Quantifying the uncertainty surrounding the implementation of a mechanistic plant hydraulic module in a dynamic vegetation model

Background/Question/Methods

Uncertainties in stomatal responses to soil water content decline are major drivers of ecosystem model predictive uncertainty of forest productivity. As such, the choice between stomatal conductance models has been highlighted as a key step required for advancing ecosystem modeling. Models that link leaf level stomatal function with soil water potential through mechanistic water transport have the potential to perform better than traditional empirically based carbon assimilation models. However, mechanistic hydraulic models also entail an increase in the number of processes and model parameters. With limited constraints on parameters there is a risk of increased predictive uncertainty as a trade off for better average model performance. The questions then stand:

• Does the model refinement outweigh the potential increases in predictive uncertainty due to increased parameter uncertainty?
• Furthermore, given the current hydraulic trait data that has been collected, can we sufficiently constrain parametric uncertainty to curtail excessive predictive uncertainty?
• If this is not currently possible, can we use uncertainty analysis to guide us towards the appropriate data collection strategy?

We addressed these questions by comparing two implementations of the Ecosystem Demography model (ED2). The first uses non-physical, empirical scheme for regulating plant water use through the ratio of transpirational demand and water supply (ED2_ORIG) while the second employs a fully mechanistic plant hydraulics module (ED2_HYDRO).

Both models were run at Barro Colorado Island, Panama, a lowland tropical forest under current and future climatic conditions. Runs were done within the Predictive Ecosystem Analyser, where we benchmarked predictions against observational data and to performed full sensitivity and uncertainty analysis, leveraging data from hydraulic plant traits meta-analysis.

Results/Conclusions

It was possible to sufficiently constrain the hydraulic parameters in ED2_HYDRO and thus to reduce model uncertainty related to water uptake and transport. This was not the case with ED2_ORIG, where hydraulic parameter uncertainty and subsequent model sensitivity to the hydraulic parameter was very high. Thus, we can conclude that the potential tradeoff for ED2_HYDRO between increase in parameters and predictive uncertainty was mitigated by the available data. However, the structural differences in ED2_HYDRO resulted in increased sensitivity to non-hydraulics related parameters such as leaf and root biomass allocation. While these parameters can be measured in the field, there is currently not enough data to sufficiently constrain them and the thus ED2_HYDRO predictive uncertainty was still quite high. However, as we had hoped, the uncertainty analysis has indeed given us direction to focus future data collection efforts.