Soil parameters such have been shown to be difficult to measure empirically or estimate from the data, due to their dependence on a complex combination of factors such as soil type, percentage organic matter and particle size distribution. These parameters are important for ecohydrological models in which they play a large role in determining soil moisture availability. Previous implementations of simple ecohydrological models have used parameter estimates either from previously published studies which may not be accurate for the current modeled system, or standard estimates for particular soil types, the general applicability of which has been questioned. We use Bayesian methodology to estimate those parameters from time series soil moisture data. A hierarchical framework is implemented in which we account for both model misfit and observation error which allows for better estimation of the parameters and soil moisture states, and characterization of the two types of error associated with ecological models in general.
We apply this method to a simple ecohydrological model which is applied to measured time series of soil moisture from Californian grasslands, and use it to characterize the difference between invasive annual grass species and native perennial species.
Results/Conclusions
The Bayesian methodology produces a distribution for each of the model parameters. The distributions obtained for the Californian grasslands are compared with standard soil parameter estimates for the soil type. All of the standard estimates are different to those shown most likely for this ecosystem, and we highlight a subset of the standard parameters which are highly unlikely to be valid.
The parameterized model shows that there are contrasting patterns of water use between the annual and perennial species in this Californian grassland, with the annuals starting to experience water stress at a higher soil moisture level than perennials, but also having a higher maximum rate of transpiration, therefore leading to much greater fluctuations in daily transpiration rates.
The Bayesian methodology that we use has wider applicability for the parameterization and characterization of simple ecosystem models. When simple state-space models such as nutrient cycling or plant competition models are applied to specific systems it is important to get specific parameter estimates and knowledge of the limitations i.e. the likely misfit of the model. This is likely to be particularly useful when the model is to be used for investigating the effects of unobserved situations for example changes in climate.
