Zoom in

In the past 50 years, under the background of global climate change, with the increase of population and economic development, Eurasian grassland has been seriously degraded. One belt, one road surface, is a key indicator of grassland quality. Its spatial temporal pattern and distribution can directly reflect the degradation of grassland. Effective assessment of grassland quality is of great significance for the sustainable development of the countries along the border and the promotion of China's "one belt and one road" strategy. In previous studies, there is room for improvement in accuracy and accuracy of spatial and temporal distribution of soil properties. With the development of geographic information system, global positioning system, various sensors and soil mapping technology, digital soil mapping has gradually become an efficient method to express the spatial distribution of soil. Based on soil landscape science and spatial autocorrelation theory, this study combined multi-source sample data and environmental covariate data, and used machine learning model to predict the spatial distribution of surface soil attributes of warm grassland in Eurasia around 2000. In order to solve the problem of soil sample standardization, the equal area spline function was used to fit the soil properties of different profiles to the soil properties of 20 cm in the surface layer, and the soil particle distribution parameter model was used to transform the classification standards of different soil textures into the United States system. In order to solve the problem of insufficient number of soil samples, pseudo expert observation points were used to supplement soil organic matter and sand content samples in under sampling area; stepwise regression combined with support vector machine model was used, and effective soil bulk density simulation samples were screened by calculating threshold. According to the characteristics of complex terrain and climate conditions, combined with multi-source remote sensing data, ngboost model is applied to mine the relationship between soil attributes and environmental landscape factors (topography, climate, vegetation, soil type, etc.) and spatial location based on sample points, and to predict soil organic matter, sand content and bulk density in the study area from 1980 to 1999 and 2000 to 2019 respectively, and the uncertainty of corresponding indicators is given Spatial distribution of sex. The spatial distribution trend of the simulated soil attribute indexes is consistent with the actual situation. Before 2000, the soil organic matter content, bulk density and sand content were 0.64, 0.35 and 0.44 respectively, and the RMSE were 0.25, 0.07 and 13.94 respectively; after 2000, the RMSE were 0.79, 0.77 and 0.86 respectively, and the RMSE were 0.2, 0.13 and 6.61 respectively. The results show that this method can effectively retrieve the soil physical and chemical properties of temperate grassland in Eurasia, and provide a basis for the evaluation of grassland degradation and the construction of grassland quality evaluation system.