Data-driven spatial modeling of global long-term urban land development: The SELECT model.

Built-up land/impervious surface expansion links urbanization and environmental change. To enable large-scale long-term spatially-explicit studies, we took a data-driven approach exploiting newly-available time series of fine-spatial-resolution remote sensing observations, and developed the Spatially-Explicit, Long-term, Empirical City developmenT (SELECT) model. Closely calibrated to observational data, SELECT functions at several spatial scales, with multiple design traits capturing local variations of urbanization, and ensuring performance for long-term extrapolations in scenario analyses (e.g. the Shared Socioeconomic Pathways). It showed low estimation residuals, explained high fractions of the response's variations, and scored well in all robustness and generalizability tests we ran. When compared with a typical spatial-interaction-based model for projecting global built-up land in 2030, SELECT allocated more new development to areas with similar characteristics to locations that exhibited expansive urban growth historically, while the example spatial-interaction-based model allocated more new development to areas with high amounts of existing built-up land. • A large-scale long-term spatially-explicit built-up/impervious land model is developed. • Newly-available time series of fine-spatial-resolution remote sensing data are used. • SELECT is a land cover change model based on statistical learning and data mining techniques. • SELECT captures local spatial variations for long-term extrapolations in scenario analyses. • SELECT scored highly in evaluations and model comparisons. [ABSTRACT FROM AUTHOR]