Abstract:
To effectively combat climate change, we need accurate predictions of how terrestrial ecosystems respond to environmental shifts. However, existing ecosystem biogeochemical models are likely not up to this task. We argue this is due to their insufficient account of scale coherence in the parameterization of biogeochemical processes. This insufficiency is manifested by their wide use of empirical relationships and the careless adoption of the multiplicative model and the law of the minimum, all of which disrupt the information flow between interacting entities essential for the functioning of ecosystem biogeochemistry. We contend that incorporating more physical rules into the parameterization will enable models to better resolve the scale coherence among biogeochemical processes. This will lead to a deeper understanding of ecosystem biogeochemistry, better-constrained model structures, and reduced model sensitivity to parametric uncertainty. We demonstrate the advantage of physical rules using three examples and provide guidance to help other researchers build a more solid foundation for ecosystem biogeochemical models used in predicting ecosystem-climate feedback.
Bio:
Dr. Jinyun Tang is a staff scientist in the Earth and Environmental Sciences Division at Lawrence Berkeley National Laboratory. He obtained his Ph.D. in Atmospheric Sciences from Purdue University, joined LBNL as a postdoctoral researcher in 2011, and has remained there ever since. His research encompasses various aspects of land surface modeling, focusing on developing theories, algorithms, and numerical codes that simulate and analyze climate-ecosystem feedback. Some of his landmark theoretical works include the equilibrium chemistry approximation theory for biogeochemistry and ecology, the chemical kinetics theory for temperature-dependent biochemical reactions, and the reaction-based theory for upscaling soil moisture dependence of biochemical reactions. Dr. Tang has been heavily involved in the development of the Community Land Model (versions 4.5 and 5.0) and the Department of Energy’s Energy Exascale Earth System Model. Currently, he is working on reformulating biogeochemical processes using physical rules derived from first principles, a new approach to improving the rigor and predictability of ecosystem-climate interactions. He now leads the development of EcoSIM for BioEPIC (the Biological and Environmental Program Integration Center) at LBNL, a new numerical code that uses physical rules to mechanistically integrate the interactions between plants, microbes, water, soil physics and chemistry, as well as ecosystem management and disturbances.
Summary:
Focus: modeling ecosystem response and interaction with the climate
Critical for understanding the overall climate
Needed to plan climate mitigation/carbon storage technologies
Challenge: predict land carbon cycle
Difficulties
Atmospheric model inaccuracy
Initial and boundary conditions inaccurate
Land model parameters are not well constrained
Land model implementations flawed
Uncertain land model structure
Symptom: land model parameterizations vary significantly across simulations
Types of ecosystem models
Multiplicative model:
Outcome scaled by multiple sub-systems
Multiply together terms that model the sub-systems
Min model:
Outcome is bottlenecked by multiple sub-systems
Min constraints from all sub-systems
Good: simple and easy to use
Bad: non-causal, inaccurate
Scale hierarchy in ecosystem modeling
Time: 10-3s - 109s
Space: 10-3m - 106m
Empirical biogeochemical models operate at the 101s x 101m scale
Missing key details, invariants of biochemical processes at finer scales
Argument: models founded at the finest scale will be more robust, causal and accurate
Approach to bring in constraints from fine scale:
Apply first principles/rules: conservation relationships, key diffusion, kinetic dynamics
Construct mechanistically: causal interactions of elementary units
Example: soil resistance for land-atmosphere volatile tracer exchange
Gas exchange between soil and atmosphere
Empirical approximations based on fitting curves to observational data
Derived fundamental differential equation based on gas flow laws and hydraulics
Error in measurement and limits in capturing experimental conditions makes empirical data fitting infeasible
A first principles model can account for all confounding factors
Example: affinity for plant-microbial substrate intake
Usually calibrated as a single constant
But actually a variable that depends on environment and spans > 3 orders of magnitude
Derived a first-principles model founded on individual bacteria -> microbial colonies -> soil layer + diffusion theory across scales
Can predict how soil respiration responds to soil properties without model calibration
Multi-nutrient regulated biological growth
Liebig’s law (law of minimum): growth of organisms depends (bounded by) on the nutrient that is least supplied
Very commonly used model of how nutrients regulate organism growth
Approach: a more first-principles model
Law of mass action to conserve mass/charge
Chemicals combine via multiple models: SU, Additive, LLM
Use experimental relationships between pairs of chemicals and organism growth
Parameterized with data from algal and plant growth experiments
Lieblig’s law is often a poor model for this data: very inconsistent values for the same parameters
The additive model is quite generalizable
First principles models are more generalizable and applicable across different ecosystems