Automatic extraction of aquaculture ponds based on Google Earth Engine

Abstract

Aquaculture is one of China's fastest-growing animal food production sectors. It accounts for the largest share in the world, mainly distributed in coastal areas. Due to the depletion of offshore resources and increasing domestic demand for aquatic products, more and more land, including newly reclaimed land, is gradually being used to build aquaculture ponds. Understanding the location, spatial pattern, scale, and other properties is critical for China's food and protein security. However, until recently, how to detect, monitor and map the aquaculture ponds with remote sensing is still a problem, which hinders the understanding of its magnitude and value, and interferes sustainable management of coastal ecosystems. Here we proposed a framework for extracting aquaculture ponds by integrating existing multi-source remote sensing data on the Google Earth Engine platform. Taking Shanghai as a study area, the Multi-threshold Connected Component Segmentation and Random Forest algorithm method were used to extract aquaculture ponds automatically. The results show that this method can effectively generate the maps of Shanghai's aquaculture ponds from 2016 to 2019, and the overall accuracy of the classification results in 2018 can reach 91.8%. This method can greatly improve the efficiency of extracting aquaculture ponds, and has a good performance in detecting non-intensive aquaculture pond areas. It can also be easily used and has high spatio-temporal transferability with the help of the Google Earth Engine platform.