GeoAI guardians: why disaster risk tools must model cascading impacts

From Texas to Japan to India, disasters continue to reveal a critical lesson: infrastructure systems do not fail in isolation.

When one system breaks down, others often follow. A new generation of artificial intelligence (AI)-enabled tools – referred to here as GeoAI Guardian– aims to anticipate not only initial failures, but also the cascading impacts on water, health, and communication systems.

In February 2021, Winter Storm Uri brought extreme cold to Texas. The electricity grid, managed by the Electric Reliability Council of Texas (ERCOT), was not designed for such conditions. Within 48 hours, more than four million households lost power. The impacts quickly spread: water treatment systems failed, natural gas supply was disrupted, and healthcare facilities relied on emergency power. At least 246 people died, many due to prolonged exposure to cold. This was not a single-system failure, but a cascade of interconnected risks.

A similar pattern occurred during the 2011 Great East Japan Earthquake and tsunami. While the hazard itself was anticipated, cascading failures led to the Fukushima Daiichi nuclear accident. Flooded backup generators disabled cooling systems, resulting in reactor meltdowns. The sequence of failures – rather than the initial hazard – drove the scale of the disaster.

In 1999, Super Cyclone Odisha caused widespread destruction in India, killing nearly 10,000 people. Power outages triggered failures in water supply, healthcare services, and communication systems, prolonging the crisis for months. Despite investments in stronger infrastructure since then, risk assessments often still focus on whether assets can withstand hazards, rather than how system failures affect people and essential services.

A persistent gap in risk modelling

These examples highlight a common limitation in disaster risk modelling: infrastructure systems are often assessed separately, despite their interdependence.

The Global Infrastructure Risk Model and Resilience Index (GIRI), developed by the Coalition for Disaster Resilient Infrastructure (CDRI), provides a comprehensive framework for assessing risk. It evaluates hazard exposure, infrastructure vulnerability, and resilience capacity to estimate potential losses. I applied it to Odisha's power sector and it produced a resilience score of 2.8 out of 10. Applied to Texas's ERCOT grid before Uri, it would have flagged winterisation gaps years in advance.

However, such models are typically static, offering periodic assessments rather than real-time insights during evolving hazard events. When Uri was 36 hours from Texas, a high-resolution assessment of the grid system was a year old. When the Tōhoku tsunami was inbound, no model was calculating how generator fuel levels at Fukushima would interact with projected seawall overtopping heights. When a Category 4 cyclone is twelve hours from Paradip, the India Meteorological Department is issuing track updates every three hours — but that intelligence is not yet flowing into the software that tells Odisha Power Transmission Corporation Limited (OPTCL) where to pre-position repair crews.

As a result, critical decisions during emergencies are often made without fully understanding how failures may cascade across systems.

From static models to dynamic systems

GeoAI-based approaches aim to address this gap by creating dynamic representations of infrastructure systems. A “GeoAI Guardian” can be understood as a digital model that integrates multiple real-time data sources to support decision-making during disasters.

These systems combine:

• Meteorological data (e.g. from national weather services);
• Satellite imagery for monitoring hazards and damage;
• Grid telemetry from Supervisory Control and Data Acquisition (SCADA) control systems to monitor, gather, and process real-time data, enabling automated control and reduced human intervention;
• AI-adjusted fragility curves, which are probabilistic tools to estimate the probability of a structure (buildings, bridges) exceeding specific damage states at given seismic intensity levels (like peak ground acceleration.

Four key features distinguish this approach

First, real-time updates allow risk assessments to evolve as hazards develop. This enables more timely and informed decisions.

When a cyclone's track shifts twenty kilometres northward at 3 a.m., the model's fragility map shifts with it — before the shift has any operational consequence.

Second, cross-sector modelling captures dependencies between systems such as power, water, telecommunications, and healthcare. This helps identify how failures in one sector may affect others.

Applied to Odisha's Paradip 220kV substation, the simulation shows a primary failure propagating to six distribution zones within two hours, exhausting telecom battery backups by hour eight, and silencing four district hospital Intensive Care Units (ICUs) within twenty-four.

Third, dynamic vulnerability assessments incorporate factors such as infrastructure condition, maintenance history, and environmental exposure, improving the accuracy of risk estimates.

In Odisha, this recalibration revealed that five of seven major transmission corridors exceed 75 percent fragility under Category 4 conditions — a threshold corresponding to near-certain failure. The worst performer, the Balasore coastal 132kV line, scores 91 percent; it is 41 years old.

Fourth, actionable outputs translate risk analysis into practical steps, such as pre-positioning repair teams, prioritising fuel supply for hospitals, or managing power distribution ahead of a hazard. This fills the gap between risk awareness and risk management.

Implications for disaster risk reduction

To realise the potential of such approaches, several changes are needed.

First, disaster risk governance should require modelling of cascading impacts across sectors. This aligns with the Sendai Framework for Disaster Risk Reduction 2015–2030, which calls for a comprehensive understanding of risk across interconnected systems; so-called cascade maps.

Second, stronger integration is needed between infrastructure operators and meteorological agencies to enable real-time data sharing and coordinated response.

POSOCO, India's grid operation corporation, already receives India Meteorological Department (IMD) alerts. The Electric Reliability Council of Texas (ERCOT) now has a winter weather programme following Uri's catastrophic exposure of its winterisation gaps. What is missing globally is the algorithmic layer that translates a weather track update into a prioritised operational response.

Third, critical services such as healthcare must plan for prolonged, interconnected disruptions. Backup systems, including emergency power, should be designed to account for extended outages and cascading failures.

Post-Uri, Texas's Public Utility Commission introduced mandatory weatherisation. The analogous reform globally is cascade-aware fuel reserve requirements for critical healthcare facilities.

From risk awareness to risk-informed action

Uri, Tōhoku, the 1999 Odisha Cyclone — in retrospect, none of these disasters were surprises. The hazards were known. The infrastructure was mapped. The fragility was measurable. What was missing was the model that connected these facts in real time and said: if that substation fails tonight, here is who dies by morning.

GeoAI-based tools do not predict disasters with certainty. Instead, they support more effective decision-making by identifying potential cascading impacts and enabling timely interventions: move this crew, fuel this generator, de-energise this line before the storm arrives.

Somewhere right now, a winter storm is deepening over the Southern Plains. A seismic sequence is building off a Pacific subduction zone. A tropical depression in the Bay of Bengal is organising itself into something worse. In each of these systems, a power substation is the first domino. The question is not whether it will fall. The question is whether we will have moved the fuel truck before it does. The GeoAI Guardian exists to make that question answerable. It is past time the world started asking it.

Strengthening the ability to anticipate and manage these cascading risks is essential for reducing disaster impacts. Moving from static assessments to dynamic, system-wide approaches can help ensure that critical services remain operational and that communities are better protected when disasters occur.

Sanjay K. Srivastava is the S. Radhakrishnan Chair Professor at the National Institute of Advanced Studies (NIAS), Indian Institute of Science campus, an Adjunct Research Professor at the United Nations University (UNU) Hub The City College of New York and an Adjunct Faculty at the Department of Emergency Medicine, KMC/MAHE.