The Devil is in the Details: Application of Emulation-Based Data Assimilation Techniques to Constrain a Dynamic Ecosystem Model

Abstract:

Exchanges of carbon, water, and energy between the terrestrial biosphere and atmosphere are major drivers of the Earth’s climate system, yet interactions between them remain the second-largest uncertainty in climate projections. A key aspect of reducing this uncertainty lies in better parameterizing and calibrating terrestrial ecosystem and land surface models using data assimilation (DA) methods. In the atmospheric sciences, DA is most often synonymous with addressing initial condition or forcing uncertainty in atmosphere and ocean components of weather forecast or climate projecting Earth System models. However, for the land surface, and for its carbon, water, and energy exchange, a larger source of uncertainty comes from representation of biological processes and most importantly, the parameters that describe those processes.

In this study, the Ecosystem Demography model version 2.2 was used for the testing and application of an emulator approach to Bayesian parameter data assimilation, with model runs occurring at a study site in Northern Wisconsin’s Chequamegon-Nicolet National Forest. This study seeks to answer two primary questions: 1.) Can model predictions of net ecosystem exchange (NEE) at a water-rich upper Midwest study site be improved through assimilating with respect to both NEE and latent heat flux, as opposed to NEE alone? and 2.) Can parameter data assimilation using an emulator approach reduce overall model predictive uncertainty in a structurally complex ecosystem model?