| Rocky desertification is considered as one of the long-term ecological and environmental problems in karst areas of southwest China.The exposed rate of bedrock and vegetation coverage rate are important indexes to evaluate the degree of rocky desertification.However,karst areas are characterized by complex terrain and broken ground features,and spectral variation and mixed pixel phenomenon are widespread.The commonly used methods of rock index,vegetation index and fixed end-member mixed pixel decomposition can’t estimate the exposed rate of bedrock and vegetation coverage rate well,and it is difficult to monitor rocky desertification in a long time series due to the limitation of data sources.In order to solve this problem,Landsat series multispectral images were used as data sources,and the extraction method of the end members of main ground objects and the estimation research of the bare rate of bedrock and vegetation coverage by the generalized linear mixed model（GLMM）considering spectral variation were carried out,and compared with the commonly used estimation methods of rocky desertification information.Finally,the grade of rocky desertification in the study area was evaluated and its temporal and spatial evolution was analyzed.The experimental results show that:（1）As for the extraction of ground object endmembers,the ground object endmembers extracted by vertex component analysis（VCA）are most similar to the corresponding ground reference endmembers,especially in the performance of bare soil spectrum extraction.The endmember spectra of ground objects extracted by each method are classified based on spectral angle classification,and the results are verified by accuracy.It is found that VCA-SAM has the best classification accuracy,with the overall accuracy reaching 95.94% and Kapaa coefficient reaching0.939,while the total accuracy of continuous maximum convex cone（SMACC）and orthogonal subspace projection（OSP）SAM classification is about 82%.The large error of bare soil classification results in the lower total accuracy of both methods than VCA.（2）The accuracy of bedrock exposed rate estimated by 2)GLMM is the best,with the total accuracy of 88.05% and Kappa coefficient of 0.845.The producer’s accuracy and user’s accuracy of different grades of bedrock rate are above 70%.However,the total accuracy of NDBI and SRI2,which perform well in other methods,is all lower than 66%,and they only perform well in areas where the exposed rate of bedrock is < 20% and >70%.By applying each method in different terrain scenes,it is found that GLMM is more stable than other methods,and the root mean square error is less than 0.093.Other methods,such as FCLSU and KBRI,have certain gaps in different scenes,and their accuracy is not high;（3）Comparing the vegetation coverage estimated by GLMM with NDVI,it is found that the total accuracy of vegetation coverage estimated by GLMM is 87.42%,while that of NDVI is 72.00%.GLMM is obviously better than NDVI.Compared with NDVI,GLMM is more stable in mapping accuracy and user accuracy of different grades of vegetation coverage,and the accuracy of most grades of vegetation coverage is higher than 87%.Although NDVI estimates the vegetation coverage at 51-70%and 71-100%,it will be on the high side in estimating areas with no vegetation coverage,low and medium vegetation coverage,and the effect is poor,with the lowest accuracy less than 25%.（4）Compared with most current methods,which rely on ground reference values for regression modeling estimation,GLMM’s direct unmixing accuracy is very close to that calculated by regression model without any ground reference values,and the difference between the estimated bare rate of different grades of bedrock and the accuracy of vegetation coverage is less than 5%.Moreover,GLMM’s estimation results can be consistent with the actual ground situation when using different time and data,and the overall stability is good.Therefore,GLMM has great potential in the estimation and application of rocky desertification information in karst areas,which can provide a method reference for the rapid and accurate evaluation of rocky desertification in this area.（5）Based on the information of rocky desertification extracted by GLMM method,the degree of rocky desertification in Weining section of Niulanjiang River Basin from 2000 to 2020 was evaluated,and its temporal and spatial evolution was analyzed.It was found that there were different degrees of rocky desertification in the study area,which were concentrated along rivers and valleys,and the closer to the valleys and valleys,the more serious the degree of rocky desertification was.In terms of time evolution,the area without rocky desertification changed the most,from 119.69 km~2 in 2000 to 349.97 km~2 in 2020.The proportion of potential rocky desertification was the highest among all rocky desertification grades,and the change was relatively stable.The areas of mild rocky desertification and moderate rocky desertification decreased steadily during this period,from the initial 155.68 km and 152.63 km to 117.45 km and 92.42 km respectively.However,the area of severe rocky desertification shows a trend of "decreasing-increasing-decreasing-increasing",but the area is generally decreasing.Judging from the changing direction and scale of rocky desertification,the changing direction of rocky desertification types is mainly from high-grade to low-grade rocky desertification,in which the area without rocky desertification changes the most,mainly from mild and potential rocky desertification,while other grades of rocky desertification mainly shift from severe to moderate,moderate to mild,mild to potential and no rocky desertification,and the area with rocky desertification decreases obviously.Overall,the ecological environment of the study area is gradually recovering. |
PDF Full Text Request