基于机器学习的综合干旱监测建模及在西南地区应用  

Integrated drought monitoring modeling based on machine learning and its application in Southwest China

作　　者：何志远 钟九生[1] 代仁丽 HE Zhiyuan;ZHONG Jiusheng;DAI Renli(School of Geography&Environmental Science,Guizhou Normal University,Guiyang 500025,Guizhou,China)

机构地区：[1]贵州师范大学地理与环境科学学院,贵州贵阳500025

出　　处：《水利水电技术（中英文）》2022年第2期43-51,共9页Water Resources and Hydropower Engineering

基　　金：国家自然科学基金项目(41661081);贵州省科技计划项目(黔科合平台人才[2017]5726-56)。

摘　　要：干旱是由降水不足引起的,受温度、蒸散发等各种环境因素的影响,导致缺水和作物歉收。传统的干旱监测方法主要侧重于气象、水文等单一因子,而对多因子综合干旱监测的研究相对有限。本文利用2001—2015年的温度状态指数(TCI)、植被状态指数(VCI)、植被供水状况指数(VSWI)、降水状态指数(PCI)、土壤湿度状态指数(SMCI)、高程(DEM)及田间持水量(AWC)等7个干旱因子为自变量,以综合气象干旱指数(CI)为因变量,利用随机森林(RF)、增强回归树(BRT)和人工神经网络(ANN)构建干旱监测模型,并以西南五省为研究区进行了评价和分析。结果表明,基于人工神经网络的干旱监测指标(ANN-CI)预测效果最好,在测试集中预测值和观测值间的可决系数(R^(2))为0.94,均方根误差(RMSE)为0.23。三种基于机器学习的综合干旱指标均在草地区表现最好,林区精度最差。在2001—2015年间,ANN-CI和植被生长状况具有显著的时空相关性(R^(2)=0.70,p<0.01)。最后选用ANN-CI对西南地区2009/2010年干旱事件的发展过程进行监测,并且与帕默尔干旱指数(PDSI)的监测结果进行对比,结果表明ANN-CI能够较好地应用于区域的旱情监测。Drought is caused by a lack of precipitation and is affected by various environmental factors such as temperature and evapotranspiration, leading to water shortages and crop failures. Traditional drought monitoring methods mainly focus on single factors such as meteorology and hydrology, while the research on multi-factor integrated drought monitoring is relatively limited. Seven drought factors, including Temperature Condition Index(TCI), Vegetation Condition Index(VCI), Vegetation Supply Water Index(VSWI), Precipitation Condition Index(PCI), Soil Moisture Condition Index(SMCI), Digital Elevation Model(DEM) and soil Available Water Content(AWC), are used as independent variables in this paper. Taking the Composite index(CI) as the dependent variable, Random Forest(RF), Boosted Regression Trees(BRT) and Artificial Neural Network(ANN) are used to construct the drought monitoring model. The five provinces in Southwest China are taken as the study area for drought monitoring and comparison. The results show that the drought monitoring index based on ANN(ANN-CI) has the best prediction effect, and the determination coefficient(R^(2)) and root mean square error(RMSE) between the predicted and observed values in the test set are 0.94 and 0.23 respectively. The three integrated drought indexes based on machine learning all perform best in the grassy area, while the accuracy in the forest area is disappointing. From 2001 to 2015, there is a significant spatial and temporal correlation between ANN-CI and vegetation growth(R^(2)=0.70, p<0.01). Finally, ANN-CI is selected to monitor the development process of drought events in Southwest China in 2009/2010, and compared with the Palmer Drought Severity Index(PDSI) monitoring results, which show that ANN-CI can be better applied to regional drought monitoring.

关 键 词：干旱监测 机器学习 卫星遥感 西南干旱 

分 类 号：P429[天文地球—大气科学及气象学]