ABSTRACT
Large-scale surface water mapping not only helps us protect, utilize and manage water resources but also contributes to the understanding of climate change and the hydrologic cycle. A recent study showed that a multilayer perceptron (MLP) neural network is an effective method to identify various surface water types from Landsat 8 Operational Land Imager (OLI) satellite imagery. We use this method to produce a surface water map of China for 2015 (SWMC-2015) at a 30 m pixel size. The accuracy of SWMC-2015 was assessed with a set of random water and not water validation points. The strengths and limitations of SWMC-2015 include: the SWMC-2015 clearly shows the major lake clusters and river networks with high mapping accuracy and the overall accuracy and kappa coefficients of SWMC-2015 are 90% and 0.78, respectively. The accuracy of SWMC-2015 can be improved from perspective of training samples representation, verification sample seasonal fluctuation and mixed pixel. The SWMC-2015 is available for free download on the remote sensing of global change website (https://vapd.gitlab.io/post/swmc2015/).
Acknowledgments
The authors thank editors and three anonymous reviewers for their valuable and careful suggestions to improve our manuscript.