采用XGBoost+SHAP揭示贵阳市地表温度的驱动力因子

Revealing the Drivers of Land Surface Temperature Variation in Guiyang City Using XGBoost and SHAP

该文在基于数理统计方法揭示贵阳市地表温度(land surface temperature,LST)年际和季节时空演变趋势的基础上,为更好地明晰影响因子及其交互影响对LST的驱动作用,协同极端梯度提升算法(extreme gradient boosting,XGBoost)和Shapley加法解释方法(shapley additive explanations,SHAP),对贵阳市LST时空演变模式背后的原因进行了诊断分析。结果表明:(1)2000-2020年间,贵阳市秋季低温区面积占比呈显著减少趋势,年均减少率约为0.068%。(2)LST增温表现出南高北低的分布特征,云岩区和南明区的亚高温区和高温区变化最为明显。(3)XGBoost可较好地刻画各季节LST与影响因子间的响应关系,获得了较高的建模精度。所构建模型在测试集上的RMSE、MAE和R^(2)分别为0.5169℃、0.3893℃和0.8950。(4)基于SHAP的分析结果表明:影响因子对LST影响的重要性存在季节差异,总体而言,国内生产总值(GDP)、高程、降水和植被覆盖对LST的影响最大;除GDP和人口密度外,各影响因子对LST变化的作用方式主要为非线性关系;高程、降水、植被和水体对研究区LST主要起到降温作用,而不透水面、裸地、农业用地、人口密度及GDP则主要起到增温作用。

Based on revealing the interannual and seasonal spatiotemporal evolution trends of the Land Surface Temperature(LST)in Guiyang City through mathematical statistics methods,a diagnostic analysis of the reasons behind the spatiotemporal evolution patterns of LST in Guiyang City was conducted.This analysis used the eXtreme Gradient Boosting(XGBoost)algo⁃rithm and the SHapley Additive ExPlanations(SHAP)method to better understand the driving effects of influencing factors and their interactions on LST.The results showed that from 2000 to 2020,the proportion of low-temperature areas during au⁃tumn showed a significant decreasing trend,with an average annual reduction rate of approximately 0.068%.The LST warm⁃ing exhibited a distribution characteristic of being higher in the south and lower in the north,with the changes in the sub-high temperature and high-temperature areas being most evident in Yunyan District and Nanming District.XGBoost was able to effectively characterize the response relationship between LST and influencing factors in various seasons,achieving high modeling accuracy.The constructed model had RMSE,MAE,and R^(2)values of 0.5169℃,0.3893℃,and 0.8950,respectively,on the validation set.The analysis results based on SHAP indicated that the importance of influencing factors on LST varied seasonally.Overall,Gross Domestic Product(GDP),elevation,precipitation,and vegetation cover had the greatest impact on LST.Except for GDP and population density,the influencing factors mainly exhibited a nonlinear relationship with LST changes.Elevation,precipitation,vegetation,and water bodies primarily had a cooling effect on LST in the study area,while impervious surfaces,bare land,agricultural land,population density,and GDP primarily had a warming effect.