Research On Fine Population Spatialization Method Based On Multi-Source Geographic Data

Population is the most important social problem.With the continuous development of urbanization,more and more people enter the city,which puts forward higher requirements for the fine management of the city.The spatial resolution of statistical population data based on the administrative units is not high,which is difficult to reflect the actual situation of population distribution,and is not conducive to the spatial visualization analysis and practical application of population data.Population spatialization is the realization of gridding population data through certain methods,which can significantly improve the spatial resolution of population data and facilitate the mastery of fine spatial distribution information of population.It can not only provide effective auxiliary data sources for traditional disaster prevention and assessment,regional accessibility analysis,natural and human factors,but also practical significance on assisting the construction of intelligent cities and the precision resource scheduling and service in the era of big data.This paper took Wuhan as the research area and studied how to obtain highprecision population spatialization data at various grid scales by fusing high-precision data,and studies the population spatialization at multi-grid scales of Wuhan’s streetlevel statistical population data in 2015.Firstly,population spatialization based on NPP/VIIRS night lighting and FROM-GLC land use data.Night lighting and land use data have the characteristics of wide range,high update frequency and simple processing.They are suitable for obtaining large-scale and high update frequency population spatial distribution data.The land use types are divided into light and nonlight areas by night lighting.On this basis,the land use data are extracted including light area,no-light area and total value of night light.The relationship between these variables and the statistical population is spatialized by using multiple linear regression model.Secondly,two kinds of fine-grained geographic elements data are extracted based on Points of Interest(POI)and national geoinformation survey.These data are difficult to obtain,slow to update and complex to process.However,this kind of data set has abundant contour and semantic information,and is more suitable for small-scale fine population spatialization.Considering the building attributes and the spatial location relationship between buildings and interest points,the random forest model is used to train and predict the extracted high-dimensional features for population spatialization.Considering the research needs of different industries or other fields,the two experiments were carried out at three grid scales: 500 m,100m and 50 m,respectively.Furthermore,the spatial autocorrelation of population in the study area is analyzed,and the filtering method is used to integrate the correlation between the population data in the adjacent grid.The results of the experiment are summarized and analyzed.The conclusions are as follows:1)The results of multi-scale population spatialization based on night lighting data and land use data are more accurate with the increase of grid scale.The results of population spatialization at 500 m scale are the best,and the accuracy is higher than that of the WorldPop data set,which is close to the experimental results of the second method at the same scale.The results show that these two types of raster data have high accuracy in population spatialization on a larger grid scale.2)The results of multi-scale population spatialization based on interest point data and housing construction have lower accuracy with the increase of grid scale.The results of population spatialization at the scale of 50 m are the best,and the accuracy is the best for all the experiments in this paper.The spatial resolution and accuracy of this method can be greatly improved,which indicates that these two kinds of fine-grained geographic elements data are more suitable for population spatialization research on smaller grid scale.3)The population of the study area has obvious spatial autocorrelation,showing the basic population distribution pattern of high concentration in the main urban area and low concentration in the far urban area.The experiment of population processing based on filtering has a certain correction effect on some scales.In summary,this paper studies two methods of multi-scale fine population spatialization,and analyses the fitting effect of different spatial auxiliary data sources on population spatial distribution at different scales,which can provide new ideas for fine population research in big data era by fusing multi-source data.