Evaluating Climate Model Emulators
Simple climate models are numerical representations of the Earth's gas cycles and climate system. They are transparent, easy to use, and have low computational needs. Because of this, they are widely used by the scientific community as tools to efficiently emulate more complex and computationally intensive Earth system models. However, simple climate models have not been fully characterized. A team of researchers led by the U.S. Department of Energy’s Pacific Northwest National Laboratory applied a set of numerical tests to quantify the fundamental responses of five simple climate models to input pulses of carbon dioxide, methane, and black carbon. These tests revealed the models have different temperature responses to radiative forcing and different gas-cycle responses to emissions.
A 2016 U.S. National Academies of Science report suggested that simple climate models be “assessed on the basis of [the] response to a pulse of emissions.” This work performs that assessment, thus filling one of the knowledge gaps identified in the report. These pulse tests can also be used as a suite of standard evaluation tools for any simple climate model, and the results reveal opportunities for improving these models.
To quantify model response, the researchers ran each one using the same emissions or concentration input assumptions to get a baseline. Then, they ran each model again with each input increased in value, or perturbed, for one year. The difference in output between these two model runs is called the impulse response. The researchers used background conditions for the baseline run that changed over time instead of idealized constant concentration conditions. A background of changing greenhouse gas emissions is more realistic than idealized constant concentration conditions; it is also a procedure that simple climate models can readily implement.
The researchers quantified the temperature and gas-cycle responses over time of five simple climate models to impulses of carbon dioxide, methane, and black carbon. They found a notable difference between more highly parameterized “idealized simple climate models” and “comprehensive simple climate models,” which contain explicit physical representations of the relevant physical processes. The idealized models showed biases, while the responses of the comprehensive models were more comparable to that of more complex Earth system models.