基于集成树类算法估算农田蒸散量  

Estimation of Cropland Evapotranspiration Based on Ensemble Tree-based Algorithms

作　　者：顾信钦 吴立峰[1] GU Xin-qin;WU Li-feng(School of Hydraulic and Ecological Engineering,Nanchang Institute of Technology,Nanchang 330099,China)

机构地区：[1]江西省南昌市南昌工程学院水利与生态工程学院,南昌330099

出　　处：《节水灌溉》2022年第5期26-32,共7页Water Saving Irrigation

基　　金：江西省教育厅科技项目(GJJ211904);江西省科技合作专项(20212BDH80016)。

摘　　要：准确估算蒸散量(ET)对水资源管理和干旱评估具有重要意义。评估了两种集成树类算法,XGBoost(XGB)和Random Forest (RF)对不同时间尺度下农田ET的表现。模型输入数据使用了通量站点的气象观测数据和MODIS卫星的叶面积指数(LAI)产品数据以及ERA再分析数据。结果表明,2个站点模型的偏差百分比(PBIAS)均在5%以内,整体上不存在高估或低估现象。在气象数据基础上增加LAI能提高模型预测精度,但气象数据与再分析数据作为输入时差异不大。在半小时尺度和日尺度下2个站点的XGB模型整体上优于RF模型。可为准确估算ET提供参考方法。Accurate estimation of evapotranspiration(ET) is critical for water resources management and drought assessment. This study evaluated the performance of two integrated tree class algorithms, XGBoost(XGB) and Random Forest(RF), on estimaing farmland ET at different time scales. The input data of the model were meteorological observation data from the flux site, leaf area index(LAI) product data from MODIS satellite, and ERA reanalysis data. The results showed that the PBIAS of the two sites was within 5%, and there was no overestimation or underestimation on the whole. Adding LAI on the basis of meteorological data could improve the prediction accuracy of the model, but there was little difference between meteorological data and reanalysis data as input. At the half-hour scale and daily scale, XGB model was superior to RF model on the whole. This study can provide a reference method for the accurate estimation of ET.

关 键 词：蒸散量 XGBoost Random Forest 机器学习 

分 类 号：S274.1[农业科学—农业水土工程]