A new map of global peatland extent as estimated by machine learning

Abstract:

Peatlands store large amounts of soil carbon and freshwater, constituting an important component of the global carbon and hydrologic cycles. However, global peatland extent is relatively poorly known, especially in low latitudes with large peatland complexes even recently being discovered. To address this issue, we used machine learning techniques to develop a global map of peatland fractional extent. This global map of fractional peatland coverage was generated on a 5 arc minute grid using remotely-sensed vegetation characteristics and climate, geomorphological, and soil information as predictors of peatland extent. Our peatland map was then used as a mask for the peatland module of the Canadian Land Surface Scheme including Biogeochemical Cycles (CLASSIC) to generate estimates of global peatland C pools and fluxes at 1 degree global resolution. We additionally ran CLASSIC using a polygon-based meta-analysis product (PEATMAP) and the histosols of the Harmonized World Soils Database as the peatland mask. Our simulations shed some light on the impact of peatland extent on both regional and global C fluxes.