Analysis of the usefulness land use and land cover maps for estimating the population of urban areas – valorisation of multi-variant population maps based on the GUS kilometre network

Abstract

The paper is a continuation and summary of a series of publications related to the dasymetric estimation of the distribution of the population of Krakow. The conversion of the population from the original census units is based on the development data from three sources, the Corine Land Cover project (CLC), the Urban Atlas project (UA) and the object classification (OBIA) of the RapidEye data. The experiment was conducted using archival statistical data from 2009 from 141 urban units (u.u.) of the city.In the first two parts of the cycle population conversion was presented on the basis of CLC, UA and OBIA maps, obtaining a total of 12 maps of Krakow's population. The obtained error distributions were presented and the calculated weights of population density for each category of residential buildings were discussed. In the third part of the cycle (Pirowski and Berka, 2019) the results were analyzed in detail by reference to the reference, high-resolution population map of the Bronowice district (north-western part of the city).In this publication, ending the cycle, population maps were verified on the basis of a kilometre grid of the Central Statistical Office (GUS), which is an aggregation of data from the National Census of Population and Housing 2011, made available by the Office in 2017. The results of high-resolution verification carried out in the Bronowice district were compared with the data of the CSO (GUS). In the GUS grid the best results were obtained for surface and weight UA methods (RMSE 908-917 people; MAPE 42-46%). The estimation of population distribution using OBIA data (RMSE 1115-2073 people; MAPE 121-184%) was found to be incorrect. After the correction of OBIA by UA data, a significant improvement in the results for surface-weighted methods was obtained (RMSE 930-1067 people; MAPE 53-68%), however, the error rate was still higher than for UA itself, which eliminates the OBIA method from practical applications in this area.A correlation was found between the RMSE and MAPE errors recorded in UC at the stage of weight selection and the RMSE and MAPE errors recorded in the GUS grid, respectively R2(RMSE)=91%, R2(MAPE)=65%. Therefore, the correlation detected indicates that the low errors obtained at the selection stage translate into reliable population estimates. The proposed weighting methodology limits the subjectivity of the method, based on the minimisation of RMSE and MAPE in the original census units. The disadvantage of the method is that it is necessary to define the boundary conditions for the selection of weights, in case of obtaining unreal weights and the possibility of occurrence of equifinality phenomenon, difficult to detect in the absence of additional reference data. Słowa kluczowe: dane spisowe GUS, modelowanie dazymetryczne, mapy pokrycia i użytkowania terenu, korelacja poziomu błędów Abstrakt Artykuł jest kontynuacją i podsumowaniem cyklu publikacji związanych z dazymetrycznym szacowaniem rozmieszczenia ludności Krakowa. Przeliczanie ludności z pierwotnych jednostek spisowyc...