India AI Impact Summit 2026

harnessing-ai-to-manage-climate-extremes-and-build-sustainable-systems

8 min read

TRACK:Resilience, Innovation, and Efficiency
SPEAKERS
  • Dev Niyogi, University of Texas at Austin, Jackson School of Geosciences
  • Dr. Akshara Kaginalkar, Atria University, Bangalore
  • Dr. Karthik Kashinath, Nvidia
  • Dr. M. Ravichandran, Ministry of Earth Sciences
  • Dr. Praphul Chandra, Atria University
  • Dr. Shivkumar Kalayanaraman, ANRF
  • Er. Manish Bhardwaj, National Disaster Mitigation Authority (NDMA)
  • Mr. Sandeep Singhal, Avaana Capital
  • Prof Amit Sheth, IAIRO & Univ. of South Carolina
SESSION OVERVIEW

The panel discussion will convene leading researchers, technocrats, investors, and policymakers to explore "AI for Sustainability and Sustainable AI." It will examine how AI can accelerate India's transition toward climate resilience and sustainable growth, particularly in managing compounding extreme events at hyper-local scales and services for impacted sectors operations. Aligned with the summit's Planet focus, the panel will discuss science driven innovation, trusted systems, and pathways to translate ideas from lab to product ecosystems.

VIDEO RECORDING

Harnessing AI to Manage Climate Extremes and Build Sustainable Systems

Abstract

The panel brought together senior scientists, technocrats, investors and policymakers to explore how artificial intelligence can accelerate India’s transition to climate‑resilient, sustainable systems. Participants examined the need for indigenous, small‑scale AI models for hyper‑local extreme‑event prediction, the role of multimodal sensing and generative AI, pathways to translate research into products, and the ecosystem of public‑private collaboration required to move from laboratory prototypes to operational services. The discussion was framed around concrete use‑cases—early‑warning systems, disaster‑response decision support, renewable‑energy grid integration and consumer‑facing resilience tools—while highlighting funding programmes, benchmark data needs and the importance of trust, validation and “Jugaad”‑style ingenuity.

Detailed Summary

The moderator welcomed the eight panelists, briefly introducing each participant and underscoring the panel’s focus on AI for Sustainability and Sustainable AI. The aim was to surface “one AI application or discovery that excites each speaker the most in the climate‑extremes domain.”

2. Vision of IRO – Indigenous, Agile AI Models

Speaker: Prof. Amit Sheth

  • The Indian Research Organisation (IRO) was conceived after a meeting with the Prime Minister (Dec 2023).
  • Goal: build original, small, agile AI models that serve hyper‑local extreme‑weather use‑cases without relying on large foundation models whose training data are opaque.
  • IRO’s roadmap includes three verticals:
    1. Earth‑science & disaster/sustainability (primary focus)
    2. Health (partnering with the Indian Pharma Alliance)
    3. Pharma (23 major manufacturers)
  • Emphasis on platform‑level integration with India’s AI infrastructure to enable rapid model deployment for specific spatial‑temporal problems.

3. Opening Question – Promising AI Applications

3.1. Hybrid Weather Forecasting (Dr. Shiv Kumar Kalayanaraman)

  • Described the current limitations of pure numerical weather prediction (NWP).
  • Advocated for a fusion of physics‑based models (good at spatial dynamics) and AI time‑series models (good at local rhythms).
  • Stressed the need for high‑impact event prediction – e.g., cloud bursts – where neither NWP nor AI alone suffices.

3.2. Trusted Early‑Warning Systems (Prof. Amit Sheth)

  • Envisioned a low‑cost, publicly accessible early‑warning service (DPG) that blends AI, sensor networks, and satellite data.
  • Highlighted the necessity of a hybrid approach and of building trust through validation.

3.3. Multimodal Sensing & Generative AI (Prof. Amit Sheth)

  • Noted the rapid fall in cost of cameras, IR, and multispectral sensors.
  • Illustrated a scenario where a simple sky‑camera, powered by generative AI, can forecast up to four hours ahead.
  • Emphasized insight‑level fusion across modalities (ground sensors, LEO satellites, radar) rather than raw‑data fusion.

3.4. Consumer‑Facing Voice Assistant (Prof. Amit Sheth)

  • Proposed a voice‑driven resilience app that translates forecasts into actionable daily guidance (“stay home”, “protect crops”, “modify travel”).
  • Integrated air‑quality alerts for urban users (e.g., Delhi).

3.5. “Jugaad” – Human‑Centric AI (Prof. Dev Niyogi)

  • Introduced the Indian concept of Jugaad (creative improvisation) as a bridge between mathematical models and human behaviour.
  • Argued that AI can capture the societal dimension missing from purely physical models, thus making predictions more accessible.

3.6. Small‑Data Fine‑Tuning & Transfer Learning (Dr. Karthik Kashinath)

  • Highlighted the challenge of fine‑tuning large foundation models with very limited local data.
  • Stressed the promise of transfer learning: leveraging data‑rich regions to improve predictions in data‑sparse locales, while respecting hyper‑local uniqueness.

4. Strengthening National AI Capabilities (Dr. M. Ravichandran)

  • Data abundance – India possesses centuries of IMD observations and satellite archives.
  • Human capital – large pool of young talent, yet under‑leveraged due to siloed approaches.
  • Key technical bottlenecks:
    1. Error propagation in NWP from assumptions – AI can reduce error via better initial conditions.
    2. Down‑scaling – need AI‑driven post‑processing to reach 1 km or finer resolution for local decision‑making.
    3. Trust, validation, verification – AI systems must meet rigorous EAML (Explainable AI, Model‑Level) standards.
  • Call for open data portals and cross‑disciplinary collaborations to bring fresh perspectives.

5. AI for Disaster‑Preparedness & Last‑Mile Impact (Dr. M. Ravichandran)

  • India’s exposure to multihazard cascades (e.g., cloudburst → landslide → flash flood).
  • Existing early‑warning capabilities (e.g., 5‑day cyclone path forecasts) have reduced mortality, but hyperlocal alerts remain weak.
  • Proposed workflow:
    1. Aggregate data from terrestrial sensors, satellites, and alert agencies.
    2. Fuse with AI models to generate granular now‑casting (sub‑kilometer) and targeted alerts.
    3. Deliver alerts to district and state authorities for timely evacuation and relief.
  • Emphasized need for startup agility to complement governmental infrastructure.

6. NRF Funding Landscape & Programme Highlights (Dr. Shiv Kumar)

  • NRF: statutory body driving research‑development funding.
  • Funding instruments:
    • Grant funding for not‑for‑profit research (academia, labs, Section‑8 entities).
    • RDI fund – a ₹1 lakh‑crore capital pool (projected to grow to ₹3‑4 lakh crore) for private‑sector scaling.
  • Ongoing programmes:
    • AI for Science & Engineering (AI‑S&E) – with an explicit Weather & Climate track.
    • Leapfrog Demonstrators for Societal Innovation – mission‑mode, outcome‑driven prototypes.
    • AI for Science Hackathon (partnered with IBM & IIT Delhi) – released curated climate datasets.
  • Collaboration model: encourage consortium‑based proposals, open IP licensing, and translational research centres that bridge academia and industry.

7. Public‑Private Partnership (PPP) Mechanics (Dr. Shiv Kumar)

  • TRL acceleration – NRF programmes push technologies from TRL 1‑2 to 5‑6.
  • Consortium‑centric calls – hub‑and‑spoke structures promote joint proposals rather than isolated bids.
  • Open IP licensing – enables startups to adopt university‑developed models quickly.
  • RDI fund – mandates industry–academia collaboration for any funded project, ensuring translation to market‑ready services.

8. NVIDIA Perspective – Benchmarks, Super‑Resolution, Hyper‑Local Modelling (Dr. Karthik Kashinath)

  1. Benchmark Datasets & Metrics – Analogous to ImageNet and ERA5, the community needs hyperlocal climate benchmarks to drive progress.
  2. Super‑Resolution – NVIDIA’s Earth‑2 program already up‑samples 25 km data to 1 km; the same techniques can be refined for 10 m–100 m scales needed for flash‑flood or landslide prediction.
  3. Model Robustness – Emphasis on physics‑aware AI, rigorous validation, and the ability to operate reliably under distribution shift (e.g., climate‑change‑driven regime changes).

9. Investment Guidance for Climate‑AI Startups (Mr. Sandeep Singhal)

  • Government partnership is essential – data access, deployment pathways, regulatory endorsement.
  • Market segmentation:
    • Public‑good services (early‑warning, disaster response) – often funded through government or philanthropy.
    • Private‑good services (energy‑grid optimisation, supply‑chain risk, insurance underwriting) – commercial revenue streams.
  • Funding sources: government schemes (NRF, RDI), philanthropic capital, and venture capital focused on climate‑tech.
  • Monetisation tip: build a dual‑track product that serves a societal need while packaging a paid, value‑added layer for enterprises.

10. AI in Energy‑Grid Integration & Sustainable Transition (Prof. Praphul Chandra)

  • India’s renewable‑energy surge (solar dominance) demands hyperlocal solar‑output forecasts for grid balancing.
  • Demonstrated a pilot linking the India Energy Stack (digital public‑infrastructure for power) to AI‑driven weather forecasts, enabling demand‑flexibility and peer‑to‑peer energy trading.
  • Highlighted the symbiosis between AI‑enhanced weather prediction and data‑center energy management – both require dynamic demand response.

11. Digital Twins & Decision‑Centric Weather Services (Prof. Dev Niyogi)

  • Weather is a “tragedy of the commons” – universally needed but under‑funded.
  • Proposed decision‑specific digital twins: lightweight, modular models that answer “what decision does the user need to make?” rather than delivering raw physical variables.
  • Stressed the importance of defining the decision pipeline, then building AI‑augmented physics to feed that pipeline (e.g., hedging agricultural contracts, routing logistics, personal outdoor activity planning).

12. Audience Query – Climate Insurance

  • The panel noted that insurance is a natural early monetisation avenue for climate‑risk analytics.
  • AI‑driven risk scores and hyperlocal hazard forecasts can feed parametric insurance products, helping insurers price policies more accurately and expanding coverage to vulnerable regions.
  • A brief follow‑up was promised after the session.

13. Closing Remarks & Announcements

  • The moderator thanked the panel and highlighted new partnerships announced at the summit: NVIDIA, Google, Qualcomm, and the Bill & Melinda Gates Foundation.
  • Reiterated the call for collaboration across academia, industry, and government to accelerate India‑centric AI solutions for climate resilience.

Key Takeaways

  • Indigenous, small‑scale AI models are preferred over large foundation models for hyper‑local extreme‑event prediction, ensuring data transparency and domain relevance.
  • Hybrid forecasting (physics‑based NWP + AI time‑series) is essential for accurate prediction of high‑impact events such as cloudbursts.
  • Multimodal sensor networks (optical, IR, LEO satellite) combined with generative AI can deliver sub‑hourly forecasts at low cost.
  • Consumer‑oriented voice assistants can translate complex forecasts into actionable everyday guidance, increasing public uptake.
  • The Indian notion of “Jugaad” highlights the need to embed human behavioural factors into AI‑driven climate services.
  • Small‑data fine‑tuning and transfer learning are critical research frontiers to adapt global foundation models to data‑sparse Indian regions.
  • Open data, benchmarking (e.g., hyperlocal climate test beds), and rigorous validation are prerequisites for trustworthy AI systems.
  • National capability building hinges on leveraging historic observation archives, nurturing interdisciplinary talent, and fostering open‑data ecosystems.
  • AI‑enhanced early‑warning systems can dramatically improve last‑mile disaster response if they deliver granular, trusted alerts to authorities and citizens.
  • NRF’s funding ecosystem (grant programmes, RDI fund, Leapfrog demonstrators, hackathons) offers concrete pathways for research‑to‑product translation.
  • Public‑private partnerships should be structured as consortiums with open IP licensing to accelerate TRL progression and market deployment.
  • Benchmarking and super‑resolution techniques (exemplified by NVIDIA’s Earth‑2) are the technical levers for achieving kilometer‑to‑meter scale forecasts.
  • Startups must align with government initiatives, clearly segment markets (public good vs. private good), and adopt dual‑track business models.
  • Energy‑grid integration of hyperlocal solar forecasts enables demand‑flexibility, supporting India’s renewable transition.
  • Digital twins that are decision‑focused rather than variable‑focused can bridge the gap between raw scientific output and actionable insight.
  • Parametric climate insurance emerges as an early commercial application, linking AI risk analytics to financial products for vulnerable populations.

Prepared by the AI Conference Summarization Team.

See Also:

Graph View

Backlinks