Using machine learning to downscale projected land conversion

Abstract

Abstract—Decisionmakers need to better understand the
effects of programs and policies impacting land conversion and
associated environmental, economic, and social impacts to
improve assessment of the costs and benefits of alternative policies.
It is not only the quantity of land conversion taking place, but
which land is being converted that determines outcomes for
endangered species habitats, biodiversity, water quality, carbon
sequestration, and other ecosystem services. However, there is a
current lack of tools that can facilitate evaluation of impacts at a
sufficiently spatially disaggregated level to adequately evaluate
them. In this study, we utilize projections of land conversion from
the U.S. Forest Service 2020 Resource Planning Assessment under
alternative socioeconomic and climate scenarios. Specifically, we
consider the projected change in cropland between 2020-2050
under (1) low emissions forcing, moderate economic growth; and
(2) high emissions forcing, high economic growth. We downscale
county-level projections to a 30x30 meter raster of cropland
expansion for the State of Louisiana using machine learning. We
then overlay spatial data on endangered species ranges to assess
impacts of land conversion to cropland. This enables us to estimate
the area of endangered species’ habitats that would potentially be
impacted by cropland expansion. The information generated can
inform the design of data collection efforts as well as policies aimed
at reducing negative impacts on critical habitats associated with
cropland expansion.
Keywords—land use change, endangered species, global climate
models, red-cockaded woodpecker, American chaffseed