Vegetation Information Extraction In Desert Areas Based On Rotation Forest Algorithm: A Case Study Of Mu Us Sandy Land

Vegetation is an important part of terrestrial ecosystems.Vegetation can maintain water and soil and protect biodiversity in the region.Vegetation in desert areas is closely related to global climate change and environmental studies.Vegetation information extraction in desert areas can objectively reflect the vegetation types and vegetation growth conditions in the region,and scientifically evaluate the ecological restoration effects of desertification areas,study the carbon cycle process of desert ecosystems,and realize the healthy development of desert ecosystems.It is of great significance.In this paper,using the rotation forest algorithm,multi-spectral remote sensing imagery and topographic data were used to calculate 98 features which are spectral features,texture features or topographic features and vegetation information extraction was completed in the desert area with Mu Us sandy land as an example.;based on the accuracy of vegetation information extraction,the classification method used in this paper and the existing classification methods in the study area are compared,and the rotation forest algorithm with higher classification performance was integrated into ENVI based on Interactive Data Language(IDL);finally,the spatial distribution of vegetation types in the study area was analyzed.The following conclusions were obtained through research:(1)When the decision tree "minimum node to be classified" is the number of 5% average category samples,the rotation forest algorithm can prevent the model from over-fitting,improve the training efficiency of the model,and ultimately improve the performance of the model.In addition,when the number of decision trees in the model is 10 and the number of features in the feature subset is 3,the model has been able to meet the accuracy requirements of remote sensing image classification.Therefore,for this study area,the above parameters are set to the optimal combination of parameters of the rotating forest algorithm.(2)In the application of vegetation remote sensing information extraction in desert areas,the rotation forest algorithm can meet the accuracy requirements of remote sensing image information extraction,and its accuracy is better than the existing hierarchical classification and multi-index combined remote sensing information extraction method.In addition,the vegetation information extraction method based on the rotation forest algorithm is simpler than the hierarchical information classification method combined with multiple indicators.The subjective factors introduced in the classification are less,and the classification results are more credible.Therefore,the vegetation information extraction method based on the rotation forest algorithm is more suitable for the extraction of vegetation information in desert areas.(3)The spatial distribution of vegetation types in each county-level unit in the study area is quite different.The vegetation coverage in the core area of Mu Us Sandy Land and the surrounding area of Mu Us Sanda is quite different.Specifically,the three Inner Mongolia county in the core area of Mu Us Sandy Land have lower vegetation coverage and simple vegetation types.The five counties of Shaanxi Province and the two counties of Ningxia in Mu Us Sandy Land have high vegetation coverage and diverse vegetation types.In summary,the vegetation information extraction based on the rotation forest algorithm can quickly and accurately obtain the vegetation distribution in the desert area,which can provide a basis and support for further study of the hydrothermal pattern and the impact of land use types on the local ecological environment.The extraction of vegetation information in desert areas provide basis for the ecological environment research in the region and global climate change research.