首页|Experimental study of population density using an optimized random forest model

Experimental study of population density using an optimized random forest model

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
Random forest model is the mainstream research method used to accurately de-scribe the distribution law and impact mechanism of regional population.We took Shijia-zhuang as the research area,with comprehensive zoning based on endowments as the modeling unit,conducted stratified sampling on a hectare grid cell,and systematically carried out incremental selection experiments of population density impact factors,optimizing the population density random forest model throughout the process(zonal modeling,stratified sampling,factor selection,weighted output).The results are as follows:(1)Zonal modeling addresses the issue of confusion in population distribution laws caused by a single model.Sampling on a grid cell not only ensures the quality of training data by avoiding the modifiable areal unit problem(MAUP)but also attempts to mitigate the adverse effects of the ecological fallacy.Stratified sampling ensures the stability of population density label values(target variable)in the training sample.(2)Zonal selection experiments on population density impact factors help identify suitable combinations of factors,leading to a significant improvement in the goodness of fit(R2)of the zonal models.(3)Weighted combination output of the popula-tion density prediction dataset substantially enhances the model's robustness.(4)The popu-lation density dataset exhibits multi-scale superposition characteristics.On a large scale,the population density in plains is higher than that in mountainous areas,while on a small scale,urban areas have higher density compared to rural areas.The optimization scheme for the population density random forest model that we propose offers a unified technical framework for uncovering local population distribution law and the impact mechanisms.

population densityrandom forest modelendowment zonesstratified samplingfactor selectionweighted output

LI Lingling、LIU Jinsong、LI Zhi、WEN Peizhang、LI Yancheng、LIU Yi

展开 >

School of Geographical Sciences,Hebei Normal University,Shijiazhuang 050024,China

Hebei Technology Innovation Center for Remote Sensing Identification of Environmental Change,Shijia-zhuang 050024,China

Geocomputation and Planning Center of Hebei Normal University,Shijiazhuang 050024,China

Hebei Key Laboratory of Environmental Change and Ecological Construction,Shijiazhuang 050024,China

展开 >

National Natural Science Foundation of ChinaNational Natural Science Foundation of ChinaNational Natural Science Foundation of ChinaSecond Tibetan Plateau Scientific Expedition and Research ProgramNatural Science Foundation of Hebei Province

4207116742201197408710732019QZKK0406D2007000272

2024

10.1007/s11442-024-2264-5

地理学报(英文版)
中国地理学会,中国科学院地理科学与资源研究所

地理学报(英文版)

CSTPCD
影响因子：1.307
ISSN：1009-637X
年,卷(期)：2024.34(8)
  • 9