| Drought,as one of the main meteorological disasters,has become a hot issue in the study of global climate change.The development of remote sensing technology provides a new way for drought monitoring.Soil moisture(SM)is a basic climatic variable affecting land-air interaction,a basic hydrological variable affecting rainfall-runoff process,a basic ecological variable regulating net ecosystem exchange,and a basic agricultural variable.Soil moisture can directly reflect agricultural drought.Therefore,it is very important to accurately obtain the trend of soil moisture change for agricultural drought monitoring.Shaanxi Province is located in the northwest of China,where frequent drought events have caused great losses to agricultural production.In this study,soil moisture was taken as a key research variable to explore the temporal and spatial changes of agricultural drought in Shaanxi Province from 2003 to 2017,and to predict the future drought.We chose soil moisture remote sensing products derived from AMSR-E(Advanced Microwave Scanning Radiometer-Earth Observing System)and AMSR2(Advanced Microwave Scanning Radiometer System 2)from 2003 to 2017.Since the spatial resolution of microwave remote sensing is coarse,we used random forest(RF)algorithm to downscale the data to meet the research needs.Precipitation,land surface temperature,evapotranspiration,vegetation index,elevation,slope,aspect and soil texture were selected as the explanatory variables of soil moisture.The downscaling model was established in four seasons,and the difference in the contribution rate of explanatory variables in different seasons was analyzed.Soil Moisture Drought Index(SMDI)was constructed based on the downscaled SM data,and the SMDI data were verified by SPI Index.In addition,the spatial and temporal variation of drought in Shaanxi Province was explored based on SMDI data.The SMDI index is fitted and predicted through the Autoregressive Integrated Moving Average(ARIMA)model and the random forest algorithm.Combined with the precipitation data,the drought situation in the future can be predicted through the prediction of the change of SMDI index.The main conclusions of this research are as follows:(1)Random forest algorithm can be used to build soil moisture downscaling model,and the accuracy of the model is improved after considering the seasonal difference of soil moisture variation.Factors such as precipitation,land surface temperature,evapotranspiration,vegetation,topography and soil texture can be used to establish the downscaling regression model of soil moisture.The downscaled data fit well with AMSR(AMSR-E and AMSR2)soil moisture products and in-situ measurements.Among the explanatory variables,precipitation plays a leading role in the change of soil moisture,and meteorology is the main factor affecting the change of soil moisture.The influence degree of each variable on soil moisture varies with the season.The influence of vegetation is more prominent in winter,while the influence of topography is more important in the other three seasons.(2)The soil moisture drought index constructed based on downscaled soil moisture data can be used to characterize drought.It can be seen from the comparison between SMDI and SPI of different time scales that the fitting degree of SMDI and SPI1 is the best.The variation trend of SMDI,SPI1 and SPI3 in different months is basically the same,the correlation between SMDI and SPI is better especially in spring and autumn.(3)Spatial and temporal analysis of drought based on soil moisture drought index shows that the inter-annual variation of SMDI values is small,but overall there is a decreasing trend.There are obvious rainy season and dry season in Shaanxi province.In terms of time,summer and autumn have a tendency to become dry,while winter and spring have a stable change in SMDI values.Summer and autumn are two key periods that affect the interannual variation of SMDI.From the perspective of monthly scale,SMDI increases slowly in spring,reaches a peak in July,remains at a relatively high level from July to September,and then decreases rapidly and remains at a relatively low level throughout the winter and early spring.From the analysis of spatial variation,it can be found that the changes of SMDI values in northern Shaanxi,Guanzhong and southern Shaanxi are also different.The soil moisture value in Northern Shaanxi is relatively low,with relatively small annual variation and sparse vegetation,which is greatly affected by precipitation.With the gradually increasing of precipitation in spring,the fluctuation of soil moisture value becomes larger,and the standard deviation of SMDI value also becomes larger.The rainy season in Guanzhong is mainly concentrated in summer.Compared with northern Shaanxi,the soil moisture value is higher,but the difference between the dry season and the rainy season is also greater.The rainy season lasts the longest in southern Shaanxi,and the soil moisture values have been maintained at a high level from late spring to early autumn.Under the influence of frequent rainfall,SMDI fluctuates most sharply in summer in southern Shaanxi,and the standard deviation of SMDI value is the largest.(4)The combination of ARIMA model and random forest model can fit the time series data well and make accurate prediction.First of all,the multivariate time series model ARIMAX with the addition of precipitation can well fit the changes of SMDI and make predictions.Secondly,the predicted results are affected by some precipitation extremes,and the residual of the predicted results is significantly correlated with the difference between precipitation and SMDI.With the help of the random forest model,this correlation can be used for further processing of residuals.When the residual of the random forest prediction is added to the SMDI predicted value by the ARIMA model,the correlation between the SMDI predicted value and the real value is better,and the accuracy is also significantly improved.Therefore,it is advisable to use ARIMA and random forest combined model to predict the soil moisture drought index. |
PDF Full Text Request