Extracting desertification from Landsat TM imagery based on spectral mixture analysis and Albedo-Vegetation feature space

Land desertification has been a worldwide environmental problem. Desertification monitoring and evaluation are very important content in desertification context. Scientific and accurate evaluation of desertification can provide scientific basis for decision making in mitigating desertification. Because of the advantage of large amount of information, short cycle and broad scope of data, less restrictions on the human and material resources and so on, remote sensing has become an important technology to monitor land desertification in the past 30 years. Desertification is the most typical and serious form of desertification in China, especially in the oasis zone distributed along inland rivers or in the lower reaches of inland rivers in northwestern China. Quantitative evaluation of the current desertification remote sensing methods used is mostly obtained through the vegetation index and vegetation cover, to gain information on the extent of desertification. As the arid and semiarid sparse vegetation cover, soil and soil moisture on the most common vegetation index have a greater effect. First, based on the spectral mixture analysis model, three kinds of endmember consisting of vegetation, water and bare soil were selected. The image dimensionality was reduced by the minimum noise fraction (MNF). The pixel purity index transformation was used to narrow the range of the endmember. On the scatter plot of MNF, three kinds of endmember were selected, and relative abundance distribution of each component was obtained by using linear spectral mixture model. Second, a spectral feature space composed of vegetation component and land surface albedo retrieved from Landsat TM Imagery was constructed to evaluate desertification present condition and degree quantificationally. Last, an empirical study was carried out taking the middle reaches of Heihe River as an example. Results indicated that this method makes full use of multi-dimensional remote sensing information, reflecting the desertification land cover, water, thermal environment and its changes, with a clear biophysical significance, and the index is simple, easy to obtain, high precision, and is conducive to quantitative analysis, monitoring and desertification assessment of desertification. It was rather ideal to assess desertification on the basis of Albedo-Vegetation feature space: correct prediction proportion of testing samples reached 90.3 %. This method was beneficial to the desertification quantitative analysis and monitoring with the characteristics of simple index, easy accessibility and high accuracy. Copyright Springer Science+Business Media Dordrecht 2013