PhD Projects

Using citizen science data as an approach to make low-cost and safe monitoring programmes in landslide zones: Cape Kidnappers and Taranaki North Cliffs case studies

This PhD research project explores the potential of citizen science data to enhance traditional landslide monitoring techniques that can be costly and pose safety risks to personnel. By studying alternative data sources, the project will develop a framework to integrate citizen science initiatives for effective landslide monitoring. The study will focus on two case sites with active rockfall hazards, Cape Kidnappers and Taranaki North, to reduce risks and improve safety for the community and

​

Deep Learning-Based Onsite Earthquake Early Warning System for Edge Devices

​

This research advances deep learning methodologies for real-time earthquake detection and rapid estimations, with a strong focus on deployment in low-cost, resource-constrained edge devices. By designing efficient neural network architectures and optimizing them for fast inference, the work enables accurate, low-latency onsite earthquake early warning using minimal computational resources. The outcomes demonstrate the practical potential of AI-driven, decentralized seismic monitoring systems for scalable and resilient hazard mitigation.

​

Crowdsourced and AI-driven approach for impact-based flood forecasts and warnings in New Zealand​

​

My research focuses on enabling impact-based flood early warnings in New Zealand by addressing key modelling challenges. A major barrier to such warnings is the computational demand of conventional phsyics-based numerical simulations. To overcome this, I develop AI-driven models for rapid, high-resolution flood inundation forecasting. This work sits at the intersection of data science, natural hazards, and emergency management, and involves integrating dynamic dataset such as hydrological and topographical data to model flood hazards and predict their impacts at fine spatial scales, including the household level. To support the dissemination of these localised forecasts, I am also developing a mobile app based platform. The app provides users with real-time, location-specific impact forecasts and enables crowdsourced reporting of local flood observations. This two-way data flow not only helps validate and improve model performance but also empowers communities with timely and actionable information.

An eco-system of low-cost ground motion sensors toward Earthquake Early Warning System using P-waves

​

This PhD research project addresses knowledge gaps in earthquake early warning systems (EEWS), particularly in implementing decentralised processing, ground motion-based algorithms, and using different types of low-cost micro-electro-mechanical systems (MEMS)-based sensors. The project aims to develop a P-wave-based decentralised low-cost EEWS using a ground motion-based algorithm and an EEWS that can work with different types of MEMS-based sensors. The research will be conducted in New Zealand, where there is a lack of a nationwide EEWS.​

A Decentralised Architecture for Low-Cost Earthquake Early Warning Systems:
Design, Simulation, and Real-World Validation

​

This PhD research project proposes a decentralised EEW framework leveraging low-cost MEMS sensor networks to enhance systemic resilience and scalability. The study begins by establishing a conceptual framework that defines the operational logic of the decentralised EEW system. In the process, utilising Model-Based Systems Engineering (MBSE), a comprehensive system model will be developed to ensure architectural coherence. To evaluate the framework's effectiveness, a step-by-step approach will be used to construct a Discrete-Event Simulation (DES) environment that incorporates stochastic variables and real-world uncertainties. The simulation results from diverse seismic scenarios will be used to refine the system's logic and performance parameters iteratively. In the final phase, the optimised framework will be deployed across the existing  MEMS sensor network in Wellington, New Zealand. This real-world implementation intends to validate the system’s operational performance in live conditions, with field findings used to calibrate and enhance the fidelity of the simulation environment.

Advancing Extreme Rainfall Forecasting Using AI and Machine Learning-Based Generative Models

​

This PhD research project addresses critical challenges in extreme rainfall forecasting, particularly in improving predictive accuracy, capturing uncertainty, and enhancing the representation of complex spatiotemporal atmospheric processes. Despite advancements in numerical weather prediction and data-driven approaches, existing models often struggle to reliably predict extreme rainfall events due to their rarity, nonlinearity, and strong dependence on dynamic climate variables.

The project aims to develop an advanced generative modeling framework for rainfall forecasting that integrates climate-scale atmospheric data with modern artificial intelligence and machine learning techniques. The research will focus on leveraging high-resolution reanalysis datasets to model the interactions between key atmospheric variables, enabling more accurate and robust prediction of extreme rainfall events.

A key objective of this research is to improve the ability of forecasting systems to quantify and represent uncertainty, which is essential for risk-informed decision-making in disaster management. The study will also explore the effectiveness of different combinations of atmospheric predictors and modeling strategies to better capture spatiotemporal rainfall dynamics.

The research will be conducted in the New Zealand context, where extreme rainfall events significantly impact infrastructure, agriculture, and communities, and where improved forecasting capabilities can contribute to more effective early warning systems and disaster risk reduction strategies.

Multimodal Generative AI For Reliable and Effective Disaster Situation Awareness Incorporating Social Media

​

The proposed project aims to enhance the validity of social media data for disaster situation extraction by utilizing multimodal data. Geo-locating and temporally aligning disaster related social media posts can work towards enhanced trustworthiness. Aiming to bridge the gaps in how different data sources and modalities such a remote sensing imagery can be used for precisely geolocating social media text and image data is studied in my work. Working with the recent advancements in AI, I aim to work towards a practicable implementation that can have a real impact in emergency management.