Accuracy comparison and driving factor analysis of LULC changes using multi-source time-series remote sensing data in a coastal area

Land use and land cover (LULC) products derived from remote sensing data are fundamental to the earth's surface ecological environment studies. Various LULC products have different data sources with large uncertainties and varying spatio-temporal resolutions. In this study, our objectives are to understand the accuracy of the current LULC products at the coastal area of Tianjin city. We conducted accuracy verification and comparative analysis of the LULC datasets available from different sources, MCD12Q1, CGLS-LC100, FROM_GLC, GLC_FCS30 and GlobeLand30 for the years of 2010, 2015, 2017 and 2020, at varying spatial resolutions 500-100-30-10 m. Using land cover dataset for these years, we conducted time-series analysis of land cover changes and driving factors. Results showed that GLC_FCS30-2020 has the highest classification accuracy (72.57%), followed by GlobeLand30-2020 (72.23%) and GlobeLand30-2010 (70.10%). The classification accuracy is related to the spatial resolution of original data source and the classification system. The verification accuracies of all products are found to be lower than their official announcements. Additionally, the accuracy comparison among different land types indicated that the classification accuracy of agricultural land, impervious surface, and water bodies is high, while that of forest, grassland, and shrubs is low. According to the spatio-temporal pattern analysis of land cover changes in Tianjin during 2010-2020, the area of agricultural land is slowly decreasing, and construction land is increasing, and the increase results mainly from the conversion of agricultural land and grassland. The rapid population growth, economic development and policy guidance are the main driving factors.