基于ERA5的自适应方法GFS短中期风速预报订正研究

Bias Correction of GFS Short- to Medium-Range Wind Speed Forecasts via an ERA5-Based Adaptive Approach

摘要: 复杂地形与快速边界层演变使近地面风速预报出现系统性偏差与时效相关误差。面向湖南区域短中期业务需求，本文在WRF (GFS驱动)基线之上，采用ERA5风矢量气压层订正(ERA5-UV)：以ERA5的U、V分量替换GFS的共用气压层风场。在气象站经纬度，按预报时效分段(12 h、1~3 d、4~6 d、7~10 d)与UTC时段块逐小时与最近整点观测配对，使用RMSE、ME、按提前量的相对改进率与UTC各个时段变化偏差进行评估；WRF (ERA5-only)作为敏感性对照。结果显示，订正并非普适：最佳窗口集中在中期前段(4~6 d)，基准平均RMSE由1.42降至1.28 m∙s⁻¹，白天与傍晚(06~11、18~23 UTC)降幅约14%~19%，ME同步向零收敛；中期后段(7~10 d)仍有局部改进，但不稳定。短临与短期总体收益有限：虽逐小时技巧多为正(约0.2~0.4)，但少数时次/站点的显著退化被RMSE平方放大，导致分段聚合RMSE可能劣化。空间上存在显著异质性，订正效果受站点局地条件制约(如1~3 d三站平均近零，实为OBS57985约+10%与OBS57853约−12%的抵消)。综上，ERA5风矢量气压层订正方案适用于中期窗口并在特定时段更稳健；订正方案的业务应用宜充分考虑站点差异和适用时间窗口，不宜采用统一的订正策略。

Abstract: Complex terrain and rapid boundary-layer evolution introduce systematic bias and lead-time-dependent errors in near-surface wind forecasts. Targeting short- to mid-range operations over Hunan, we evaluate an ERA5 wind-vector pressure-level correction (“ERA5-UV”) built on a WRF (GFS-driven) baseline by replacing the GFS common‐level U and V fields with those from ERA5. At station coordinates, forecasts are paired hourly with the nearest top-of-the-hour observations and evaluated by forecast segments (12 h, 1~3 d, 4~6 d, 7~10 d) and UTC diurnal windows using RMSE, mean error (ME), relative improvement by lead, and UTC-hour–resolved bias. WRF (ERA5-only) serves as a sensitivity control. The correction is not universal. Its most effective window is 4~6 d, where the segment-mean RMSE decreases from 1.42 to 1.28 m∙s⁻¹, and reductions of ≈14%~19 % occur during 06~11 and 18~23 UTC, with ME moving toward zero. In 7~10 d, improvements persist locally but are less stable. Benefits are limited in nowcasting and 1~3 d: although hourly skill is often positive (≈0.2~0.4), a few hours/sites degrade markedly, and RMSE aggregation amplifies these penalties. Strong spatial heterogeneity is evident; performance depends on site conditions (e.g., the 1~3 d three-site mean near zero reflects offsetting changes of +10% at OBS57985 and −12% at OBS57853). Overall, ERA5-UV is most suitable for the mid-range window and specific UTC periods; operational use should account for site variability and applicable time windows rather than applying a uniform correction.