Long-Term Landscape Classification Model for the Ebi Lake Basin Constructed Using Google Earth Engine and Multidimensional Feature Sets
The study focuses on the Ebi Lake Basin in Xinjiang,integrating historical remote sensing imagery data with field sur-face feature investigations.The landscape types of the Ebi Lake Basin were classified into six categories:cultivated land,forest land,grassland,water bodies,construction land,and unused land.Using Landsat imagery provided by the Google Earth Engine(GEE)platform for the years 2002,2008,2013,2018,and 2023 as the primary data source,we estab-lished landscape classification as an evaluation indicator.Spectral indices(normalized difference vegetation index,en-hanced vegetation index,normalized water index,and normalized construction index),terrain features(elevation,slope,and aspect),and texture features(angular second moment,contrast,correlation,entropy,covariance,inverse differ-ence,and mean of sum)were used as influencing factors to construct and optimize the multidimensional feature model from three perspectives:spectral indices,terrain features,and texture features.Four machine learning algorithms-classification and regression trees(CART),random forest(RF),gradient boosting trees(GBT),and support vector machines(SVM)-were employed to automatically classify long-term landscape information in the Ebi Lake Basin and to compare the classification accuracy of the four algorithms to identify the most suitable landscape classification model for the region.The results showed:(1)The Google Earth Engine cloud computing platform efficiently processed,analyzed,and visualized landscape classification results for nearly 22 years in the study area.(2)The importance of the three feature types ranked from highest to lowest was as follows:spectral indices(31.52%),terrain features(17.18%),and texture features(3.83%).Among individual features,terrain features exhibited the highest classification accuracy.The combination of spectral features and texture features showed improved classification accuracy for the study area.Meanwhile,models that integrated spectral indices,terrain features,and texture features could effectively enhance the classification accuracy of the landscape model regardless of the feature combination used.(3)Among the four classification algorithms that fused spec-tral,terrain,and texture features,the classification accuracy from highest to lowest was as follows:random forest,gradient boosting trees,classification and regression trees,and support vector machines.The random forest algorithm achieved a to-tal classification accuracy of no less than 93%across all five periods(2002,2008,2013,2018,2023)with a Kappa co-efficient of no less than 0.92,making it the most suitable model for landscape classification in the Ebi Lake Basin.
Landscape classificationEcosystemEcological environmentEbi Lake watershed
万方数据
维普
