Search

Man-made reservoirs are key components of terrestrial hydrological systems. Identifying the location and number of reservoirs is the premise for studying the impact of human activities on water resources and environmental changes. While complete bottom-up censuses can provide a comprehensive view of the reservoir landscape, they are time-consuming and laborious and are thus infeasible on a global scale. Moreover, it is challenging to distinguish man-made reservoirs from natural lakes in remote sensing images. This study proposes a convolutional neural network (CNN)-based framework to recognize global reservoirs from Landsat 8 imageries. On the basis of the HydroLAKES dataset, a Landsat 8 cloud-free mosaic of 2017 was clipped for each feature (reservoir or lake) and was resized into 224 × 224 patches, which were collected as training and testing samples. Compared to other deep learning methods (Alexnet and VGG) and state-of-the-art traditional machine learning methods (support vector machine, random forest, gradient boosting, and bag-of-visual-words), we found that fine-tuning the pretrained CNN model, ResNet-50, could reach the highest accuracy (91.45%). Application cases in Kansas (USA, North America), Mpumalanga (South Africa, Africa), and Kostanay (Kazakhstan, Asia) resulted in classification accuracies of better than 99%, which showed the applicability of the proposed ResNet-50 model to the extraction of reservoirs from a vast amount of moderate resolution images. The framework that was developed in this paper is the first attempt to combine remote sensing big data and the deep learning technique to the recognition of reservoirs at a global scale. [ABSTRACT FROM AUTHOR]

National Oceanography and Atmosphere Administration's Advanced Very High Resolution Radiometer (AVHRR) (1970s–present, ∼1 km) and China's GaoFen-1 (GF-1) image series (2013–present, ∼16 m), can both provide extremely useful information of land use and land cover, due to the long history and, wide scanning width of AVHRR and the high resolution of GF-1. However, both data sets have apparent weaknesses of large and inconsistent geometric and geolocation errors (AVHRR: 1–20 pixels, GF-1: 1–10 pixels), prohibiting them from many interdisciplinary applications. This paper, based on feature point detection method, presents a new approach to improve the georeferencing of AVHRR and GF-1 data using inland water bodies. No need of a prior knowledge of the error source and only based on the image information and the base map, this method is able to automatically search and find the optimal water threshold value (reflectance), conduct feature point detecting and matching, and perform affine transformation. In the case study of the Tibetan Plateau, water body (i.e., lake) with its lower reflectance and relatively stable location and shape, is used as the referencing mask (i.e., base map) for detecting feature point pairs. The accuracy of the geometric correction is found within subpixel level, with average RMSE = 0.76 for AVHRR and RMSE = 0.91 for GF-1. The developed georeferencing method can be potentially applied to other less accessible areas that are difficult to derive ground control points for numerous applications with improved location and geometric accuracies. [ABSTRACT FROM AUTHOR]

• Nearly 40 years of remote sensing-based water area data from Lake Chad are generated. • The total inundation area lags behind the open water area by approximately 2 – 3 months. • An interactive online application for inspecting yearly and monthly areas of Lake Chad is provided for the general public. As the second largest lake in Africa, the extent of Lake Chad has over 50% seasonal variations with large parts of water under aquatic vegetation. Although a great area shrinkage since the 1960s divided the lake into two parts, namely, the northern and southern basins, it still feeds millions of people from four countries around it. Given sparse in situ measurements, remote sensing with seamless spatial coverage has served for acquiring the long-term area series of Lake Chad. This study retrieves the open (unvegetated) surface water area series of Lake Chad and the total inundation area, including open water and water under macrophytes, combining multiple remote sensing data from around 1980 to 2020 with satellite-based water level and in situ observations at the Bol gauge as validation and auxiliary. Results show that the total inundation area continuously recovered at a rate of 145 km2/year from 1982 to 2020 with large annual fluctuations. Approximately two- to three-month time lags between the open surface water and the total inundation area reveal the hydrological asynchrony of Lake Chad, which can be attributed by the large area of rooted macrophytes. The overspill of the southern surface water flowing over the Great Barrier into the northern lake also has a one-month lag with the open surface water of the southern lake. The Google Earth Engine-based online application can be easily shared for use and data downloaded by the general public, and the open code can be adjusted for any other lakes worldwide. [ABSTRACT FROM AUTHOR]