Assessing Structural and Parameter-Based Uncertainty in High-Latitude Terrestrial Biosphere Models through combination of Causal Loop Analysis and Ensemble Modeling
Abstract
Changes in arctic ecosystem dynamics alter carbon budgets and may shift the region from a historic carbon net sink to a source. Projections from high-latitude terrestrial biosphere models (TBMs) can differ, resulting in ambiguity surrounding the possible timing of this shift. Since these differences are caused by model uncertainties, reducing them is an important step in making TBMs more precise and useful for climate change risk management.
Two of the most influential sources of model uncertainty are structural and parameter uncertainty, the latter of which is more feasible to address across models. Therefore, we are aiming to address parameter uncertainty, while keeping structural processes in mind. As a first step, we identified the mismatch between parameters needed by the three models in this comparison, DVM-DOS-TEM, SIPNET, and ED2, and the data available to parameterize them. Since focusing solely on these data gaps ignores structural uncertainty, we created a causal loop diagram (CLD) of the arctic and boreal ecosystem that includes unquantified, and thus unmodeled, processes. By mapping the models to the CLD, we assessed the vulnerability of parameters in the system as described by the network structure of the CLD. We combine the data availability with the level of vulnerability to create a "Pa-factor" which we then use as a first measure of which parameters need additional data. To further understand whether a parameter needs more measurements, we assess how sensitive the models are to the parameter's uncertainty. We use the Predictive Ecosystem Analyzer (PEcAn), which treats parameters as probability distributions, to vary parameters over the course of an ensemble of model runs and assess how their uncertainty propagates through the models. Due to the large number of parameters in each model, we are only able to vary a selection of them simultaneously. Therefore, we concentrate on parameter groups with a high Pa-factor, like those describing the impact of temperature on microbial activity. Additionally, we incorporate the CLD to help select parameter groups that are likely to be affected by each other and vulnerable to change. Through all three methods combined, we expect to have a heightened understanding of how parameter uncertainty causes model uncertainty in the Arctic and how to address it.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.B32C1373M