A rules-based approach for estimating high-resolution population distributions for climate studies (California)

A population distribution model is developed and tested with California as the study area. The model—which distributes population spatially at 30-arc seconds—is based on “habitability,” the propensity for a place to sustain a fixed population. Individual environmental parameters are examined, as are composite habitability functions comprised of combinations of environmental parameters. The environmental variables are elevation, topographic accessibility, river proximity and annual water surplus. Both the total population of California and the populations of individual counties are used as boundary conditions. Model-estimated population distributions are compared to 1990 California census data at the block group level.; Results show that the model can realistically distribute population based on environmental variables when constrained by both the State's total population and the individual-county populations. When constrained by the State's total population, the elevation-based model returned the lowest Mean Absolute Deviation (MAD) error of the individual-variable models; the annual-water-surplus-based model returned the highest. More of the relative spatial variance ( V) is represented by river proximity than by any other variable. Composite habitability functions performed better than individual ones. The model returning the lowest MAD and highest V combines elevation, topographic accessibility, and river proximity. It represents approximately 50% of the relative variance. Composite functions also consistently return higher Vs when constrained by the State's total population; this indicates the likely performance of the model(s) when only coarse census data are available. Constraining the composite models with county total populations reduced the MADs by approximately 20%. Using county populations, however, had mixed effects on the individual-variable Vs.; Incorporating annual water surplus generally degrades the performance of the habitability models. This suggests that population may be distributed adequately, without using a hydroclimatic variable.