The Met Office developed a prototype high‐resolution, hourly‐cycling NWP forecasting system using four‐dimensional variational data assimilation, covering the southern half of England. This system was known as the Met Office Nowcasting Demonstration Project ( NDP ), and ran in real time from June 2012 to March 2013 (covering the period of the London 2012 Olympics and Paralympic Games). It was principally developed to assess the abilities of a frequently updated NWP ‐based nowcasting system to forecast precipitation at short range such as convective storms for flood forecasting. This article provides a comprehensive assessment of the precipitation forecast skill of the NDP and its comparison against the Met Office operational nowcasting system. It was found that the NWP ‐based nowcasting system becomes more skilful than an advanced extrapolation‐based nowcasting system from T + 1.5 to T + 2 h depending on weather type.

As the societal impacts of hazardous weather and other environmental pressures grow, the need for integrated predictions that can represent the numerous feedbacks and linkages between sub-systems is greater than ever. This was well illustrated during winter 2013/2014 when a prolonged series of deep Atlantic depressions over a 3 month period resulted in damaging wind storms and exceptional rainfall accumulations. The impact on livelihoods and property from the resulting coastal surge and river and surface flooding was substantial. This study reviews the observational and modelling toolkit available to operational meteorologists during this period, which focusses on precipitation forecasting months, weeks, days and hours ahead of time. The routine availability of high-resolution (km scale) deterministic and ensemble rainfall predictions for short-range weather forecasting as well as weather-resolving seasonal prediction capability represent notable landmarks that have resulted from significant progress in research and development over the past decade. Latest results demonstrated that the suite of global and high-resolution UK numerical weather prediction models provided excellent guidance during this period, supported by high-resolution observations networks, such as weather radar, which proved resilient in difficult conditions. The specific challenges for demonstrating this performance for high-resolution precipitation forecasts are discussed. Despite their good operational performance, there remains a need to further develop the capability and skill of these tools to fully meet user needs and to increase the value that they deliver. These challenges are discussed, notably to accelerate the progress towards understanding the value that might be delivered through more integrated environmental prediction.

Wivenhoe and Somerset Dams are operated with the dual purpose of providing drinking water supply to South East Queensland and also to provide flood mitigation benefits for communities along the Brisbane River downstream of Wivenhoe Dam. Both dams have flood gates and the dams are operated during flood events in accordance with a Flood Operations Manual that provides strategies to reduce the impact on communities in Brisbane, Ipswich and smaller rural communities along the river. The flood operations rules for the dams consider flows in the catchments both upstream and downstream of the dams. The overall catchment area of the Brisbane River is 13,500 km2, of which 7,039 km2 is upstream of Wivenhoe Dam. For any individual flood in the Brisbane River catchment, the flooding outcome along the river downstream of Wivenhoe Dam depends upon the volume, peak flow and timing of the flood hydrographs generated from the individual catchments upstream of Somerset Dam, between Somerset and Wivenhoe Dam, and from the tributaries of the Brisbane River downstream of Wivenhoe Dam, including Lockyer Creek and the Bremer River.

Intense rain falling on rapidly responding catchments has been responsible for some notable flooding across England, with examples including Boscastle in 2004 and Ottery St Mary in 2008. Flash floods pose a significant risk to lives and livelihoods. In 2008 the need for significant improvements to the detection and forecasting of flash floods was highlighted in a comprehensive review into the major flooding that affected large swathes of England and Wales in 2007. Aspirations were further raised after the flash flooding affecting Devon and Cornwall in December 2010, and many locations during 2012.

There are significant uncertainties inherent in precipitation forecasts and these uncertainties can be communicated to users via large ensembles that are generated using stochastic models of forecast error. The Met Office and the Australian Bureau of Meteorology developed the Short Term Ensemble Prediction System (STEPS) was developed to address these user requirements and has been operational for a number of years. The initial formulation of Bowler et al. (2006) has been revised and extended to improve the performance over large domains, to include radar observation errors, and to facilitate the combination of forecasts from a number of sources. This paper reviews the formulation of STEPS, discusses those aspects of the formulation that have proved most problematic and presents some solutions. The performance of STEPS nowcasts is evaluated using a combination of case study examples and statistical verification from the UK. Routine forecast verification demonstrates that STEPS is capable of producing near optimal blends of a rainfall nowcast and high resolution NWP forecast. It also shows that the spread of STEPS nowcast ensembles are a good predictor of the error in the control member (unperturbed) nowcast.

High-resolution precipitation estimates from weather radar and radar-based precipitation forecasts are key inputs to hydrological applications and, in particular, to flood forecasting models. This paper examines the processes applied to the radar-measured reflectivity data from the UK weather radar network in order to derive products useful for hydrological applications. This starts with the quality control of the reflectivity scan data then looks at processes to convert the measured reflectivity into estimates of precipitation rate close to the ground. The approaches applied operationally at the UK Meteorological Office are compared with other operational approaches. In order to use radar data for hydrological applications, it is important to understand the likely error characteristics of the precipitation estimates. Two different approaches to representing this uncertainty are outlined. The first considers a quality index, formed by combining a number of different components, representing different sources of error, multiplicatively. The second approach considers the generation of ensemble precipitation estimates which represent the likely spread of errors. The use of the precipitation estimates in generating short-period probabilistic precipitation forecasts is discussed. The methodology adopted in the short-term ensemble prediction system is outlined. Characteristics of these radar products are illustrated with a precipitation event.

Several techniques for the generation of ensembles of radar observations are described and evaluated. These have been combined to generate ensemble estimates of surface precipitation rate for use in conjunction with the Short Term Ensemble Prediction System. STEPS is an operational, quantitative precipitation nowcasting algorithm developed jointly by the Met Office and the Australian Bureau of Meteorology. It generates ensemble nowcasts of precipitation rate and accumulation by scale-selectively blending a weather radar-based, extrapolated analysis of surface precipitation rate with a recent precipitation forecast from a high-resolution configuration of the Unified Model, and a time series of synthetically generated precipitation fields (noise) with space time statistical properties inferred from radar. Currently, STEPS incorporates an observation uncertainty algorithm based upon on analysis of Z-R errors. In this paper, the performance of STEPS precipitation nowcast ensembles, generated using radar ensembles, is compared with that of operational STEPS precipitation nowcasts, produced using unperturbed observations.

As radar becomes more influential to those responsible for water management and short range precipitation nowcasting, an estimate of the space-time uncertainty associated with the radar precipitation estimates is needed. The Met Office radar precipitation rate product is generated every 5 minutes at 1, 2 and 5 km resolution, as described by Harrison et al (2000 and 2009). The 2km product is used by The Met Office Short Term Ensemble Prediction System (STEPS) (Bowler et al., 2006) to generate precipitation nowcasts out to T+6 hours. The current uncertainty product that accompanies radar data gives an estimate of data quality (a function of beam height); however this does not give a quantitative estimate of the error. Recently, ensemble generation has been suggested as a method of quantifying uncertainty. These models fall into two categories: modelling individual sources of error, (Lee et al., (2007), Lee and Zawadzki (2005a, 2005b, 2006), Jordan et al (2003), Berenguer and Zawadzki (2008)) and statistical descriptions of the total error in the radar data (Germann et al (2009), Ciach et al (2007), Llort et al (2008)). The difficulty with the first approach is that the error structure is complex and inter-dependent. The difficulty with the second approach is the requirement to have a reference field with adequate resolution, which is usually based upon a dense network of rain gauges. STEPS already incorporates an observation uncertainty algorithm based upon an analysis of the error in the assumed Z-R relationship. This is used to generate an ensemble of perturbation fields. The perturbations are correlated in space and time to replicate the correlations in the random fluctuations about an assumed Z-R relationship. The advantage of statistical models based on the error measured by comparing radar estimates to a reference, is the whole error is measured. Not just the error in the assumed VPR and Z-R relationship. Germann et al (2009) (hereafter GM09) looked at the location dependence of the errors in radar data as compared to gauge measurements. This has advantages where location dependent variables affect the measurement (such as beam shielding or orographic enhancement). Here two models for ensemble generation, one falling into each category have been built and tested. Section 2 describes the two models, and the results of the verification of the models against rain gauge ground truth measurements are given in Section 3. In order to be used within STEPS, the ensemble generator needs to generate an ensemble time series of fields of instantaneous rain rate at time intervals less than 15 minutes suitable for the following purposes: a) for use as initial conditions for generating STEPS ensemble nowcasts of rain rate and accumulation; b) to generate ensembles of T+0 rainfall accumulations for use in flash flood forecasting and other downstream user applications; c) for wider use to generate a range of QPE uncertainty products tailored to customer requirements. STEPS currently issues extreme rainfall alerts (ERAs) to the Environment Agency (EA) when rainfall amount and accumulation time thresholds are exceeded (a part of (a) and (b)). The statistics used for the comparison will focus on this requirement.

An ensemble-based probabilistic precipitation forecasting scheme has been developed that blends an extrapolation nowcast with a downscaled NWP forecast, known as STEPS: Short-Term Ensemble Prediction System. The uncertainties in the motion and evolution of radar-inferred precipitation fields are quantified, and the uncertainty in the evolution of the precipitation pattern is shown to be the more important. The use of ensembles allows the scheme to be used for applications that require forecasts of the probability density function of areal and temporal averages of precipitation, such as fluvial flood forecasting-a capability that has not been provided by previous probabilistic precipitation nowcast schemes. The output from a NWP forecast model is downscaled so that the small scales not represented accurately by the model are injected into the forecast using stochastic noise. This allows the scheme to better represent the distribution of precipitation rate at spatial scales finer than those adequately resolved by operational NWP. The performance of the scheme has been assessed over the month of March 2003. Performance evaluation statistics show that the scheme possesses predictive skill at lead times in excess of six hours.

In collaboration with the Bureau of Meteorology (Melbourne, Australia), the Met Office (Joint Centre for Hydro-Meteorological Research, UK) has developed a stochastic precipitation nowcast scheme, designed to model and predict the PDF of surface rain rate and rain accumulation in space and time. Here we demonstrate the range of probabilistic products generated by the scheme, and their potential applications for fluvial flood forecasting and warning.With the aid of a hydrological model (the PDM), we consider the use of ensembles of predicted catchment rain accumulation in evaluating the range of possible river flow responses from a given catchment. When employed in conjunction with a catchment specific, cost-based decision-making model, we highlight the value of PDFs of forecast catchment rainfall accumulation and river flow as an aid to objective decision making within the flood warning process. (C) Crown Copyright 2005. Reproduced with the permission of Her Majesty's Stationery Office. Published by John Wiley & Sons, Ltd.

A new advection-based nowcasting scheme for precipitation has been developed for the Gandolf system. The method employed removes the need to split the radar analysis into contiguous rain areas (CRA's) and uses a smoothness constraint. Thus, it is similar to the Variational Echo Tracking approach. Optical flow ideas are used to diagnose the advection velocity of blocks within rain analyses, which involve the direct solution of the Lagrangian persistence equation. In block-based approaches, the radar analyses are partitioned a-priori and a CRA may be split over a number of blocks. This scheme is compared with the old Gandolf advection scheme, which is based on CRA's, and the new scheme performs better-both in cases associated with severe flooding and over a continuous verification period of 3 months. The benefit is conjectured to be due mainly to the difficulty in unambiguously identifying CRA's, which is particularly troublesome on small domains, such as the UK. Thus, it is concluded that block-based methods are likely to be superior to object-based methods in the majority of cases.