An Integrated Remote Sensing Drought Monitoring Model Based on Multi-source Information

Abstract

In order to solve the problem of the traditional remote sensing drought index focuses on the monitoring of a single response factor and lacks a complete analysis of drought.In this paper， we selected TVDI， RVI， PDI， and GVMI daily products estimated from remote sensing data as independent variables， and MCI calculated from meteorological data at the adjacent moments of satellite transit as dependent variables， and uses the Random Forest Regression （RFR） model to construct a integrated remote sensing drought monitoring model.The results show that the accuracy of RFR model is better than that of the Ordinary Least Squares （OLS） model in bothtraining data and test data.The R value of the RFR training data is 0.97， the RMSE is 0.33， the R value of the RFR test data is 0.90， and the RMSE is 0.53.The R value of the OLS training data is 0.78， the RMSE value is 0.73， the R value of the OLS test data is 0.76， and the RMSE value is 0.79.The comparisons of RFR and OLS model in R and RMSE show that the RFR model is superior than the OLS model in the characterization of regional drought.In the application of drought monitoring in Southwest China in 2022， the RFR results are consistent with the spatiotemporal distribution of the MCI index， which can better characterize the spatial and temporal dynamics of the regional drought， reflecting the practicality of the RFR model in the actual drought monitoring process.However， the accuracy of RFR model is related to the number of regional stations and the spatial distribution of stations， and the accuracy of the RFR model is higher in areas with a large number of stations and uniform distribution of stations.

Keywords

• drought
• remote sensing
• random forest regression
• ordinary least squares