Pluvial flooding is on the rise as more cities are challenged by a changing climate and local drivers: increased urbanisation and inadequate sewer systems. Sustainable flood risk management requires a hybrid of structural and non-structural measures to ensure water-hazard resilient cities. In this regard, flood forecasting and early warning systems have been proposed as a “low regret” measure to reduce flooding and increase preparedness through forecast-based actions. However, many cities, especially in data-scarce regions, lack the capacity to produce high-quality rainfall forecasts and well-calibrated flood predictions (including timing, water levels, extent and impact). This limitation creates a cascading effect, hindering the ability to make reliable decisions due to uncertainties in the forecast or inaccuracies in the input data. For example, decisions in anticipatory flood management become problematic due to their dependencies on knowledge derived from uncertain data and the consequences of incorrect predictions and/or actions.

Previous studies have focused on improving the accuracy of predictions or models, but there has been comparatively less focus on addressing the complexities of flood forecasts and the decision support chain, particularly under conditions of limited data and uncertainty. Probabilistic forecast knowledge is beneficial for quantifying uncertainty and has been acknowledged to support decision-making, but there is no consensus on the most suitable and effective way to incorporate it for decision-making.

This thesis aims to explore how pluvial flood forecasting and decision support at local scales in data-scarce cities can be improved, with a focus on using available data, method suitability and incorporating uncertainty into the decision-making process for anticipatory flood action. This research was motivated by practitioners’ need to take action ahead of time, despite limited and uncertain data, which is so often the case in cities in the Global South. The research was carried out in the coastal city of Alexandria, Egypt, which experiences annual flooding due to winter storms.

The research adopted an incremental and sequential approach, in which each research outcome is built upon in the next step and used as input for the next objective, thereby integrating the data, models and knowledge. First, the research investigated how available data can be combined for flood forecasting using a rainfall threshold method, crowd-sourced information and global ensemble forecasts. Next, the study evaluated the suitability and limitations of using a downscaled high-resolution Weather Research and Forecasting model for deterministic rainfall forecasts at multiple spatial scales for different urban flood forecasting approaches. A high-resolution limited-area ensemble rainfall forecasting model was then configured and simulated. The resulting probabilistic distributions were incorporated into a Bayesian decision framework to translate probabilistic rainfall forecasts into actionable information.

Findings show that, despite limited data, using rainfall forecasts alongside non-traditional data sources can effectively predict flooding and improve decision-making. Although this method introduces some complexity compared to widely used deterministic rainfall forecasts, it offers a balance between a simplified approach and forecast uncertainty when limited data exist.

Regarding the use of higher-resolution forecasts, the results showed that, despite the need for very high-resolution forecasts for urban flood forecasting, there are limits when using downscaled models, as outcomes vary depending on the model configuration, parameterisation and the flood forecast approach. These findings highlight the need to consider the interdependencies between meteorological forecasts and their application to flood forecasting.

Finally, the thesis demonstrated the value of using a Bayesian decision framework for incorporating knowledge of flood vulnerability and local flooding at the neighbourhood scale for triggering equitable flood early actions. The proposed probabilistic framework showed the potential benefits of a user-preference parametric loss function that assigns weights to different vulnerability classes. The Bayesian decision framework served as a valuable decision support tool which incorporates uncertainty, decision makers’ risk-averse attitudes and prioritises equitable anticipatory flood actions to reduce flood damage.

The primary aim of this research was not to make forecasts more accurate, but rather to develop a greater understanding of the usefulness of imperfect data, forecasts and models in decision-making for pluvial flood forecasting and local anticipatory flood management. The research raises themes of fit-for-purpose data and creating value from imperfect information by examining the use and limitations of data and the interdependencies of the flood forecasting and decision-making chain. This approach supports uncertainty-aware flood forecasting and decision-making instead of creating an illusion of certainty.

High-resolution rainfall forecasts are essential for forecasting urban pluvial floods and providing local decision support before an event when using rainfall or model-based approaches. Regional Numerical Weather Prediction (NWP) models increase the resolution and extend forecasts by several days when initialised with global models. These models produce forecasts at higher spatial and temporal resolutions but are computationally demanding and do not necessarily result in more accurate forecasts especially in data-scarce cities. This research evaluated how rainfall forecasting model scale dependencies satisfy high-resolution hazard forecasting requirements with two different flood forecasting approaches. Rainfall forecasts of different spatial resolutions, cumulus schemes and lead times from a high-resolution Weather Research and Forecasting (WRF) model were first evaluated and then used as input in a rainfall threshold and 1D MIKE urban drainage model for flood forecasting in the data-scarce city of Alexandria City, Egypt. Results indicate the flood forecast severity class and flood model simulation results vary with the neighbourhood size, forecast horizon, and chosen cumulus configuration but in general the smallest resolutions evaluated did not improve the hazard estimation for both flood forecasting approaches. Therefore, trade-offs must be made regarding model configurations, resolution, lead times and how the forecast output will be used. This study demonstrates the opportunities and limitations for better integrating high-resolution meteorological for the development of a rainfall threshold-based and model-based flood forecasting in cities with similar conditions. It also highlights the need to align the selected model configuration with the goals of the flood forecasting application which is critical for effective early warning systems and anticipatory flood management.

Rapidly expanding cities in the Middle Eastern and North African (MENA) region are at risk of flooding due to heavy rainfall, insufficient drainage capacity, a lack of preparedness and insufficient data to conduct required studies. A low regret Early Warning Systems (EWS) using rainfall thresholds is proposed as a cost-effective short-term solution. This study aims to utilise a probabilistic approach to characterise and predict urban floods by assessing critical rainfall thresholds likely to cause flooding combined with ensemble precipitation forecast in Alexandria, Egypt. Rainfall thresholds were inferred by associating observed rainfall and historical flood information sourced from social media and newspapers. Floods were classified in a colour-coded hazard matrix as no flood (green), minor flood (yellow), significant flood (orange), and severe flood (red). Probability of occurrence of hazard classes was derived by incorporating ensemble rainfall into the hazard matrix to jointly evaluate likelihood and hazard severity. Results from this study showed that three of four severe events analysed could have been predicted with a high likelihood up to 24 hr before. The presented approach supports decision making to issue warnings and flood control actions with limited data and is a model for other data scare regions.