The CEMS-flood sub-seasonal and seasonal products are essentially the same across the two systems and similarly the same across EFAS and GloFAS. Below, the available products and their main features are introduced. The example snapshots below are from EFAS sub-seasonal and seasonal forecasts, but the styling is the same for GloFAS as well, other than the different domain of the maps.
Product colours
The forecast signal is shown by colouring on the map product layers, either for individual river pixels or larger basins (see the further product details below in the subsequent sections). Each of these river pixels or basins are coloured by the expected anomaly category and the uncertainty category defined for the actual forecast.
Five main anomaly categories (from 'Extreme low' to 'Extreme high') and 3 uncertainty categories (from 'Low' to 'High') were defined. The 5 anomalies with the 3-level uncertainty representation means, there are in total 15 forecast signal categories that could be represented with distinguishable colours on the forecast products. Each of the 5 anomaly categories is shown by a distinct colour ranging from red ('Extreme low') to blue ('Extreme high'), with the 'Near normal' category shown by the neutral grey colour. In addition, the 3 uncertainty sub-categories are highlighted by different intensities of the same colour, going from dark to light as the uncertainty increases (see Figure 1).
To help users monitor the anomalies in more detail, especially in the longer ranges where they tend to be much smaller in the forecast, a 7-value version of the anomaly categories was also implemented by splitting the middle 'Near normal' anomaly into three sub-categories of 'Bit below', 'Normal' and 'Bit high'. These are only shown in the Forecast probability table in the popup product for the reporting points, represented as numbers and inheriting the grey shade colouring of the 'Near normal' category.
The forecast signal is based on the extremity level of the 51 ensemble forecast members in the 100-value climatological distribution and the mean (for the anomaly) and standard deviation (for the uncertainty) of these 51 rank values. The details of the computation methodology is described in: CEMS-flood sub-seasonal and seasonal forecast anomaly and uncertainty computation methodology.
Figure 1. List of the anomaly and uncertainty categories defined with the colours used on the map products.
River network map
The river network map layer shows the combined expected forecast anomaly and uncertainty signal in a simplified way for each forecast lead time (Figure 1). The lead time is weekly (always Monday to Sunday with the weekly average river discharge) in the sub-seasonal and monthly in the seasonal (always calendar month with the monthly average river discharge).
The users can navigate between the different lead times by clicking on the chosen weeks (in sub-seasonal) and month (in seasonal) in the lead time controller (see Figure 2a bottom left corner). This way the users can check the individual signal for each lead time, which currently is 5 or 6 weeks for the sub-seasonal (5 weeks for Tue/Wed runs and 6 weeks for Mon/Thu/Fri/Sat/Sun runs) and always 7 months for the seasonal.
The forecast signal is shown by colouring of all river pixels above a certain minimum catchment area (currently 250 km2 in EFAS and 1000 km2 in GloFAS). The 15 pre-defined anomaly and uncertainty category combinations are used with the 15 colours, as described in the previous section (see Figure 1).
Figure 2 shows examples, which highlight some river sections with the explanation of the assigned colours and the corresponding anomaly and uncertainty levels. All three include the colour legend with the 15 categories and the corresponding colours (same as in Figure 1), for easier interpretation. a) and b) are from the EFAS seasonal, while c) is from the EFAS sub-seasonal, but the difference in appearance will only be the navigation panel in the bottom left corner, which will have the corresponding lead times.
The river network map also contains the reporting points, which are labelled in Figure 2b. These are river locations, where detailed information is provided about the evolution of the forecast signal over the forecast horizon. These reporting points are either fixed points, which are also used in the medium-range flood products or the basin-representative points, which are selected locations on a one point per basin basis. Further details about the basins and the representative points are available here: CEMS-flood sub-seasonal and seasonal basins and representative stations.
a)
| b)
| c)
|
Figure 2. Example snapshots of the river network maps with the reporting points, lead time navigation indicated and categories and colours explained. The examples in a) and b) are from the EFAS seasonal, while c) from the EFAS sub-seasonal.
Reporting point pop-up window
At the predefined reporting point locations (either fixed or basin-representative) further detailed information is provided about the evolution of the forecast signal.
Point information table
Figure 3. Example snapshot of the reporting point pop-up window product's Point information table.
The first table in the popup window is the 'Point information', which provides metadata information of the station (Figure 3). The 'Station ID' is the station identification number, while the 'Point ID' is an internal product generation related process number. In addition, the station name (if available), country, basin and river names are also available. In addition, the point's location is described by two sets of lat/lon coordinates and upstream area. The first 3 values describe the real location, that is either a river gauge (if it is an observation point) or a location requested by users. The second 3 values are the LISFLOOD river network equivalent, the coordinates and upstream area that the hydrological simulations have. For the basin-representative points, however, only the LISFLOOD coordinates and upstream area are available, as these points were defined solely on the simulated LISFLOOD river network, therefore the first 3 numbers will be identical to the second 3 numbers. Apart from the different appearance with either black dot or rectangles, the basin-representative points' 'Point ID' will start with 'SR', while the fixed reporting points' will start with 'Sl' in the metadata table, respectively.
Hydrograph
Next in the popup window is the hydrograph, which graphically summarises the climatological, antecedent and forecast conditions (see Figure 4).
The left half of the plot, left of the horizontal dotted line (as forecast start date), is the antecedent condition section. This part includes some black dots (connected by black line), which show the average calendar weekly (sub-seasonal) or monthly (seasonal) discharge from the so-called water balance, the proxi observations, which are produced as a LISFLOOD simulation forced with either gridded meteorological observations in EFAS, or ERA5 meteorological reanalysis fields in GloFAS. This represents the simulated reality, as close as it can go to reality in EFAS and GloFAS. It always includes 6 lead time periods of the past on the hydrographs, either the 6 calendar weeks before the first week of the forecast in the sub-seasonal, or 6 calendar months before the first month of the forecast in the seasonal (as in Figure 4a for the seasonal and 4d for the sub-seasonal).
The black dots with the observation-based conditions are added to the hydrographs retrospectively, after each calendar week (in sub-seasonal) or month (in seasonal) passes and the weekly or monthly mean proxi-observed river discharge values become available. For the seasonal forecast hydrograph, the water balance is known for all 6 past months at the time of the forecast hydrographs are produced, while for the sub-seasonal the last week (for some run temporarily it can even be 2 weeks) of the antecedent section will not have a water balance black dot, as that is not available yet at the time of the production (but will be plotted usually few days later, depending on the date of the forecast run). The users are encouraged to go back and check previous forecasts to see how well the earlier forecasts predicted the anomalies.
The right half of the plot covers the forecast part, in the example in Figure 4a-c this means 7 lead times of 7 calendar months from August to February (next year), while for the sub-seasonal example in Figure 4d it will be either 5 weeks or 6 weeks, depending on the forecast run date (with 6 weeks in the actual example). The forecast distribution is indicated by boxplots (box-and-whiskers), displaying the minimum and maximum values in the ensemble forecasts of all the 51 members, the lower and upper quartiles (25th and 75th percentiles) and the median (the 50th percentile).
The colour-shaded background in the hydrograph is the model climatology (see Figure 4b for explanation). The climatology is generated using reforecasts over a 20-year period. From the climatology, the 5 main anomaly categories are coloured, the below 10th percentile zone ('Extreme low' with red), the 10th to 25th percentile zone ('Low' with orange), the 25th to 75th percentile zone (extended 'Near normal' with grey), the 75th to 90th percentile zone ('High' with cyan) and the above 90th percentile ('Extreme high' with blue). Further information on the climatologies and their generation is given at: Description of the real time forecasts, reforecasts and climatologies as components of the CEMS-flood sub-seasonal and seasonal forecasts and CEMS-flood sub-seasonal and seasonal forecast anomaly and uncertainty computation methodology.
In the forecast half, for each forecast period the equivalent climatology is plotted with the appropriate lead time from one climatological set. For the seasonal, where the climatology is static and produced only once for each month of the year, the climatology with the same month start date is displayed. For example, for an August 2024 seasonal forecast run, all reforecasts with August run dates (from 20 years) are taken and the climatological percentiles for month1 (August), month2 (September) ..., month7 (February) of those reforecasts are generated and shown on the hydrographs. For the sub-seasonal, on the other hand, the climatologies are produced dynamically for every 4 days (fixed days as 1, 5, 9, 13, 17, 21, 25, 29 of the months, except 29 Feb). So, for the forecasts, the specific climatological set can be chosen which has the run date closest to the forecast run date. So, for example, for a sub-seasonal forecast of 14 December 2024, the climatology produced for 13 December will be chosen, which is the closest to the 14th from the available climate run dates of 1, 5, 9, 13, 17, 21, 25, 29 of December. In addition, the week-1 forecast period (the first Mon-Sun period) will have the lead time of days3-9, week-2 will be days10-16, week3 days 17-23, week4 days24-30, week5 days31-37 and finally week6 is days 38-44.
In the antecedent half of the hydrograph, the climatology with the shortest possible lead time will always be shown. For the seasonal, this will be month-1 for each of the 6 months. For the sub-seasonal, on the other hand, it will be the days1-7 climate lead time. If one of the available climate dates coincides with the Monday of the water balance week, then that will be directly shown. While, if the Monday falls in between two climate dates, then we show the weighted average (by the distance in days from Monday) of those two climatologies. In the above example of the sub-seasonal run on the 14th of December 2024, which was a Saturday run, the 6 water balance weekly periods will be 4-10 Nov, 11-17 Nov, 18-24 Nov, 25 Nov - 1 Dec, 2-8 Dec and 9-15 Dec, all Monday-Sunday calendar weeks. The last of these (9-15 Dec), the one before the week-1 forecast, will not be known by the time the forecast signal is generated and will only be added later retrospectively. For this 9-15 Dec week, the Monday of 15 Dec falls in between two climate dates of 13 and 17 Dec, so the plotted climate will be the average of the two with 50%-50% weights (as both 13 and 17 Dec are 2 days away from 15 Dec Monday).
a)
| b)
| c)
| d)
|
Figure 4. Example snapshot of the reporting point pop-up window product's hydrograph, with different interpretation information added for the seasonal example (a-b-c) and the sub-seasonal (d).
The seasonal hydrograph indicates a property of the model climatologies, which is highlighted in Figure 4c. The seasonal hydrograph is designed to have exactly 13 (12+1) monthly periods, which guarantees that the last month of the forecast (February in Figure 4) will feature both as a month-7 forecast climatology and as a month-1 forecast climatology in the past period, as the oldest month plotted. This way, the comparison between the left-most and right-most background colouring of the hydrograph gives a visual impression of the drift in the river discharge reforecasts. Drift in this context means, the month-7 climatology could occasionally be very different to the month-1 climatology. There can be a noticeable shift or drift in the forecast behaviour, represented by the change in the reforecast behaviour from shorter to longer ranges (see Figure 4c for visual indication of this).
The example in Figure 4 highlights that the 'Extreme low' and 'Low' categories shift only very little from month-1 to month7, with the median being very stable. However, the higher percentiles (75th and 90th) are noticeably higher in the month-7 climatology, indicating a noticeable drift for larger values in the reforecasts for longer ranges. While for month-1, the 90th percentile is about 20 m3/s, so about 10% of the time the monthly mean river discharge can exceed this value, in the longer month-7 range the same 90th percentile increases to 25 m3/s. So, the seasonal forecasts are more likely to show larger values in the longer ranges than in the shorter range.
Unfortunately, this feature of contrasting the climatological behaviour at shorter and longer lead times, does not work for the sub-seasonal, as it is not feasible to plot 52+1 weeks in the hydrograph, which would be necessary to see the same week appearing as a week-1 climatology of the past period and as a week-6 climatology of the forecast.
Probability evolution table
The last part of the reporting point popup window is the probability evolution table. It shows the extended 7-value version with all the 7 anomaly categories (from 'Extreme low' to 'Extreme high' as left to right) and the related probabilities for all the forecast lead time periods and from all the most recent forecast runs that verify during the most recent forecast horizon. For the sub-seasonal, this means 5 or 6 calendar weekly forecast lead time periods, depending on which day of the week the forecast run (Tue/Wed only 5, for all other run dates 6 weeks), and thus how many calendar weeks the 46-day lead time in the forecast can cover; and 7 calendar monthly periods for the seasonal. For the seasonal forecasts, there is always 7 rows with the most recent 7 seasonal forecast probabilities (as Figure 5 shows). While for the sub-seasonal, with including all the daily (00 UTC) forecast runs verifying in the forecast horizon, there can be 34 to 40 rows. Always as many, as many forecasts can verify in the forecast horizon of the actual forecast, and it again depends on which day of the week the forecast run. The bottom right corner of the probability evolution table is blank, as those lead times are not available from the earlier forecast runs.
Figure 5. Example snapshot of the reporting point pop-up window product's Probability evolution table from the seasonal and the sub-seasonal forecasts.
The ‘expected to happen’ anomaly category (one of the seven categories) is indicated by cell colouring in the tables, having one coloured cell for each lead time and run date (all other cells are left white). The colour is dependent on the anomaly, but also on the related forecast uncertainty category (how uncertain the forecast is; either low, medium or high). The expected anomaly category and the related uncertainty are defined by the ensemble member ranks in the 100-value climatology, based on the rank mean and rank standard deviation (of all the 51 ensemble members' ranks), respectively, as described in CEMS-flood sub-seasonal and seasonal forecast anomaly and uncertainty computation methodology.
The used colours in the table are the same as in the river network and basin summary maps (see Figure 1), representing the 15 categories with 15 colours (5 anomaly categories combined with the 3 uncertainty sub-categories), explained above in the 'Product colours' paragraph. The coloured cells will have the exact same colour that the corresponding pixels (that represent the reporting point location in the river network) have on the 'River network' map layer. In addition, the coloured cell's probability value is displayed bold to be more noticeable.
The coloured cell's probability is often the highest of the 7 categories as well, but not always, like in many forecasts in Figure 4. For example, in the latest February forecast run for March, the 'Normal', 'Bit high', 'High' and 'Extreme high' categories all have about the same (22-25%) probability, with the 'High' one being the most likely with 25%. However, given that more than half of the ensemble members are in lower anomaly categories than 'High', some even in the 'Bit low' and 'Low', the most likely 'High' should not really be the expected one, and 'Bit high' really makes it a more reasonable choice, which outcome is guaranteed by the rank-mean-based approach. In addition, the real disadvantage of the most likely category being the expected one is the jumpiness. For example, the May forecast above, again in the February run, shows a stable expected anomaly situation with 'Normal' category. However, the most likely category is 'Low' in the November run and 'High' in the December run, which are very different, even though the underlying distribution of ensemble member ranks is obviously not that different. This jumpiness of the most likely category between forecast runs is likely to happen if the uncertainty is high in the forecasts.
Basin summary map
The basin summary map is the equivalent of the river network map averaged onto the larger basin scale. The basins are predefined, as described in CEMS-flood sub-seasonal and seasonal basins and representative stations, with 204 basins in the EFAS domain and 942 basins globally in GloFAS.
Figure 6 shows an example snapshot in the same area as Figure 1a above. One can compare how the variable signal on the river network averages into the basin signal. The averaging is done from the river pixel information. A rank mean and a rank std value will be calculated for each basin and with those the exact same method is used (as for the river pixels on the river network map) to define the expected forecast anomaly category and the uncertainty category and thus the colour of the basin (see the inset figure with the colour legend in Figure 4). The colours used are the same as for the river network map, except there is a slight shift in transparency, in order to allow visibility for all river pixels, even for those with the same colour as the basin colour.
The basin rank mean and rank std values are determined using all the large enough river pixels in the basin. Currently, only pixels above 250km2 are used in EFAS and pixels above 1000 km2 are used in GloFAS in the averaging. The basin rank mean and rank std values are calculated as an arithmetic average of the rank mean and rank std values of the individual river pixels, weighted by the square root of the upstream area, as described below:
The weighting by the square root of the catchment size, as opposed to the simple arithmetic average (which essentially gives proportionally larger weight to the small rivers as they have usually many more pixels in the basins), gives more weight to the larger rivers in the area averaging. Therefore, in a basin with a single large river near the outlet (such as the Danube or Amazon, etc.) and lots of comparably very small rivers, the large river will have noticeable contribution to the area average.
Figure 6. Example snapshot (from EFAS seasonal) of the basin summary maps with the lead time navigation indicated and the categories and basin colours explained.
Lead time navigation
The forecasts can be advanced with the lead time control panel, both on the river network and the basin summary map layers (see Figure 2a and Figure 6 bottom left corner). The users can check the individual forecast signal for each lead time, which currently is 5 or 6 weeks for the sub-seasonal (depending on which day of the week the forecast run) and always 7 months for the seasonal.