PNNL

The Science
Future changes in Earth system state will impact agricultural yields and, through these changed yields, have profound impacts on the global economy. The results of past crop modeling efforts require too much computation to interact dynamically with the Global Change Assessment Model (GCAM). Previously, researchers have had to rely on public archives of pre-run scenarios provided by crop modelers, making exploration of feedbacks among socioeconomics, Earth system changes, and crop yield changes impossible. Persephone v1.0 evaluates arbitrary Earth system changes and produces yield changes in response by emulating the Coordinated Climate-Crop Modeling Project (C3MP) sensitivity test data set from the Agricultural Model Intercomparison and Improvement Project (AgMIP). The breadth of data included in the C3MP data set also allows Persephone v1.0 to provide a measure of uncertainty associated with a yield change.

The Impact
With the Persephone yield responses, a new variety of agricultural impact experiments will be open to GCAM and other economic models. Because Persephone v1.0 doesn’t need pre-computed Earth system inputs to produce yield changes, it can interact directly with GCAM to examine new, more complex questions. For example, Persephone can be used with GCAM to examine the economic impacts of a multi-year drought in a key agricultural region and how those economic changes can, in turn, impact the drought itself.

Summary
To examine the feedback loop among socioeconomics, Earth system changes, and crop yield changes, rapidly generated yield responses with some quantification of crop response uncertainty are desirable. The Persephone v1.0 response functions presented in this work are based on the AgMIP C3MP sensitivity test data set and are focused on providing GCAM and similar models with a tractable number of rapid-to-evaluate dynamic yield response functions corresponding to a range of C3MP data set yield response sensitivities. This is one solution to the challenge of linking detailed, process-based crop models to global, multi-region, multisector models such as GCAM in a computationally tractable way.

Using given training data, Persephone employs a Bayesian framework to estimate the distributions of agricultural response function parameters. Future versions of Persephone can use this Bayesian framework to explore different functional forms and different crop yield training data sets as more become available. This approach is relatively novel in the field of crop yield response functions and was noted as a strength by reviewers.  Further, future projections of yield changes from Persephone were placed in the context of more detailed process-based crop modeling results. It was found that the Persephone results were largely consistent with past crop modeling efforts. This emulator is supported by an open-source R package (https://github.com/JGCRI/persephone).

PI Contact
Mohamad Hejazi, Pacific Northwest National Laboratory, Mohamad.Hejazi@pnnl.gov  

Funding
This research was supported by the U.S. Department of Energy Office of Science, Biological and Environmental Research through the Multisector Dynamics, Earth and Environmental System Modeling Program. Additional support was provided by the NASA Climate Impacts Group under the Modeling, Analysis, and Prediction Program.