Probabilistic Flood Forecasting in England and Wales – can it give us what we crave?

Abstract

Probabilistic flood forecasting has a number of benefits. It represents the inherent uncertainties associated with flood forecasts and can improve the utility of forecasts for flood warning in situations of greater uncertainty, such as in convective storms or for longer lead-times. Overall, it allows certain actions to be taken earlier in a more informed way and provides a more complete picture of potential flood risk as an event develops. In response to drivers such as the Pitt Review, various probabilistic flood forecasting techniques have been developed and applied with some success in recent years. This paper outlines some techniques that have been developed and trialled in the UK and discusses those showing most promise for operational flood forecasting. An over-arching framework for assessing uncertainties in fluvial forecasting in a risk-based manner is also presented. This framework aims to help practitioners in operational flood forecasting to select and implement appropriate techniques which can add value to the forecasting process and are robust enough to use operationally. The paper ends with a brief review of how probabilistic flood forecasting techniques can be of most value to practitioners. Both the opportunities and the barriers to be overcome are discussed in relation to the wider use of probabilistic techniques in operational flood forecasting. This leads to the conclusion that, if used appropriately, probabilistic techniques can add valuable information to operational flood forecasting. However, application of such techniques should not be seen as a ‘universal panacea’ addressing all current challenges in flood forecasting. Introduction The Environment Agency provides a Flood Warning and Forecasting Service to people at risk from flooding in England and Wales. For fluvial flood forecasting, rainfallrunoff, channel flow routing and hydraulic river models have traditionally been used. These are often combined into linked model cascades (‘integrated catchment models’) and utilise various forms of data assimilation to help improve forecasts. Outputs from forecasting models are currently deterministic with one model run delivering the flood forecast which is assumed to be the best representation. Forecasting Duty Officers then have to use experience and judgement to provide guidance on the likely error range when interpreting forecasts in the context of decision-support for flood warning and incident management. The accuracy of flood forecasts is influenced by the accuracy of the input data together with uncertainties in the structure, states and parameters of the model. Having a sound understanding of these modelling uncertainties is vital to assess and improve the flood forecasting service the Environment Agency provides. In recent years, the probabilistic treatment of modelling uncertainties has advanced from research to a range of near-operational techniques and provides an opportunity to quantify and ultimately reduce uncertainty in flood forecasting and modelling. For example, in the UK progress has been made in quantifying the uncertainties associated with quantitative precipitation forecasting (QPF) and how to propagate these through flood forecasting models (Environment Agency 2010a; Golding, 2009; Bowler et al., 2006). In addition, QPF ensembles, such as MOGREPS (Bowler et al, 2008), have been successfully linked to storm surge forecasting models to provide longer lead times for coastal forecasting whilst also allowing an assessment of confidence in these forecasts by capturing the uncertainty around them (Hawkes et al, 2009; Flowerdew et al, 2008) Historically, practitioners in flood risk management have applied a number of techniques to account for major sources of uncertainty. Examples are the use of factors of safety, sensitivity testing and assumptions analysis (what if 40 mm of rain falls in 6 hours), taking a precautionary approach, and of course improving the underlying data and models. Probabilistic modelling techniques are another way to capture uncertainties when modelling or forecasting for complex systems. This provides an understanding of the ‘robustness’ of the forecast results to the underlying assumptions made. Here, we define a robust probabilistic method as giving acceptable values of POD (probability of detection) and FAR (false alarm rate) across a wide range of forecasting situations, and better values than for deterministic forecasting. The major advantage of probabilistic techniques compared to more traditional deterministic approaches is that they not only allow investigation of the range of possible outcomes but also they assign a probability. If these probabilities can be formulated such that they reflect the expected natural occurrence, information is available on forecast reliability, and provided that the underlying assumptions are understood, they provide a good basis for risk-based management and decision-making. This allows the user to take account not only of the probability but also the potential consequences of flooding (which might affect the level of acceptable risk in deciding on suitable