Markov chain Monte Carlo parameter estimation for testing multiple hypotheses of the drivers of seasonality in Sphagnum gross primary production

Peatlands harbor vast stockpiles of carbon (approximately 1⁄5 - 1⁄3 of global soil carbon), which are susceptible to recent and future climate change. Sphagnum gross primary production (GPP) is a major entry point of carbon into peatland ecosystems, making it a central component of peatland carbon cycling. This study evaluates alternative mechanistic hypotheses, represented in a process-based model, for the drivers of seasonality in Sphagnum GPP. To rigorously evaluate alternative hypotheses, parameters were estimated against Sphagnum GPP data using a Markov chain Monte Carlo (MCMC) algorithm developed in a flexible modeling software, the Multi-Assumption Architecture and Testbed (MAAT). Predictions from the optimized models (hypotheses) were then evaluated against a validation dataset. Data were collected at the Spruce and Peatland Responses Under Changing Environments (SPRUCE) experiment sited in the Marcell Experimental Forest in northern Minnesota. Sphagnum magellanicum GPP fluxes were calculated from hourly measurements of gas exchange in LI-8100s situated in hollows throughout the growing seasons from 2014 - 2018. This study applied the developed MCMC algorithm to compare model fit between two alternate Sphagnum GPP models that represent two hypotheses - constant Shoot Area Index (SAI) and dynamic SAI. In this analysis, SAI is the photosynthesizing tissue area per unit ground area. In contrast to constant SAI, dynamic SAI assumes an interaction between photosynthesizing tissue surface area and fluctuating water table height, reflecting the idea that submerged tissue is not photosynthetically active. The MCMC parameter estimation process implemented in MAAT formally shows that the dynamic SAI hypothesis better explains the seasonal dynamics in the GPP. Thus, this study demonstrates that accurate models of Sphagnum GPP at the Marcell Forest should incorporate the interaction between changing water table levels and the Sphagnum surface. Overall, the parameter estimation methods developed by this study enable the discovery of the most parsimonious model of Sphagnum GPP from a candidate set of models (hypotheses). By determining which Sphagnum GPP model is better equipped at describing a set of observed data, the most parsimonious model can subsequently be chosen for use in a predictive setting.