Comparative Study Of Population Spatialization Models Based On PCA-GWR And PLS-GWR

The research on population spatialization is an important part of demography,geography and GIS.It plays an important role in exploring the law of population spatial distribution,resource allocation and urban structure optimization.However,there are some advantages in the traditional population spatialization research,such as single model and redundant population distribution influence factors.Repeated expression of factor information can easily reduce the accuracy of spatialization results.In response to the above problems,the geographical conditions,nighttime lights,digital elevation,model and points of interest data were used for this paper,and there are many factors of population spatial distribution were extracted such as forest and grass coverage area and planting land area and so on.Principal component analysis(PCA)and partial least square(PLS)were used to extract the factor combinations that have significant impact on the spatial distribution of population,and then coupled with geographically weighted regression(GWR),and the PCA coupled GWR(PCA-GWR)population spatial model and the PLS coupled GWR(PLS-GWR)population spatial model were constructed.In this paper,the detailed population distribution data of the 100-megabit grid-scale in Jiangsu in 2018 and the and the accuracy of the simulation results were carried out.The results are as follows:(1)The PCA model was used to reduce the dimension of 14 factors.On this basis,four principal components were extracted to represent four kinds of information,including human settlements,ecology,terrain and road.The four principal components were used for geographically weighted regression to build a population spatial model based on PCA-GWR,and the goodness of fit of the model was 0.642.(2)The PLS model was used to extract 14 factors step by step,and two components were extracted on the premise of meeting the cross test.The two components were used for geographically weighted regression to construct the population spatialization model based on PLS-GWR.The overall goodness of fit of the model is better than PCA-GWR model,and the goodness of fit was 0.781.(3)The results show that compared with World Pop data set,the two spatialization results have higher simulation accuracy,and the accuracy of PLS-GWR model was better;The proportion of relative error absolute value less than 30% of population spatialization results based on PLS-GWR model is 61.930%,which is 8.311% higher than PCA-GWR model,indicating that PLS-GWR model can simulate population distribution more precisely.