NIU Zengyi, DING Jianli, LI Yanhua, WANG Shuang, WANG Lu, MA Chengxia
Soil salinization, an important component of the land degradation forms, usually appears in the arid and semi-arid regions where the climate is drought, soil evaporation is very fierce, and the water table is high and contains rich soluble salts. Soil salinization has impeded the development of oasis agriculture in Xinjiang due to its severe effects on agricultural productivity and sustainable development. In order to understand the threat level of soil salinization in oasis agriculture, and ensure the sustainable development of oasis agriculture in the arid and semi-arid regions, it is very necessary to study the method of soil salinization monitoring. At present, remote sensing technology has been widely used to monitor soil salinization and is very useful. However, monitoring soil salinization is mainly based on the low-resolution satellite images, which is not enough for monitoring soil salinization in details. In this study, the domestic GF-1 PMS image was adopted to extract the salinization information based on the advanced object-oriented method. First, fractal net evolution approach was used to segment image and build classification rules for salinization information extraction. Then, by using the maximum likelihood method, soil salinization information was extracted from the domestic GF-1 PMS image and Landsat OLI image in the same region, respectively. Finally, the results of two different methods and different latest sensors for salinization information extraction were compared. The results show as follows: (1) The overall accuracy of object-oriented method for salinization information extraction based on GF-1 PMS image is 92.94% and the kappa coefficient is 0.91. The overall accuracy of maximum likelihood method for salinization information extraction based on GF-1 PMS image is 87.78% and the kappa coefficient is 0.77. Compared with the maximum likelihood method, object-oriented method is better for soil salinization extraction based on GF-1 PMS image and the accuracy is overall improved by 5 percentage points. This illustrates that the object-oriented method is more suitable for GF-1 PMS image when monitoring soil salinization. In addition, object-oriented method can make full use of the relationship between pixels through scale segmentation technology, and can more fully utilize the information contained in the image to improve the accuracy of salinization extraction from high-resolution remote sensing image information. (2) The overall accuracy of Landsat8 OLI image for soil salinization extraction is only 63.47%. Compared with Landsat OLI image, the overall accuracy of GF-1 PMS image is improved by 30 percentage points. The ability of GF-1 image for soil salinization extraction is stronger than that of Landsat image. We can extract the vegetation that is affected by soil salinization, which is meaningful for study of agricultural field scale salinization. This illustrates that GF-1 images has great potential for salinization monitoring on agricultural field scale. In this study, we use domestic GF-1 image to extract soil salinization information based on the advanced object-oriented method for the first time. The result is positive. This shows domestic GF-1 image can be one of data sources to monitor soil salinization in the arid and semi-arid regions.