arXiv：cs.LG（机器学习，全量分类）· Ahmed Marey, Henry Lu, Abhishek Gaur, Sherif Goubran, Malek Aloui, Theodore Potsis, David Rolnick, Alex Hernandez-Garcia, Liangzhu Leon Wang·· 5 小时前AI 评分44
少即是多：极端高温公里级降尺度数据高效性的误差-距离缩放关系
Less is more: error-distance scaling relation for data-efficient kilometer-scale downscaling of extreme heat
AI 导读
研究提出 CASPER——一种带结构保持损失的 U-Net，可将 32 km 再分析数据降尺度至 1 km 的温度、湿度与风速，并发现留出误差随与训练数据的气候距离线性增长（RMSE = 0.83 + 2.95 d），可解释 90% 的方差，而数据量仅解释 7%。
正文
Abstract:Extreme heat is where urban adaptation needs kilometer-scale data the most, but the simulations training a downscaler can cost more than they save, and how much is needed has not been identified. We measured it with CASPER, a U-Net with a structure-preserving loss downscaling 32 km reanalysis to 1 km temperature, humidity and wind, across 24 configurations of one to eight months. Held-out error grows linearly with climatological distance to the training data, RMSE = 0.83 + 2.95 d, explaining 90% of its variance against 7% for volume and predicting unseen months in advance. On held-out extreme summer weeks CASPER preserves the fine-scale structure and cross-variable physics that matched-budget baselines degrade, and matches station observations during documented heat waves to within 1.8 K. Transfer to a new region degrades geographically; 11 days of local simulation cuts Vancouver's held-out error from 3.8 to 1.3 K. Training periods should span the target climate: the same accuracy for four times less simulation, putting kilometer-scale downscaling of extreme heat within reach of groups without large computing facilities.
| Subjects: | Atmospheric and Oceanic Physics (physics.ao-ph); Machine Learning (cs.LG) |
| Cite as: | arXiv:2609.40140 [physics.ao-ph] |
| (or arXiv:2609.40140v2 [physics.ao-ph] for this version) | |
| https://doi.org/10.48550/arXiv.2609.40140 arXiv-issued DOI via DataCite |
Submission history
From: Ahmed Marey [view email]
[v1]
Wed, 30 Sep 2026 16:50:36 UTC (7,475 KB)
[v2]
Thu, 1 Oct 2026 17:01:30 UTC (7,476 KB)
来源：arXiv：cs.LG（机器学习，全量分类） · arxiv.org