Flood forecasting – the potential for the scientist and decision maker

I was struck by the presentation of Carlos Velasco-Forero at the recent EGU General Assembly when he referred to a ‘critical need to develop defensible flood forecasts’ in Australia. He mentioned the uncertainties involved when dealing with decision making when rainfall and flood forecasts rapidly unfold and the challenges associated with communicating these uncertainties with responders. So what were some of the take home points from presentations across EGU19 and how can they support the scientist and decision maker through improved flood forecasting?


Presenting at the EGU 2019 PICO session was an excellent way to bridge the gap between science and practice in operational forecasting for different water-related natural hazards.

Integrating the disciplines of meteorological and hydrological science to improve the end-to-end framework for forecasting intense rainfall and flash flooding was presented by Linda Speight. Linda summarised the work that has been delivered as part of the recently completed NERC and Met Office funded Flooding from Intense Rainfall (FFIR) programme which included improved knowledge of catchment vulnerability to flash flooding, improved rainfall (radar) and river observations during events, improved forecasting of convection, and real time flood inundation modelling.

Taking over from FFIR as a research initiative attempting to improve flash flood forecasting is the French ANR PICS project which will run through to 2021. As presented by Olivier Payrastre, the project aims to couple the disciplines and actors involved in flash flood nowcasting, from the meteorologist through to the decision maker. The aim is to integrate short-range forecasting chains incorporating high resolution precipitation forecasts, distributed rainfall-runoff modelling across ungauged catchments, DTM-based modelling of impacts including dynamic population exposure and vulnerability.


Presentation by Annegian Tijssen on ‘from forecast to action: a focus on end-user information needs during a disaster.’

Whilst these highlighted papers were concerned with the forecasting timescales in the short range, Louise Arnal described the potential role for seasonal forecasting in flood early warning and an ‘autopsy’ technique to understand the relative contributions of various hydro-meteorological variables.  Whilst unlikely to offer immediate operational benefit for most rivers, it did present some suitability for understanding the contribution to flooding on large-scale river systems such as the Danube. On a similar scale, Gabriela Guimaraes Nobre – who was awarded an outstanding student poster and PICO Award – described that whilst there had been great advancements in flood forecasting, it remains a challenge to provide useful impact-based forecast information and presented an approach for linking large-scale indices of climate variability and flood losses.

But back to the theme of flash flooding – Steven Boeing presented work on dealing with the uncertainties in urban scale surface water flood predictions.  As part of the iCASP initiative (Integrated Catchment Solutions Programme), work is being carried out to understand how probabilistic rainfall forecasts can be linked to high resolution hydrodynamic modelling to produce hyper-local forecasts over short lead times. One key element of the work is to understand how responders are able to respond to such complex information.

As for the Australian case study, forecasting science developments are now supporting the availability of STEPS (Short-term Ensemble Prediction System) every few minutes compared to the current operational set-up of updating the hydrological forecast a few times per day which is now providing a framework for a probabilistic approach to flood warnings. However, as Carlos concluded (and matching a theme across other hydrological forecasting presentations at EGU19), whilst increasing the resolution of flood forecast information can be seen as a positive development, it in turn creates its own challenge of where science development meets the operational reality of delivering flood forecasting services.

Flood Forecasting and Warning – a retrospective

Chris Haggett has worked in the UK water industry for over 40 years, including with the Greater London Council (Department of Public Health Engineering), Thames Water (Rivers Division), National Rivers Authority, Environment Agency and RAB Consultants. He has experience in hydrology, flood forecasting and warning, emergency management, and benefits and business change management. Chris has worked as Flood Forecasting and Warning Development Manager within the Environment Agency’s National Flood Warning Centre, and has experience in managing many major flood incidents, including Easter 1998, Autumn 2000 and Summer 2007 floods, at both a regional and national level. Chris has also worked as a Flood Warning Manager and was responsible for delivering the day-to-day flood forecasting and warning service for the Thames catchment. Chris looks back at four decades of development in flood forecasting and warning.

I started work as a graduate hydrometric assistant working for the Greater London Council (GLC), Rivers Division in 1975. My first job was assisting with the aftermath of the Hampstead Storm when 170.8mm of rain fell in just 155 minutes on 14 August 1975 (Keers and Wescott, 1976). One man died in the storm, four of London’s main-line railway stations were flooded and closed, a large part of the London Underground was brought to a standstill as tunnels were inundated and the electricity supply failed (figure 1). Within 5 minutes of the start of the storm over a hundred homes were flooded to more than a metre. And yet just 6km or so away from the storm centre, there was almost no rain at all.

This storm highlighted a number of fundamental issues including summer storms in urban areas can be destructive and life threatening; they are difficult to detect, even with a dense network of recording rain-gauges; they are challenging to forecast and almost impossible to warn against due to very short lead times.


In 1977, another severe storm affected north-west London when a maximum of 113mm of rain fell in a 9-hour period at Ruislip, Middlesex (below). No deaths were reported but some 1,100 houses, 20 factories, 20 shops, several schools and churches, many roads and several railways were flooded to depths of up to 1.5m (Haggett, 1980). Although a rudimentary flood warning scheme was in place at the time, a Government inquiry highlighted that the system was not effective in giving any warning to the police or the public (Prickett, 1978).

The Prickett Report made several important recommendations to improve flood forecasting and warning for fast responding urban floods:

  • Flood detection and forecasting: information from detection gauges needs to be captured by a central processing unit with facilities for automatic interrogation and the issue of alerts to duty staff; the prospects for measuring and forecasting the location, duration and intensity of summer storms sufficiently for warning purposes could be improved with the aid of weather radar.
  • Flood warning dissemination: the most feasible method for alerting householders is for messages to be conveyed to nominated persons or flood wardens who would undertake to inform their neighbours; to be effective in the short time available, some river alarms might need to go direct to occupants or via flood wardens. Local audible warnings may be the only possibility in some situations.
  • Emergency management and public awareness: the operational arrangements should be clearly documented including objectives, equipment, communications, staffing, duties and duty rosters; occupants of property in flood susceptible areas should be regularly reminded of the warning arrangements and action which can be taken to minimise damage including flood proofing methods and other forms of self-help.

Flooding from the River Brent, Ealing, London, August 1977

It is interesting to see how these recommendations have been acted upon and what improvements have been made to flood forecasting and warning in urban areas in the 40 years since these devastating storms in London in the 1970s.

Advances in Flood Detection, Forecasting and Warning in the 1980s and 1990s

Flood Detection. Following the storms and floods of the 1970s and early 1980s, there was a great emphasis placed on developing an integrated flood detection, forecasting and warning system for fluvial rivers draining Greater London. The resultant CASCADE system (a Catchment Assessment System Concerned with the Accurate Dissemination of Effective flood warnings) included a large element of automation, was simple to operate, reliable and accessible to users 24 hours a day. The design and operation of this system was carefully planned as the nature of catchments involved meant that there is often very little time for the issue of a warning before the onset of flooding.


2km radar image of the “Lewisham Storm” of 10 June 1997

Initially, CASCADE incorporated data from a network of telemetered rain and river level gauges and from the London Weather Radar at Chenies. To facilitate the detection of localised urban storms 2km radar data displays were created which updated every 5 minutes (above). At the time, the Met Office did not routinely disseminate 2km radar data to users.

In the 1980s and 1990s disparate telemetry systems were in operation and progress was made to link, integrate and extend these systems, allowing operational access to any individual outstation via regional computer systems. At the same time, comprehensive alarm handling software was developed to alert duty staff of a developing situation based either on pre-defined rainfall intensities or river levels. Telemetry data displays allowed duty staff to rapidly assimilate near real-time information and this facilitated operational decision making (below).

Figure 4

Hyetogram/Hydrograph of the “Edgware Flood” of 22 September 1992

Although weather radar provided an unparalleled source of information on rainfall variations over urban areas, the accuracy of the rainfall estimates was on occasions poor and it was clear that improved calibration was needed. In the late 1980s an operational calibration technique was developed which made use of data from the London Weather Radar and a network of telemetered rain gauges to provide more accurate estimates of spatial rainfall. It was demonstrated that this technique provided a 22% improvement in accuracy relative to that obtained by radar data without calibration (Moore, et al., 1989a).

Flood Forecasting. Despite the importance placed on developing effective flood detection systems, it was soon recognised that to improve warning lead times further, more emphasis would be needed on introducing operational forecasting methods (Haggett et al., 1993 & 1995). This involves the prediction of rainfall, river levels and river flows to forecast the time of occurrence and duration of a flood. Methods of achieving this varied in technical complexity from rudimentary correlation techniques to sophisticated mathematical models capable of predicting rainfall, in-bank flows and flood plain inundation. For fast responding urban catchments development work focussed on applying rainfall forecasting and rainfall/runoff forecasting techniques.

In the 1990s, three approaches to quantitative rainfall forecasting were evaluated operationally:

  • FRONTIERS/NIMROD – making use of data generated by the Met Office giving rainfall forecasts out to 6 hours ahead on a 5km grid, updated every 15 minutes (Golding, 1994);
  • Local Radar Rainfall Forecasting – regional based systems providing forecasts up to 2 hours ahead on a 2km grid, updated every 15 minutes;
  • Convective Storm Forecasting (GANDOLF) – a system for generating automated nowcasts of convective precipitation (Collier, et al., 1994a).

Although FRONTIERS rainfall forecasts were very useful for flood forecasting, the technique did not fully meet the specific operational requirements needed for fast responding catchments. This prompted the development of a local rainfall forecasting system producing higher resolution forecasts. It was found that the local forecasting procedure achieved a superior performance to FRONTIERS up to 2 hours ahead (Moore et al., 1992). However, beyond lead times of two hours the utility of the local system, which is based on a single radar, is reduced as the radar field is advected beyond the area of interest (below).

Figure 5

Local Radar Rainfall Forecast Sequence for 14.00 to 16.00 GMT on 30 September 1993

As May (1994) demonstrated, both FRONTIERS and the Local Forecasting Technique had difficulties in forecasting in convective situations under certain circumstances. Where there is rapid storm development and erratic motion, for instance, both procedures failed to produce credible predictions. As the Hampstead Storm demonstrates, serious flooding incidents can be generated from convective storms which develop rapidly and become stationary over a location for a period. This prompted development work by the Met Office on the GANDOLF system based on an object-oriented convective precipitation model designed to exploit the predictive capabilities of a conceptual model of convective cell evolution (Pierce, et al., 2000).

In the 1990s much work was devoted to developing, calibrating and implementing rainfall-runoff models which made use of data from rain gauge networks, weather radar and short-period rainfall forecasting models. Studies were commissioned to assess the different rainfall/runoff model types, real-time updating methods and the effects of data time interval and spatial resolution. A study in the Thames catchment (Moore, 1993), showed that the Local Radar Rainfall forecasting scheme applied to locally calibrated radar data provided the best performance overall especially in small catchments. It illustrated that the simpler rainfall/runoff models had difficulty producing credible forecasts in catchments with complex responses.

Following studies of this type, decisions were made to use the more complex conceptual models for forecasting and to integrate them with operational detection and forecasting systems such as CASCADE. This led to the development of automated flood forecasting systems which ran every 15 minutes with new data and auto selection of the appropriate rainfall forecast or combination of rainfall forecasts for given circumstances.

Flood Warning. Despite the progressive improvement of detection and forecasting capability in the 1980s and 90s, flood warning failures continued. As Parker (1987) observed, this was often due to weaknesses in the dissemination phase of the process upon which the efficacy of the entire forecasting and warning system rests.

In England and Wales, the traditional method of disseminating flood warnings via the police and local authorities was becoming strained due mainly to resource constraints and the lack of clarity on mandatory responsibilities for warning dissemination. Consequently, a number of initiatives were instigated to improve the effectiveness of this important part of the flood warning process: automation of warning dissemination to the agencies; identification of areas of flood risk and better targeting of warnings; the use of flood wardens; the use of the media.

Figure 55

Flood Warning Dissemination System for the Colne Catchment (1994)

An example of warning automation can be afforded in the Thames Region where the fluvial Thames and each of its tributaries were divided into operational flood warning reaches. By nominating a specific reach or group of reaches, the duty officer was able to trigger the automatic creation and issue of warning notices to relevant agencies via telex and fax. This computer-controlled facility radically reduced the amount of time needed to generate warning messages to the emergency services, local authorities and utility agencies (above), (Haggett, 1995).

To improve the targeting of flood warning messages within each of the flood warning reaches, records stretching back to 1965 were collated and attributed to each reach. Information including property details, frequency of flooding and the level of flood risk was supplied to each police force and local authority in the Thames Catchment. It was hoped that armed with this information, the agencies would be able to target their resources more effectively during flooding incidents.

Several flood warden schemes were established in fast responding river catchments in and around Greater London where traditional dissemination methods failed to respond in sufficient time to be effective. In the early 1980s the GLC introduced the “watchdog” scheme at 22 high-risk locations where river level warning devices were installed which automatically alerted local authorities and residents when pre-defined levels were exceeded. Recipients of an alert monitored river levels locally and warned their neighbours of impending flooding. The scheme had mixed success with technological problems causing large numbers of false alarms to be generated in the early stages. In areas where the number of subsequent flooding incidents was low, the level of public interest soon waned and in some localities the scheme was withdrawn due to lack of support. Nevertheless, the scheme did signal a turning point, for the first time an attempt was being made to address the issue of flood warning dissemination for fast responding rivers in London by installing systems that would alert the public directly (Haggett, 1997).

There was also a conscious effort to involve the media in flood warning dissemination. Notices were passed to local radio stations for immediate broadcast during times of flood which identified the areas at risk. Advice and situation updates were given throughout an event via live and recorded interviews on radio and television and through the press. It was recognised at the time that the media would feature more extensively in flood warning dissemination in future years.

Research undertaken by Middlesex University in the 1980s and 90s showed that the passing of warning messages to the police and local authorities was very effective. However, these warnings often failed to reach the at-risk public in time for action to be taken to reduce flood losses. The research recommended that improvements should be made to the effectiveness of warning dissemination and that a multi-route approach should be developed involving a range of techniques that should work in parallel to ensure that the warning message reaches its intended target. (Hyde, 1992), (Fordham and Haggett, 1994).

Advances in Flood Detection, Forecasting and Warning from 2000

Flood Detection. Major UK floods over the last two decades have motivated significant technological and scientific advances in operational flood forecasting and warning.

The telemetered networks of river and rain gauges have been extended and integrated further, within and across government agencies. Access to data from such networks in near real time has been made publicly available for the first time.

The Environment Agency, for example, has extended its river level monitoring service to cover 36,000 kilometres of rivers in England. In addition, the Agency has approximately 1,000 real time rain gauges which are connected by telemetry and the data from some gauges is combined with data from the Met Office’s network to operationally calibrate rainfall radar data to improve accuracy.

Data from river level and rain gauge networks are published openly on the internet as it has the potential for a wide range of uses externally such as flood forecasting, farming, and recreation; flood forecasts are also being trialled online.

The weather radar network now consists of 15 operational radars across the UK providing real-time information to help monitor and forecast heavy rainfall (figure 8). In 2016, the Met Office and the Environment Agency launched the network renewal project which will provide major investment and improvements in new weather radar systems. The latest technology will be used to increase radar capability, such as dual polarisation which will improve quality control and rainfall rate accuracy. This investment will extend the operational life of the network by a further 15-20 years.

New technologies are being adopted more widely that allow incident staff to see what’s happening on the ground and to respond faster and better when floods threaten. This includes the use of mobile technology and mapping systems to facilitate the capture and sharing of real-time data.

Flood Forecasting. One of the major changes in flood forecasting during this period was the establishment in 2009 of the national Flood Forecasting Centre (FFC), a partnership between the Environment Agency and the Met Office. The FFC, for the first time, brought together the experts who predict the weather with those who forecast floods. This has led to the production of a daily, national Flood Guidance Statement (FGS) to the emergency response community.

The FGS is based on a Flood Risk Matrix approach that is a function of potential impact severity and likelihood. It has driven an increased demand for robust, accurate and timely forecast and alert information on fluvial and surface water flooding along with impact assessments. The Grid-to-Grid (G2G) distributed hydrological model has been employed across England, Scotland and Wales at a 1km resolution to support the FGS.

At a more local level, there was a move to bring together regional systems into a single national system to improve forecasting consistency across the country. The National Flood Forecasting System (NFFS) provided a single software shell capable of using a standard set of models that improved forecast quality and allowed extension of the service.

Flood Warning. Floodline Warnings Direct (FWD) was deployed in 2006 to replace and enhance legacy warning systems. There was a requirement to deliver a single, consolidated and robust national solution for warning and informing the public of flooding. Using the latest technology, warnings were targeted more effectively increasing the number of people receiving appropriate and timely warnings (Haggett, 2015).

FWD is the first integrated multi-media warning system deployed in the UK and it can now warn 1.2m people of risk of flooding at the touch of a button. The number of communication channels has steadily increased as social media grows and in December 2013 Twitter was used for the first time to send a severe flood warning.


Over the last 40 years there have been huge strides taken to develop the various elements of the flood forecasting and warning process. Improved flood detection using telemetry, weather radar and computer technology; enhanced forecasting using operational flow prediction models and quantitative rainfall forecasting techniques; more effective warning dissemination using computer technology to target warnings directly to the public; and timely, efficient and coordinated response by the relevant agencies. However, it is important to remember that flood forecasting and warning systems work on the principle of reducing flood losses through effective action by those at risk prior to a flood. To do this effectively all elements of the process must operate smoothly and in an integrated way to have any significant impact on flood losses. To deliver accurate, reliable and timely flood warnings the total warning system must operate efficiently.

Technology has played and is continuing to play an important role in integrating the total warning system, but there is a need to ensure that the service does not become technology driven. Other factors need to be considered which influence the performance and integration of flood warning systems, such as coordinated institutional arrangements, the flow of information and knowledge between organisations and communities, and human interaction and expertise.


Collier, C.G, Walsh, P.D. and Haggett, C.M. (1994a). The use of radar-based rainfall forecasts in operational flood warning. In: Merriman, P.A. and Browitt, C.W.A. (eds) Natural Disasters: Protecting Vulnerable Communities, Thomas Telford, London, 210-224.

Fordham, M. H. and Haggett, C. M. (1994). Flood forecasting, warning and response systems: the problem of warning dissemination. Workshop on Integrated Radar Estimates of Rainfall in Real-time Flood Forecasting. 25-26 July 1994, Monselice, Italy.

Golding, B.W. (1994). NIMROD and its relevance to the NRA. Internal Report. Restricted circulation.

Haggett, C.M. (1980). Severe storm in the London area – 16/17 August 1977. Weather. 35: 2-11

Haggett, C.M., May, B.C. and Crees, M. A. (1993). Advances in Operational Flood Forecasting in London. Proceedings of the Second International Symposium on Hydrological Applications of Weather Radar. Hannover, 7-10 September 1992.

Haggett, C.M., (1995). Operational fluvial flood forecasting and warning in the Thames Catchment. BHS Occasional Paper No.5. British Hydrological Society.

Haggett, C.M. (1997). Aims and objectives of flood warnings from the perspective of the flood forecaster, in Handmer, J.W. Flood Warnings: Issues and Practice in Total System Design. London: Middlesex University, Flood Hazard Research Centre.

Haggett C.M. (1998). An integrated approach to flood forecasting and warning in England and Wales. Journal of the Chartered Institution of Water and Environmental Management. 12: 425–432.

Haggett, C.M. (2015). Advances in Flood Warning Dissemination and the Challenges for Flood Forecasting. Real Time Flood Forecasting – Developments and Opportunities. Institution of Civil Engineers.

Hyde, V. M. (1992). The flood warning systems and flood-events of two catchments within the Eastern Area of the NRA. Thames Region. London: Middlesex University, Flood Hazard Research Centre,

Keers, J.F. and Wescott, P. (1976). The Hampstead Storm – 14 August 1975. Weather. 31: 2-16.

May, B.N. (1994). Integration of radar-based forecasting methods into CASCADE flood forecasting and warning system. COST-75 Seminar 1994, Brussels.

Moore, R.J., Watson, B.C., Jones, D.A., Black, K.B., Haggett, C.M., Crees, M.A., Richards, C.I. (1989a). Towards an improved system for weather radar calibration and rainfall forecasting using rain gauge data from a regional telemetry network. New directions for surface water modelling, IAHS Publication No. 181. International Association of Hydrological Sciences, pp13-21.

Moore, R.J., Austin, R.M., Jones, D.A. and Black, K.B. (1992). London Weather Radar Local Rainfall Forecasting Study: Final Report. Contract Report to the National Rivers Authority, Thames Region, Institute of Hydrology, September 1991, 124.

Moore, R.J., Austin, R.M., Carrington, D.S. (1993). Evaluation of FRONTIERS and Local Radar Rainfall Forecasts for use in Flood Forecasting Models: Final Report. Contract Report to the National Rivers Authority, R&D Note 225.

Parker, D. J. (1987). ‘Flood warning dissemination: the British experience’, in Handmer, J. (ed) Flood Hazard Management: British and International Perspective. Geobooks, p 169-190

Parker, D.J. and Haggett, C., 2001. The development of flood warning technologies in England and Wales, Paper presented at the International Emergency and Disaster Management Conference (TIEMS), Oslo, Norway, June.

Pierce, C.E., Hardaker, P.J., Collier, C.J. and Haggett, C.M. (2000). GANDOLF: a system for generating automated nowcasts for convective precipitation. Royal Met. Soc., Meteorological Applications, 7: 341-360.

Prickett, C.N. (1978). Departmental investigation into flood warning arrangements in north-west London. Ministry of Agriculture, Fisheries and Food.





Developing multi-hazard early warnings across Europe

ANYWHERE (enhANcing emergencY management and response to extreme WeatHER and climate Events) is a H2020 innovation programme developing tools to support decision makers in real-time coordination of emergency management operations. Started in 2016, the programme seeks to capitalise on advances in observation systems and in forecasting models in anticipation of natural hazards such as flash floods, landslides, droughts and heat waves.

Flooding in Catalunya

Mataro near Barcelona, affected by flooding in 2016. Source: La Vanguardia

The programme includes 31 organisations from a mixture of experts from the hydrometeorological forecasting community, research scientists, early warning system developers and responders. Those from the Civil Protection Authorities are providing a valuable role in shaping the end-user product requirements of the Multi-Hazard Early Warning Systems (MH-EWS).

The recent 6th programme meeting was hosted by the Fire and Rescue Department of North Corsica: Sevice d’Incendie et de Secours de Haute-Corse (SIS2B). SIS2B have a specific role in providing emergency response to victims of accidents, incidents and have a specific interest in the developing tools to aid operational response to forest fires and are now supporting the development of a MH-EWS for the region.

With the programme set to complete in 2019, the new platforms are currently being trialled. Under the working name of ‘A4’, various regional and local pilot demos are being developed for multi hazards. The pan-European A4EU utilises ECMWF and EFAS products such as the Flash Flood Index and the Extreme Forecast Index. More regionally applied applications include modelling snow load impacts on electricity supply in Finland, modelling traffic disruption due to severe weather in Catalonia and providing early warnings to Schools due to flooding in Genoa.

One application is A4CAMPSITE which is specifically targeted at increasing self protection in campsites located in flood prone areas in Catalonia. There is a high risk of flooding for many hundreds of campsites in the region and there are several legal restrictions being based on the flood plain including ensuring all campsites have adequate flood risk plans.

The EWS is using radar rainfall nowcasting product linked to warning triggers within a self contained warning system.  These are linked to flood plans, with configurable actions (such as evacuation plans) and linked to SMS alerts for the campers.  A4CAMPSITE is being supported by new ANYWHERE partners developing innovative hydrometeorological applications and is being implemented on 13 campsites along the Tordera river with a full trial over the late-summer rainy season.