Bayesian estimation of a parsimonious wetland hydrology model with remote sensing data

A major impediment to large-scale studies of non-floodplain wetlands is the absence of a suitable hydrological model that can be calibrated using remotely sensed hydrology data. We address this gap by developing a Bayesian representation of an upland-embedded wetland with parameters which can be reliably estimated from noisy remote sensing data. This model operates at a monthly timescale and was fit to data from three wetlands at the Cottonwood Lake Study Area in North Dakota, USA using Markov chain Monte Carlo. We find that the model is an effective representation (NSE > 0.5) of hydrological dynamics during dry periods but that wetland-groundwater interactions make it unsuitable (NSE < 0.0) during extended periods of elevated precipitation. This study highlights the combination of noisy observational data with mechanistic prior knowledge to conduct statistical inference with hydrological systems.