Divergence in land surface modeling: Linking spread to structure

C.R. Schwalm, K. Schaefer, J.B. Fisher, D. Huntzinger, Y. Elshorbany, Y. Fang, D. Hayes, E. Jafarov, A.M. Michalak, M. Piper, E. Stofferahn, K. Wang and Y. Wei


A well-documented problem in terrestrial carbon cycle modeling is the divergence in model estimates, both within ensembles of models and between models and observations. Here we explored the role of model structural differences in explaining the observed divergence. We found that a model’s initial conditions have a dominant impact on the final estimates produced, and that these initial conditions are themselves the consequence of model structure. This study highlights the necessity of having a mechanism that maps models’ skill in producing accurate estimates to specific aspects of model structure.


Figure: Relationship between GPP (gross primary productivity), LAI (leaf area index) and latent heat from functional benchmarking. Panels show functional relationships between GPP and LAI (top), GPP and latent heat (middle), and latent heat and LAI (bottom). Red lines show individual models. Thick black line is based on covariation between reference datasets; upscaled FLUXNET for GPP and latent heat as well as AVHRR-based LAI. Data values are binned satellite-era long-term averages by pixel, using the x-axis, across global vegetated land surface at 0.5 m2 m−2 increments for LAI and at 10 Wm−2 increments for latent heat. Where red lines continue past the black reference line indicates physical simulated values.


Divergence in land carbon cycle simulation is persistent and widespread. Regardless of model intercomparison project, results from individual models diverge significantly from each other and, in consequence, from reference datasets. Here we link model spread to structure using a 15-member ensemble of land surface models from the Multi-scale synthesis and Terrestrial Model Intercomparison Project (MsTMIP) as a test case. Our analysis uses functional benchmarks and model structure as predicted by model skill in a machine learning framework to isolate discrete aspects of model structure associated with divergence. We also quantify how initial conditions prejudice present-day model outcomes after centennial-scale transient simulations. Overall, the functional benchmark and machine learning exercises emphasize the importance of ecosystem structure in correctly simulating carbon and water cycling, highlight uncertainties in the structure of carbon pools, and advise against hard parametric limits on ecosystem function. We also find that initial conditions explain 90% of variation in global satellite-era values—initial conditions largely predetermine transient endpoints, historical environmental change notwithstanding. As MsTMIP prescribes forcing data and spin-up protocol, the range in initial conditions and high levels of predetermination are also structural. Our results suggest that methodological tools linking divergence to discrete aspects of model structure would complement current community best practices in model development.

Schwalm, C.R., K. Schaefer, J.B. Fisher, D. Huntzinger, Y. Elshorbany, Y. Fang, D. Hayes, E. Jafarov, A.M. Michalak, M. Piper, E. Stofferahn, K. Wang, Y. Wei (2019) "Divergence in land surface modeling: linking spread to structure," Environmental Research Communications , 1, 111004, doi:10.1088/2515-7620/ab4a8a.