发表机构
Jua.ai AG（Jua.ai AG）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本研究验证11种物理与AI气象模型，发现极端天气下相对技能缺失是特定模型的特性而非AI气象模型整体的问题，部分AI模型在极端场景表现优于数值模型。
AI 中文摘要
据报道，第一代AI气象模型在极端天气场景下表现往往不佳，这类结论大多基于对基于再分析数据的确定性回归系统的评估。本研究针对欧洲天气站、太阳能站和雨量站，在10个月内对11种物理预报系统和AI预报系统进行验证，评估变量包括10米风速、2米温度、小时短波累积量和小时降水量，将平均绝对误差（MAE）与ECMWF IFS在ERA5 1991-2020气候态下的表现进行对比评分。结果显示，这些系统中AI模型在分布尾部并未表现出统一的相对技能缺陷：Jua EPT-2.1 Europa在全条件风速预测中领先（+8.4%），Jua EPT-2 HRRR在整体温度预测中领先（+12.1%）且在高温态下领先（+19.6±2.2%）；EPT-2.1 Europa和DWD ICON Global在大风速预测中领先；Jua EPT-2.1 Helios在整体太阳能预测中领先（+10.2±1.7%）、阴天条件下领先（+16.4±3.4%）且在晴天尾部领先（+24.8±5.4%）；对于降水，3种Jua模型在中等强度下提升14-15%，在P75-P95分位数区间提升9-11%，EPT-2 Reasoning在P95以上区间仍领先（+1.7±0.5%）。模型特有的失效表现包括：ECMWF AIFS在高温尾部表现下降4.9±2.0%，NOAA GFS在该区间下降22.8±2.0%。所有模型（包括数值气象预报系统）均存在向观测分布中心集中的共同条件偏差，且模型间差异远小于共同信号。因此，极端天气下相对技能的缺失并非AI气象模型整体的特性，而是特定AI模型和物理模型的特性。
英文摘要
AI weather models are often reported to underestimate extremes, but most evidence concerns deterministic regression models verified against reanalysis. We evaluate twelve physical and AI forecast models against ECMWF IFS using ten months of European station observations. The evaluation covers 10 m wind, 2 m temperature, solar radiation, and precipitation within regimes defined from a fixed ERA5 1991-2020 climatology. We find no uniform AI-specific deficit in the tails. Several AI models remain more accurate than IFS under extreme conditions, while others deteriorate markedly; comparable variation occurs among physical models. Every model nevertheless exhibits a common conditional-error pattern, overpredicting low observations and underpredicting high observations. Attenuation of extreme values therefore does not imply a uniform loss of relative skill: tail performance depends on the model, variable, and evaluation setting rather than on whether the forecast is produced by AI or physical numerical modelling.
