Quantitative Retrieval Of Vegetation Cover For Kangbao County-A Desert Area Based On Remote Sensing

Desertification not only is of a major environmental concern in the world, but also has caused most serious ecological problems in China. Remotely sensed images that provide multi-temporal and real-time large coverages of the Earth surface have gradually become the primary source data and a powerful tool used for monitoring and assessing the dynamics of land desertification. Accurately mapping spatial distribution of vegetation and understanding its change for desert areas is critical for desertification monitoring and assessment. As an important parameter to quantify the vegetation coverage information of the Earth surface, the vegetation cover percentage is a direct and effective indicator for monitoring and assessing the degree of land desertification.In this study, Kangbao County of Hebei Province was selected as the study area to monitor and assess desertification and its dynamics by generating spatially explicit estimates of vegetation cover percentage. The monitoring and assessment were conducted by collecting and combining ground-based digital photos,2014’s Landsat 8 images and field measurements of vegetation cover percentage from sample plots. The spatial distribution of vegetation cover percentage was generated using both ground-based digital photos and field measurements. Various vegetation indices were derived and utilized. The used spatial interpolation methods included multivariate linear stepwise regression, linear spectral mixture analysis and k nearest neighbors (kNN) algorithm. By comparing the results from the methods, following conclusions were drawn:(1) Estimation of vegetation cover percentage using ground based digital photosKangbao County was characterized by sparse vegetation cover and complex topographic features. The use of ground based digital photos by the combination of computer automatic classification and manual visual interpretation led to a coefficient, 0.7884, of determination between the estimated and observed values, implying accurate estimation of vegetation cover percentage. This method has the features of quickly obtaining information, time- and labor-saving, and improved the retrieval of vegetation cover information.(2) Estimation of vegetation cover percentage using multivariate linear stepwise regressionWhen the selected vegetation indices were separately used to develop linear regression models of vegetation cover percentage, NDVI and SAVI had the most significant contributions (R2=0.604), and soil adjusted factor L showed insignificant effect on the estimation of vegetation cover percentage in the areas of Kangbao County in which there were low vegetation cover percentage and limited soil types, and slight effects in the areas with moderate vegetation cover percentage. Multivariate linear stepwise regression resulted in the selection of three vegetation indices including SAVI0.5, SRN-R, EVI and a determination coefficient of 0.7193 and root mean square error (RMSE) of 0.2416 and accuracy of 86.69%.(3) Estimation of vegetation cover percentage using linear spectral mixture analysisBecause of sparse vegetation cover, and a lot of mixed pixels in Kangbao County, it was difficult to find purity pixels of endmembers. Geometric vertex endmember selection method was superior to the pure pixel index (PPI) endmember selection method. Although accuracy of estimation of vegetation cover percentage using linear spectral mixture analysis was inferior to that of using multivariate linear stepwise regression, this method is acceptable. Since it does not rely on the collection of field data, is easy to use and low cost, and can be regarded as a practical and feasible method for desertification monitoring.(4) Estimation of vegetation cover percentage using kNN algorithmThe significant vegetation indices obtained from multivariate stepwise regression analysis were used as spectral space variables of kNN algorithm to measure spectral distance of each unknown location to sample plots and estimate its vegetation cover percentage. It was found that as the number k of nearest neighbors increased, the RMSE and average error decreased gradually and the overall estimation accuracy increased eventually, which, to some extent, reduced the impact of similar spectra for different vegetation types on the estimation accuracy in the desert area, as k take 10, the highest prediction accuracy is up to 73.67%. Therefore, kNN algorithm is a promising method to estimate vegetation cover percentage in the desert area.(5) Comprehensive evaluation and analysis of results from methodsWhen the vegetation fraction from linear spectral mixture analysis was directly added into the set of the independent variables obtained after multivariate linear stepwise regression, or used together with all the vegetation indices to conduct multivariate linear stepwise regression, the estimation accuracy of vegetation cover percentage, to some extent, increased. The increase was 5.7% for multivariate linear stepwise regression and 0.24% for kNN algorithm. Therefore, the result of linear spectral mixture analysis, vegetation fraction which is the sensitive indicator of estimation of vegetation cover percentage, deserves effective utilization.