| Lakes and reservoirs are crucial in the global hydrological cycle.At the same time,they also play an important role in material transfer and energy exchange between land and ocean,as well as global climate and ecological changes.Accurately and quickly monitoring the long-term spatiotemporal changes of lakes and reservoirs are very important for the scientific management and related research of their water resources.As a comprehensive earth observation technology,satellite remote sensing has the advantages of wide coverage and periodic repeated monitoring compared with traditional observation methods.It has become the main technology used in global and regional surface water monitoring.At present,optical remote sensing images like MODIS,Landsat and Sentinel-2 imagery have become a common data source for monitoring the area and dynamic changes of lakes and reservoirs because of their long time series and free access.Using optical remote sensing images to monitor the dynamic change of lake and reservoir surface water area is mainly limited by two key factors.On the one hand,optical remote sensing data is easily affected by cloud coverage,which reduces the actual monitoring frequency;On the other hand,remote sensing data with long time series and high spatial resolution are still very scarce,limiting the research on the long-term development process of lakes and reservoirs.To solve the above problems,the following aspects are studied in this paper.(1)Mapping water bodies under cloud cover using a spatiotemporal dependence model.In order to solve the problem of cloud contamination of optical remote sensing images in the process of water monitoring,a category-based approach is proposed to reconstruct the water distribution in cloud-contaminated images using a spatiotemporal dependence model.The proposed method predicts the class label(water or land)of a cloudy pixel based on the neighboring pixel labels and those at the same position in images acquired on other dates according to historical spatiotemporal water-distribution data.The method was evaluated through eight experiments in different study regions using Landsat and Sentinel-2 images.The results demonstrated that the proposed method could yield high-quality cloud-free classification maps and provide good waterextraction accuracy and consistency in most hydrological conditions,with an overall accuracy of up to 98%.The accuracy and practicality of the method render it promising for applications across a wide range of future research and monitoring efforts.(2)Monitoring surface water area variations of reservoirs using daily MODIS images by exploring sub-pixel information.Moderate Resolution Imaging Spectroradiometer(MODIS)imagery is an attractive data source for the routine monitoring of reservoirs,however,the accuracy is often limited due to the negative impacts associated with its coarse spatial resolution and the effects of cloud contamination.Methods have been proposed to solve these two problems independently but it remains challenging to address both problems simultaneously.To overcome this,a new approach that aims to monitor reservoir surface water area variations accurately and timely from daily MODIS images by exploring sub-pixel scale information is proposed.The proposed approach used estimates of reservoir water areas obtained from cloud-free and relatively fine spatial resolution Landsat images and water fraction images by spectral unmixing of coarse MODIS imagery as reference data.For each MODIS pixel,these reference reservoir water areas and their corresponding pixel water fractions were used to construct a linear regression equation,which in turn may be applied to predict the time series of reservoir water areas from daily MODIS water fraction images.The proposed approach was assessed with 21 reservoirs,where the correlation coefficients between reservoir water areas predicted by the common pixel-based analysis method and altimetry water levels were all less than 0.5.With the proposed sub-pixel analysis method,the resultant correlation coefficients were much improved,with eleven values larger than 0.5 including six values larger than 0.8 and the highest value of 0.94.The results show that the proposed sub-pixel analysis method is superior to the pixel based analysis method.The proposed method makes it possible to directly estimate the whole reservoir water area from,potentially,an individual cloud-free MODIS pixel,and is a promising way to improve the accuracy in the usability of MODIS images for the monitoring of reservoir surface water area variations.(3)Monitoring surface water area variations of lakes and reservoirs using water occurrence data.In view of the problems that the method in(2)uses too little band information when extracting water fraction data from MODIS imagery,and cloudless Landsat data may be difficult to obtain in some areas,this study improves the MODIS pixel water fraction extraction algorithm by replacing the input single-band MODIS imagery with multi-band MODIS imagery,and using the statistical relationship between pixel water fraction and the total surface water area of lakes and reservoirs to monitor the long-term dynamics of surface water area of the target lake or reservoir more precisely.The proposed method can monitor the long-term surface water area variations of lakes and reservoirs with high accuracy and robustness.It can be applied to lakes and reservoirs with narrow water-level-fluctuating zone which are difficult to monitor accurately by existing methods.This research uses the proposed method to extract the long-term time series of daily water areas of 30 main lakes in the Tibetan Plateau from 2000 to 2018,and evaluates the efficiency of the proposed method by calculating the correlation coefficient between the resultant surface water area time series and the altimetry data.Based on the resultant lake surface water area series,the dynamic changes of 30 main lakes in the Tibetan Plateau from 2000 to 2018 are analyzed.The results show that there are 26 experimental lakes with a correlation coefficient higher than 0.5,among them there are 13 lakes with a correlation coefficient higher than 0.8,up to 0.94.It reveals that the predicted surface water area series of most experimental lakes have high accuracy,showing the fine application prospect of the proposed method in monitoring long-term dynamic change of surface water area of lakes and reservoirs.Generally,this paper aims to solve the problem of being easily affected by cloud cover when using optical remote sensing images to monitor the dynamic change of lakes and reservoirs,and the mutual restriction between the temporal and spatial resolution of remote sensing images.The paper proposes three different methods to solve the above problems,including the method of mapping water bodies under cloud cover using a spatiotemporal dependence model(STD method),monitoring surface water area variations of reservoirs using daily MODIS images by exploring sub-pixel information(MSPI method),and monitoring surface water area variations of lakes and reservoirs using water occurrence data(WOD method).The STD method mainly solves the problem of cloud coverage when using optical remote sensing data;while the MSPI method and WOD method are mainly to solve the problem of mutual restriction between temporal and spatial resolution of remote sensing data.The above three methods have achieved good results in the experiments,and can be used to monitor the long-term dynamics of lakes and reservoirs.They are important parts of the global lake and reservoir monitoring methodology system. |
