective recommendation unless new status appears |
| M3R / M3Rain | 2026-07-02 | retrospective related/direct review | processing scripts mentioned; release completeness not verified here | arXiv 2026-04 / ICME 2026 author note in prior report context | formal page, license, release, weights, or reproducible metrics appear | keep as retrospective recommendation unless new status appears |
Pointwise is Pointless? |
2026-07-03 | carry-forward direct paper | no new code/status in latest report | already covered; not a new 2026-07-03 item | new version, code, weights, formal venue, or benchmark update appears | do not repeat as daily new item without state change |
FlashBack Memory |
2026-07-03 | carry-forward direct/related paper | no new code/status in latest report | already covered; not a new 2026-07-03 item | new version, code, weights, formal venue, or benchmark update appears | do not repeat as daily new item without state change |
TA-SmaAt-UNet |
2026-07-03 | carry-forward / retrospective | no new code/status in latest report | already covered as high-value review candidate | code, data, formal page, or status change appears | use in review section, not same-day new item |