Development of AI-powered, low-cost, onsite earthquake early warning systems

Bridging AI and seismology to build real-time earthquake early warning systems that run directly on low-cost sensors.

Research Profile

Danuka Ravishan is a PhD researcher at CRISiSLab, Massey University, working at the intersection of Artificial Intelligence and Earthquake Early Warning (EEW) systems. His research focuses on designing and deploying deep learning models for real-time seismic detection and rapid magnitude and intensity estimation, specifically optimized for low-resource, decentralized edge devices.

His work aims to enable scalable, low-latency, and cost-effective EEW solutions tailored to New Zealand’s unique seismic environment, contributing toward next-generation national seismic resilience infrastructure.

Research Overview

Danuka’s doctoral research develops a first-of-its-kind practical implementation of an AI-based onsite EEW system capable of operating entirely on low-cost sensing hardware. The research integrates advances in deep learning, edge computing, and seismology to address key challenges in real-time earthquake warning:

Ultra-fast P-wave detection using lightweight deep neural networks
Rapid S-wave intensity and magnitude estimation from minimal early signals
End-to-end deployment on resource-constrained devices such as Raspberry Shake sensors

This work demonstrates that high-performance AI models can operate under strict computational constraints, enabling real-time decision-making directly at the sensor level, significantly reducing latency compared to centralised systems.

Key Contributions

Lightweight Deep Neural Network for Real-Time P-Wave Detection

Developed a compact CNN-based model achieving:

~97% accuracy
Sub-7 ms inference time
Real-time deployment on edge devices
Published in Nature Scientific Reports

Download Paper

Mamba-Based Early Magnitude Intensity Estimation

Designed a novel state-space (Mamba) architecture that:

Estimates ground shaking intensity and magnitude
Uses only ~2 seconds of post P-wave data
Improves real-time decision accuracy significantly
Manuscript under submission (Geophysical Research Letters)

Fully Deployed Onsite AI-EEW System

Integrated detection + estimation into a cascaded real-time system:

Running continuously on distributed sensors
Validated using real New Zealand earthquake data
Demonstrates the practical feasibility of decentralised EEW

Vision

This research represents a shift from theoretical models to deployable AI systems for real-world seismic risk reduction. By enabling intelligent processing directly on low-cost sensors, this work:

Reduces warning latency
Improves accessibility of EEW systems
Supports scalable nationwide deployment

The long-term vision is to contribute to a distributed, intelligent EEW ecosystem for New Zealand and the global community, where an AI-driven sensor provides faster and more reliable earthquake warnings.

Conference Procedings

Contact

Open to research collaborations and discussions in:

AI-powered onsite earthquake early warning.
Deep learning for real-time signal analysis.
Edge AI for distributed sensing systems.
Rapid seismic signal processing methodologies.

Reach at

LinkedIn: https://www.linkedin.com/in/danukaravishan/

Email- dmudiyan1@massey.ac.nz