| Rocky desertification is a special desertification landscape formed by social and natural factors in the karst area,and it is also a complex dynamic process of land surface.Rocky desertification will decrease the stability of the regional ecosystem,and even lose the value of agricultural use,which seriously restricts the sustainable development of local ecology and the social economy.It has become the primary ecological problem in southwest China.To grasp the information of rocky desertification quickly and accurately is the premise of carrying out the control of rocky desertification.Remote sensing images have grown up to be important data source for monitoring rocky desertification due to its advantages of high temporal resolution,low cost and large regional coverage.However,researchers at home and abroad mainly utilized singlephase remote sensing data in low rainfall season to extract the extent of rocky desertification in a region,and fewer studies consider using multi-phase data in low rainfall season.Therefore,this paper took the key control engineering area of typical karst rocky desertification in Southeast Yunnan as the research object,and used Sentinel-2 L1 C remote sensing images in 2020 as the data source to construct remote sensing time series data of the degree of rocky desertification.The FastTime Weight Dynamic Time Warping（FastTWDTW）algorithm was used to identify the degree of rocky desertification in the study area（extremely severe,severe,moderate,mild,potential and none）,and the extraction results were compared and analyzed with other methods.The research results are as following:（1）The two constructed remote sensing time series data of rocky desertification degree（EVI-NDRI2 and PC1）can effectively represent rocky desertification degree information.In rocky desertification areas,the EVI index ignores non-vegetation information,and NDRI2 easily ignores vegetation information.Therefore,the EVINDRI2 rocky desertification degree index is constructed by integrating vegetation information and non-vegetation information.Negative correlation,that is,the smaller the EVI-NDRI2 index value,the more severe of rocky desertification.Rocky desertification has the characteristics of high brightness.According to the size and shape of the feature vector,Principal Component Analysis（PCA）can predict that the target land cover is a dark pixel or a bright target object in the corresponding PCA image,and the first principal component（PC1）has the highest contribution rate,so choose PC1 as another feature of rocky desertification,and the results show that PC1 is positively correlated with the degree of rocky desertification,that is,the greater the value of PC1,the more severe the degree of rocky desertification.J-M distance was used to analyze the effect of time series data of rocky desertification characteristics formed by different time-phase combinations on the separability of rocky desertification degree.Data of 8 time phases were used to construct EVI-NDRI2 remote sensing time series data of rocky desertification degree and PC1 time series index.（2）The improved sequential similarity algorithm not only improves the matching between sequential data,but also improves the efficiency of the algorithm.On the basis of DTW algorithm,time weight,constraint and data abstraction are combined to obtain FastTWDTW algorithm.Compared with DTW algorithm,FastTWDTW algorithm improves the similarity of matching between remote sensing time series data of rocky desertification degree,helps to correctly classify rocky desertification degree,and effectively improves the efficiency of algorithm,which only takes 1/6 of DTW algorithm.（3）Based on the FastTWDTW algorithm,the information extraction indicators such as rocky desertification identification and degree were optimized.The classification accuracy of rocky desertification obtained by the combination of EVINDRI2 and PC1 based on the FastTWDTW algorithm is 85.68%,and the Kappa coefficient is 0.81.The classification accuracy is better than other rocky desertification characteristic time series data,CART and comprehensive analysis methods.The comparison methods were divided into two categories.One was based on the FastTWDTW algorithm,which compared different time series indices of rocky desertification characteristics（the combination of EVI-NDRI2 and PC1,EVI-NDRI2,EVI,and NDRI2）to extract the effect of rocky desertification,that is,multi-temporal phase.Compared with the multi-temporal comparison,the result of rocky desertification extracted by the combination of EVI-NDRI2 and PC1 is better than the extraction results of EVI-NDRI2,NDRI2 and EVI,indicating that PC1 feature is sensitive to rocky desertification information,and the combination of EVI-NDRI2 and PC1,it can effectively improve the extraction accuracy of rocky desertification.The second category was to compare the classification results obtained by the combination of EVI-NDRI2 and PC1 based on the FastTWDTW algorithm with the commonly used rocky desertification extraction methods（the rocky desertification extracted by CART and comprehensive analysis methods are all single-phase results）,that is,multitemporal and in the comparison between single temporal phases,the classification results extracted by the combination of EVI-NDRI2 and PC1 are better than CART and comprehensive analysis methods,that is,the classification result of rocky desertification degree based on time series data is better than those extracted from single temporal data results.The comparison of classification methods fully reflects that the combination of EVI-NDRI2 and PC1 time series data has advantages in the extraction of rocky desertification types. |
