基于随机森林模型的“生产-生活-生态”空间识别及时空演变分析——以郑州市为例

基于2007年和2017年郑州市POI数据,采用随机森林模型和样方比例法识别其城市内部的“生产-生活-生态”空间,并利用核密度等方法分析研究区“生产-生活-生态”空间的时空演变格局。结果表明：① 随机森林作为新兴的机器学习算法,能够识别“生产-生活-生态”空间且具有较高的精度。② 郑州市“生产-生活-生态”空间分布格局与城市功能分区相匹配,生产空间集聚分布在产业集聚区,生活空间在城市中心城区内呈面状分布,生态空间整体呈点状分布。③ 随着郑州市城镇化建设和基础设施的完善,10年间郑州市“生产-生活-生态”空间的空间分布格局更加合理,生产空间向产业集聚区集聚,生活空间逐渐分散,生态空间分布更加均衡。基于POI数据,利用随机森林模型对城市“生产-生活-生态”空间的识别方法更加有效,识别结果更加精准,能够在更小的尺度上为国土空间规划提供数据与方法支撑。

Recognition and spatio-temporal evolution analysis of production-living-ecological spaces based on the random forest model: A case study of Zhengzhou city,China

Based on the POI data of Zhengzhou city in 2007 and 2017, the “production-living-ecological” spaces within the city was identified by using random forest model and quadrat proportion method, and the spatial-temporal evolution of “production-living-ecological” spaces in the study area was examined by using nuclear density and other methods. The results show that: First, as a new machine learning algorithm, random forest model can identify “production-living-ecological” spaces with high accuracy. Second, the spatial distribution pattern of “production-living-ecological” spaces in Zhengzhou matched with the urban functional zoning. The production space was concentrated in the industrial agglomeration area, the living space was located in the central urban area with a plane shape, and the ecological space was distributed in a scatter pattern as a whole. Finally, with the development of urbanization and the improvement of infrastructure in Zhengzhou, the spatial distribution pattern of “production-living-ecological” spaces in the city was more reasonable in the past 10 years. The production space was concentrated in the industrial agglomeration area, the living space was gradually dispersed, and the ecological spatial distribution was more balanced. Based on POI data, the method of random forest model to identify “production-living-ecological” spaces within the city was more effective, and the recognition results were more accurate, which can provide data and method support for territorial spatial planning on a smaller scale.