基于随机森林算法高分土壤湿度产品的构建与评估

Construction and evaluation of high spatial resolution soil moisture product based on random forest algorithm

以湖南地区干旱精细化监测业务需求为牵引,基于多源卫星遥感、实时地面观测、中国气象局模式等资料,应用随机森林学习算法,建立土壤湿度因子与降水、地温、地表反照率、蒸发、植被覆盖等多种卫星遥感影响因子的降尺度模型,研制一种高分辨率土壤湿度产品(1 km×1 km)。对建模形成的高分辨率产品进行评估,结果表明,该产品能够很好地模拟湖南地区0~10 cm土壤湿度时空变化,平均偏差为-0.01 m^(3)/m^(3),相关系数为0.9,在旱情较重的夏秋季准确性较高。所研制产品能够为农业干旱精细化监测提供可行性方案,对于地广人稀且监测站点相对较少的地区有广阔的应用前景。

Driven by business needs of the precision drought monitoring in Hunan area,based on multi-source satellite remote sensing,real-time ground observation,China Meteorological Administration model,and other data,the random forest method is used to establish the downscale model for soil moisture factors and various satellite remote sensing influencing factors such as precipitation,ground temperature,surface albedo,evaporation,vegetation cover,etc.,and develop a high-resolution soil moisture product(1 km×1 km).The modeled high spatial resolution product is evaluated,and the results show that the product can reflect temporal and spatial variation characteristics of the soil moisture at the depth of 0 to 10 cm in Hunan area,with an average deviation of -0.01 m^(3)/m^(3) and a correlation coefficient of 0.9.It has higher accuracy in the summer and autumn with severe drought conditions.The developed product can provide a feasible solution for precision monitoring of agricultural drought,and has broad application potential for areas with sparse population and relatively few monitoring stations.