All-weather And Superpixel Water Extraction Methods And Application Based On Multisource Remote Sensing Data

Water is one of the most important resources on the earth,and small water bodies account for a large proportion in water resources,and they play an important role in maintaining the stability of climate balance and hydrological cycles.In water monitoring,the high spatial and temporal resolution of water body data can be employed to quickly identify water bodies,especially small water bodies,and to accurately locate affected areas,which is significant for disaster monitoring and assessment.However,optical images are easily affected by weather factors such as clouds,rain and fog,and the weather is cloudy both before and after floods occur,which limits the application of optical images.Besides,there are many mixed pixels in the fused image that are affected by the resolution of the remote sensing image.As a common phenomenon,mixed pixels exist in remote sensing images of different resolutions,which has a great impact on the classification accuracy for ground objects.In this paper,water extraction methods of the multisource data fusion model（MDFM）and superpixel water extraction model（SWEM）are proposed,in which the first part is the MDFM.First,the optical image and SAR image are preprocessed to build a MDFM based on random forest.The SWEM extract water boundary on the basis of the MDFM,and the FCLS method was used to perform superpixel decomposition to construct a model of superpixels based on multisource data fusion.The results show:（1）From the perspective of water extraction model,the accuracy of random forest algorithm(r=0.90、0.55,P_area=0.86,0.45)are higher than that of support vector machine algorithm(r=0.89、0.53,P_area=0.84,0.44)and Wat Net algorithm(r=0.83、0.48,P_area=0.76,0.36)when only optical images are used to extract water,no matter with or without clouds.From the perspective of remote sensing data source,the accuracy(r=0.90 and 0.83;P_area=0.87 and 0.79,respectively)of the MDFM in the without clouds and with clouds are higher than those when only optical and only SAR images were used(r=0.90,0.55,and 0.78;P_area=0.86,0.45,and 0.63,respectively).（2）The correlation coefficient（r）and area accuracy(P_area)of the MDFM are improved by 38.46%and 70.59%（without clouds）,respectively,and 27.69%and 54.90%（with clouds）,respectively,compared with the global surface water product of the European Commission Joint Research Centre’s Global Surface Water Explorer（JRC-GSWE）,the MDFM+SWEM are improved by 41.54%and 85.09%（without clouds）,respectively,and 32.31%and 84.31%（with clouds）,respectively,compared with the global surface water product of JRC-GSWE.（3）The water data set based on the MDFM and the MDFM+SWEM are generated.The temporal and spatial resolution of the water data set based on MDFM and MDFM+SWEM are 10m、6 days at the shortest and less than 10 m,and 6 days at the shortest,respectively,which are higher than JRC-GSWE（30 meters and 30 days）.（4）The MDFM was used to extract water in Changbai Mountain area from 2016to 2021 to analyze the temporal and spatial distribution changes.The correlation analysis was applied to the water area of the time series and precipitation data,and the correlation coefficient r=0.87,indicating a good consistency between the water sequence change and the change of precipitation data.（5）The MDFM and the MDFM+SWEM were applied to three severe flood disasters in Yongji County,Jilin Province in 2017,and the water area before,during and after the disaster were calculated respectively.The correlation analysis was applied to the water area and the corresponding precipitation data,and the correlation coefficient were both 0.92.Meantime,the relative error of the affected area extracted by MDFM+SWEM is 3.9%,which proves the effectiveness of the method.（6）The water area and water level of Xingxingshao Reservoir in Yongji County from 2020 to 2023 were analyzed by regression.The R² was 0.79,and the difference between the predicted water level and the measured water level before and after the disaster was 0.09m and 0.24m,which had a good effect on water level prediction.The algorithm in this paper extracts water with all-weather and super-pixel,and improves the temporal and spatial resolution,which has obvious advantages.Meanwhile,the method and dataset can be well applied to the area assessment and reservoir level prediction,providing data support for disaster monitoring and prevention.