A predictive algorithm for wetlands in deep time paleoclimate models

Methane is a powerful greenhouse gas produced in wetland environments via microbial action in anaerobic conditions. If the location and extent of wetlands are unknown, such as for the Earth many millions of years in the past, a model of wetland fraction is required in order to calculate methane emissions and thus help reduce uncertainty in the understanding of past warm greenhouse climates. Here we present an algorithm for predicting inundated wetland fraction for use in calculating wetland methane emission fluxes in deep-time paleoclimate simulations. For each grid cell in a given paleoclimate simulation, the algorithm determines the wetland fraction predicted by a nearest-neighbour search of modern-day data in a space described by a set of environmental, climate and vegetation variables. To explore this approach, we first test it for a modern-day climate with variables obtained from observations and then for an Eocene climate with variables derived from a fully coupled global climate model (HadCM3BL-M2.2; Valdes et al., 2017). Two independent dynamic vegetation models were used to provide two sets of equivalent vegetation variables which yielded two different wetland predictions. As a first test, the method, using both vegetation models, satisfactorily reproduces modern day wetland fraction at a course grid resolution, similar to those used in paleoclimate simulations. We then applied the method to an early Eocene climate, testing its outputs against the locations of Eocene coal deposits. We predict global mean monthly wetland fraction area for the early Eocene of 8×106 to 10×106 km2 with a corresponding total annual methane flux of 656 to 909 Tg CH4 yr−1, depending on which of the two different dynamic global vegetation models are used to model wetland fraction and methane emission rates. Both values are significantly higher than estimates for the modern day of 4×106 km2 and around 190 Tg CH4 yr−1 (Poulter et al., 2017; Melton et al., 2013).