Production of historical classification products based on existing land cover classification products and Google Earth Engine platform

Land cover classification plays a crucial role in the detection of changes (such as cities, forests, vegetation changes) and the extraction and analysis of information. However, traditional land cover classification methods always spend a lot of time on data processing, which leads to inefficient production of classified products. Furthermore, most of the existing methods rarely consider the application of the sample in time, especially the lack of basic data for building long time series. In this study, we used historical high-precision classification data to create a sample set to train the classifier on the Google Earth Engine, we applied the classifier to other years in the same region. To some extent, the classifier's expansion in time is realized. The classified data mainly comes from the Landsat satellite series data available by Google Earth Engine. The classification effect is evaluated by the confusion matrix, and the results show that the overall accuracy exceeds 93.0% and kappa coefficient is 0.9 the accuracy of other years is also higher than 85%, it shows that we have a better and stable realization of sample migration and the expansion of classifier in time.