Research On The Spatialization Method Of Beijing Population Based On Multi-source Data And Considering Spatial Differences

Population resources are the most important resource for urbanization development,and urbanization construction must adhere to the basic principle of "people-oriented,fair sharing".Reasonable urbanization planning is inseparable from the support of accurate and precise population data.At the same time,the refined management of urban society also puts forward an urgent need for high-precision population data.Demographic data with administrative divisions as the basic unit has been suffering from problems such as low temporal and spatial resolution,failure to effectively reflect the differences in the spatial distribution of population within administrative units,and difficulties in the joint analysis of natural,economic and social data caused by the "MAUP",Demographic data is difficult to reflect the reality and meet actual needs.The spatialization of population data transforms statistical data from administrative units to regular grid units according to certain rules.It helps to expand the application depth and breadth of population data.At the same time,it is of great significance to disaster risk assessment,urban planning,and resource allocation.However,population spatialization currently faces two problems: On the one hand,the accuracy of population spatialization results based on nighttime light is low due to pixel overflow,economic development level,industrial structure and other reasons.Researchers has drawn attention to the integration of nighttime light and other data,while the existing population spatialization research lacks the method of integration of nighttime light and location-based service data;on the other hand,the existing population spatialization model ignores the multi-scale effect of the population spatial distribution factors,and the spatial scale of the population spatial distribution factors is identical,while there is still a gap with the real world.The solution of these two problems is conducive to deepening the understanding of the influence mechanism of population spatial distribution,and helps to improve the accuracy of population spatialization.In response to the problems mentioned above,this article completed the following work from two aspects:(1)Research on population spatialization method of nighttime light image and location-based service dataNighttime light data has the characteristics of comprehensively reflecting human activities and location-based service data can reflect the real-time location characteristics for microscopic individuals.LJ1-01 nighttime light data and We Chat positioning data are emerging representatives of these two types of data,respectively.This thesis firstly optimized the bandwidth of the kernel density estimation of We Chat positioning data,and then combined the two data with geometric mean.Next,the gridded population data was simulated by geographically weighted regression(GWR)model.Finally,the accuracy evaluation and error analysis was performed.The experimental results show that: 1)Kernel density estimation of We Chat positioning data is greatly affected by bandwidths,and the bandwidth is positively correlated with the degree of dispersion of spatial points;2)After fusion by geometric mean for these two types of data,the correlation coefficients between statistical population and simulated population are 0.565(LJ1-01 only),0.795(We Chat only)and 0.839(data fusion by geometric mean),respectively,and the mean relative error(MRE)were 35.86%,23.18%and 19.14%,correspondingly;3)The MRE of this fusion method in the densely populated urban area is only 13.1%,which has the best performance among all subareas.This method has certain practical significance in the study of population spatial distribution for Beijing,which has an urbanization rate of more than 80%.(2)Research on multi-scale effects of factors affecting population spatial distributionFirst,several indicator factors affecting population distribution—land coverage,topography,climate,economy,etc.,were selected and then optimization of variable selection was conducted.Second,a population spatialization model based on multiscale geographic weighted regression(MGWR)at the township level in Beijing was built.Next,the accuracy evaluation and the comparison for scale differences between the classical geographic weighted regression(GWR)and the MGWR results were carried out.Finally,this thesis discussed the spatial pattern and rationality of the local parameters,and realized the gridded population.The main results are as follows: 1)The accuracy of MGWR results is generally superior to that of GWR;2)The influencing factors of the population distribution in Beijing have obvious spatial non-stationarity.Among them,the spatial non-stationarity of cropland and low-rise buildings is the strongest,and the scale of spatial operation is local.The spatial non-stationarity of multistorey buildings,artificial excavation and land surface temperature is moderate,and they play a role at a regional scale,while the traffic network and slope have littleto-no spatial non-stationarity,which has an impact on the whole study area;3)MGWR results are more consistent with conventional cognition,but the GWR results are more unreasonable with an artificially high accuracy.The above research indicates that the spatial distribution of population is the result of multiple factors at multiple scales,not the result of multiple factors at a single scale.In summary,this thesis studies the population spatialization method based on LJ1-01 nighttime light image and We Chat positioning data,and conducts a study on the scale difference that takes into account the influence factors of the population spatial distribution,which can provide some new methods and new perspectives for future population spatialization research.The former is the realization of emerging data on population spatialization research methods,and the latter is related to the multi-scale effect of population spatial distribution.The two are relatively independent and unified.