Reduce methane estimation uncertainty in earth system models by including eco-hydrological patch types sub-grid representation coupled with HLS-derived within-wetland patch distribution
Theresia Yazbeck, Ohio State University, yazbeck.3@osu.edu (Presenter)
Gil Bohrer, Ohio State University, bohrer.17@osu.edu
Yang Ju, Ohio State University, ju.116@buckeyemail.osu.edu
Oleksandr Shchehlov, Ohio State University, aleshcheglov@gmail.com
Madeline Scyphers, Ohio State University, scyphers.3@buckeyemail.osu.edu
Justine Missik, Ohio State University, missik.2@osu.edu
Eric Jason Ward, USGS, eward@usgs.gov
Robert Bordelon, University of Louisiana at Lafayette, robert.bordelon1@louisiana.edu
Diana Taj, University of Louisiana at Lafayette, diana.taj1@louisiana.edu
Jorge Villa, University of Louisiana at Lafayette, jorge.villa@louisiana.edu
Qing Zhu, Lawrence Berkeley National Laboratory, qzhu@lbl.gov
William J Riley, Lawrence Berkeley National Laboratory, wjriley@lbl.gov
Wetlands are considered to be the largest emitters of biogenic methane, yet, they represent the highest source of uncertainty in global methane estimations in Earth System Models (ESMs), largely attributed to the small-scale spatial and temporal heterogeneity of biogeochemical and hydrological processes driving methane production, oxidation, and transport. Due to their coarse scale, ESMs do not explicitly resolve within-wetland variability of ecosystem conditions and biogeochemical processes, which are usually represented at the whole-grid level. In addition, these variabilities are usually underepresented in remote sensing images due to their coarse spatial and temporal resolution. In this study, we are using the Exascale Earth System Model (E3SM) Land Model (ELM), where we are developing a wetland land-unit similar but separate from the existing model construct of a vegetation land-unit. Resolving wetland land units allows the model to simulate multiple eco-hydrological patches (i.e., different vegetation communities) within a wetland at the subgrid level, where distinct ecological, microbial, and hydrological parameters represent each patch type. The patch cover distribution is being incorporated as an input into ELM using global, high-resolution, Harmonized Landsat Sentinel-2 (HLS) multispectral products. We are using seasonal time-series of HLS-derived NDVI, which provide distinct seasonal temporal “fingerprints” used to classify HLS pixels to specific patch types and infer the corresponding plant cover distribution within the wetland. Two study-sites in Louisiana were used to validate our results, where regular Eddy-Covariance, chamber flux, and pore-water concentrations are sampled. Our results show a match between observed and modelled carbon and methane fluxes and methane concentration in the soil after optimizing vegetation photosynthetic rates, respiration rates, and methane production and oxidation parameters using a Bayesian Optimization approach. Our findings also show a higher precision when simulating multiple patches compared to single patch representations, thus, emphasizing the role of wetland sub-grid representation coupled with HLS-derived within-wetland patch distribution in reducing models uncertainty.
Poster: Poster_Yazbeck_2-49_37_35.pdf
Poster Location ID: 2-49
Presentation Type: Poster
Session: Poster Session 2
Session Date: Wed (May 10) 5:15-7:15 PM
CCE Program: TE