Remote Sensing Drought Monitoring Model Based On Random Forest And Its Application

The drought monitoring based on remote sensing data in large area has been one of important methods, however, the conventional remote sensing methods mainly focused on single drought response factors such as soil moisture or vegetation status. But the study of integrated drought monitoring of multiple factors is relatively limited. Random forest is a machine learning method, with many advantages such as being accurate, handy, fast and stable and has been used in many fields in recent years. In this paper, a remote sensing drought model was developed based on random forest algorithm and multi-source remote sensing data, including MODIS, TRMM and SRTM-DEM of 2001-2010 in Henan province. Meantime, he drought events in 2010 were simulated and the frequency and intensity of drought were explored. The main conclusions are as follows:1.Vegetation Condition Index(VCI),Temperature Condition Index (TCI), Land Cover types (LC),TRMM-Z,DEM, slope, aspect and Available Water Capacity(AWC),which were extracted from remote sensing data and other soil data, were used as independent variables, and the comprehensive meteorological drought index(CI) were used as dependent variable. Based on random forest algorithm, a remote sensing drought model was built. Model took into account a variety of hazard factors, which extracted from remote sensing, provides a new way for regional integrated drought monitoring.2.1n order to test the accuracy and validate the model, the correlation test and drought grade consistent rate was carried out with model and comprehensive drought index (CI),standardized precipitation evapotranspiration index(SPEI), relative humidity in 10 cm depth. The correlation coefficients of CI and training set results were above 0.96, and the test set has reached more than 0.74. According to the CI classification standards, the model results were classified grades. The total consistent rate of test set and CI of the reached more than 69%. The correlation coefficient of SPEI and model results in each month was between 0.6-0.9. And the overall grade consistency rate of two more than 60% in each season. The correlation between the relative humidity of 10cm depth soil and the model results was a very significant correlated, and the correlation coefficient of the other stations was more than 0.6.3. To examine the differences of drought monitoring capability with the model and existing remote sensing drought indexes, the model and temperature vegetation drought index (TVDI) were compared and analyzed. Model applicability in all seasons and all terrain area were relatively better than TVDI. The model and TVDI could reflect the changes in crop yield, but the correlation coefficient of the model and the change of yield were slightly higher than TVDI; which mean the model could monitor agricultural drought better, and had stronger adaptability in different seasons and terrain. The correlation coefficient of SPEI and model was more significant than TVDI, which indicated the model had better meteorological drought monitoring capacity.4. The drought events in Henan in May to July and September to November 2010 were simulated by the model and the results could reflect the actual drought situation and its spatial variation. Therefore, this method could be well applied to monitor regional drought. The results of frequency and intensity of drought analysis in Henan Province during the whole year and in season show, the average drought frequency was 31%, and the frequency of drought in most areas of the province is between 25-40%, which had three obvious drought prone center in the north, central and western. The average drought intensity of the province was 0.985, which was light drought. The drought in northern was frequent in winter and spring, but the severe drought appeared less. The drought in central region was frequent in all seasons, annual drought intensity grade was light -moderate drought, and spatial difference was small. Expect the Nanyang region, the drought frequency in western was low and the average drought grade was light drought, the drought in Nanyang was frequent and drought intensity grade was moderate drought. Drought in southern rarely appeared but severe drought had a high frequency.