Remote Sensing Retrieval Of Soil Moisture Content Based On Ensemble Learning

Soil moisture content(SMC)is a key parameter in the fields of hydrology,meteorology,agriculture,and ecology.Monitoring of SMC provides effective information for soil drought analysis,regional flood warning,land degradation forecast and ecological environment assessment.The development of Remote Sensing technology has provided an effective way to obtain large-scale and long-term information on SMC.Remote sensing retrieval refers to the process of predicting the real-time state parameters of ground target based on its electromagnetic characteristics,that is,the process of transforming remote sensing data into land surface practical parameters.At present,remote sensing retrieval models for soil moisture content can be divided into empirical models and physical models: empirical models are simple in construction and easy to practice,but they have limited monitoring accuracy;physical models have a solid theoretical foundation,but they are not easily applicable because they always involve too many parameters.This paper combines the advantages of empirical models and physical models,chooses the Tibet Plateau(TP)as the study area,and uses MODIS products as the main data source to construct a remote sensing retrieval model of SMC based on ensemble learning.The main work and conclusions of this paper are as follows:(1)Extraction of surface parameters related to soil moistureCollected and collated the measured data from the Tibetan Plateau observatory of plateau scale soil moisture and soil temperature(Tibet-Obs)and the remote sensing data;based on the spectral reflectance characteristics of the soil,use MODIS surface reflectance product MOD09A1 as the source data to retrieve the soil moisture content related parameters such as vegetation index,vegetation coverage and leaf area index.(2)Reconstruction of land surface temperature based on random forestReconstructed the MODIS surface temperature product MOD11A1 based on random forest algorithm to weak the interference of vegetation and terrain,and supplement the missing values.The verification results show that the reconstructed surface temperature has a good correlation with the measured surface temperature,the accuracy and spatial continuity of the results are improved.(3)Evaluation of soil moisture based on temperature-vegetation drought index(TVDI)methodBased on the NDVI retrieval results and the reconstructed surface temperature,the NDVI-LST feature distribution space is constructed,and the dry and wet edge equations are fitted by considering the frequency of each discrete point for the problem of scatter interference at the body boundary in the feature distribution space.The temperaturevegetation drought index(TVDI)is obtained,and the soil moisture status of the QinghaiTibet Plateau is evaluated based on this.The results show that TVDI can reflect the general distribution of soil moisture to a certain extent,but the accuracy is still limited.(4)Remote sensing retrieval of soil moisture content by integrating multiple modelsXGBoost and extreme random trees are used as the primary learners to construct the soil moisture retrieval model;The stacking method is used to combine the prediction results of the two primary learners with the TVDI index inversion results,thereby constructing a multi-model fusion soil water content inversion algorithm.Verification results show that the accuracy of the multi-model fusion algorithm is significantly higher than that of the single model.Based on the inversion of the multi-model fusion algorithm,the soil moisture content of the Tibet Plateau is consistent with its terrain and climate distribution characteristics.