ECMWF has included two new products in ecCharts, based on ensemble forecasts of instantaneous precipitation type: a map showing the most probable precipitation type and a meteogram showing probabilities of different precipitation types for a user-selected site. These new products, created in the framework of the EU-funded ANYWHERE project and inspired by a similar bar chart product from the Hungarian national meteorological service (OMSZ), exploit the probabilistic information provided by ensemble forecasts. This is especially useful in more challenging situations, where there is a risk of snow or freezing rain. In designing these new products, we have incorporated the instantaneous precipitation rate variable in a new way, to define for each type a cut-off between precipitating and dry. This helps to eliminate the misleading impression that miniscule precipitation rates can give in a forecast.
The new meteogram-style product depicts the temporal evolution of probabilities (in percentages) for a specific location in bar chart format. Probabilities are calculated from the instantaneous precipitation type variable, which has seven different categories: dry, rain, sleet, wet snow, snow, ice pellets, and freezing rain. The categories are represented by different colours. Different hues provide details on the probability of different instantaneous precipitation rates for different precipitation types, which can be key for determining the severity of, for example, potential freezing rain or snowfall events. Three different rate categories are used: <0.2, 0.2 to 1 and >1 mm/hour. On the meteogram, the bars are stacked in such a way that the nominally most hazardous type (freezing rain in the high intensity category) is shown at the bottom, and the least hazardous (low intensity rain) at the top. Blank areas at the very top denote dry. The temporal resolution is 3-hourly for T+0h to T+144h and 6-hourly for T+144h to 168h, matching the standard ecCharts meteogram style.
The map product shows which of the six precipitation types, represented by different colours, is most probable whenever it is more likely than not to be precipitating – i.e. the probability of any type is >50%. The hue is used to denote what the probability of the type denoted by the colour is, in three ranges: <50, 50–70 and >70%. In order to expand on this, particularly for longer lead times when high probabilities are less common, we also use grey shading to denote two further categories: probability of any type of precipitation = 10–30 or 30–50%. Another design aspect is that, whenever the lightest shade of a given colour (except grey) appears on a map, the user immediately knows that more than one precipitation type has been predicted for that time, which can serve as an alarm bell for uncertainty.
Bias correction and verification
Over a training period comprising four winter months, we computed the frequencies, amongst weather reports from European SYNOP weather station observations, of different precipitation types, and also type frequencies seen in the corresponding model forecasts for these sites. By adjusting the threshold of minimum permissible instantaneous precipitation rate in the latter, one can achieve a frequency bias of 1 (i.e. unbiased) for each type. The thresholds thus computed for model cycle 43r3, which are now used for generating both products described here, were 0.12, 0.1 and 0.05 mm/h, respectively for rain, sleet and all other types. We expect that these values may change as the model physics is upgraded in future cycles, and indeed perform routine monitoring to check this. After applying these filters, verification of both products showed that they are highly skilful in forecasting rain and snow, but only moderately skilful for sleet and freezing rain (little useful skill beyond day 2 overall), whilst predictive ability for ice pellets is negligible. Another revealing result was that the thresholds required to deliver unbiased forecasts turned out to be fairly consistent across the lead time range, encouragingly suggesting that there is little model drift.
Examples: how to use the products
From the night of 12 January 2017 to the morning of 13 January 2017, the areas of Lombardy and Emilia-Romagna (Italy) suffered a rare freezing rain event, related to the passage of a cold front over cold near-surface air trapped south of the Alps. This caused dozens of traffic accidents including the overturning of a bus near Bologna. In the 24-hour forecast map valid at 00 UTC on 13 January, an area with probabilities of freezing rain greater than 70% is observed in northern Italy, matching quite well with observations from the nearest SYNOP stations. The freezing rain area was smaller in earlier forecasts but there was a signal of a risk even at 48-hour and 72-hour lead times.
One of the most affected places was the town of Parma, corresponding to the black circle in the top left map of northern Italy. The meteogram indicates high probabilities of freezing rain during the first half of 13 January, in forecasts initialised one to three days in advance. Most ensemble members also show moderate rates, between 0.2 and 1 mm/hour (medium red). The small signal of probabilities under 10% observed in a forecast initialised four days in advance may not be not enough for decision-making, but it would alert forecasters to the need to pay close attention in the following days.
Other familiar atmospheric structures can often be seen on the map product in winter. These include the formation of a freezing rain band between rain and snow areas related to a depression (marked ‘1’ in the forecast chart for the North Atlantic–Europe region), the transition from rain to sleet to wet snow and then snow in a region of warm advection (2), and a geographically focused distribution of rain probabilities for an Atlantic front for which the timing is uncertain (3).
In practice we recommend that users generally start with the map product to identify possible events, then investigate in more detail the actual probabilities and rates using the meteogram product, noting also how model site altitude (in the meteogram title) matches the true site altitude, and interpreting accordingly. Users may also wish to compare the charts with the high-resolution forecast (HRES) precipitation type product that was already available in ecCharts. All such products can help with decision- making for local or regional warnings.
For further information, please consult an article on the new products by Estíbaliz Gascón et al. in Weather and Forecasting (doi:10.1175/ WAF-D-17-0114.1).