| Sandy desertification is the main form of land degradation in northern China,causing a significant impact on China’s economic development and environmental security. Monitoring the transformation from "non-sandyland" to "sandyland" within a certain time simply and rapidly, will be very important for the prevention of sandy desertification.This paper is based on the project of Geography Census of Gansu Province "Dynamic Monitoring of Oasisification and Desertification in Hexi Corridor", taking Minqin, a most typical region of desertification as a study area, and Landsat satellite images in 1994 and 2014 as the data source, this paper aimed at exploring a simple method to extract sandy desertification area directly through the research on method of automatically extracting sandy desertification area based on Landsat satellite images. Work carried out and conclusions drew in this paper are as follows:1. A rapid method of extracting desertification area based on difference principal component transformation had been explored.It was found that the positive and negative values of the first component of difference principal component transformation with a combined use of two phases Landsat satellite images, could represent the direction of change:the positive value represent desertification, while the negative value represents oasification. Based on the above discover and the combined utilization of difference principal component transformation and OTSU method, this paper proposed a rapid method of extracting desertification area based on multi-temporal Landsat satellite images,which was feasible and easy to use. Image contrast of many different regions showed that this method had a high accuracy.2.NDMI(Normalized Difference Moisture Index) was evaluated as the best indicator to reflect the characteristics of sandyland.Eight indicators, respectively NDMI (Normalized Difference Moisture Index), NDWI (Normalized Difference Water Index), NMDI (Normalized Multiband Drought Index), MSAVI (Modified Soil Adjusted Vegetation Index), Albedo, Mean of 3×3,5×5 and 7×7 window were quantitatively retrieved, and their suitability for one-time extraction of different types of sandyland were evaluated. Finally, NDMI was evaluated as the best indicatior of sandyland, with the best homogeneity in different types of sandyland and the best heterogeneity between sandyland and other landuse.3.A method of extracting sandyland based on multi-segmentations was proposed.Inpired by the principal of image binarization, this paper proposed a method of determining threshold value through multi-segmentation based on OTSU, and then extracted sandyland using this threshold. The advantage of this method was that it did not depend on the training samples, which was well known for having a great impact on classification, was very difficult to choose and would varies from different people. The results of experiment in Minqin showed that this method could effectively distinguish sandyland and abandoned land information, saline-alkaline land and gobi information, which were all very easily misclassified into sandyland. The overall accuracy was 82%, however, disatisfactory was that it was difficult to distinguish mountain, whose NDMI value was similar to sandyland.4. A easy method of identifying desertification area firstly and then extracting sandyland, and thus obtaining sandy desertification area directly was summerised.As we all know, sandy desertification area was a subset of desertification area, so sandy desertification area could be obtained though the intersection of desertification area and the sandyland. Through field investigation and image contrast, the experimental results showed that using this method to extract sandy desertification area in Minqin from 1994 to 2014 had a high accuracy, the sandy desertification area was very close to the objective reality, fully showed that this method had a good applicability. |
PDF Full Text Request