Abstract: Combined with social economy and population changes, the rapid extraction of urban construction land from medium and high resolution remote sensing images based on cloud platform can be used to effectively and accurately monitor the dynamic changes of urban construction land expansion in a large range and a long time series, thereby providing reference for urban management and planning. Based on Google Earth Engine (GEE) cloud platform, the urban area was extracted by NPP/VIIRS (suomi national polar-orbiting partnership/visible infrared imaging radiometer suite) annual average night light data and threshold segmentation. We obtained 3 142 Landsat images covering the cities of the three provinces of Northeast China. On the basis of the original spectral band, we constructed exponential features, texture features, and terrain features, and used SEaTH algorithm for these features optimization. Based on the JM distance, the number of features was reduced from 20 to 12. Within the urban area, combined the optimal features with random forest (RF) algorithm, the Landsat monthly composite images were reclassified, and more accurate extraction of construction land was obtained. The experimental results show that the average overall accuracy and Kappa coefficient of urban construction land expansion in the three provinces are 96.19% and 0.92, the method is more efficient and reliable. The urban construction land of the three provinces had expanded by 49.07% from 1989 to 2019. Among the provincial capitals, the expansion rate of Shenyang is the fastest, followed by Changchun, and Harbin is the slowest. Population and economy are the main factors that promote the expansion of urban construction land.

Key words: Google Earth Engine, feature optimization, random forest algorithm, construction land expansion, driving mechanism