A method for automatically extracting long-term and large-scale land use and land cover information

Accurate and comprehensive Land Use and Land Cover (LULC) information is essential for applications such as urban planning and crop monitoring. However, traditional manual sampling is labor-intensive and time-consuming. This study presents an automated approach, implemented on the Google Earth Engine platform, to extract LULC information by identifying stable areas where land use types remain unchanged over the years. Long-term sample points were collected within these areas and used with Landsat imagery and a Random Forest classifier to produce annual maps from 1987 to 2022 for four regions: Arizona (USA), central-western Brazil, the Beijing–Tianjin–Hebei region (China), and Portugal. Improvements in imagery quality and sample quantity led to steadily increasing classification accuracy over time. After post-processing and validation through visual interpretation, the KAPPA and OA values for the study areas ranged from 0.79 to 0.87 and 0.82 to 0.89, respectively. These metrics indicate a high level of classification accuracy, comparable to the claimed overall accuracy of other regional-scale LULC products. The results demonstrate that the automated method can extract long-term LULC information with sufficient coverage from remote sensing imagery on a global scale, significantly saving researchers time and manpower costs.