A Hindcast Experiment Using the GCAM 3.0 Agriculture and Land-Use Module

TitleA Hindcast Experiment Using the GCAM 3.0 Agriculture and Land-Use Module
Publication TypeJournal Article
Year of Publication2017
AuthorsCalvin, Katherine, Wise Marshall, Kyle Page, Clarke Leon, and Edmonds Jae
JournalClimate Change Economics
Volume8
Number1
Pages1750005
Date Published03/2017
Abstract / Summary

We report results of a “hindcast” experiment focusing on the agricultural and land-use component of the Global Change Assessment Model (GCAM). We initialize GCAM to reproduce observed agriculture and land use in 1990 and forecast agriculture and land use patterns on one-year time steps to 2010. We report overall model performance for nine crops in 14 regions. We report areas where the hindcast is in relatively good agreement with observations and areas where the correspondence is poorer. We find that when given observed crop yields as input data, producers in GCAM implicitly have perfect foresight for yields leading to over compensation for year-to-year yield variation. We explore a simple model in which planting decisions are based on expectations but production depends on actual yields and find that this addresses the implicit perfect foresight problem. Second, while existing policies are implicitly calibrated into IAMs, changes in those policies over the period of analysis can have a dramatic effect on the fidelity of model output. Third, we demonstrate that IAMs can employ techniques similar to those used by the climate modeling community to evaluate model skill. We find that hindcasting has the potential to yield substantial benefits to the IAM community.

URLhttp://www.worldscientific.com/doi/abs/10.1142/S2010007817500051
DOI10.1142/S2010007817500051
Funding Program: 
Journal: Climate Change Economics
Year of Publication: 2017
Volume: 8
Number: 1
Pages: 1750005
Date Published: 03/2017

We report results of a “hindcast” experiment focusing on the agricultural and land-use component of the Global Change Assessment Model (GCAM). We initialize GCAM to reproduce observed agriculture and land use in 1990 and forecast agriculture and land use patterns on one-year time steps to 2010. We report overall model performance for nine crops in 14 regions. We report areas where the hindcast is in relatively good agreement with observations and areas where the correspondence is poorer. We find that when given observed crop yields as input data, producers in GCAM implicitly have perfect foresight for yields leading to over compensation for year-to-year yield variation. We explore a simple model in which planting decisions are based on expectations but production depends on actual yields and find that this addresses the implicit perfect foresight problem. Second, while existing policies are implicitly calibrated into IAMs, changes in those policies over the period of analysis can have a dramatic effect on the fidelity of model output. Third, we demonstrate that IAMs can employ techniques similar to those used by the climate modeling community to evaluate model skill. We find that hindcasting has the potential to yield substantial benefits to the IAM community.

DOI: 10.1142/S2010007817500051
Citation:
Calvin, K, M Wise, P Kyle, L Clarke, and J Edmonds.  2017.  "A Hindcast Experiment Using the GCAM 3.0 Agriculture and Land-Use Module."  Climate Change Economics 8(1): 1750005.  https://doi.org/10.1142/S2010007817500051.