Cloud Removal Algorithm Research And Application Based On MODIS Snow Products

Optical remote sensing image is sensitive to cloud cover,and it is not possible to obtain information of snow cover under clouds,exploring the cloud removal algorithm has great significance for restore the snow condition under the cloud.Northern Xinjiang,one of the three largest snow covered areas in China,is critical to accurately acquire the dynamic snow cover changes for understanding the climate change,water resources investigation,water cycle,the development of animal husbandry and prediction of snow disaster.Therefore,this study proposed new MODIS Snow product cloud removal algorithm and studied the temporal and spatial changes of snow in Northern Xinjiang.In order to fill the spatiotemporal gaps of NDSI and binary snow products caused by cloud cover,the new cloud removal method is proposed.The spatial and temporal adaptive gap-filling method predicts the NDSI value of the pixels under the cloud by establishing a weighted cloud-free similar pixel function.First of all,the combination of Aqua and Terra data and adjacent day synthesis is used to remove part of the cloud pixels.Then,the spatial and temporal adaptive gap-filling method is implemented by using the long time interval data to fully recover the NDSI gaps under the cloud.The conditional probability interpolation method based on a space-time cube is used for cloud removal of binary snow products.This method can completely recover the snow information under the cloud after the combination of Aqua and Terra data.Firstly,the conditional probabilities of the center pixel and the neighborhood pixel have the same snow cover condition in the space-time cube are counted.Then,the probability of cloud pixel reclassification into snow is calculated.Finally,the snow probability is used to reconstruct the snow condition of cloud pixel.Based on method of cloud removal,daily MODIS cloud-free snow cover extend from 2001 to 2018 is produced in Northern Xinjiang of China,and we use the produced cloud-free snow products to analyze the snow cover variation trend in Northern Xinjiang of China.The results show that:(1)The spatial and temporal adaptive gap-filling method proposed in this paper can completely remove the cloud gaps,and has high accuracy.The results show that the average correlation coefficient(r),mean absolute error(MAE)and root mean square error(RMSE)of the restored NDSI with true value are 0.95,0.06 and 0.08respectively;(2)The daily cloud-free snow cover products generated by the conditional probability interpolation method based on the space-time cube completely eliminates the cloud cover,and can improve the precision of snow cover monitoring effectively.The results show that this method can completely remove different proportions of clouds,and the overall accuracy is 97.44%.The results of the cloud-free MODIS Snow cover products and Landsat 8 OLI snow cover classification have high consistency,and the overall accuracy is 90.34%;(3)In Northern Xinjiang of China,snow cover presents obvious seasonal characteristics and is closely related to altitude.Snow cover from winter to summer gradually decreases,and the proportion of annual coverage fluctuates is relatively large in the period of snow melting and accumulation.In different altitudes,the snow cover area increases with increasing altitude,and the trend of snow melting period moving backward and snow accumulation period moving forward is obvious.In the area with altitude better than 3000 m,the lowest snow cover rate in summer can still reach 23.88%.In addition,69.70% of the areas in Northern Xinjiang has altitude less than 1500 m,with more than 80% snow cover in winter and nearly zero in summer;(4)In the spatial distribution of snow cover from 2001 to 2018,there is a positive correlation between the average number of snow cover days and the altitude(r = 0.76),the number of snow cover days gradually increases with increaseing altitude.Altay Mountains and Tianshan Mountains have distributed the biggest snow cover days.Through Theil-Sen median and Mann Kendall method,we analyzes the change trend of snow cover days in Northern Xinjiang of China from 2001 to 2018 at pixel scale,which shows that the snow cover days has a descend tendency in 34.39%of the area,mainly distributed in Junggar basin,in 0.86% of the total area is significantly reduced.The 59.96% of the total area has an increasing trend of snow cover days,of which 3.80% showed a significant increasing trend,mainly distributed in the Tianshan Mountains and Altay Mountains.