A Two-Step Machine Learning Assessment of Rain Area and Intensity Errors in Summer WRF Precipitation over Hainan Island
Year: 2026
Type: article
Abstract: Accurate hourly precipitation assessment over tropical islands requires separate treatment of rain area placement and wet-hour intensity. Hainan’s summer precipitation combines coastal, monsoon, and terrain influences, while WRF errors vary across space and time of day. We assessed a two-stage correction of June–August (JJA) 2016–2022 WRF precipitation against a station-derived spatial precipitation analysis field using a rain area model followed by rainfall band LightGBM intensity models. Rain area transfer was evaluated with 2016–2020 fitting, 2021 cutoff selection, and a fixed 2022 test, while the complete chain was also diagnosed retrospectively for 2022. In the held-out rain area evaluation, the threat score (TS) and the fractions skill score (FSS) at the 0.1 mm h−1 threshold increased relative to raw WRF, although the predicted area remained approximately twice the analysis field area. In the retrospective complete chain diagnosis, threshold skill improved from the rain/no rain level through 25 mm h−1, and RMSE decreased modestly. However, the 50 mm h−1 threshold remained unhit. The results show the clearest benefit in rain area detection and moderate intensities, whereas broad masks contributed to nighttime and central mountain wet biases. The two-stage design therefore separates occurrence and intensity errors for diagnosis, while a later complete system evaluation should freeze all choices before testing a new season.
Source: Atmosphere
Authors: Zilin Ren(Data curation, Formal analysis, Methodology, Software, Visualization, Writing – original draft), Mingxin Li(Formal analysis, Methodology, Writing – review & editing), Chunxu Han(Data curation, Investigation, Software, Visualization), Chenxiao Shi(Validation, Writing – original draft), Jie Ma(Conceptualization, Project administration, Resources, Supervision), Qi Li(Funding acquisition, Validation, Writing – original draft), Jianmei Wu(Investigation), Lei Bai(Conceptualization, Funding acquisition, Resources, Supervision)
Language: English
Subfield: Atmospheric Science
Field: Earth and Planetary Sciences
Domain: Physical Sciences
SDG: Climate action
Open Access status: gold
Estimated APC paid: $2,896
Awards: G6015713199, G6969612347, G9001396680
OpenAlex ID: W7236016056
