Intelligent interpolation for population distribution modeling
Kim, Hwa Hwan
MetadataShow full item record
Dasymetric mapping is an intelligent interpolation method to accurately disaggregate population distribution with assistance of ancillary data describing the underlying pattern of geographical phenomena. Many studies have demonstrated that the dasymetric mapping method can substantially improve accuracy. Despite the significant performance advantages of the dasymetric mapping method, it has not been widely adopted amongst the broader geography community because of its relative complexity to implement and the difficulties to acquire high quality ancillary data. This research aims to investigate how to elaborate the method of dasymetric mapping for population distribution modeling while minimizing the effort for data acquisition and processing, so as to encourage more users to take advantage of the dasymetric mapping method for applications involving population distribution data. This dissertation addresses two questions related to efficient implementation of dasymetric mapping for population distribution modeling. First is how to improve the performance of dasymetric mapping method. Second is what kind of public-domain land cover data could be utilized. Regression-based population estimation models and three dasymetric mapping methods are briefly reviewed and tested with the National Land Cover Dataset (NLCD). Although, the correlation between residential land cover and population density is clearly proved, the relative performance of the three dasymetric methods (binary, three-class, and limiting variable) is inconclusive. A hybrid dasymetric method integrating the pycnophylactic interpolation and the dasymetric mapping significantly outperforms the other methods (areal weighting interpolation, binary dasymetric mapping, and pycnophylactic interpolation method). Sensitivity analysis shows that the hybrid method can be further improved with appropriate selection of search radius size. Geographical weighted regression (GWR) modeling performs very well to estimate population density weight for each land cover class of the NLCD 2001 data. GWR based multi-class dasymetric method outperforms all other methods (areal interpolation methods including areal weighting interpolation, pycnophylactic interpolation, binary dasymetric method, and globally fitted ordinary least squared (OLS) regression based multi-class dasymetric method. This is attributed to the fact that spatial heterogeneity is accounted for in the process of determining density parameters for land cover classes.