Building Interpretable Climate Emulators for Economics

Abstract

This paper introduces a framework for developing efficient and interpretable carbon-cycle emulators (CCEs) within Integrated Assessment Models (IAMs). Our framework allows economists to custom-build CCEs that are accurately calibrated to advanced climate science and integrates pattern scaling techniques to convert emulator-based global mean temperature changes into spatially heterogeneous warming patterns. We present a generalized multi-reservoir linear box-model CCE that maintains key physical quantities and can be customized for specific applications. Three CCEs are showcased: the 3SR model (replicating DICE-2016), the 4PR model (incorporating the land biosphere), and the 4PR-X model (accounting for dynamic land-use changes such as deforestation that affect the reservoir's storage capacity). Evaluating these models within the DICE framework reveals that land-use changes in the 4PR-X model significantly influence atmospheric carbon levels and temperatures, highlighting the necessity of tailored climate emulators. Additionally, pattern scaling demonstrates the impact of regionally varying absolute temperatures and future temperature changes, along with their associated uncertainties, on potential economic outcomes. By offering a transparent and adaptable tool for policy analysis, our framework enables economists to more accurately evaluate the economic impacts of climate policies.

Keywords: climate change, social cost of carbon, carbon taxes, environmental policy, deep learning, integrated assessment models, DICE-2016 JEL classification: C61, E27, Q5, Q51, Q54, Q58