The primary purpose of the Global Flash Flood Guidance (GFFG) system is to provide real-time guidance products to forecasters worldwide. These products pertain to the threat of potential small-scale flash flooding over large regions with high resolution. The system provides the necessary products to support development of warnings for precipitation-induced flash floods. It uses real-time in situ and remotely-sensed data, numerical spatially distributed land-surface hydrological models and mesoscale numerical weather prediction models.
The GFFG system consists of regional systems (referred to as Flash Flood Guidance Systems (FFGSs)) that allow incorporation of local information into system products and development of regional cooperation in hydrometeorological forecasting. Six regional FFGSs – Black Sea and Middle East, Central America, Central Asia Region, Mekong River Commission, South Africa Region and South East Europe – have been completed and have become fully operational, covering 41 countries. Four – Haiti and Dominican Republic, South Asia, Chiapas and Southeastern Asia-Oceania – are under implementation, covering an additional 17 countries. Another two FFGSs are being designed. One is a stand-alone system for an individual country, while another is for Northwest South America (emanating from the Zarumilla River Basin pilot application), which will likely include three countries. The system has also been successfully implemented at a subnational scale.
Real-time information on precipitation is provided by gauge-adjusted, satellite-based rainfall estimates and, when available, radar-rainfall estimates. An extensive training programme – designed to allow forecaster to adjust system products in real time based on local experience and local up-to-the-minute information – complements the system. The design aims to reduce loss of life and human suffering from the devastation caused by flash floods, and is consistent with an end-to-end forecast response process.
To demonstrate use of products of the GFFG system, imagine a hypothetical situation in which an operational forecaster in Panama begins a shift at 1 p.m. local standard time on 21 November 2015. The forecaster is told that it has been raining in western Panama. The first question the forecaster is likely to ask is: “what is the rainfall forecast for the next 6 hours?” At that time, FFGS, based on the Weather Research and Forecast mesoscale numerical weather prediction model, indicates the precipitation forecast (Figure 1a) for the region.
The forecaster sees that the western mountainous region of Panama is forecast to have significant rainfall in the next 6 hours, exceeding 100 mm/6 hours in some areas. Observing from the operational forecaster logs that there has already been rainfall in that same region, the forecaster would like to know the current saturation level of the upper soils. The upper soil water deficit provides buffers against the production of surface runoff as flash flooding from future rainfall. Based on the GFFG system component that covers the region (the Central America Flash Flood Guidance (CAFFG) system), the forecaster sees saturation levels of 1 or close to 1 in the upper soils for a very significant portion of western Panama (Figure 1b). Consequently, there are very small buffers to absorb the forecast rainfall in the next 6 hours, which is a cause of concern.
The next likely question in the forecaster’s mind is: “which specific areas will experience flash floods?” Using the FFGS Flash Flood Threat (FFT) index (Figure 1c), the forecaster notes that several basins exhibit high FFT indices (values beyond what is considered a normal index variability range, appearing in the yellow part of the index scale).
 |
 |
 |
| Figure 1a. Rainfall forecast for Panama from the Central America Flash Flood Guidance (CAFFG) system |
Figure 1b. Upper soil saturation fraction for small Panama basins from the CAFFG system |
Figure 1c. FFT index for small Panama basins from the CAFFG system |
The forecaster then communicates with local contacts responsible for the high-threat areas (either the Panama weather service or the disaster management agencies of Panama) in order to validate the system estimates of soil saturation and antecedent rainfall. The forecaster also evaluates the rainfall forecasts to make adjustments to FFGS products before deciding whether to issue a warning for high-threat areas. In this case, if the hypothetical forecaster issued a warning, it would have been well verified by what actually happened).
Design fundamentals
WMO and the American Meteorological Society provide definitions of flash floods. These highlight that flash floods are characterized by short response times and small spatial scales. This implies that to gain a lead time allowing for effective response to this hazard, meteorological predictions of rainfall and hydrological predictions of soil water and stream bankfull deficits are needed (and in small scales at that). In addition, local and up-to-the-minute information and data are very useful for issuing a local flash flood warning.
 |
| Figure 2a. Interactive forecaster interface for the GFFG system referring to the Southern Africa regional system |
The Global Initiative for Flash Floods was established in response to recognition that flash floods are prodigious killers and that no reliable warnings existed for them in many countries. Its goal was to support National Meteorological and Hydrological Services (NMHSs) worldwide to provide reliable and effective flash flood warnings and improve disaster management efficiency. With the participation of WMO, the United States Agency for International Development (USAID) Office of Foreign Disaster Assistance (OFDA), the National Oceanic and Atmospheric Administration (NOAA) and the Hydrologic Research Centre (HRC), the initiative focuses on flash flood prone regions worldwide, especially in countries with sparse real-time observations.
In each region, a regional centre with well-developed computational and communications capability is designated by participating countries. Data communication and computational facilities at the regional centres are used to host FFGS computers, which provide secure Internet sites to disseminate information and products to NMHSs of individual countries. Extensive online courses and hands-on training sessions conducted in the regions and at HRC enable country forecasters to use the system products effectively. They can also develop skills for making real-time adjustments as necessary. To achieve such system implementation for each region, computational and data handling components are involved at regional centres and at country forecast agencies. The design of these components is such that it can accommodate global data, regional data and local data through the computational system and can allow for in-country adjustments by forecasters prior to issuing warnings.
The computational servers that perform data ingestion, quality control and model processing are housed at the regional centres where the communications infrastructure is sufficient to handle the requirements. For product dissemination, the national forecasters access the FFGS Dissemination Server at the regional centre through secure Internet connections. After making any needed adjustments using local data and information, they provide warnings to the national response agencies – disaster management agencies, civil defence, etc. This approach creates a general paradigm where the regional system comprises a regional centre Computational Core, and an in-country Adjustments and Warning Core.
The key term “Flash Flood Guidance” (FFG), used for implementation of the GFFG system, refers to a rainfall threshold concept consistent with long-standing meteorological forecast concepts associated with severe rainfall events. According to such a traditional rainfall exceedance concept, a warning is issued when the forecast rainfall exceeds a certain fixed amount of heavy rainfall. The latter is selected from experience or past history to represent a lower bound to amounts that are likely to cause significant damage. The additional value of FFG stems from the fact that it is not a fixed rainfall threshold. Instead, it is a time-varying rainfall threshold based on time behaviour of the soil water deficit and the unfilled bankfull storage of the streams of the small flash flood prone basin of interest. Thus, for high saturation, even modest rainfall amounts can produce flash floods.
An FFG index is defined for each small drainage basin in the region as the amount of rainfall of a given duration and over the small basin that is just enough to cause bankfull flow at the outlet of the draining stream. It is a mean areal rainfall threshold over the basin of interest. If it is exceeded by forecast rainfall, it signifies the onset of overbank flooding at the outlet of the basin draining stream. The concept is used only to signify the occurrence of potential flash flooding for a particular basin and not the magnitude of the overbank flow. The location of occurrence is at the outlet of the small basin.
 |
| Figure 2b. Zoomed-in detail of the regional system of Figure 2a |
The definition of FFG indicates that it is necessary to have soil water accounting (and in cold regions, snow accumulation and melt accounting) in each flash flood prone drainage basin. This is necessary for continuous production of this index. To estimate FFT out to 3–24 hours, it is also necessary to have mesoscale precipitation prediction capability with sufficient resolution (2–4 km2) for the region.
There is a snow water accumulation and melt model, a soil water accounting model (soil moisture model), a model to estimate FFG using information from the soil water accounting model and a component that estimates stream storage deficit from bankfull (threshold runoff model). All of these models are running continuously for all the small basins in the region with varying temporal resolutions depending on the region and available input data. Most systems run with a 6-hourly update cycle, but there are some implementations in convective environments that use an hourly update cycle. The spatial resolution of the small basins delineated for each region depends mainly on the available precipitation data (satellite precipitation is associated with resolutions of order 100 km2, while radar precipitation is associated with spatial resolutions down to tens of square kilometres).
There is input data quality control and, specifically for the remotely sensed precipitation data, there is bias adjustment effected in real time on the basis of the available automated on-site gauges. The adjustment has a climatological factor that is constant over a season and a time-varying, real-time component that uses an adaptive Kalman filter to process real-time gauge observations.
There is a mesoscale numerical weather prediction component, forced by global atmospheric model input that provides high-resolution gridded precipitation forecasts for the region with a maximum lead time of 48 hours. FFGS allows for precipitation forecasts from up to five precipitation models and computes FFT for each such model (a multi-model ensemble) for forecaster review.
A recent rendering of the product interface disseminated through secure Internet is shown in Figures 2a–2b. The interface allows the forecaster to select what they want to see and to overlay other information pertaining to vulnerability to flash floods (e.g. roads or municipalities, and, if available, externally-created vulnerability maps). The forecaster may also zoom in for specific places where there is a potential threat of flash flooding and obtain product information for individual small basins (Figure 2b). The forecaster can also see time series of products (not shown) to compare the present values to those of the recent past before they make a warning assessment.
There is also a five-step training programme to build capacity for flash flood hydrometeorologists in the regions where FFGS has been implemented. The goal of the programme is to train forecasters to effectively use of the system. Those forecasters that exceed a certain level of performance earn WMO certification to become the GFFG system trainers in their countries. Steps 4 and 5 involve such trained and certified in-region trainers.
The development of performance metrics for evaluation and adjustments of the computational component and the development of warnings are recommended to be established for several of the regional systems at least once per year after the wet season. Metrics include the probability of detection of flash flood occurrence and the probability of false warning as well as the probability of a miss. The results obtained so far by the various systems and forecasters point to the value added by the forecasters in real-time operations and preparation of warnings, and the generally reliable seasonal performance of systems during the wet season. In most cases, the sources of highest uncertainty are the numerical weather prediction model forecasts.
Advances and challenges
Each regional component of the GFFG system includes the ability to use input from several mesoscale numerical weather prediction models to develop threat indices for each model for forecaster review. It is notable that forecasters requested that all such model input be distinctly preserved in the products series rather than combine these through the production of a statistical blend of such forecasts. The forecaster’s ability to choose in real time which model to use as input for a particular region underlies such a request and guided the design.
Advances of the basic GFFG system capability include the ability to produce real-time nowcasts of landslide occurrence based on precomputed high-resolution susceptibility maps and real-time estimated thresholds of the FFG-produced precipitation and soil water. This is currently under evaluation for one of the regional systems. Reports are positive, and indicate that this is being used effectively.
FFGSs are also being enhanced with riverine routing and reservoir simulation capabilities. They are now able to provide simulated and forecast hydrographs for pre-specified locations on large regulated rivers of a region. Such information, currently under evaluation in a few locations, is useful for large riverine flood warnings.
Another recent application of FFGS component models pertains to the seasonal ensemble forecasting of snow water equivalent, and combined runoff from snowmelt and rainfall with 6-hourly resolution. This was done for Tajikistan using the data and models of the Central Asia Flash Flood Guidance system (another regional implementation of the GFFG system) in collaboration with Tajik Hydromet.
The challenges experienced during implementation of FFGSs worldwide are listed below, with emphasis on data and information:
- Data ingest. This concerns the great variety of data formats, the availability of public versus private data, the reliability of data delivery to the system, the asynchronous arrival of data and a mix of space–time data resolutions.
- Measurement/forecast uncertainty. This refers to the uncertainty characterization as climatological versus time varying, and the uncertainty implications of the availability of only short records for fine-tuning of system reliability.
- Timely product/warning generation. The main issues here are the regional centre computers (and the related communication requirements and constraints) and the opportunity for timely forecaster adjustment and warning generation coupled with effective response. For example, decisions to not use ensemble prediction products in the initial implementations were made on the basis of these constraints in regional centres.
- Products easily accessible and searchable by NMHSs. This refers to the in-country interface and database requirements, the constraints in communications to access the system products in a timely fashion by certain developing countries, the use of local versus regional data storage, and the requirement to use free and open source software for developing countries (e.g. end-user geographic information system (GIS) software).
- Education and training in product interpretation and communication with disaster management agencies. The issues here arise because of diverse forecaster backgrounds, the necessary interdisciplinary and multidisciplinary nature of the assessment process that leads to the generation of warnings, and the cultural and socioeconomic diversity in the perceived value of and response to warnings by forecasters, disaster managers and the public.
- The presence of the GFFG system worldwide and the experience being accumulated by its use in a variety of situations, constitute a catalyst to turn these challenges to opportunities to further improve the utility and reliability of the GFFG system and to enhance hydrometeorological information and collaboration worldwide.
Acknowledgements: Implementation of the GFFG system regional components was done over the last 15 years with the support of USAID/OFDA, WMO and NOAA. The support and guidance provided during these years by Sezin Tokar, Curt Barrett, Paul Pilon, Ayhan Sayin, Claudio Caponi, Wolfgang Grabs and Dan Beardsley of the partner agencies is gratefully acknowledged. The willing and proactive participation of the staff of regional centres and the country forecasters constituted the essential foundation that made the development of useful regional GFFG system components feasible. Contributions of HRC staff by area (in no particular order) were as follows: Theresa Modrick-Hansen in hydrometeorological modelling and GIS analysis, Eylon Shamir in soil water and snowmelt modelling and evaluation studies, Rochelle Graham in training programme development and development of links to disaster management agencies, Ari Posner (now at the US Bureau of Reclamation) in landslide assessments, Zhengyang Cheng in riverine routing and urban flood prediction, Jason Sperfslage in computational systems engineering and software design, Cris Spencer in operational software design and development, Randall Banks in interactive Internet interface development, and Robert Jubach in programme management and linking of warnings to disaster management.
