A map of global peatland extent created using machine learning (Peat-ML)

Abstract. Peatlands store large amounts of soil carbon and freshwater, constituting an important component the global hydrologic cycles. Accurate information on extent distribution peatlands is presently lacking but needed by Earth system models (ESMs) to simulate effects climate change balance. Here, we present Peat-ML, a spatially continuous map peatland fractional coverage generated using machine learning (ML) techniques suitable for use as prescribed geophysical field in ESM. Inputs our statistical model follow drivers formation include distributed climate, geomorphological data, remotely sensed vegetation indices. Available maps 14 relatively extensive regions were used along with mapped ecoregions non-peatland areas train model. In addition qualitative comparisons other literature, estimated error two ways. The first estimate training data blocked leave-one-out cross-validation strategy designed minimize influence spatial autocorrelation. That approach yielded average r2 0.73 root-mean-square mean bias 9.11 % −0.36 %, respectively. Our second was comparing Peat-ML against high-quality, extensively ground-truthed Ducks Unlimited Canada Canadian Boreal Plains region. This comparison suggests be comparable quality mapping products through more traditional approaches, at least boreal peatlands.