Sensitivity of tropical forest carbon and energy fluxes to Community Land Model parameters
Incomplete knowledge about ecological processes contributes to substantial uncertainties in climate change projections. Global land surface models and Earth system models used to make climate projections are computationally expensive and contain large numbers of parameters representing ecological processes, limiting the feasibility of uncertainty quantification (UQ) studies that require large model ensembles. Fortunately, even relatively small model ensembles can yield useful information about key model parameter sensitivities. Model parameters are often linked to key ecosystem traits. Knowledge about how these traits are distributed and how they drive carbon and water cycles is of key importance to not only the modeling community, but also the measurement community as this knowledge can inform observation strategies targeted at reducing prediction uncertainties. In particular, tropical forest ecosystems are highly diverse and contribute a large fraction of global land-surface carbon and latent heat fluxes. In this study, we determine the ranking of parameter importance by applying the Morris method on 100 Community Land Model (CLM) model parameters at 3 tropical forest eddy covariance sites. We then perform a variance-based analysis to provide a way of analyzing the contributions (linear and non-linear) by each parameter and its interactions with other parameters to the variance of the model output. Particular model outputs of interest are biomass, net carbon fluxes and energy fluxes.
We find that the ranking of parameter importance strongly depends on climate variables (e.g. dry season length, mean annual temperature and precipitation). The parameterization of mortality, rooting depth, photosynthesis temperature response and stomatal conductance is particularly important and regulates total biomass and seasonal patterns of carbon exchange in these tropical ecosystems. We also evaluate the performance of the model against measured eddy covariance data and determine which combinations of parameters are most likely given the observations. This information can be used to improve model performance, and to guide measurement strategies in tropical forest systems.