Background/Objectives
The emergent consequences of climate change are increasingly apparent, affecting a multitude of human activities. From first-order impacts such as heat waves and droughts to second- and third-order effects like crop failures, water scarcity, social upheaval, and political instability, the complex interplay between these phenomena presents significant challenges for accurate short- to medium-term (3-24 months) forecasting. Effective planning and preparation for these impacts necessitates the simultaneous consideration and modeling of multiple domains and data sources. Further, an analytical framework that quantifies the inherent uncertainties in such multi-domain, multi-source analyses is essential for providing effective decision support.
Approach/Activities
We are pursuing initial activities as a proof-of-concept for this work:
- Establish a UQ (Uncertainty Quantification) framework that enables the evaluation and integration of many different data sources with differing certainty profiles. This will combine vastly different types of data: climate model outputs, domain-specific indicators and reports, news sources, civil infrastructure information, remote sensing/geospatial data, and survey responses.
- Create a forecasting model of regional fish abundance built on the predicted climate indicators.
- Develop an LLM-based NLP system capable of extracting and tagging information related to specific social domains that can be integrated into forecasts and analyses.
- Design mockups of an initial User Interface (UI), demonstrating how the downstream effects arising from the uncertainty in source data may be presented to enhance decision support for future end users.
Results/Lessons Learned
We are in an early phase of research and development but have already built initial pipelines for many data modalities, including different forms of climate, text, and geospatial data. We have also identified a UQ framework that will allow the uncertainty to propagate across data modalities, providing the flexibility to answer a variety of questions and maintain a distribution of likely outcomes. While our framework is designed to be adaptive to forecast impacts ranging from first-order to third-order, we will use our fishing abundance model as an illustrative example and highlight some initial UI development.
