A Multimodal Causal Framework for Large-Scale Ecosystem Valuation: Application to Wetland Benefits for Flood Mitigation

Climate change is poised to alter wetland ecosystems through changes in temperature and precipitation patterns, compounding the already pronounced influence of human-driven wetland development. In this context, policymakers and environmental managers would benefit from accurate wetland valuations to guide their decision-making, as their choices regarding this critical natural resource directly impact flood mitigation efforts, biodiversity conservation, and economic activity. This paper introduces a novel multimodal causal framework for producing location-specific ecosystem valuations at a national scale to be used in cost-benefit policy analysis. It leverages recent advances in estimating heterogeneous treatment effects to flexibly determine how the expected impact of ecosystem-level changes---such as wetland loss via development---varies conditional on high-dimensional and multimodal measures that characterize the complex interactions between human and natural systems such as aerial satellite imagery, weather sequence data, land cover classifications, and water surface networks. From this effort, we aim to create a national database of location-specific wetland valuations in an approach that can be readily extended in estimating the effect of other interventions on ecosystems. We also plan to generate open-source feature embeddings for each U.S. wetland, embeddings that can be used to address other climate-related causal questions as well.