Thoughts on land modeling philosophy

[AlexisRenchon]

Land models, in the context of climate, are mostly needed to inform land-atmosphere exchanges (source or sink) of water (evaporation and transpiration), energy (latent and sensible heat flux, shortwave and longwave radiation) and greenhouse gases (e.g., CO2, CH4, N2O).

Water and energy fluxes are important at short time scale (hour, days) whereas greenhouse gas exchange are important at larger time scale (sink or source over decades: e.g., how much carbon will be in the atmospheric pool vs. land pools such as plant biomass or soil carbon).

Transpiration is coupled to photosynthesis, as both are regulated by the conductance of gas through leaves opening (stomata), the more stomata are open, the more transpiration and photosynthesis can occur from air exchange between the insides of leaves and the atmosphere.

Modeling photosynthesis and transpiration can be excellent when we prescribe structural information such as leaf area index (the area of leaves per area of ground), as we have good theory and data to inform parameterisation of instantaneous fluxes.

While change in pools (i.e., carbon pool in plant biomass) over long time scale is equal to the integration of short time scale fluxes, it remains very challenging to model the evolution of vegetation structure (biomass, leaf area index, roots, etc.), as there is less theory and less data, and great complexity in processes happening at microscales. Small errors in instantaneous fluxes, not only random, propagates when integrated over larger time scale, leading to large uncertainties for change in vegetation pool sizes in space and time.

Current model attempt to capture this complexity, leading to heavily parameterized models with no observation to inform those parameters, especially their spatio-temporal variability. This leads to large disagreement between models - large uncertainties.

Similarly, change in soil carbon is very complex and difficult to parameterize, as there is almost no observation of change in soil carbon. Nevertheless, models are becoming more and more complex, with no parameterization framework (because no observation to constrain the parameters), and no spatiotemporal generalization.

Despite decades of research, land models perform poorly when it comes to change in carbon pools (plant biomass, soil carbon), which is unavoidable - we have no observations to constrain models at large scale, especially belowground. More complexity will always remain a dead end as long as there is no data. While above-ground observation is increasing from remote sensing (e.g., SIF, NDVI, LiDAR), the period of data is often too short relative to climate models, and below-ground remain unobservable (change in roots depth and density, change in soil carbon).

I think ClimaLand should aim for intelligent but as simple (and generalizable) as possible carbon pool models, for now, until we have more data. Such models could be informed by optimality framework, simple processes (mass conservation - growth, litter fall, transport ...), and empirical relationship informed by data and parameterisation (for example, growth in transient and equilibrium at steady state of known vegetation structure in given environmental condition - mean temperature, mean precipitation, their distribution, soil properties, etc.). We should avoid over-complicated models, even if eventually we can add them to inform synthesis, using our modular framework.

Note: I plan to edit this text, and I am happy to discuss more.