PhD in Bridging Physics and Earth Observation: Geospatial foundation models for real-time flood mapping.

About the project

About the Project Overview of the Project Flooding is one of the most significant natural hazards in the United Kingdom, with increasing frequency and severity driven by climate change and urban expansion. Accurate and timely mapping of flood extent is critical for disaster response, infrastructure resilience, and long-term planning. Traditional flood modelling approaches fall into two broad categories. Physically based hydrodynamic models, such as CityCAT, simulate flood processes with high accuracy but are computationally expensive and difficult to deploy at national scale in real time (Neal et al., 2012; Glenis et al., 2018). In contrast, remote sensing approaches using satellite imagery, particularly from Sentinel-1, enable rapid and large-scale flood detection but are limited by issues of generalisation, dependence on labelled data, and difficulty capturing underlying hydrological processes (Schumann and Bates, 2018). Within the UK, operational flood mapping by the Environment Agency combines modelling and observations but still relies heavily on precomputed flood risk maps and does not fully exploit real-time satellite data for nationwide dynamic prediction. Recent advances in deep learning have improved performance in flood mapping tasks, yet these models often remain event-specific and struggle to transfer across geographic regions (Bentivoglio et al., 2022). At the same time, the emergence of geospatial foundation models trained on large-scale Earth Observation data offers a new paradigm for environmental modelling. These models learn general-purpose representations, or embeddings, that can be adapted to multiple downstream tasks with minimal supervision (Bommasani et al., 2021; Mañas et al., 2021). While early studies demonstrate their potential for flood mapping and other environmental applications, their integration with physically based hydrodynamic models remains largely unexplored (Xiao et al., 2024; Tulbure et al., 2025). This presents a critical opportunity, to bridge the gap between data-driven and physics-based approaches by learning representations that encode both the observed appearance of flooding and the underlying processes that govern it, particularly within the diverse hydrological regimes of the UK. Research questions This project aims to develop a national-scale flood mapping system for the UK that integrates satellite-derived embeddings with hydrodynamic simulations to enable real-time flood extent prediction anywhere in the country. The project has three core research questions: How can a shared latent representation be learned that aligns Earth Observation data with hydrodynamic processes for flood prediction? To what extent can physics-informed surrogate models replace computational hydrodynamic simulations for real-time, national-scale flood mapping? How can multi-scale geospatial data (from Sentinel to sub-meter imagery) be integrated to improve generalisation and urban flood prediction? Anticipated outcomes & benefits for the sponsoring organisation and other stakeholders The project is expected to make several key contributions. Scientifically, it will introduce a novel framework for learning physics informed geospatial embeddings, bridging the gap between Earth Observation and hydrodynamic modelling. It will demonstrate, for the first time, the feasibility of a unified flood prediction model that generalises across the diverse hydrological and geographical conditions of the UK, including upland catchments, lowland floodplains, and urban environments. The integration of multi scale satellite data, including very high resolution commercial imagery, will further advance understanding of how fine scale urban features influence flood dynamics at larger scales. Technically, the project will advance methods in multi modal representation learning, surrogate modelling of physical systems, and scalable geospatial data processing. By translating complex environmental data into real-time, accessible flood insights, this work empowers social scientists and decision-makers to anticipate and respond to flooding earlier, reducing impacts on communities and infrastructure. Start Date 18th January 2027 Application Deadline 19th October 2026 How to apply Use Apply to Newcastle Portal Once registered select ‘Create a Postgraduate Application’. Use ‘Course Search’ to identify your programme of study: · Search for the ‘Course Title’ using programme code: 8040F · Research Area: Civil Engineering · Select ‘PhD in Civil and Geospatial Engineering discipline’ as the programme of study. You will then need to provide the following information in the ‘Further Details’ section: · ‘Personal Statement’ - upload a document or write a statement directly into the application form. · ‘Research Proposal’ - when prompted for how you are providing your research proposal – select either ‘Write Proposal’ or ‘Upload document’. · Studentship code ENG163 in the ‘Studentship/Partnership Reference’ field. In the ‘Supporting Documentation’ section please upload: · your CV You must submit one application per studentship; you cannot apply for multiple studentships on one application. Contact Details Dr Maria Valasia-Peppa maria-valasia.peppa@newcastle.ac.uk

Research areas