基于Poisson对数线性模型的居民点与地理因子的相关性研究

Study on Correlation between Residential Points and Geographical Factors based on Poisson Logarithm Linearity Model

在基于图斑的地理因子库基础上,通过样本采样、数据预处理、建立Poisson对数线性模型、模型估计、统计检验和假设检验等一系列处理过程,研究全国范围内居民点个数与地貌类型、表土质地、高程带、土地利用类型、年降水量和平均气温等地理因子的相关关系,定量地揭示地理气候条件对居民点分布的影响。从而对挖掘具有特定地理气候特征的地理单元内居民点分布的规律,进而推演目标区域内居民点分布的特性并估算该区域内居民点个数打下理论和数据基础。本文是地理学与统计学交叉研究。运用本文的结论,结合不同人口或经济发展水平等级的居民地的研究,将对区域内人口、资源与环境的协调发展做出贡献。

In this paper, based on map patches of geographical factors in China, Poisson logarithm linearity model is applied in studying the correlation between distribution of residential points and geographical factors. Here, map patches are defined as such regions that are the outcome after intersected by several polygons representing certain type or class of geographical factors, and geographical factors include physiognomic types, soil types and land - use types, altitude belt classes, annual rainfall classes and average air temperature classes and so on in the area of China. Because of Distribution particularity of residential points is found in map patches. On the one hamd, from all of the map patches, mere then 98 percents have few residential points in them. On the other hand, in this paper, residential points are belonging to countable data. The two points makes the authors apply Poisson logarithm linearity model. Just like other models, before building the model, several processes are performed. Sampling and pretreatment of data are necessary. Sometimes certain variables ought to be eliminated from the model for high relativity with other variables. And statistical proof- test and hypothesis proof- test are also essential, which will testify the validity of the model. Certainly, the model fulfilled all of the proof- tests. In other words, the model is valid. From the analysis, several important results are given. Among the last five factors, annual rainfall is the most important factor which affects the distribution of residential points. In order to turn the amount of annual rainfall into class data, they are classified into 9 levels ranging from 1 to 9, and symbol 1 represents low rainfall and 9 denotes high rainfall. With the increase of annual rainfall by levels, the amount of residential points also increases. Another important conclusion is that if the other factors are kept invariable, with the amount of annual rainfall jumping one level, the number of residential points will raise several times. The second important factor to affeet residential points is altitude belt. From the coefficient of - 1. 1333, we can draw the conclusion that with the heightening of altitude, the distribution of residential points will be less and less. And the number of residential points will drop many times when altitude belt adds one level in keeping other variables unchanged. Among the five factors,the area of map patches is the most unimportant one, and its effect is too small to be ignored compared to the other four factors. Just because of this, we could carry out this research based on map patches instead of kilometer grids. Then only with 1.5 percent data, the same result can be obtained. The influence of geographical conditions put to the distribution of residential points is confirmed quantificationally. It will ground by theory and data in mining the laws of distribution rules of residential points in geographical cells with special geographical characteristics, then deducing the distributive specialization of residential points in n objective region, and estimating the number of residential points in this region.