Irrigation Mapping In Winter Wheat Region Of China Using Multi-Source Remote Sensing Data

In China,the precipitation is much lower than the water demand of winter wheat during its growing season.The lack of water resources has become the main restricting factor in our nation’s winter wheat breadbasket(WWB).Artificial irrigation is usually used to improve the soil water status.The study of irrigation mapping technology and the exploration of the spatial irrigation distribution in China’s WWB are the technical support for the dual strategy of food security and water resources security.However,there are still some shortcomings in the current mapping methods and products for irrigation areas in China.There are few thematic map products about irrigation in WWB and most of them rely on traditional singlesource remote sensing(RS).Therefore,this study proposes a fast irrigation mapping method based on multi-source RS fusion,which will provide reliable mapping products for the research of irrigation patterns in China’s WWB.Firstly,the irrigation characteristics were extracted by the time series distribution deviations between the soil moisture microwave observation data product and ERA-Interim simulation data.Soil moisture change is the most direct indicator of irrigation activities.Low correlations among different soil moisture time series data are effective indicators to detect irrigation signals.On the basis of the analysis of the seasonal characteristic deviations among the irrigation pixels and non-irrigated pixels of three microwave soil moisture observation products(AMSRE,ASCAT,ESA CCI)and ERA-Interim soil moisture simulation products,Dynamic Time Warping(DTW)and Kolmogorov Smirnov(K-S)test algorithms were used to measure similarities among different time series data and then to map irrigation information.The overall accuracy of irrigation mapping is 84.78%,and kappa coefficient is0.6951.Secondly,after the 16 day NDVI time series data of MODIS are smoothed and filtered,the optimal phenological parameters are selected by combining artificial interpretation with intelligent algorithm analysis to extract irrigation characteristics.Significant differences of wheat growth characteristics after watering,can be reflected by several key phenological characteristic values of NDVI time series derived from optical RS.After S-G filtering,13 phenological parameters were extracted from the reconstituted NDVI curve of winter wheat.Principal component analysis(PCA)and random forest regression(SFR)were used to screen seven main optimal features and SVM to distinguish irrigated pixels and non-irrigation pixels.The overall classification accuracy is 83.81% and the kappa coefficient is 0.6715.Compared with the irrigation mapping products based on all the 13 phenological features,the data compression is achieved and the identification accuracy of irrigation pixels is improved by3.07% in this study.The classification results are good,but the classification error of nonirrigation pixels is 20.43%,which affects the accuracy of irrigation mapping.Lastly,the crop phenological parameters and K-S parameters of soil moisture products were integrated into a multi feature classification vector as the input of classification model,which is used to extract high-precision irrigation pixels and to map the spatial distribution of irrigation signals in WWB with spatial resolution of 1km.The identification accuracy of irrigation pixels improves a lot with the irrigation mapping method based on multi-source RS fusion.The overall classification accuracy is 89.89% and the kappa coefficient is 0.7962.The overall identification accuracy of irrigation mapping products in this study improves 5.08%compared with those based on soil moisture characteristics,and also improves 6.04%compared with others based on crop phenology characteristics.The irrigation mapping method based on multi-source RS fusion,decreases the quantities of overestimating irrigation pixels in optical RS irrigation mapping,solves the problem of coarse spatial resolution in microwave RS irrigation mapping,and shows the irrigation spatial distribution characteristics in China’s WWB.The selected optimal phenological characteristics not only compress the total amount of data and improve the mapping efficiency,but also improve furtherly the accuracy of irrigation mapping.The irrigation signal is reserved in maximum and the ability of extracting irrigation information is enhanced,after the fusion of the three microwave soil moisture characteristic values with the maximum synthesis algorithm.