Advancing Edge AI for Earthquake Early Warning

Excited to share CRISiSLab's latest publication in Nature Scientific Reports, where Danuka Ravishan leads a project exploring how ultra-lightweight deep learning models can enable real-time earthquake detection directly on low-cost, low-power, MEMS-based edge devices such as Raspberry Shake or similar.

Citation: Ravishan, D., Prasanna, R., Herath, P. et al. Lightweight convolutional neural network for real-time earthquake P-wave detection on edge devices in New Zealand. Sci Rep 16, 11536 (2026). https://doi.org/10.1038/s41598-026-42568-y

Earthquake early warning systems traditionally rely on centralised processing and computationally heavy deep learning models. While landmark works such as PhaseNet, Generalized Phase Detection, and EQTransformer have significantly advanced automated seismic analysis, their architectures are often designed for high-performance computing environments rather than always-on embedded sensing networks.

This project takes a different perspective: instead of prioritising model complexity, it focuses on extreme efficiency for operational deployment. A key contribution of this research is the development of an ultra-compact convolutional neural network for real-time P-wave detection specifically optimised for edge computing environments. Compared with widely used deep-learning seismic pickers such as PhaseNet, GPD, and EQTransformer, which prioritise large-scale accuracy and generalisation, the proposed model demonstrates that high detection performance can be achieved with dramatically reduced computational complexity. The architecture contains ~38K parameters, operates on 2-second waveform windows, and achieves sub-7 ms inference on Raspberry Pi-class hardware, enabling detection directly at sensor nodes rather than relying on centralized servers. By training and validating the system using New Zealand strong-motion data, the work demonstrates how lightweight AI models can support scalable, distributed, and cost-effective earthquake early warning networks, particularly in regions where dense seismic instrumentation remains challenging.

Why this matters:

• Enables AI directly on seismic sensors

• Supports distributed EEW architectures

• Reduces reliance on centralised processing infrastructure

• Opens pathways for low-cost community-scale sensing networks

As extreme events continue to challenge traditional infrastructure, edge AI combined with low-cost sensing technologies may play a critical role in the next generation of disaster early warning systems.

Look out for some exciting outcomes from our EdgeAI work, building on the work published in this paper...!!