# Improving ESS approaches to evapotranspiration partitioning through data fusion

Active Dates | 9/1/2023-8/31/2025 |
---|---|

Program Area | Environmental Systems Science |

Project Description

This project aims to address the uncertainty in modeled estimates of T/ET, which is the ratio of transpiration (T) to
evapotranspiration
(ET). Previous research has shown that models often exhibit total ET estimates that match observations well, while also exhibiting strongly divergent T/ET ratios. This uncertainty in T/ET ratios limits the utility of
Earth System Models
in applications that heavily rely on this partitioning, including investigations of soil moisture dynamics, vegetation dynamics and productivity, food security, and
watershed
hydrologic response. This projects aim is to provide new insight for mechanistic modeling of ET partitioning within Earth System Models and improve future predictions by addressing three objectives.

The first objective is to create a benchmark dataset of T/ET values and associated uncertainties at select AmeriFlux locations, using established T/ET methods. Specifically, we will investigate four broad and overlapping categories of methods to estimate T/ET: theoretical models, high- frequency approaches, remote sensing, and geochemical models. We will synthesize these different approaches through colocation analysis, a diagnostic approach to assess the uncertainty in a targeted quantity by analyzing multiple inputs of the same quantity when the ‘truth’ is not well known. The second objective is to improve parameterization of transpiration and evaporation processes within an advanced Earth Systems Model through a combination of surrogate modeling and Bayesian optimization. This approach uses machine learning methods and detailed uncertainty characterizations from Objective 1 to find parameters that result in model output best matching observational data. An ensemble of Energy Exascale Earth System (E3SM) Land Model (ELM) simulations will be run at select AmeriFlux sites used to create a data fusion product to evaluate patterns and uncertainties in T/ET estimates. The third objective is to forecast trends in T/ET in the future under different scenarios using the posterior parameter distributions from Objective 2. ELM will be run with future forcing conditions to determine if T/ET will decrease in the future. This will address key uncertainties in future ecohydrologic conditions relevant to carbon, water, and energy cycling.

In summary, this project will evaluate different T/ET partitioning methods and use them to produce benchmark T/ET data products at long-term research sites and in an upscaled estimate, global T/ET estimates derived from calibrated ELM model simulations under current and future conditions. In doing so, the project will provide fundamental advancements in the characterization of process-based model uncertainty, improvement of modeling with T/ET fusion estimates, and estimation of future declines in T/ET under changing climates. The proposed research will leverage data from AmeriFlux networks and other networks and aligns with the DOE Model-Experiment (ModEx) paradigm and broad DOE Environmental System Science (ESS) objectives to improve understanding across the water, carbon, and energy cycles.

The first objective is to create a benchmark dataset of T/ET values and associated uncertainties at select AmeriFlux locations, using established T/ET methods. Specifically, we will investigate four broad and overlapping categories of methods to estimate T/ET: theoretical models, high- frequency approaches, remote sensing, and geochemical models. We will synthesize these different approaches through colocation analysis, a diagnostic approach to assess the uncertainty in a targeted quantity by analyzing multiple inputs of the same quantity when the ‘truth’ is not well known. The second objective is to improve parameterization of transpiration and evaporation processes within an advanced Earth Systems Model through a combination of surrogate modeling and Bayesian optimization. This approach uses machine learning methods and detailed uncertainty characterizations from Objective 1 to find parameters that result in model output best matching observational data. An ensemble of Energy Exascale Earth System (E3SM) Land Model (ELM) simulations will be run at select AmeriFlux sites used to create a data fusion product to evaluate patterns and uncertainties in T/ET estimates. The third objective is to forecast trends in T/ET in the future under different scenarios using the posterior parameter distributions from Objective 2. ELM will be run with future forcing conditions to determine if T/ET will decrease in the future. This will address key uncertainties in future ecohydrologic conditions relevant to carbon, water, and energy cycling.

In summary, this project will evaluate different T/ET partitioning methods and use them to produce benchmark T/ET data products at long-term research sites and in an upscaled estimate, global T/ET estimates derived from calibrated ELM model simulations under current and future conditions. In doing so, the project will provide fundamental advancements in the characterization of process-based model uncertainty, improvement of modeling with T/ET fusion estimates, and estimation of future declines in T/ET under changing climates. The proposed research will leverage data from AmeriFlux networks and other networks and aligns with the DOE Model-Experiment (ModEx) paradigm and broad DOE Environmental System Science (ESS) objectives to improve understanding across the water, carbon, and energy cycles.

Award Recipient(s)

- Oregon State University, Corvallis (PI: Good, Stephen)