基于CLDAS数据和机器算法模型的大清河流域地表土壤湿度降尺度研究

CLDAS soil moisture downscaling using MODIS products through machine learning in Daqinghe River Basin

土壤水分是“四水”转换的纽带,农业生产的基础,传统的野外定点测量土壤水分的方法难以实现空间上的展布,现代微波遥感数据可以得到大尺度的土壤水分,但分辨率低。本文利用CLDAS数据,将机器算法应用到遥感影像指数运算中,开展土壤水分的降尺度研究。论文分别采用OLS算法、Bagging算法、BRT算法和随机森林算法模型建立MODIS光学遥感数据(LST、Albedo、NDVI、ET)与土壤水分的关系模型。研究结果表明:四种算法中随机森林算法的拟合效果更优(R 2=0.96128,RMSE=0.00699)。利用该算法算出降尺度后的土壤体积水分,可以得到大尺度且空间分辨率更高的土壤水分数据。大清河流域西北部土壤含水量高于东南部,土壤含水量差异可达0.2 mm 3/mm^3,在流域土壤含水量空间分布的季节变化显著,3月土壤水分低至0.16mm 3/mm^3,9月土壤水分高达0.33mm^3/mm^3。

Soil moisture is the tie that water transformation and the function of agricultural production.The traditional method of field measuring soil moisture is difficult to achieve space distribution.At the same time,microwave remote sensing data can get large scale but low resolution.In this paper,soil moisture of CLDAS and machine learning algorithm is applied to the remote sensing image index operation.The study on soil moisture downscaling paper respectively using OLS algorithm,Bagging algorithm,BRT algorithm and random forest algorithm model is established between the MODIS data(LST,Albedo,NDVI and ET)and soil moisture.It is used to calculate the soil moisture by the random forest algorithm with better fitting effect(R 2=0.96128,RMSE=0.00699),so as to obtain large-scale soil moisture data with higher spatial resolution.The soil moisture in the northwest of Daqing river basin is higher than that in the southeast,and the difference of soil water content can reach 0.2 mm^3/mm^3.The spatial distribution of soil moisture varied significantly with different seasons.The soil water content is as low as 0.16 mm^3/mm^3 in March and it is as high as 0.33 mm^3/mm^3 in September.