Fine spatial resolution residential land-use data for small-area population mapping: a case study in Riyadh, Saudi Arabia

Rapid growth in the world’s urban population presents many challenges to planning and service provision. Conventional sources of population data often fail to provide spatially and temporally detailed information on changing urban populations. While downscaling methods have helped bridge this gap, use of fine spatial resolution data coupled with object-based image analysis (OBIA) methods is relatively novel, and few studies exist outside the western, developed world. This article presents a study in Riyadh, Saudi Arabia, in which population distribution estimates were obtained by downscaling using detailed residential land-use classes derived from the application of OBIA to fine spatial resolution remotely sensed imagery. To assess the utility of these data for population downscaling, three statistical regression models (using built area, residential built area, and detailed residential built area) and two dasymetric areal interpolation models (using residential built area and detailed residential built area) were applied to downscale the density of dwelling units, prior to estimating the population distribution through a simple transform. The research suggests that, for regression, the proportion of residential land use (Model 2) increased the accuracy over built area proportion (Model 1), and, in a multivariate extension, the proportions of six separate residential land-use classes (Model 3) increased the accuracy further, thereby demonstrating the value of the fine spatial resolution imagery. For example, the actual number of dwelling units was 7771 and the estimated numbers of dwelling units of Models 1 and 3 were 10,598 and 8759, respectively. Moreover, the root mean square error (RMSE) was 5.9 for Model 1 and 2.6 for Model 3. Additionally, six-class dasymetric mapping was evaluated in comparison to the conventional binary dasymetric mapping approach. The six-class dasymetric mapping approach was found to be slightly more accurate than binary dasymetric mappin