基于多源数据的天山北坡典型草地植被覆盖度遥感反演与变化特征研究OA

[Objective]This study aimed to construct remote sensing inversion models for typical grassland vegetation coverage based on multi-source remote sensing data including UAV imagery,Landsat 8 OLI,and Sentinel-2.[Methods]Combining vegetation indices such as NDVI,MSAVI,RVI and PVI,the inversion models were optimized using pixel binary model,generalized linear regression,and random forest regression method.[Results]For the inversion accuracy in non-growing season:the MSAVI regression model achieved an R² of 0.78 with an RMSE of 6.4%.After random forest optimization,the R² of the ensemble model(MSAVI+PVI)increased to 0.83 and RMSE dropped to 5.8%,significantly enhancing the identification capability for low-coverage grasslands.For the comparison of inversion accuracy in growth season:Sentinel-2 data outperformed Landsat 8 OLI in accuracy,with its random forest model achieving an R² of 0.821 versus Landsat 8 OLI's 0.794.Support Vector Machine(SVM)model demonstrated optimal performance,with SVM-RF(random forest feature selection)achieving and R² of 0.856 and an RMSE of 4.2%on Sentinel-2 data,a 12.3%improvement over the traditional method.For the spatio-temporal variation characteristics:the grassland coverage showed an overall decline from 2019 to 2024,with high-coverage areas(>60%)decreasing by 23.5%,low-coverage areas(25%to 35%)increasing by 18.7%,and medium-coverage areas(40%to 50%)rising by 11.2%.[Conclusion]The inversion accuracy of grassland coverage in arid regions can be enhanced by multi-source data fusion and machine learning algorithms.Data support was provided by the study for ecological monitoring and degradation management.