The use of Google Earth Engine and Geographical Detectors to monitor farmland changes in Fujian province

Abstract

Detailed information regarding the status and dynamics of farmlands is essential to protect food security, particularly for areas with a limited portion of farmlands like Fujian province. Since Fujian has a mountainous terrain and subtropical climate, it is challenging to acquire cloud-free remote sensing images to extract land information from remote sensing images. This study investigates the use of Google Earth Engine (GEE) to monitor farmland changes from remote sensing images, and Geographical Detectors to analyze the underlying driving factors in the Fujian province. We trained an online Deep Neural Network (DNN) model via Google Colaboratory, and then applied to it classify land-use types from Landsat images during 2000-2020 derived from the GEE platform. The obtained changes of land-use types were then inputted to Geographical Detectors, integrated with population, meteorological, and socio-economic data, to analyze their driving forces. The results showed that the proposed method produced land use maps at intervals of three years with higher overall accuracy (0.91±0.01). Based upon Geographical Detectors, we found that the farmland changes mainly caused by factors regarding elevation, population and slope, followed by factors of soil type, temperature and gross domestic product (GDP). We conclude that the GEE platform combined with deep learning models is of high potential to extract land cover and multi-temporal land use maps over large regions, and the Geographical Detectors are suitable for analyzing land changes using remote sensing derived products.