Identification Of Land Use Change In Continuous Time Series Based On GEE

Land use remote sensing monitoring is an important technical means for land use change related research,especially the emergence of cloud computing platforms such as Google Earth Engine(GEE),which provides new ways and methods for obtaining land use change information.Supervised classification is a commonly used method for obtaining land use information,which can achieve high accuracy.However,for long time series land use classification,it is difficult to update samples year by year,and the accumulation of false changes year by year also seriously restricts the accuracy of mapping.To solve the above problems,this study proposes a method for achieving high-precision continuous time series land use change recognition.The core steps include automatically updating each training sample year by year based on remote sensing feature indices and statistical filtering methods,as well as using moving windows and intra year and inter year information to perform time consistency testing and false change correction.Taking the burning area of the Great Khingan Mountains as the research area,the Landsat dataset and auxiliary data for consecutive years were obtained based on the GEE platform.2333 Landsat scene data from 1986 to 2020 in the research area were selected to identify complex and rapid changes in land use after the fire,and the method was applied and validated.The main conclusions of this article are as follows:(1)The robust Mahalanobis distance is used to realize the rapid update of training samples for each region year by year,establish a complete training sample database,and improve the efficiency of image semi-automatic interpretation;(2)The random forest method was used to realize the initial interpretation of land use classification from 1986 to 2020.The average accuracy of the initial classification was about 90.05%,which laid a good foundation for the subsequent time consistency check.The study also found that within a year,when using both Landsat ETM+ and Landsat OLI images,the classification accuracy is relatively stable.The number of Landsat images also affects the classification accuracy,as the more scenes in a year,the higher accuracy can be achieved during the classification process.At the same time,it was found that most of the scene data was collected from summer.(3)The use of time consistency test corrected the pseudo change of land type mutations,effectively improving the recognition accuracy of the change area.The average accuracy before correction was about 85.41%,and after correction,it increased to 93.62%,ranging from 88.1% to 96.5% in different years;At the same time,the overall accuracy of the change area before and after correction has improved by about 5.1%.The above indicates that the method proposed in this study can quickly and accurately identify land use changes year by year.