Version 2.0 includes the following major features:

• Version numbering system for the GloFAS cycles
• Calibrated LISFLOOD routing component
• Updated reference discharge climatology
• Improved initialisation of the real-time forecasts
• Available GloFAS datasets

When will the GloFAS upgrade happen?

It is staged in two phases:

8 October 2018, the version 2.0 discharge forecast reruns will be made available upon request. This will enable users to conduct their own skill assessments, including defining decision-making thresholds, and to start the adjustments in their downstream applications necessary for a smooth transition to version 2.0. More information is provided in the "requesting GloFAS data" section.

14 November 2018, real-time GloFAS flood forecasts displayed on www.globalfloods.eu and distributed to the users will be calculated from version 2.0. On the same day, the current operational production chain version 1.0 will be stopped.

Why an upgrade?

Every real-time forecasting service implements regular cycle upgrades for a continuous improvement of the service. In April 2018, when GloFAS became an operational service, significant changes were made in the way forecasts were calculated beyond 16 days lead time. Extended hydro-meteorological forecasts were introduced to force the model from day 16 to day 30 for more realistic discharge predictions, particularly in small to medium-sized river basins.

Since then, preparations were made for a major upgrade in the forecasting chain, including calibration of the hydrological routing component to improve the flood wave simulation worldwide and change in the configuration chain for improved initial conditions. In response to requests from users, the upgrade also includes the introduction of a version numbering system for full traceability of GloFAS flood forecasts, and the publication of more than 20 years of discharge forecast reruns, so that users can conduct their own evaluation of the skill of the forecasts over a long period.

Requesting GloFAS data

Access to real-time forecasts and forecast reruns from GloFAS is possible through a dedicated ftp service set-up by the GloFAS team upon request. Data is provided either as point time series or 2D grid series, both in netCDF format. Switch-over from version 1.0 to version 2.0 will be transparent to the user, with real-time forecasts from version 2.0 being automatically transferred from the release date, the 14th November 2018.

For many users, access to forecast reruns is very important. The version 2.0 forecast reruns will be made available to users from 8 October 2018. Whilst the GloFAS team undertakes rigorous performance assessment of the system, and will publish those through the GloFAS website, assessment of the forecast performance for local applications is very important and strongly encouraged before GloFAS forecasts are used in a decision-making context.

Request of real-time forecasts or forecast reruns from version 2.0, or any other inquiry and feedback on the service can be made to info@globalfloods.eu. The team endeavours to respond to all requests as soon as possible within 10 working days. Please provide the following information when sending a request for data:

• Requesting data for stations/points (only for river sections with upstream area larger than 1000 km²):
• Station ID(s)
• Name of station(s)
• Name of River(s)
• Name of country
• Station latitude(s)/longitude(s)
• Station upstream area(s) from provider (if available)


• Requesting data for areas:
• Latitudes/longitudes of the bounding box region

Please contact us!

For any other inquiry, feedback on the service or collaboration opportunities, contact info@globalfloods.eu.

Version numbering system for the GloFAS cycles

When GloFAS transitioned to an operational service, a 2-digit version numbering system was introduced, to uniquely identify the discharge produced by GloFAS:

• The first number relates to major upgrades that involve changes to the hydrological modelling chain (rainfall-runoff and routing)
• The second number relates to any other change in the system’s configuration that have an impact on the discharge (e.g. change of the reference climatology, change of the forecast reruns or an upgrade of the real-time ECMWF Integrated Forecast system (IFS) data input)

Calibrated LISFLOOD routing component

GloFAS produces daily discharge forecasts based on the HTESSEL land surface model (the land surface component of ECMWF IFS, Balsamo et al., 2009) and off-line LISFLOOD hydrological routing model (developed at the Joint Research Centre of the European Commission, van der Knijff et al., 2010).

Version 1.0 uses a LISFLOOD configuration based on expert-defined parameters which are uniform across the entire geographical domain, regardless of land surface characteristics.

For version 2.0, a major calibration exercise of the LISFLOOD hydrological routing parameters was conducted for better accounting for the variability in the hydrological processes (Hirpa et. al., 2018). The calibration used ECMWF re-forecast forcing (surface and sub-surface runoff) from June 1995 to June 2015, and was conducted on 1287 stations worldwide with available daily discharge observations. The calibration was performed using an evolutionary optimization algorithm with the Kling-Gupta Efficiency (KGE, Gupta et al., 2009) as objective function. As a result of the calibration, the version 2.0 discharge simulation skill has improved for the majority of stations (increased KGE score relative to the version 1.0 reference discharge climatology), and it is expected to improve the real-time version 2.0 discharge and flood forecast skill.

Calibration is only possible if good quality river flow observations are available, and the team is grateful to all data providers who have given us access to their data.

Updated reference discharge climatology

GloFAS reference discharge climatology is a long deterministic hydrological simulation, providing modelled daily river discharge over the GloFAS river network. It is used to evaluate the skill of the system, and to define the reference discharge thresholds which provide the context to identify the severity of flood events in real-time forecasts.

The version 1.0 reference discharge climatology is based on ERA-Interim/Land data, a surface-only version of ERA-Interim, the previous ECMWF atmospheric reanalysis. ERA-Interim/Land data shares the same atmospheric forcing as ERA-Interim (~80 km horizontal resolution) although it benefits from a better description of land surface processes, which is similar to the scheme used operationally in the ECMWF IFS. The version 1.0 reference discharge climatology covers the 1980-2015 period.

The version 2.0 reference discharge climatology is based on ERA5, the new atmospheric reanalysis product of ECMWF. ERA5 has a horizontal resolution of ~32 km and benefits from substantial model improvements compared with ERA-Interim (see Hersbach and Dee 2016 for further reading). The version 2.0 reference discharge climatology covers the 1996-2017 period.

Improved initialisation of the real-time forecasts

The GloFAS configuration determines the hydrological conditions that are used at the start of the real-time discharge forecasts. This is an important aspect of the modelling chain as the initial conditions contribute much to the evolution of discharge, especially in the first few days of the forecast period.

Real-time daily initialisation: When GloFAS runs in forecast mode, it requires a set-up defining in real-time the initial states (atmosphere, land surface) from which to start the ensemble discharge forecast. This is provided by a daily update of the river state, every day for the past 5-day period. ERA5 is used as input in the first three days of this period (or as long as ERA5 is available). As ERA5 is available in near-real-time, with an approximately 2-day latency, the forecast initialisation for the most recent two days, prior to the forecast run date, is based on day-1 ECMWF IFS ensemble control forecasts.

The version 2.0 configuration ensures the highest possible consistency between the reference discharge climatology and real-time forecasts. This is to make sure the return period thresholds, calculated using the reference climatology, are consistent with real-time forecasts. This aspect is crucial for producing reliable flood forecasts, and addresses a known weakness of version 1.0.

Available GloFAS datasets

Version 2.0 includes 30-day forecast simulations performed for a set of past dates. These "discharge forecast reruns" can be used to analyse GloFAS performance systematically over a long period (see the following table for a summary).

Table 1: Summary of the available GloFAS datasets with the period covered, the frequency at which they are produced, the number of members in the ensemble and the ECMWF IFS versions that were used to produce the runoff input for GloFAS. Note that GloFAS datasets only include daily discharge time series.

	Period	Frequency	Ensemble members	ECMWF IFS version
Reference climatology	1996-2017	Daily	1	41r2
Re-forecasts	Jan 1997 - Dec 2016	Twice per week	11	43r1: 1 Jan 2017 - 11 July 2017
Hindcasts	1 Jan 2017 - 30 Jun 2018	Twice per week	51	43r3: 12 Jul 2017 - 5 Jun 2018
Real-time forecasts	1 Jul 2018 - real-time	Daily	51	45r1: 6 Jun 2018 - real-time

The version 2.0 re-forecasts are discharge forecast reruns based on the available ECMWF IFS re-forecasts for the 1997-2016 period, initialised using the version 2.0 reference discharge climatology. The IFS re-forecasts are forecasts for past dates created using the same version of the ECMWF IFS system that is used to produce real-time IFS forecasts. IFS re-forecasts are available twice per week for Mondays and Thursdays and only contain 11 members (against 51 members for real-time forecasts).

The version 2.0 hindcasts are discharge forecast reruns based on the available real-time ECMWF IFS forecasts from 1 January 2017 to 30 June 2018, initialised using the version 2.0 reference discharge climatology. The real-time IFS forecasts are available daily and contain 51 members. The version 2.0 hindcasts were produced twice per week (on Mondays and Thursdays), for consistency with the version 2.0 re-forecasts, but using all 51 ensemble members.

From 1 July 2018, the version 2.0 system has been running in pre-operational mode, using the real-time version 2.0 configuration and the daily initialisation. These real-time version 2.0 forecasts will be made available to the end users after the implementation of the GloFAS upgrade in November.