Key Focus Areas
Dive into our Key Focus Areas¶
Earth embeddings for sustainability solutions, Earth embeddings, a type of compression of the original data source, keep viable information from environmental data while preparing it for use in AI models to make predictions. Break out groups can use cutting-edge Earth embeddings already developed by the ESIIL team and partners, e.g., the MOSAIKS (Multi-task Observation using SAtellite Imagery and Kitchen Sinks) embedding method tailored to multispectral satellite imagery or computer vision models that identify wildlife in images. We seek to inspire teams to develop a critically-needed library of approaches for embeddings that leverage the variety of environmental biology data (e.g., acoustic data, satellite data, landscape data, camera-trap data, eDNA data, and many others) that will enable us to use AI to ask questions about how species and ecosystems will respond to future global change.
Continental-scale digital twins for the U.S.: There is a rapid movement to create planetary digital twins, but initiatives may lack the domain knowledge and critical data inputs to make them robust and useful. Digital twins, or digital replicas of key Earth system processes that leverage and learn from contemporary information, could be beneficial tools in understanding and defining sustainability strategies. Breakout groups will focus on building frameworks and identifying key environmental biology data, analytical models, cyberinfrastructure, and domain needs for sustainability-focused digital twins. We need near-real-time predictions enabled by edge computing and longer-term forecasting systems that will derive from the interaction of domain experts and decision-makers with digital twins. But we need our environmental data science community to help ground these frontier AI-driven technologies in appropriate scales, data, domain knowledges, and management objectives. This theme tackles the opportunity to create digital twins for sustainable management and conservation of our nation’s biodiversity and ecological heritage.
Best practices for using large language models for sustainability: Large language models provide powerful new capabilities for scientific synthesis, enabling researchers and stakeholders to rapidly integrate information, generate hypotheses, and construct analytical workflows that accelerate environmental discovery and solutions generation. As conservation challenges intensify, LLMs will become important tools for exploring environmental futures—helping decision makers compare alternative trajectories and conduct spatial planning, evaluate new text-derived data about threatened species, and assess global change impacts through natural language processing of social media. The rapid uptake of LLMs in the context of sustainability, however, raises critical questions about how to quantify uncertainty, detect and mitigate bias, and ensure environmentally responsible and energy-efficient workflows. Break out groups will explore these emerging challenges and opportunities, helping to develop community-based best practices for using LLMs to advance both environmental understanding and sustainability-oriented decision making.
AI and causal inference—understanding ecological mechanisms & levers for decision-making: In environmental science, ecology, and resource management, we often seek to understand cause-and-effect relationships, such as identifying the drivers of tree mortality or quantifying the effect of fisheries management on coral reef health. The field of causal inference leverages large observational datasets to detect and quantify causal relationships in complex social-ecological systems. Recent advances in machine learning/AI can advance causal understanding by assessing heterogeneity in causal effects, capturing nonlinear relationships and complex interactions, and expanding the types of data available for causal inquiry. Yet all causal inference relies on fundamental domain knowledge of the social-ecological system of study. Breakout groups will seek methodological advances in the fusion of causal inference with machine learning and AI, and identify research questions that advance understanding of sustainability.