{"paper":{"title":"Longwang: Zero-Shot Global Spatiotemporal Precipitation Downscaling with a Latent Generative Prior","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Longwang enables zero-shot downscaling of global precipitation to daily 10 km fields from monthly 100 km inputs by combining a context-conditioned latent generative prior with posterior sampling.","cross_cats":["cs.LG"],"primary_cat":"physics.ao-ph","authors_text":"Daniele Visioni, Yue Wang","submitted_at":"2026-05-17T19:01:47Z","abstract_excerpt":"High-resolution precipitation information is essential for climate impact assessment, yet global climate models remain too coarse to resolve key small-scale processes. Existing machine learning downscaling methods often require paired low- and high-resolution data for supervised learning, are tied to fixed regions or scale factors during inference, and can be computationally expensive to train and run in physical space. Here we introduce Longwang, a zero-shot latent generative framework for global spatiotemporal precipitation downscaling. Longwang learns a context-conditioned latent generative"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"On ERA5 reanalysis, Longwang outperforms standard posterior sampling with an unconditional generative prior in reconstructing fine-scale spatial patterns, preserving temporal coherence, and recovering extreme precipitation intensities. The framework further generalizes to historical climate simulations and future climate projections under substantial distribution shift.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that a context-conditioned latent generative prior learned in an unsupervised or self-supervised manner can be effectively combined with a physically informed observation operator to produce accurate posterior samples that generalize across significant distribution shifts in climate data.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Longwang enables zero-shot downscaling of global precipitation to daily 10 km resolution from monthly 100 km data by learning a context-conditioned latent generative prior and using posterior sampling with a physical observation operator.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Longwang enables zero-shot downscaling of global precipitation to daily 10 km fields from monthly 100 km inputs by combining a context-conditioned latent generative prior with posterior sampling.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"5455636188c91b4d40cfd48ff83d5181717b0856113e7a9466ddea2388df0f5a"},"source":{"id":"2605.17603","kind":"arxiv","version":1},"verdict":{"id":"425fa5b5-ad77-467f-85b9-fff3b485ea48","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T22:12:10.458984Z","strongest_claim":"On ERA5 reanalysis, Longwang outperforms standard posterior sampling with an unconditional generative prior in reconstructing fine-scale spatial patterns, preserving temporal coherence, and recovering extreme precipitation intensities. The framework further generalizes to historical climate simulations and future climate projections under substantial distribution shift.","one_line_summary":"Longwang enables zero-shot downscaling of global precipitation to daily 10 km resolution from monthly 100 km data by learning a context-conditioned latent generative prior and using posterior sampling with a physical observation operator.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that a context-conditioned latent generative prior learned in an unsupervised or self-supervised manner can be effectively combined with a physically informed observation operator to produce accurate posterior samples that generalize across significant distribution shifts in climate data.","pith_extraction_headline":"Longwang enables zero-shot downscaling of global precipitation to daily 10 km fields from monthly 100 km inputs by combining a context-conditioned latent generative prior with posterior sampling."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.17603/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T22:31:19.539857Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T22:21:37.797342Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T21:33:23.573811Z","status":"skipped","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T21:21:57.502847Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"e0ac4210f94fa1e9de3bc9a5cff050b2310310cf28c06ee239ceb166667cdac8"},"references":{"count":47,"sample":[{"doi":"","year":2024,"title":"Zhang, W., Zhou, T. & Wu, P. Anthropogenic amplification of precipitation variability over the past century.Science385, 427–432 (2024). 20","work_id":"11924c99-a4fa-48c4-8d7a-f30b067ec91c","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2017,"title":"Schneider, T.et al.Climate goals and computing the future of clouds.Nature Climate Change7, 3–5 (2017)","work_id":"93c2cb33-f27c-4dab-bf84-4ead2c2627c3","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2015,"title":"& Gutowski Jr, W","work_id":"dcfe2748-2c8b-410e-ab07-b59c3f6ca245","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Kotz, M., Levermann, A. & Wenz, L. The effect of rainfall changes on economic production.Nature601, 223–227 (2022)","work_id":"a0b45f3f-838e-4428-af1b-003a4ebf5ed0","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Hess, P. & Boers, N. Deep learning for improving numerical weather predic- tion of heavy rainfall.Journal of Advances in Modeling Earth Systems14, e2021MS002765 (2022)","work_id":"c5d5c8c9-936e-4e90-bb02-07973563ec7d","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":47,"snapshot_sha256":"80ec32ce5f1f6cab2d6bc70255f97db5c95369173aec07f27ed7625ceaa7cdb7","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"620f613e9e23eafe639ab2090835ece024b094023204b24849c8c6b608e12778"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}