| According to the fourth national monitoring of desertification and sandy land, the phenomenon of desertification and sandification in our country have been controled preliminary, but some regions still keep deteriorating and expanding. In addition, technology for sandy land prevention is in the low level, and technology support is not enough, the problems need urgently to be solved in the sand prevention project at present. Remote sensing technology has played a positive and effective role in the sandy land evaluation and monitoring, but the traditional classification of visual interpretation method was mainly used in the past work, this method needs a lot of manpower and material resources, and its efficiency is relative lower, especially for ground object recognition and classification in the large area. In addition, the transition sandy land is difficult to extract because of the influence of vegetation cover, and its boundary is difficult to define exactly. These problems cann’t be solved by traditional classification methods, so that it is neccessary to further explore quantitative method for sandy land evaluation and monitoring based on remote sensing technology, it will be more conducive to the well-off implementation of sand prevention engineering in our country.Against the problems such as transitional sandy land is difficult to extract, its boundary is hard to define exactly, efficiency of qualitative classification is low, and evaluation index for sandy land is differ, taking Zhenglan Banner in Xilingol league of Inner Mongolia as the study area, and GF-1 and Landsat TM as the main sources of remote sensing data, combining with the ground survey data, this study adopted the decomposition of mixed pixels and soil particle size threshold methods to extract sandy land information quantitatively, and analyzed the changes of vegetation coverage between each month, and determined the best vegetation coverage period for sandy land quantitative evaluation, and monitored sandy land dynamic change in the study area for ten years. The conclusions of this study as follows:(1) Endmumber selection method of Pure Pixel Index(PPI) is better than the Sequential Maximum Angle Convex Cone(SMACC) and geometric vertex of scatterplot. The results showed that if the number of endmumber was greater than 6, it was prone to generate noise and error, if the number was less than 5, the mixed pixel couldn’t be decomposed effectively, so that when the endmumber number was 5 or 6, the decomposition result would be more accurate in the process of sandy land extraction. If sand abundance accounted for more than 50% in the remaining endmumber abundance except for vegetation, or if it took up less than 50% but it was the maximum in the pixel, the pixels would be determined as sandy land. It was verified that the extraction precision of this method was higher, the total accuracy was 86.42%, and the transitional sandy land with high vegetation coverage could be also extracted accurately. This method can solve the effect of vegetation effectively, and define the sand boundary accurately.(2) The composition content of soil particle size obeyed the normal distribution better, silt content was verified by T test with the most significant difference in sandy and unsand land, silt content in the unsandy land was approximately three or four times more than that of sandy land, it was the best indicator for sandy land, clay took the second place. Based on Partial Least Squares Regression(PLSR) analysis and modeling, the model effect weight of silt was the largest, and silt had a high correlation with the independent variables, Rb3 and Rb4 were determined as the independent variable for modeling, because their VIP value were greater than 1. By Leave One Out(LOO) validation, the model error PRESS was the minimum when the independent variable number was 2. As the results shown that silt content threshold 3.5% was reasonable, which meant that when the silt content was less than 3.5%, the pixel was divided into sand land, and if silt content was greater than 3.5%, it was divided into the unsandy land. Total accuracy of sandy land extraction based on this method was 80.86%, the result was relatively accurate, but it was second to the decomposition of mixed pixels. Compared with mixed pixels decomposition, this method was slightly insufficient to solve the vegetation effects.(3) In the process of vegetation coverage extraction, mixed pixel decomposition with multi-endmumber was better than dimidiate pixel model and Soil Adjust Vegetation Index(SAVI). After entering the growing season, vegetation coverage gradually increased with the time, and the maximum coverage appeared from the end of July to the beginning of August, then it declined with different velocity. Vegetation in sandy land kept growing with a gently speed from the end of may to the middle of July, but the annual ephemeral plant in sandy land growed rapidly in the last ten-day of July because of the influence of precipitation until the first ten-days of August. Differences of annual precipitation conditions leads to a great change of vegetation coverage in sandy land, it needs to eliminate the instable influence of ephemeral plant to evaluate sandy land and monitor its dynamic change in different year. It was thought that vegetation coverage in the beginning of August is more reasonable to evaluate sandy land.(4) The total areas of sand land in Zhenglan Banner were 5524 km2 and 4109 km2 respectively in 2005 and 2014, reduced by 25.6%, the reduced sandy land accounted for approximately 13.9% of Zhenglan Banner area. In both two years, the area of fixed sandy land was the most, the area of semi-fixed sandy land was second to it, and the shifting sandy land area was the smallest. After 10 years’ management and prevention, fixed and semi-fixed sandy lands and shifting sandy land were reduced by 22.1%, 4.7% and 31.1% respectively, the declined level of shifting sandy land was the largest. Under the condition of improvement for sandy land management, there were still some areas with an sandy degree increase. In the future work, it needs to strengthen protection in the key regions, and to further improve the ecological and social benefits of the sand prevention engineering.In the past researches on the sandy land extraction based on the decomposition of mixed pixels, most of the studies only achieved the sandy land abundance by decomposing the mixed pixel, but did not do further research on sand abundance. The innovation point of this research was to further analyze the relationship between the sand abundance and other endmumber abundance. This study defined a reasonable threshold for sandy land abundance, and proposed a quantitative method for sandy land extraction based on remote sensing tecnology, which solved the influence of vegetation cover effectively, distinguished the transitional sandy land accurately, and defined the sand boundary accurately. In addition, the effective indicator for sandy land was put forward through the analysis of soil particle size composition differences in the sandy and non-sandy land, and a reasonable threshold of the silt content was defined to extract sandy land quantitatively, the result had proved that the method was valid and feasible. This study can improve the efficiency and provide referable basis and technical support for the future sandy land evaluation and dynamic monitoring, and also can provide service and decision-making basis for relevant departments to carry out sandy land prevention work, and make the quantitative extraction, evaluation and dynamic monitoring for sandy land enter a new stage at some extent. |
PDF Full Text Request