频繁的旱灾对甘肃省经济、农业生产造成严重危害,因此利用先进方法建立准确可靠的农业干旱监测模型对该省防旱减灾十分重要。本文基于随机森林（RF）、BP神经网络和支持向量机（SVM）3种机器学习方法,利用甘肃省2002—2019年4—10月多源遥感数据得到的植被状态指数（VCI）、温度状态指数（TCI）、植被供水指数（VSWI）、降水状态指数（PCI）以及DEM、土壤有效含水量（AWC）和气候类型作为自变量,气象站点以3个月时间尺度的标准化降水蒸发指数（SPEI_3）为因变量,构建3种不同的农业干旱监测模型,分析比较出适用于监测甘肃省农业干旱的最佳模型,同时进一步探究了机器学习方法构建的模型在不同环境下的适用性。结果表明：构建的3种机器学习模型中,随机森林模型的R²平均值高（0.86）且误差小（RMSE为0.40,MAE为0.31）,农业干旱的监测效果要优于BP神经网络和支持向量机模型;干燥和湿润两种环境下分别构建的3种机器学习模型在湿润环境中监测能力表现均更优异（R²>0.82）,而随机森林模型在两种环境中监测干旱的表现比其他两种模型强。研究结果为甘肃省的农业干旱监测与评估提供了新的科学方法,对农业干旱研究具有重要意义。

Frequent droughts have caused serious harm to the economy and agricultural production of Gansu Province. Therefore, it is very important for the province to use advanced methods to establish accurate and reliable agricultural drought monitoring models. Three machine learning methods (random forest, BP neural network, support vector machine) are used to construct an agricultural drought monitoring model using remote sensing factors and meteorological factors. Analysis of the best model for agricultural drought monitoring in Gansu Province. At the same time, the applicability of the model constructed by machine learning in different environments was further explored. The results show that among the models constructed by the three machine learning methods, the RF model has a high coefficient of determination (0.86) and small errors (RMSE 0.40, MAE 0.31). The monitoring effect of agricultural drought is better than the model constructed by BP and SVM; The three machine learning method models constructed in the two environments, respectively, performed better in the wet environment, while the model constructed by the random forest method performed better than the other two models in monitoring drought in the two environments. The research results provide a new scientific method for agricultural drought monitoring and evaluation in Gansu Province, and are of great significance to agricultural drought research.