The Different Schemes And Error Evaluation On Spatialization Of Statistical Population Data

The statistical population data are generally recorded by administrative units. Therefore, inside an administrative zone, population can only be regarded as uniformly distributed though the actual situation is not so. Population data with this structure is not suitable to comprehensive analysis among different study areas. However, the spatialization technology of statistical population data effectively solves this problem. Since the concept of spatialization of population data was proposed, domestic and foreign researchers have done a lot of research in the fields of spatialization of population data, and put forward various kinds of methods or models. Most parts of these methods or models are based on the relationship between population distribution and land use, land cover, night lights or other related factors, this kind of situation looks particularly obvious for domestic researcher in past ten years.As a kind of method of geoscience data processing and products out putting, spatialization of statistical population data inevitably results in errors like other methods. So far, the analysis or evaluation of errors on spatialization of population data generally occurred at specific specialization models or specialization results that the researchers developed themselves. However, systematic researches, analysis, evaluation and comparison of errors among different spatialization schemes have not been found, for example, errors of spatialization models based on different parameters or scales of statistical samples. It is beneficial to preferring models of spatialization, reducing errors and improving precision of spatialization product to analysis and research systematically errors resulting from different schemes of spatialization of population data. In view of facts mentioned above. The author used national land cover map at scale of 1 to 250,000, national administrative division map and statistical population data at county level in 2005 as basic data to carry out the following researches:(1) Design of the spatialization schemes.(a)Design of scale of statistical samples: According to the relationship between population distribution and land cover types, selected the area of land cover at level I including forest, grassland, farmland, urban settlements, rural settlements, wetland & water, desert as dependent variables of a model, used the population data as independent variables of the model, established two models of population to types of land cover through multi-variables linear regression with statistical samples at county and prefecture levels respectively.(b) Design of addition or reduction of independent variables: With the area of farmland, urban settlements, and rural settlements, which have close relationship with population distribution, as mandatory independent variables, the other parameters including the area of forest, grassland, wetland & water and desert as optional independent variables, establish 28 models with statistical samples at county and prefecture levels respectively.(2) Index and method of errors evaluation. Selected correlation coefficients, fitted values, relative errors and absolute errors as evaluation index of errors for spatialization models and spatialization outputs. Besides indicators in the above, the ratio of amount of grid cells with negative value in spatialization outputs to entire study area was also considered as an evaluation index of errors for spatialization outputs.(3) Analysis of spatialization errors.(a) Compared two spatialization schemes with county-level and prefectural-level statistical population data as analysis samples respectively. It was concluded that the scheme of spatialization based on county-level samples is better with a correlation coefficient of 0.797 and a relative error of 6.2%.(b) Compared 28 spatialization schemes based on different parameters at county and prefecture levels. A conclusion was drawn that the spatialization schemes with farmland, urban settlements, rural settlements, wetland & water as dependent variables and county-level statistical data as analysis samples was best, the correlation coefficient of which was 0.797 and relative error 8.3%.(4) Selection and optimization of spatialization schemes. According to the result of analyzing error to select the best scheme, then it optimizes the scheme.(a) Optimized the spatialization scheme with county-level statistical data as analysis samples by zoning the whole county into 7 sub-regions. Optimized models of each sub-regions based on correlation coefficients, regression coefficients and scatter diagram that delete the abnormal value of scatter diagram. The optimized models had a correlation coefficient of greater than 0.92 for each sub-regions and a relative error of 0.134%.(b) Optimized the spatialization model with farmland, urban settlements, rural settlements, wetland & water as dependent variables and county-level statistical data as analysis samples. The numbers of parameter are kept unchanged when it was optimized. The model was optimized based on correlation coefficient, regression coefficient and scatter diagram that delete the abnormal value of scatter diagram. The optimized model had a correlation coefficient of 0.955 and a relative error of 0.131%.(c) A conclusion was drawn from analyzing and comparing spatialization results of two spatialization models(a) and(b) in the above that the spatialization model with farmland, urban settlements, rural settlements and wetland &water body as dependent variables and county-level statistical data as analysis samples was optimal scheme.In this paper, the author established an optimal scheme of population statistical data spatialization through analysis and comparison of different schemes. The scheme can express actual distribution of population in China. Therefore, the ideas, thoughts and final results from this paper will have a certain guidance and reference value on spatialization of statistical population data in the future.