Estimating the spatial distribution of the population of Riyadh, Saudi Arabia using remotely sensed built land cover and height data

• Main Author: Alahmadi, Mohammed
• Published: University of Southampton 2013
• Subjects: 304.601
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.664981

Timely and accurate data on the spatial distribution of the population for small urban areas are a key requirement for sustainable development. Unfortunately, population data in many countries is limited by coarse spatial and temporal resolution. The aim of this research was to develop, refine and validate a model that uses remotely sensed satellite sensor data, demographic data and field survey data to obtain an accurate small-area population map. It uses the specific case study of Saudi Arabia, where few studies have attempted to estimate population distributions at fine scale. Moreover, it is desired to better understand how (i) different image algorithms affect the accuracy of land cover data and Cii) different spatial resolution satellite sensor data affect the discrimination of urban characteristics to be used as input to a predictive model. This research investigated a range of models of the relationship between dwelling unit density and urban remote sensing covariates, with a view to predicting dwelling unit density at a finer spatial resolution than currently available and subsequently to transform this distribution to estimate population. The main contribution of this research was to demonstrate the gradual refinement of the predictive model using both Ci) a variety of explanatory variables including building height and the number of floors and a choice of resolutions and Cii) a set of alternative models including global regression, regional regression, geographically weighted regression and dasymetric mapping. The fitted models varied in terms of their accuracies. The set of models that used detailed residential land use classes obtained from fine spatial resolution satellite sensor data were the most accurate compared with the set of models that used general land cover classes obtained from coarse spatial resolution satellite sensor data. 6-class dasymetric mapping was the most accurate model of all those tested.