British Hydrological Society - water@leeds Annual Webinar

The British Hydrological Society (Pennines Branch) and water@leeds present the third of our annual webinars showcasing current work from the next generation of water researchers.

Wednesday 13 December, 12.00

Register here to receive the zoom link.

This year's presentation is from two postgraduate researchers based in the School of Civil Engineering.

Thitipoom Chailert

Title: Enhancing flash flood forecasting accuracy via machine learning
Abstract: This talk by Thitipoom Chailert will focus on flash floods. They usually occur in steep, small catchment areas, making them difficult to predict due to their rapid onset. Utilizing local hydrological data with machine learning models offers a potential solution for forecasting. These models learn patterns from large local hydrological datasets to generate forecasts. Results can be compared to traditional models like The Probability-Distributed Model (PDM).

Maria Luisa Taccari

Title: Deep Learning for Groundwater Modelling 
Abstract: This talk by Maria Luisa Taccari, a final-year PhD candidate delves into the development of deep learning surrogate models in groundwater flow simulation. Surrogate models serve as efficient approximations of complex numerical groundwater models, like MODFLOW, which are traditionally crucial for water resource management but come with high computational demands. Through her research, Maria Luisa has illustrated that deep learning surrogate models, encompassing standard computer vision models, transformers, and neural operators, can substantially reduce computational demands. These data-driven methods have shown the potential to mimic the capabilities of numerical techniques accurately and efficiently. Finally, the talk also touches upon the application of these models with real-world data, which present unique challenges such as accounting for time-dependent problems, sparse data, and true-world variability.