Harnessing machine learning to enhance plant coexistence in a vegetation demographic model
Image by Justin Clark: Harmonious plant coexistence in a vibrant tropical forest ecosystem.
The Science
Tropical forests are critical components of the global carbon, water, and energy cycles that feature the highest biodiversity on Earth. However, modeling the coexistence of different plant types—a key feature of biodiversity—in these forests remains challenging. Researchers used a vegetation demographic model, the Functionally Assembled Terrestrial Ecosystem Simulator (FATES), integrated with the Energy Exascale Earth System Model land model (ELM) to improve modeling plant coexistence. The team employed advanced machine learning (ML) techniques to optimize key trait parameters in FATES, resulting in a remarkable enhancement in simulating plant coexistence. The ML approach also improved the accuracy of FATES simulations of water, energy, and carbon fluxes and aboveground biomass.
The Impact
By harnessing the power of ML, this study significantly enhanced scientists’ ability to model the coexistence of different plant types in tropical forests. Artificial intelligence-enhanced ecosystem models hold the potential to accurately predict the effects of environmental changes on diverse ecosystems, fostering effective strategies for sustainable development, carbon sequestration, and achieving carbon-neutral and net-zero emissions goals. Moreover, this research highlights the need for advancing vegetation demographic models to refine the simulation of coexisting plants to capture intricate ecosystem interactions.
Summary
A research team employed two approaches to optimize trait parameters in FATES: 1) leveraging field-based plant trait relationships, and 2) utilizing ML surrogate models. Ensembles of FATES experiments were conducted of a tropical forest site near Manaus, Brazil, in the Amazon basin. The ML-based surrogate models were used to optimize the trait parameters in FATES to improve plant functional type (PFT): sets of plants that have similar environmental responses and ecosystem roles, coexistence, and achieve better model-observation agreements. Considering only observed trait relationships improved the water, energy, and carbon simulations, but degraded PFT coexistence in ELM-FATES simulations. The ML approach significantly enhanced PFT coexistence in the FATES experiments, increasing its occurrence from 21 to 73 percent. After applying observation constraints to identify small simulation biases, the ML-guided simulations retained 33 percent of the coexistence experiments, showing a 23.6-fold increase in PFT coexistence compared to the default experiments. The ML approach also improved FATES simulations of water, energy, and carbon fluxes, as well as aboveground biomass. Based on these results, researchers proposed a reproducible method that utilizes ML to improve model fidelity and PFT coexistence in vegetation demography models. This research highlights the potential of using ML in Earth system modeling of ecosystem dynamics and their response and feedback to climate change impacts.
Contact
Ruby Leung
Pacific Northwest National Laboratory
Ruby.Leung@pnnl.gov
Funding
This research was supported by the Department of Energy’s Biological and Environmental Research program as part of the Terrestrial Ecosystem Science program through the Next-Generation Ecosystem Experiments-Tropics project.
Publications
Li, L., Y. Fang, Z. Zheng, M. Shi, M. Longo, C. D. Koven, J. A. Holm, R. A. Fisher, N. G. McDowell, J. Chambers, and L. R. Leung. “A machine learning approach targeting parameter estimation for plant functional type coexistence modeling using ELM-FATES (v2.0),” Geosci. Model Dev., 16, 4017–4040 (2023). [DOI: 10.5194/gmd-16-4017-2023]
Related Links
https://egusphere.copernicus.org/preprints/2023/egusphere-2022-1286/