Comparing approaches to quantify urbanization on a multicontinental scale

Abstract

Urbanization is an increasingly prevalent driver of environmental, ecological, and evolutionary change in both terrestrial and aquatic systems, and it is important that our sampling designs accurately capture this urban environmental change. Common approaches to sampling urban environments include: urban-nonurban transects, which sample along urbanization gradients; random points, which sample locations at random within an area of interest; and systematic points, which sample locations based on a regularly-spaced grid from a predetermined starting point within the area of interest. Presently, we lack a comparative analysis of the efficacy of these different sampling designs in capturing variation in urban environments. Here, we compare the environmental variation captured by transect- and point-based sampling designs in 136 cities across six continents. We quantified and compared a common set of environmental variables for each sampling design, with variables capturing landcover, climate, and socioeconomic facets of urban environments. Mean landcover and socioeconomic metrics consistently differed among sampling designs, in contrast to climate variables that primarily varied among cities. Additionally, changes in environmental variables with distance from the city centre depended on the sampling design, with this distance-by-sampling design interaction present in 27%-51% of cities, depending on the environmental variable. This implies that the rate of environmental change along urban-nonurban gradients frequently depends on the sampling design used. We also examined potential causes of deviations between transect- and point-based sampling designs and identified human population density and city area as common predictors of deviations between transect- and point-sampling designs. Our results show that sampling design can dictate how the urban environment is characterized, with sampling design as important - or more important - as the selected environmental variable. We further developed R code so researchers can implement these methods as they develop and validate sampling designs in novel or unstudied urban environments.