CEMS-Flood reporting points and dynamic point generation algorithm

Reporting points have two main goals: they provide an overview of the floods over the CEMS-Flood domain over the forecast time horizon and also give access to local detailed forecast information. They are determined based on the maximum flood signal in the forecast horizon.

But how are the location of the dynamic points determined, and how the severity of the signal, which drives to the colour of the reporting points and the river cells in the flood summary layers, is computed?

River network

Figure 1 highlights how the rivers are represented in CEMS-Flood. It follows a simplified and discretised way by converting the real rivers into a sequence of cells, where the water flows from one cell to the next downstream.

In the dynamic reporting point generation algorithm, only cells with a large enough upstream drainage area are considered, the upstream area threshold depending on the hydrological model grid used, so cells with very small drainage areas can not become reporting points. In the following, the reporting point generation is demonstrated on this small idealised river network area.

Figure 1. River network schematic.

Threshold exceedances

Next step is to compute the flood threshold exceedance values. CEMS-Flood primarily uses three main flood severity levels, the 2-year, 5-year and 20-year return periods. We determine for each of these
thresholds whether the predicted maximum river discharge over time exceeds the threshold value for each individual forecast member (and deterministic forecasts for EFAS). We assign 1 if they exceed and 0 if they do not. This process is represented in Figure 2 by the coloured river cells, which show that the maximum river discharge over time exceeds the flood thresholds. Please note the triviality that if the 20-year threshold is exceeded for any cell, than both the 5-year and 2-year must be also exceeded for those cells, i.e. more and more cells are coloured (indicating exceedance), as we go from 20-year to 2-year level.

Figure 2. Schematic to represent the flood threshold exceedance for one ensemble member for each of the 2-, 5- and 20-year severity levels.

Exceedance probabilities

We then compute the threshold exceedance probabilities (0-100%) for all the three thresholds of 2-year, 5-year and 20-year. Figure 3 demonstrates the process of the exceedance probability computation for 5-year return period (RP). Depending on how many of the 51 members exceed the thresholds, the probability is computed and represented by the different shades of red, as the darker the red the higher the probability.

Figure 3. Schematic representation of the exceedance probability computation based on the 51 ensemble members for the example 5-year return period (RP).

For EFAS, in which the system is composed of multiple forecasting system, the exceedance probabilities of the forecasts are combined into a total exceedance probability, using a weighted average over the forecasts.

Merged probabilities

In the following step we combine the probabilities of the three severity levels and create one probability map that hierarchically combines the three probabilities.

Firstly, we convert each probabilities into binary (yes/no) information by applying a minimum threshold value, (typically 30%). If the exceedance probability exceeds this minumum threshold value, we consider this flood category forecasted. This is represented by thicker black borders around the river cells in Figure 4a.

Then, the three probabilities are combined hierarchically (Figure 4b), using three probability categories of 30-50% (shown by light colour), 50-75% (shown by medium colour intensity) and 75-100% (shown by dark colour). If the 20-year probability exceeds 30%, then those cells get the 20-year severity level (purple) and one of the three subcategories (of of the purple colours) depending on the probability value. For cells which has less than 30% 20-year probability, if the 5-year probability exceeds 30%, then they will get the 5-year (red) severity level, again with one of the three red intensities depending on the 5-year probability value. Then if the 5-year probability is below 30%, then it is 2-year (yellow) severity, etc. Finally, if even the 2-year probability is below 30%, then the cell is characterised as 'no flood' and will be left uncoloured.

An example of the product that is available on the CEMS-Flood website for GloFAS ('Flood summary for days 1-30') based on the merged probabilities is shown in Figure 4c.

a	b	c

Figure 4. Representation of how the three probabilities are combined into one map with three subcategories (30-50%, 50-75% and 75-100%) in each of the 2-, 5- and 20-year severity levels.

Fixed reporting points

The fixed reporting points will always appear in CEMS-Flood (Figure 5a). They will be shown by one of the three colours, yellow, red and purple, representing the three flood severity levels. This level comes from the combined probability map, generated in the previous step (i.e. the map shown as 'Flood summary for days 1-30' on the GloFAS website). This means, whichever (highest) of the three severity levels is exceeded by at least 30% will define the colour of the fixed reporting points (if none of the severity levels are exceeded the points are shown in grey). In our examples one of the points will remain grey (i.e. below 30% 2-year probability, so no flood) and the other will be shown as purple (20-year flood level).

a	b

Figure 5. Schematic representation of the fixed reporting points and how the flood severity with the colour is determined.

Dynamic reporting point generation

Motivation: The dynamic points are generated on river sections that do not have any fixed reporting points and show some expected flood conditions (the 2-year exceedance probability is at least 30%). These points help the users to monitor the expected evolution of the hydrological state with detailed information (i.e. hydrographs) at these river sections (Figure 6a).

Filtering: In order to find the optimal representation of these 'no-fixed-point' areas, we first filter all the very short and small river sections (these are represented by grey colour in Figure 6b). Currently, all flooded river clusters with at least 6 contiguous cells will be kept and the smaller ones removed. This means we disregard all the short rivers or those longer rivers that only flooded on a short section.

Pre-selection: Then a list of cells are pre-selected to become dynamic reporting points from the filtered flooded areas based on some rules (Figure 6c). This is done independently for each flooded cluster (contiguous cells, which are sometimes only a short river section, other times potentially a very large river system, depending on the forecast situation) and within these for each of the three severity levels separately, we select the following cells:

most downstream cell
every 10th cell going upstream from the most downstream cell
the maximum probability cell

The maximum exceedance probability cell is chosen as an extra point (in case it is different to the already selected cells), as often that will carry an important extra piece of information about the flood situation. If the maximum exceedance point is too close to the most downstream one, only the maximum exceedance point is kept. This process will deliver in our simplified example in Figure 6c the selection of two 5-year and 2-year cells (darker red and darker yellow) and three 20-year cells as the 20-year over-30% probability river section is long enough to have the every-10 rule giving another point.

Final selection: Finally, those preselected points are discarded which are too close to the fixed reporting points or too close to the dynamic points defined in the previous forecast run. This latter condition is added in order to avoid jumpiness of the dynamic point locations in subsequent runs, depending on the geographical distribution of the probabilities. This guarantees that the users can continue monitoring the expected flood conditions at these dynamic locations as long as the flood signal is there. This means, if the probability decreases below 30% in a forecast, then the dynamic point is removed. Even if in the next forecast the flood signal comes back, in which case a new dynamic point position will likely be defined.

The removal of the pre-selected dynamic points is done using the so-called kill-zone (Figure 6d). This is a short river section both upstream and downstream (5 cells both directions) from either the fixed points or yesterday's dynamic points. The kill-zone is represented by light grey colouring in Figure 6d. The kill-zone only covers the main river channel and does not extent into the tributaries. It stops at the confluence cell when two relatively similar (in drainage area size) rivers meet. This is the case for the purple fixed point going downstream in the top-middle map and the yellow yesterday's dynamic points going downstream in the top-right map in Figure 6d. In case a bigger and a smaller rivers meet, like when going upstream from yesterday's yellow dynamic point in the top-right map in Figure 6d, the kill-zone will not stop at the confluence.

For our idealised river network, the end result will not include three of the preselected dynamic points. This means we remove one purple point, as it is too close to the purple fixed point and we also remove both the red dynamic points, as they are too close to yesterday's yellow dynamic point. This yesterday's point will also change colour from yellow to red today, as the flood probabilities have increased from yesterday to today.

In total, we will end up with three purple (20-year), one red (5-year) and two yellow (2-year) reporting points in today's forecast. These points will be displayed in the 'Reporting points' layer on the GloFAS website. they will all show consistent information with the 'Flood summary for days 1-30' layer, that has the combined probabilities for each river cell.