Day 1: Climate Data Science Grant Awardee Presentation - Salman Khan | ADIA Lab Symposium 2025

Value of the Information & Strength of the Argument

The presentation offers a clear and well-structured vision for applying agentic AI to climate challenges, building on the presenter’s prior work in multimodal earth observation models. The argumentation is logical, moving from problem definition to proposed solution, and acknowledges limitations such as hallucination and the need for uncertainty quantification. However, it remains at a proposal stage, with no empirical results or comparative analysis to substantiate the claimed advantages over existing siloed models.

The title accurately reflects the content: a grant awardee presentation on climate data science, delivered by Salman Khan at the ADIA Lab Symposium 2025.

7/10

Presentation by a recognized researcher in computer vision, with references to published works (EarthDial, CopperFM) and established datasets (ERA5, CMIP6, GRACE). However, it is a proposal, not peer-reviewed results, and lacks detailed methodological validation.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:00Introduction and acknowledgment of ADIA Lab award and collaborators.
• 00:30Presentation of the MENA twin crisis: water depletion and crop stress, with example of Mosul Dam lake.
• 02:00High-level overview of the proposed AI agentic pipeline, integrating satellite, IoT, climate models, and reports.
• 05:00Core hypothesis: LLMs as orchestrators to reason across fragmented modalities and models.
• 08:00Data pillars: satellite remote sensing (GRACE, Sentinel-2, MODIS), streamflow, climate projections (ERA5, CMIP6), and reports.
• 10:00Prior work: EarthDial and CopperFM as foundations for multimodal understanding.
• 12:00Agentic pipeline: decomposition of complex queries, use of tools, and explainable step-by-step reasoning.
• 15:00Challenges: hallucination, uncertainty quantification, tool chain audit, and benchmarks.
• 18:00Potential impact: water authorities, farmers, policymakers, and scientific community.
• 20:00Q&A: interpretability, data security, and collaboration with Destination Earth.

• EarthDial — Mentioned as prior work presented at CVPR, a conversational model for earth observation data.
• CopperFM — Mentioned as prior work, best paper candidate at ICCV, expanding input modalities.
• Aurora model — Mentioned as a weather forecasting model with extensive variables.
• ERA5 — Mentioned as reanalysis data product for historical climate data.
• CMIP6 — Mentioned as long-term climate projections.
• GRACE — Mentioned as satellite mission for groundwater estimation.
• Sentinel-2 — Mentioned as multispectral satellite data source.

• EarthDial — Prior work by the presenter, supporting the feasibility of multimodal EO models.
• CopperFM — Prior work by the presenter, supporting the feasibility of multimodal EO models.