A Tool for Sharing Empirical Models of Climate Impacts

Scientists, policy advisors, and the public struggle to synthesize the quickly evolving empirical work on climate change impacts. The Integrated Assessment Models (IAMs) used to estimate the impacts of climate change and the effects of adaptation and mitigation policies can also benefit greatly from recent empirical results (Kopp, Hsiang & Oppenheimer, Impacts World 2013 discussion paper). This paper details a new online tool for exploring, analyzing, combining, and communicating a wide range of impact results, and supporting their integration into IAMs. The tool uses a new database of statistical results, which researchers can expand both in depth (by providing additional results that describing existing relationships) and breadth (by adding new relationships). Scientists can use the tool to quickly perform meta-analyses of related results, using Bayesian techniques to produce pooled and partially-pooled posterior distributions. Policy advisors can apply the statistical results to particular contexts, and combine different kinds of results in a cost-benefit framework. For example, models of the impact of temperature changes on agricultural yields can be first aggregated to build a best-estimate of the effect under given assumptions, then compared across countries using different temperature scenarios, and finally combined to estimate a social cost of carbon. The general public can better understand the many estimates of climate impacts and their range of uncertainty by exploring these results dynamically, with maps, bar charts, and dose-response-style plots. Front page of the climate impacts tool website. Sample "collections" of models, within which all results are estimates of the same fundamental relationship, are shown on the right.

Simple pooled result for Gelman's "8 schools" example. Pooled results are calculated analytically, while partial-pooling (Bayesian hierarchical estimation) uses posterior simulations.