Study On The Regional Distribution Of Soil Organic Matter Based On Spatial Regression Model

As ecotone in terrestrial ecosystem, the accumulation and changes of soil organic matter in plain-hills transition belt have been attracting increasingly concerns for its influences on soil productivity and global climate change. Investigations on the spatial distribution characteristic of soil organic matter could shed light on acquiring detailed soil nutrition information, and they could also pave the way for understanding the balance of the carbon budget in terrestrial ecosystem. Soil organic matter has a strong spatial dependence and heterogeneity, and its spatial pattern was always affected by climate, terrain, soil parent material, soil types, land use et al. For the purpose of promoting the level of managing resource and environmental information by a quantitative and automatic way, spatial regression model was reasonable used to model the spatial distribution of soil organic matter.In order to improve the accuracy of predicting the spatial distribution of soil organic matter, in this paper, relevant methods and theories from soil-landscape models, statistical and spatial statistical models were employed. Firstly, the spatial characteristics of soil organic matter in plain-hills transition belt were studied. Secondly, the spatial distribution of soil organic matter in plain-hills transition belt was predicted mapped. Last, influence scopes of soil types and land use were determined. In the paper, some significant results were shown as follows:1. Local spatial clusters of soil organic matter in low relief area, hills area and the transition belt were analyzed. Consequently, significant high-low and low-high clusters were detected on the plain-hills transition belt, which could be used to feature the conversions on environmental gradient. Features and capabilities in predicting the spatial distribution of soil organic matter were compared between spatial autoregressive model and geographically weighted regression model. Both residuals presented weak spatial autocorrelation, but geographically weighted regression residuals were smaller. Basically, residuals of geographically weighted regression model shown non-existent of spatial autocorrelation at the distance of 800 m, whilst spatial autocorrelation regression residuals still showed a weak autocorrelation at the distance of 1km. Plus, each residual generated by the above models expressed no spatial at the distance of 5km.2. Topographical factors, as auxiliary, were employed to construct regression kriging model and geographically weighted regression model, results shown as follows:(1) the globally scope of soil organic matter ranged from 3.80 g/kg to 69.40 g / kg. Coefficient of variation was 39.59%, which represents a moderate degree of variation. Interpolation accuracies of regression and geographically weighted regression kriging were increased by 25.84% and 27.61% than ordinary kriging, respectively.(2)The ranges of soil organic matter estimated by both models were underestimated. Additionally, accumulation of soil organic matter was detected in key transition belt.(3) In the case with the presence of spatial autocorrelation in residuals, polygons of soil organic matter generated by geographically weighted regression kriging model shown richer details than the others.3. Differences between soil types and land use degree were quantified, and both have significant correlations with soil organic matter. Variabilities of soil organic matter differed in different soil types. Meanwhile, high-high cluster of land use degree was noted in the low-relief area. By constructing geographically weighted regression model, the extent of soil types and land use degree as dominant controls on variability of soil organic matter were determined, this indicated that soil types exert significant influence globally, while land use degree only paid in the low-relief area.