Spatio-temporal-spectral Fusion Method And Its Application In Monitoring Agricultural Drought On Complex Landscapes

Agricultural drought can not only reduce crop output,but also cause secondary disasters such as diseases and insect pests,which is one of the important factors restricting the development of agriculture and social economy.In the coastal plain areas with large land use intensity and complex landscape patches in the mountainous areas in the south of China,the high spatial and temporal resolution remote sensing data can not only reflect the time change process of agricultural drought,but also reflect the spatial change details of agricultural drought,which has an urgent need in remote sensing monitoring of agricultural drought.Taking zhangzhou in fujian province as an example,based on the analysis of agricultural drought hazard-formative factors of the process and its monitoring mechanism,on the basis of low spatial resolution of the normalized difference vegetation index and surface temperature,temperature vegetation dryness index and vegetation supply water index and meteorological drought composite index as independent variables,relative soil moisture as the dependent variable,using the random forest regression model building agricultural drought monitoring model;On this basis,by spatiotemporal fusion,spatial downscaling and spatio-spectral Fusion,generating high spatio-temporal resolution of normalized difference vegetation index,surface temperature,temperature vegetation dryness index and vegetation supply water index,combining with the spatial interpolation of meteorological drought composite index,generates high spatio-temporal resolution of agricultural drought monitoring data,the main conclusions are as follows:(1)Taking Sentinel-2 as the data source and zhangzhou as the research area,a high spatial resolution band was generated or selected from the four 10m spatial resolution bands by four methods: the maximum correlation coefficient,the central wavelength nearest neighbor,the pixel maximum and the principal component analysis.We fused the one high spatial resolution band produced and six multispectral bands with 20m spatial resolution by the five fusion methods of PCA,HPF,WT,GS and Pansharp to produce six multispectral bands with 10m spatial resolution and the fusion results were evaluated from two aspects: qualitative and quantitative(information entropy,average gradient,spectral correlation coefficient,root mean square error and general image quality index).The results show that fusion quality of HPF with the maximum correlation coefficient is better than other fusion methods As far as zhangzhou is concernedthe.(2)in view of the fact that when the Fit-FC algorithm was proposed,there was no real Sentinel-3 data verification and the fusion image boundary was fuzzy predicted by the algorithm.In this paper,the fitting process of the Fit-FC algorithm was improved by using the thin plate spline,multiresolution Segmentation and the change information of the low-spatial resolution image(the improved algorithm is referred to as IFit-FC for short).The influence of the fitting window size on the Fit-FC algorithm,the influence of the segmentation scale on the IFit-FC algorithm,and the comparison between the optimal fusion results of the Fit-FC and IFit-FC algorithms and the accuracy of FSDAF and STARFM algorithms,The results show that the accuracy of the reflectance and surface temperature predicted by the IFit-FC algorithm is better than other algorithms in terms of zhangzhou.(3)Taking the normalized difference vegetation index and surface temperature,temperature vegetation dryness index and vegetation supply water index and meteorological drought composite index of low spatial resolution every day from march to July of 2013 to 2017 as independent variables,and relative soil moisture as the dependent variable,the agricultural drought comprehensive monitoring model was constructed by using the random forest regression model according to the seasons,and the model was verified by the daily low spatial resolution data and relative soil moisture from march to July 2018,the results show that the spring and summer training set and testing set of correlation coefficient is above 0.7,In addition,mean absolute relative error of the test sets in spring and summer is lower than that of the training sets.From the results of the classification of drought,the consistency rate between the test sets and the actual drought grades is over 75%.(4)using IFit-FC algorithm,spatial downscaling and spatio-spectral Fusion to generate high spatial resolution of the normalized difference vegetation index and surface temperature,temperature vegetation dryness index and vegetation supply water index,combining with meteorological drought composite index acquired by IDW and random forests regression model to build the agricultural drought monitoring model to generate high spatial resolution of agricultural drought monitoring data,and by using the ground site relative soil moisture and model monitoring results of MODIS data for validation,The results show that the consistency rate between the drought grade in the high spatial resolution data and the actual drought grades is 78.95%,and it also is consistent with the monitoring results of MODIS data.