高级检索

基于随机森林的长江中下游流域综合干旱指数构建与应用

Construction and application of an integrated drought index for the middle and lower reaches of the Yangtze River Basin based on the random forests

    摘要
  • 摘要: 综合考虑土壤水分与降水、气温、辐射、归一化差值植被指数（normalized difference vegetation index，NDVI）、地表温度（land surface temperature，LST）和饱和水汽压差（Vapor Pressure Deficit，VPD）等要素的非线性关系，本文构建了2000—2023年长江中下游流域综合干旱指数（Integrated Drought Index，IDI），并对其精度进行了评价，进一步从时间和空间尺度识别了该流域2022年的干旱事件及其各因素的重要性，结果表明：1）训练集和测试集与实测土壤水分的决定系数（R2）分别为0.91和0.50，均方根误差（Root Mean Square Error，RMSE）的值分别为0.016和0.032，表明基于随机森林构建的IDI预测效果较好；2）14个站点土壤水分与IDI的R2在0.16至0.52之间（P < 0.01）。1—12月，站点土壤水分与IDI的R2为0.17~0.48（P < 0.01），表明本文构建的IDI精度较高；3）综合干旱指数距平表明2010年10月—2011年4月、2013年7—12月、2019年7—10月以及2022年6—10月长江中下游流域发生了干旱事件；4）空间分布特征显示，2022年6月干旱首先发生在流域西北部，7—9月干旱程度加剧，超过60%的区域出现重旱，10月北部干旱缓解但南部旱情持续至11月，与实际情况一致；5）特征重要性分析结果表明，VPD、LST和降水是影响长江中下游流域干旱的主要驱动因子。本研究成果可为长江流域的干旱监测与预警提供新的技术支撑与科学依据。

     

    Abstract: Considering the nonlinear relationships among soil moisture, precipitation, air temperature, radiation, normalized difference vegetation index (NDVI), land surface temperature (LST), and vapor pressure deficit (VPD), this study constructed an Integrated Drought Index (IDI) for the middle and lower reaches of the Yangtze River Basin from 2000 to 2023 using a random forest model. The accuracy of the model was evaluated, and the IDI was further applied to identify the temporal and spatial characteristics of the 2022 drought event and the importance of its influencing factors. The results show that: 1) The coefficients of determination (R2) between the predicted and observed soil moisture were 0.91 for the training dataset and 0.50 for the testing dataset, with corresponding root mean square error (RMSE) values of 0.016 and 0.032, indicating good predictive performance of the model. 2) The R2 values between soil moisture and IDI across 14 stations ranged from 0.16 to 0.52 (P < 0.01). On the monthly scale (January to December), R2 values ranged from 0.17 to 0.48 (P < 0.01), demonstrating the high accuracy of the constructed IDI. 3) The anomalies of the integrated drought index were negative during October 2010 to April 2011, July to December 2013, July to October 2019, and June to October 2022, indicating that drought events occurred in these periods. 4) The spatial distribution characteristics show that drought began in the northwestern part of the basin in June 2022, intensified from July to September when over 60% of the region experienced severe drought, and was alleviated in the northern part in October, while drought in the southern part persisted until November, consistent with actual conditions. 5) The feature importance analysis indicates that VPD, LST, and precipitation are the main driving factors influencing drought in the middle and lower reaches of the Yangtze River Basin. This study provides new technical support and a scientific basis for drought monitoring and early warning in the Yangtze River Basin.

     

    • HTML全文
    • 参考文献(58)
    • 相关文章
    • 施引文献
  • 资源附件(0)

/

返回文章
返回