Intership Projects 2024-25

This study develops an AI model for flood impacts detection, leveraging remote sensing data. It also incorporates news reports to combine image and text data, aiming for improved accuracy in detecting flood impacts

This project structured historical flood impact data from unstructured web sources using the DigitalNZ API, Python-based ETL workflows, and a Large Language Model (GPT-4). Extracted impact information tagged with date and location, was normalized and stored in PostgreSQL. The resulting dataset enables future research, risk analysis, and disaster resilience decision-making

This project develops a New Zealand earthquake catalogue using MEMS strong motion data from Raspberry Shake stations. By capturing waveforms based on GeoNet information, it addresses the lack of a MEMS earthquake database in New Zealand. This catalogue is vital for CRISI-Lab's low-cost EEW solutions.

This project focuses on enhancing earthquake detection capabilities through innovative applications of deep learning and generative models. Key efforts included adapting the PhaseNet architecture for short-window input, optimizing its performance for deployment on edge devices.

This project aims to deploy a Meshtastic LoRa network and management system to provide emergency communication in the event of a natural disaster that disables or destroys our communication infrastructure. Since the start of this project we have identified

This project centers on the backend and firmware development for a Meshtastic LoRa network, providing emergency communication when natural disasters disrupt infrastructure. Key efforts involve optimizing firmware upgrades for existing LoRa devices to enhance information and network capabilities, alongside creating a backend server-side application for comprehensive network management. The system aims to deliver critical, resilient communication.

This study developed a landslide detection method using satellite imagery and machine learning on Google Earth Engine. Applied in Sri Lanka, it achieved over 94% accuracy. Combining optical, SAR, and topographic data improved performance, highlighting slope and soil index as key predictors for scalable landslide mapping in data-scarce regions.

This short talk outlines a practical image-processing workflow for detecting changes in coastal cliffs using citizen-taken photographs. By aligning and comparing greyscale images within a defined region of interest, significant differences—such as landslides—are clearly highlighted. The method supports community-based monitoring of coastal hazards through low-cost, replicable visual analysis.