Identification and Analysis of Industrial Land in China Based on the Point of Interest Data and Random Forest Model

Abstract

The purpose of this study was to provide a new concept and technical method for the large-scale identification of industrial land and analyze the distribution characteristics of industrial land in China. The following research methods are employed using the point of interest data and random forest model based on data accessibility, this study selected 2015 data on Wuhan and Luoyang as training samples to identify the industrial land of China. Then, the proportion of industrial land in all 334 prefecture-level cities on the Chinese mainland was calculated, and the spatial pattern was analyzed. The results show that: 1) by comparing multiple experiments and robustness analysis, the optimal parameter setting of the random forest model is obtained. According to the test of actual industrial land distribution in Wuhan city and Luoyang city, the identification of industrial land in different scale cities by random forest model is accurate and effective. 2) From the perspective of spatial patterns, industrial land shows a “large aggregation and small scattering” distribution. 3) From the perspective of spatial distribution, the proportion of industrial land in these cities shows spatial aggregation. High–high aggregation areas were mainly distributed in North and Northeast China, and low–low aggregation areas were mainly located in West China. 4) From the perspective of related factors, industrial land was close to rivers, highways, and railway stations and had a relatively low correlation with the distribution of airports. Industrial land was located within approximately 10–60 km distance from the municipal government office. In terms of the proportion of industrial land, the proportion of industrial land is higher in the cities where the industrial land was closer to railway stations. However, when the industrial land in cities was closer to four other types of related factors (waters and lakes, major highways, airports, and municipal government stations), the share of industrial land is lower. In conclusion, the method based on the point of interest data and random forest model can accurately and effectively identify large-scale industrial land.