Scholarlink

<p>To enhance the performance of precipitation simulation in regions where multiple climate systems and complex underlying surface interactions occur, and to mitigate prediction uncertainty. In this study, a multi-source merged precipitation integrated learning method (MSMP) is constructed using topographic data and Coupled Model Intercomparison Project Phase 6 (CMIP6) model precipitation data in Yunnan Province. Furthermore, the Shapley Additive Explanations (SHAP) method is used to interpret the MSMP model. Compared with the multi-model ensemble mean (MME) and the ERA5 reanalysis dataset, the MSMP method demonstrated good performance in capturing the spatiotemporal characteristics in previpitation, which reduced the relative error of the regional annual average precipitation during the validation period 2005–2014 from an overestimation of 47.31% and 40.12% (of MME and ERA5, respectively) to 3.00%. Results from MSMP show that relative to the historical baseline period 1995–2014, the percentage change (PC) of precipitation in Yunnan under the SSP1-2.6, SSP3-7.0, and SSP5-8.5 scenarios exhibits increasing trends of 0.05%/year, 0.06%/year, and 0.09%/year, respectively. Among these, the most rapid increases occur in the mid-term and long-term periods under the SSP5-8.5 scenario, where the trend reaches 0.12%/year. In terms of spatial distribution, the annual mean PC of precipitation across different scenarios and periods maintains a relatively consistent pattern with the baseline, with increased precipitation observed in the central and northwestern regions of Yunnan and a decrease in other areas. The largest increase in precipitation is approximately 20%. Additionally, the SHAP reveals that the MSMP method can effectively identify the potential location signals hidden in precipitation data.</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

Integrating CMlP6 and terrain factors via MSMP: improved precipitation simulations and SHAP interpretation

  • 期刊论文

  • 发表日期：

    2025年11月7日

查看全文
  • Wenxi Lu,
  • Hao Zhang,
  • Yunfeng Duo,
  • Jinne Li,
  • Yuexi Yao,
  • Hongming Yan

摘要

To enhance the performance of precipitation simulation in regions where multiple climate systems and complex underlying surface interactions occur, and to mitigate prediction uncertainty. In this study, a multi-source merged precipitation integrated learning method (MSMP) is constructed using topographic data and Coupled Model Intercomparison Project Phase 6 (CMIP6) model precipitation data in Yunnan Province. Furthermore, the Shapley Additive Explanations (SHAP) method is used to interpret the MSMP model. Compared with the multi-model ensemble mean (MME) and the ERA5 reanalysis dataset, the MSMP method demonstrated good performance in capturing the spatiotemporal characteristics in previpitation, which reduced the relative error of the regional annual average precipitation during the validation period 2005–2014 from an overestimation of 47.31% and 40.12% (of MME and ERA5, respectively) to 3.00%. Results from MSMP show that relative to the historical baseline period 1995–2014, the percentage change (PC) of precipitation in Yunnan under the SSP1-2.6, SSP3-7.0, and SSP5-8.5 scenarios exhibits increasing trends of 0.05%/year, 0.06%/year, and 0.09%/year, respectively. Among these, the most rapid increases occur in the mid-term and long-term periods under the SSP5-8.5 scenario, where the trend reaches 0.12%/year. In terms of spatial distribution, the annual mean PC of precipitation across different scenarios and periods maintains a relatively consistent pattern with the baseline, with increased precipitation observed in the central and northwestern regions of Yunnan and a decrease in other areas. The largest increase in precipitation is approximately 20%. Additionally, the SHAP reveals that the MSMP method can effectively identify the potential location signals hidden in precipitation data.