Optimal model complexity for terrestrial carbon cycle prediction

Abstract

Abstract. The terrestrial carbon cycle plays a critical role in modulating the interactions of climate with the Earth system, but different models often make vastly different predictions of its behavior. Efforts to reduce model uncertainty have commonly focused on model structure, namely by introducing additional processes and increasing structural complexity. However, the extent to which increased structural complexity can directly improve predictive skill is unclear. While adding processes may improve realism, the resulting models are often encumbered by a greater number of poorly determined or over-generalized parameters. To guide efficient model development, here we map the theoretical relationship between model complexity and predictive skill. To do so, we developed 16 structurally distinct carbon cycle models spanning an axis of complexity and incorporated them into a model–data fusion system. We calibrated each model at six globally distributed eddy covariance sites with long observation time series and under 42 data scenarios that resulted in different degrees of parameter uncertainty. For each combination of site, data scenario, and model, we then predicted net ecosystem exchange (NEE) and leaf area index (LAI) for validation against independent local site data. Though the maximum model complexity we evaluated is lower than most traditional terrestrial biosphere models, the complexity range we explored provides universal insight into the inter-relationship between structural uncertainty, parametric uncertainty, and model forecast skill. Specifically, increased complexity only improves forecast skill if parameters are adequately informed (e.g., when NEE observations are used for calibration). Otherwise, increased complexity can degrade skill and an intermediate-complexity model is optimal. This finding remains consistent regardless of whether NEE or LAI is predicted. Our COMPLexity EXperiment (COMPLEX) highlights the importance of robust observation-based parameterization for land surface modeling and suggests that data characterizing net carbon fluxes will be key to improving decadal predictions of high-dimensional terrestrial biosphere models.

Optimizing Carbon Cycle Parameters Drastically Improves Terrestrial Biosphere Model Underestimates of Dryland Mean Net CO2 Flux and its Inter-Annual Variability

“…Given we chose weak prior constraints in the assimilation, the fact that some posterior parameters are hitting their bounds suggests that the optimization may be aliasing model structural error onto the parameters (as demonstrated in MacBean et al, 2016;Wutzler & Carvalhais, 2014) and/or that the model cannot improve further via parameter optimization. This suggests that further model developments are likely needed to address structural uncertainties and missing processes, which will then need to be followed up with additional parameter DA experiments to ensure increasing complexity does not degrade model skill (Famiglietti et al, 2021). We know for example that certain important processes for sparsely vegetated, mixed shrub-and grass-dominated dryland ecosystems, such as wildfires (Exbrayat et al, 2018;Lasslop et al, 2016;Whitley et al, 2017) and biological soil crust C cycling (Belnap et al, 2016), are currently not represented in most TBMs.…”

Potassium-limitation of forest productivity, part 1: A mechanistic model simulating the effects of potassium availability on canopy carbon and water fluxes in tropical eucalypt stands

“…Here, the model additions were clearly necessary since the CASTANEA model, into which we developed the K modules, was initially incapable of reproducing the effect of K limitation on GPP and no mechanistic model of the effect of K on plant productivity at the stand level existed. This development also broadly followed several of the guidelines posited by Famiglietti et al (2021) in their paper addressing the question of models' structural complexity: 1) the use of datasets (here multiple experiments over multiple rotations) to constrain model parameters, 2) the new developments led to increased forecast ability (since no forecast of K deficiency was previously possible), and 3) we sought to calibrate unmeasured parameters. We adopted a reductionist approach, typical of the development of mechanistic model, by formulating and parameterising the model on dedicated experiments conducted at the organ scale.…”

Resolving scale-variance in the carbon dynamics of fragmented, mixed-use landscapes estimated using Model-Data Fusion

“…At the core of our model-data fusion framework sits DALEC, an intermediate complexity model of the terrestrial C cycle (Williams et al, 2005;Bloom and Williams, 2015;Smallman et al, 2017;Famiglietti et al, 2021). DALEC is a mass balance model of the C cycle with carbon moving through different pools based on parameterised fluxes (Figure 3).…”

“…DALEC is a mass balance model of the C cycle with carbon moving through different pools based on parameterised fluxes (Figure 3). A number of variants of DALEC have been created representing ecosystem carbon dynamics with varying degrees of complexity (Famiglietti et al, 2021;Smallman et al, 2021). The specific version of DALEC used here corresponds to the C6 model outlined in Famiglietti et al (2021) which combines the C-cycle structure from Bloom and Williams (2015) with the revised photosynthesis model from Smallman and Williams (2019).…”

“…In typical spatially distributed MDF applications, available observations are aggregated to pixel-level "community averages", prior to inversion. There are usually sufficient degrees of freedom in ecological process models to fit the observed temporal changes in aggregated stocks and fluxes, based on available observation constraints (Beven et al, 2006;Famiglietti et al, 2021). Nevertheless their ecological fidelity may be limited in heterogeneous landscapes, for which the parameters retrieved by these "community-average" models provide intermediate representations of the distinct ecosystems present.…”