CEMS-Flood sub-seasonal and seasonal forecasting

The CEMS-Flood sub-seasonal and seasonal range forecast products follow the same design from EFAS version 5.4 (implemented on 12 March) and GloFAS version 4.3 (implemented on 04 June 2025), described on this page.

Key features

• CEMS-Flood sub-seasonal products are produced every day, showing calendar week averaged river discharge anomalies up to 6 weeks divided into 5 main categories (extended to 7 categories in some products), from extreme low to extreme high compared with a forecast-range dependant climatology.
• CEMS-Flood seasonal products are produced every month, showing monthly averaged river discharge anomalies up to 7 months divided into 5 main categories (extended to 7 categories in some products), from extreme low to extreme high compared with a forecast-range dependant climatology.
• Both sub-seasonal and seasonal products are generated for the EFAS and GloFAS systems based on the same methodology.
• Products are shown as maps over the full domain (river network and basin summary versions) with additional information available for reporting points, such as:
  • Hydrographs: showing the antecedent conditions of the recent past and the forecast evolution of the future, together with the underlying model climatology to indicate the anomalies.
  • Probability evolution tables: showing the forecast probabilities for the different anomaly categories at the different lead times from the most recent forecast runs.
• Map products are plotted for each lead time individually, with the possibility to navigate between them.

Background

In a sub-seasonal and seasonal forecast, especially at the the longer ranges, the day-to-day variability of the river flow can not be predicted due to the very high uncertainties in numerical weather prediction. What is possible, is to rather give an indication of the river discharge anomalies and the confidence in those predicted anomalies. As the forecast range increases, the uncertainty also generally increases and with it the sharpness of the forecasts will gradually decrease; subsequently, the forecasts will more and more replicate the climatologically expected conditions with some possible positive or negative shift (i.e. anomaly). 

The generation of CEMS-Flood sub-seasonal and seasonal forecast signals and related products is reflective of this and was designed to deliver a simple-to-understand categorical information for both the anomalies and the associated uncertainties present in the forecast, very much relative to the underlying climatological distribution.

Workflow

The generation of the sub-seasonal and seasonal forecast signal relies on few major steps, illustrated by a flowchart in Figure 1, with some further details described in Figure 2 and 3. The forecast anomaly and uncertainty signals are derived by comparing the real time forecast (top left section in Figure 1) to the 99-value percentile climatology (bottom left section in Figure 1). The real-time forecasts are based on the ECMWF sub-seasonal and SEAS5 seasonal meteorological forecasts (Figure 2), while the climatology is generated using reforecasts (from the same two systems) over a 20-year period, which provides range-dependent climate percentiles that change with the lead time (Figure 3).

Figure 1. Flowchart of the sub-seasonal and seasonal anomaly and uncertainty signal generation methodology.

Figure 2. Flowchart of the details in the real-time forecast generation for the sub-seasonal and seasonal forecasts.

Figure 3. Flowchart of the details in the model climatology generation for the sub-seasonal and seasonal forecasts.

The CEMS-Flood sub-seasonal products cover calendar week periods (i.e. always Monday-Sunday), while the CEMS-Flood seasonal products are valid for whole calendar month periods. The forecast signal is derived from the relationship between the calendar-weekly- or monthly-averaged river discharge and the climatological distribution of the historical calendar-weekly- or monthly-averaged values. While calendar months naturally allow consistency checks, the fixed calendar-weekly lead-times in the sub-seasonal were introduced to allow users to directly compare forecasts from different forecast runs, as the verification period is fixed onto the calendar weeks, with different corresponding lead times from the different forecast runs, depending on the day of the week the forecast did run. 

Definition of anomaly and uncertainty signal

The anomaly and uncertainty signals are determined by comparing the real time forecast ensemble members to the model climatology (see the middle-right section in Figure 1). This identifies how extreme the ensemble members of the forecast are in the context of the climatological behaviour, which is represented by the 99 percentiles and the corresponding 100 equally likely bins of the climatological range. Each ensemble member gets a rank from 1 to 100, after sorting them into one of the 100 climatological bins. The real time ensemble forecasts then has 51 rank values from 1 to 100.

In order to display the information content of these 51 ensemble rank values (mainly maps and hydrographs), we use a simplified anomaly representation. For this, 5 main anomaly categories were defined by the 10^th, 25^th, 75^th and 90^th percentiles, representing 5 categories of larger anomalies as 'Extreme low', 'Low, 'Near normal', 'High' and 'Extreme high'. However, the probability evolution table in the reporting point pop-up window product includes seven anomaly categories, in order to give more details by sub-dividing the 'Near normal' category into 'Bit low, 'Normal' and 'Bit high' with the 40^th and 60^th percentiles.

The overall expected anomaly category of the forecast is defined by the mean of the 51 ensemble member ranks (rank mean), which represents a robust anomaly value without any random jumpiness. The uncertainty about these anomalies is defined using the standard deviation of the 51 ensemble ranks (rank std). Based on this rank standard deviation value, one of three uncertainty categories (low / medium / high) is assigned to the forecast.

Product overview

Two web layers are available for both the CEMS-Flood sub-seasonal and seasonal systems on the EFAS and GloFAS websites. These are the river network map, which also includes the reporting points and the related popup windows delivering extra information about the forecast evolution, and the basin summary map (see the last column on the right side of Figure 1).

River network map

The river network map highlights the combined forecast anomaly/uncertainty signal on the rivers, including all river pixels above 250 km² in EFAS and 1000 km² in GloFAS. The forecast signal (or overall expected anomaly for a given lead time) is given by colouring of the river pixels. The five main anomaly categories are shown with different colours, with the uncertainty level provided by the colour intensity. 

From the river network summary map, the reporting point pop-up products can be accessed by clicking on the point markers on the river network layer (See more on the reporting points at CEMS-Flood diagnostic and web reporting points and  CEMS-flood sub-seasonal and seasonal basins and representative stations). The pop-up window product contains metadata information about the stations, a hydrograph with the evolution of the climatological, antecedent and latest forecast conditions, and the probability evolution table which shows discharge probability for the extended 7-value anomaly categories from the most recent forecast runs, and highlights the overall expected anomaly category by colouring one of the 7 cells. Note: the probability table is divided in 7 anomaly categories with the larger 'Near normal' category shown on the river network and basin summary maps and the hydrograph extended into 3 categories. 

Basin summary map

The basin summary map is a geographical summary of the river network map, after averaging the signal in the predefined basins (see CEMS-flood sub-seasonal and seasonal basins and representative stations) and synthesising the more complex and geographically variable information presented on the river pixels in the river network layer.

Forecast evolution

The river network and basin summary forecasts are provided for each lead time individually. Users can navigate through lead-times by using a control panel in the bottom left of the map viewer providing a menu for the available forecast dates. For each forecast date, 6 maps (which for some days of the week decreases to 5) are available for the sub-seasonal products (from week-1 to week-6) and 7 maps are available for the seasonal products (from month-1 to month-7).