Tuesday, April 9, 2019 - 4:00pm
Assistant Professor, Department of Earth System Science at Stanford University
Understanding the Effects of Parameter Uncertainty in Water and Carbon Models
Models of ecosystems or water at the land surface represent a large number of uncertain processes. Over time, these models have become increasingly complex, capturing ever more processes. However, the model parameters in these process representations often vary dramatically in space but are assigned according to simple groupings (soil type or plant functional type, for example) that do not capture the full variability of parameters. Observational data can be used to help constrain these parameters - and their effect on model uncertainty- in so-called model-data fusion approaches using Markov Chain Monte Carlo methods. In this talk, I'll present two studies that demonstrate the power of model-data fusion in ecological and hydrological settings, as long as sufficient observational constraints are available. I'll first discuss our ability to estimate global carbon fluxes in the recent past (e.g. 2001-2010). This will be done by comparing conventional ensembles-of-opportunity such as TRENDY and CMIP5 with fluxes from CARDAMOM, a model-data fusion system with a relatively simple set of process representations but which uses a suite of remote sensing observations to optimize parameters and their uncertainty. In most regions of the world, CARDAMOM ensemble members are able to match the variability of TRENDY models as well as TRENDY models can match each other, suggesting the simplified structural representation in CARDAMOM is outweighed by its greater ability to capture parametric uncertainty. A second study focuses on modelling evapotranspiration in densely vegetated areas, where plant water response strategies are often particularly poorly parametrized. In this highly non-linear setting, effective stand-scale plant hydraulic properties can be estimated from flux tower evapotranspiration and soil moisture alone only at a subset of sites. Nevertheless, even when relatively uncertain parameters are used, it is clear that accounting for plant hydraulic strategies is necessary for capturing vegetation responses to water stress.