Population Estimation At The Building Level Based On Random Forest And Nighttime Light Data

Under the background of rapid urbanization,urban population has shown a continuous growth trend.The monitoring of refined population distribution information within the city is of great significance for urban planning,resource allocation,and disaster warning.On the basis of the previous research results,advantages and research gaps of existing research can be drawn.This study used rapid and accurate population estimation as the research goal,and established population estimation model at the building level based on random forest regression algorithm using multi-source data.In this study,Shanghai Huangpu District was used as the study area.POI(Point of Interest)data,Luojia 1-01 nighttime light remote sensing image,population statistical data and building vector data were used in the model.The population distribution data at the building level of Huangpu District were obtained by the random forest population estimation model.The accuracy of the model was evaluated and the model was explained and analyzed in detail.The main achievements are as follows:(1)Extracting feature vectors from POI data,Luojia 1-01 nighttime light remote sensing image,and building vector data to build model multi-dimensional feature library after feature engineering and feature screening.Based on the random forest regression algorithm and multi-dimensional feature database,the random forest population estimation model was established.The population estimation model with the best estimation effect is obtained by adjusting the model parameters.The model output population data at the building level of Huangpu District.Compared population estimation result with the population statistical data at the building level,we found that the coefficient of determination(R~2)is 0.68,which represents the model performed well.(2)After completing the model training and results output,through the random forest interpretation method based on decision path estimation,quantificationally analyze the importance of each feature in the population estimation model,and the impact and of features on population estimates.The two indicators of feature importance and feature contribution were used in the analysis.From the several perspectives,which are the relationship between feature value and feature contribution,the composition of individual sample's feature contribution,and the feature contribution distribution of different functional areas,the feature contribution is analyzed in detail.In this process,the potential causes of error and the direction of model improvement were explored.The building-level population estimation method proposed in this paper can obtain more accurate population distribution data based on easy-to-obtain data,enrich the access methods of population distribution data,and provide new methods and new ideas for population distribution monitoring.