Improving Simulations of Carbon Cycle Feedbacks Through Integration of ELM with Observations and Experiments in Vulnerable Ecosystems
Authors
Daniel Ricciuto* (ricciutodm@ornl.gov), Jiafu Mao, Anthony Walker, Xiaoying Shi, Xiaojuan Yang, Melanie Mayes, Paul Hanson
Institutions
Environmental Sciences Division and Climate Change Science Institute, Oak Ridge National Laboratory, Oak Ridge, TN
URLs
Abstract
Carbon cycle feedback parameters estimated by the current generation of Earth system models may be underestimated due to missing processes and uncertain parameters. One key shortcoming is that most land-surface models do not represent peatlands, which cover a relatively small fraction of Earth’s surface but contain a much larger proportion of terrestrial carbon stocks that may be vulnerable to environmental change. The Spruce and Peatland Responses Under Changing Environments (SPRUCE) whole-ecosystem warming and elevated carbon dioxide (CO2) experiment began in 2015 in an ombrotrophic bog in Northern Minnesota and will continue through 2025. Results to date indicate strong releases of carbon (C) from warming. This results from a combination of factors including increased heterotrophic respiration, increased methane (CH4) emissions, and a loss of productivity and coverage of Sphagnum mosses. Upscaling SPRUCE to better understand regional-scale feedbacks requires a process-based modeling approach. In addition to missing peatland and wetland processes, many parameters regulating C feedbacks in surrounding upland systems are also uncertain. The most vulnerable C stores of boreal and temperate ecosystems may be those in vegetation or organic soils located near ecological boundaries, including at SPRUCE, which is at the southern edge of the boreal zone. Researchers introduce a new version of the Energy Exascale Earth System Model (E3SM) land component that is augmented to include peatland processes (ELM-Peatlands), which is initially tuned using SPRUCE observations and scaled regionally using a combination of remote sensing and ground-based observations. Researchers use a neural network-based surrogate modeling approach to constrain ELM parameters for five major plant functional types using observed carbon and energy fluxes from long-record AmeriFlux wetland and forested upland sites. Researchers present initial simulations using the constrained ELM-Peatlands over a regional domain covering central North America and compare feedback parameters estimated in this model to those in the CMIP6 ensemble.