Hydrology in the 21st century: Challenges in science, to policy and practice

Over the last four decades, hydrological science has made notable strides, yet the prediction of hydrological extremes, adaptation to climate change, and effective water resource management continue to pose significant challenges. Artificial Intelligence (AI) and Machine Learning (ML) technologies have emerged as powerful tools with the potential to address these issues, though they also introduce new challenges.

In her talk, Professor Slater will delve into how AI/ML innovations are reshaping hydro-meteorological modelling, forecasting, and analysis. She will highlight examples that demonstrate ML's strengths, such as its capacity to integrate extensive datasets from varied sources and improve the prediction of extremes, including in ungauged basins. Professor Slater will explore various approaches to model interpretability, where ML can offer novel insights into the complex dynamics driving hydrological changes. Alongside these areas of success, she will also discuss various challenges related to ML in hydrology such as model biases, physical consistency, and uncertainty quantification.

The presentation will conclude with a forward-looking perspective on the potential of AI/ML to overcome these barriers and the opportunity to drive future advances. This includes the ability to improve early warning systems, flood projections, and water resources management in a rapidly changing world.

Professor Louise Slater, University of Oxford, UK

Professor Louise Slater, University of Oxford, UK

Louise Slater is Professor of Hydroclimatology and Tutorial Fellow at Hertford College, University of Oxford. Louise leads the Hydro-Climate Extremes Research Group, which develops computational approaches to detect, attribute and predict how changes in climate and land cover may affect water-related extremes and society. She is currently a Future Leaders Fellow of UK Research and Innovation. Her project DRIFT: the Dynamic Drivers of Flood Risk aims to develop a more comprehensive understanding of flood non-stationarity by drawing on a holistic set of flood drivers across various timescales.

While it may be easy to simulate the rise and fall of a flood hydrograph, it is surprisingly challenging to model the magnitude-frequency relationships of peak flow rates, flood volumes and levels under historic or current conditions, particularly over a range of spatial and temporal scales within a single catchment. Under climate change, these challenges only get harder. Dealing with a non-stationary hydroclimate imposes new standards of evidence for the defensibility of our risk-based hazard modelling. We need to develop new methods to characterise flood behaviour that are appropriately informed by both historical observations and projected outputs from climate models. For example, we have historically done a poor job of characterising the propagation of uncertainty through the rainfall-hydrology-hydraulics modelling chain, and we need to ensure that industry has access to modelling tools that take some pragmatic account of the different sources of aleatory and epistemic uncertainty that confound their parameterisation, application, and interpretation. To be useful, climate model projections need to be provided in a processed form at spatial and temporal scales relevant to the rainfall extremes of interest, and they need to be based on sufficient ensembles to characterise the variability and indicative uncertainty of the projections. Further, our treatment and understanding of risk needs updating to suit changing engineering design and planning requirements. We need to overhaul our existing approach to reporting flood risk in terms of annualised probabilities or return-periods. Decisions will increasingly need to move to an adaptive risk management framework, where attention is given to the changing likelihood of failure over given planning periods. We might argue about the efficacy of our current flood estimation practices, but global warming represents a Grand Challenge that is rapidly undermining our ability to both understand and estimate flood risks. 

Professor Rory Nathan, University of Melbourne, Australia

Professor Rory Nathan, University of Melbourne, Australia

Rory Nathan is a hydrologist who has focussed much of his 40-year career in industry evaluating flood risk for public and private sectors. He is co-editor and leading author of the Australian national flood guidelines, and he has undertaken many expert advisory and review roles for government, legal, and private entities. He has published over 130 papers in refereed journals, and his research at the University of Melbourne is focussed on the impacts of climate change on floods and environmental systems.  In 2000 he was recognised as national 'Civil Engineer of the Year' by Engineers Australia, and as the 2018 Munro Orator he is recognised as being one of Australia’s most influential engineers eminent in the field of water resources.