Scholarlink

<p>Precipitation forecasting plays a critical part in proactive management. Focusing on the July 2021 extreme precipitation in Henan Province of China, this paper performs an intercomparison of artificial intelligence (AI) forecasts and numerical weather predictions. Specifically, two AI weather models, GraphCast and FuXi, are run autoregressively to generate precipitation forecasts spanning lead times from 0 to 10 days; their performance is compared against that of the high resolution forecast (HRES) from the European Centre for Medium-range Weather Forecast (ECMWF) using the mean error, Pearson correlation coefficient, root mean square error and threat score; furthermore, the forecasts of atmospheric circulation and moisture transport are also verified. The results show that GraphCast forecasts initialized at 00:00 on 15 July 2021 capture the temporal evolution of precipitation for this event with the correlation coefficient being 0.68 but underestimate the accumulated precipitation by 49.95&#xa0;mm. In the meantime, GraphCast and FuXi forecasts tend to capture the large-scale patterns of the enhanced western North Pacific subtropical high, Typhoon In-Fa and moisture convergence associated with this event, albeit some underestimations of the magnitude. In addition, GraphCast forecasts tend to capture the correlation between precipitation and the magnitude of vertically integrated moisture flux convergence in this event. The correlation coefficients are respectively 0.81 and 0.88 for GraphCast forecasts and HRES initialized during July 14–16. Overall, the intercomparison facilitates valuable insights into applications of AI forecasts for hydroclimatic extremes and also provides forecast verification practices readily applicable to other cases of extreme precipitation.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Intercomparison of AI forecasts and numerical weather predictions for the July 2021 extreme precipitation event in Henan, China

  • 期刊论文

  • 发表日期：

    2026年5月22日

查看全文
  • Qiang Li,
  • Tongtiegang Zhao

摘要

Precipitation forecasting plays a critical part in proactive management. Focusing on the July 2021 extreme precipitation in Henan Province of China, this paper performs an intercomparison of artificial intelligence (AI) forecasts and numerical weather predictions. Specifically, two AI weather models, GraphCast and FuXi, are run autoregressively to generate precipitation forecasts spanning lead times from 0 to 10 days; their performance is compared against that of the high resolution forecast (HRES) from the European Centre for Medium-range Weather Forecast (ECMWF) using the mean error, Pearson correlation coefficient, root mean square error and threat score; furthermore, the forecasts of atmospheric circulation and moisture transport are also verified. The results show that GraphCast forecasts initialized at 00:00 on 15 July 2021 capture the temporal evolution of precipitation for this event with the correlation coefficient being 0.68 but underestimate the accumulated precipitation by 49.95 mm. In the meantime, GraphCast and FuXi forecasts tend to capture the large-scale patterns of the enhanced western North Pacific subtropical high, Typhoon In-Fa and moisture convergence associated with this event, albeit some underestimations of the magnitude. In addition, GraphCast forecasts tend to capture the correlation between precipitation and the magnitude of vertically integrated moisture flux convergence in this event. The correlation coefficients are respectively 0.81 and 0.88 for GraphCast forecasts and HRES initialized during July 14–16. Overall, the intercomparison facilitates valuable insights into applications of AI forecasts for hydroclimatic extremes and also provides forecast verification practices readily applicable to other cases of extreme precipitation.