基于空间统计和多元统计的耕地影响因素及回归模型研究——以重庆市石柱县为例

目的]通过空间自相关和多元回归分析,揭示耕地空间分布规律,为土地开发复垦及整理提供快速的评价方法。方法]以耕地面积占比为空间变量,运用空间自相关及马塞克图分析耕地分布整体特征,通过距离、地形、NDWI和人口密度共9个因素对耕地空间分布进行多元回归分析,模拟耕地分布适宜性并进行了检验。结果]空间自相关分析结果表明,距离和地形因素对耕地空间分布具有显著影响,空间自相关分析Moran’s I值为0.701 5,研究区耕地分布主要为不显著、LL(低空间自相关)和HH(高空间自相关)类型,其中不显著类型占研究区总面积的65%以上;基于多远回归分析结果表明:回归模型具有较高拟合优度和可靠性(R~2=0.846),模拟得到的耕地分布适宜性图与现有耕地分布基本吻合。结论]研究区耕地空间分布总体上呈现较强的正相关关系,且受距离、地形因素影响明显;回归模型能够较好地揭示研究区耕地空间分布规律;研究区具有一定耕地补充潜力;将回归模型应用于土地开发复垦以及整理工作中,有利于提高补充耕地质量,减弱水土流失以及优化区域土地结构。

A Study on Influencing Factors of Cultivated Land Based on Multivariate Regression and Spatial Statistics-A Case Study of Shizhu County, Chongqing City

Objective] We aimed to reveal the regularity of spatial distribution of cultivated land by spatial autocorrelation analysis and multivariable regression, in order to provide a rapid evaluation method for land development, reclamation, and consolidation. Methods] Coverage ratio by cultivated land was as response variable, and methods of spatial autocorrelation and mosaic plot were utilized to demonstrate its spatial pattern. Nine factors such as Euclidean distances, terrain, NDWI, population density, simulated cultivated land distribution suitability, etc. were used as independent variables, and multivariate regression of them with the response variable was conducted to test the distributional suitability of cultivated land. Results] The Euclidean distances and terrain have significant impacts on the spatial distribution of cultivated land, and the Moran''s I index is 0.701 5. In addition, the main types of local indicators of spatial association(LISA) distribution are not significant. L-L(low spatial autocorrelations) and H-H(high spatial autocorrelations) and insignificant types are three of the main types, especially the third type covered over 65% of study area. Multivariate regression behaved well in the distribution suitability simulation of cultivated land, it was remarkably coincided with the present distribution of cultivated land. The regression model was testified reliable and had goodness of fit (R²=0.846). Conclusion] (1) The spatial distribution of cultivated land in the study area generally exhibits a strong positive correlation. And the distribution of cultivated land is affected by distance and terrain significantly. (2) The regression model can well reveal the spatial distribution of cultivated land in the study area, showing that the study area has a potential for cultivated land supplement. (3) We can improve the quality of additional cultivated land, reduce soil erosion, and optimize the land utilization structure if under the guidance of the regression model for land development, reclamation and consolidation.