Predicting Rain. Harder than you think.

“When it is rainy, your forecast is always wrong. The rain starts early and is heavier than expected, so why should we continue to use MeteoGroup’s forecast?”

A blogpost by Wim van den Berg, Meteorological Researcher

Maybe you're a consumer using our forecasts before an outdoor activity, or maybe you're one of our professional users, either way, you might well recognize the question above. In this post, we will explain why forecasting rain – or “precipitation”, to use a broader term – is so difficult, and how our approach using a blended forecast is the most valuable.

Let’s start with the easiest case: a widespread band of rain approaches. Rain radar will capture the actual situation quite well, “upper” air winds will push the rain forward in a downwind direction, while global and regional numerical weather forecast models capture this large-scale feature. So, what can go wrong?

First of all, the radar is less accurate than you might expect, as proven by the following cases. The radar beam will miss precipitation in the lowest 1km of the atmosphere, such as drizzle from low clouds, but may see precipitation that evaporates on its way down to the surface (in which case it simply stays dry). The radar cannot see upwind beyond 150-200km, so when a rain belt is already moving at 50kmph after 3 hours, we have no information about whether the rain will continue or not. We can correct this using surface observations, but by then the weather has already passed. That’s why, at MeteoGroup, we mix forecasted radar with Model Output Statistics (MG-MOS), our station-based weather forecasts.

Secondly, the radar does not show the most likely precipitation type. By combining radar data with other parameters, such as the temperature profile, we are able to distinguish between rain, sleet, snow, freezing rain, and hail.

This MG-MOS forecast is the basis of our forecasts . We use this station-based forecast in 20,000 locations, while in 50,000 other locations we downscale forecasts using a smart selection of nearby MG-MOS points.

The MG-MOS system is trained, location by location, on 2-3 years of recent observations and numerical weather forecasts for several models. The resulting blended forecast for any weather parameter - temperature, wind speed, solar radiation or precipitation - is optimized to get the smallest mean absolute error. Day to day changes in the forecast therefore are smooth, but as a consequence, possible extreme events 10 days ahead, are weakened until they become more certain.

Now, let’s discuss a more difficult case: scattered showers are expected. Compared to a rain belt, a shower is short-lived (sometimes lasting only 30 minutes) and local (it may remain dry just 3km away). Showers are also strongly dependent on the daily cycle and local terrain. A shower forecast must therefore be described as “a risk” or “a possibility”. Showers will occur in some areas, however the when and the where are both uncertain. The rain radar will of course detect showers, and we will see the showers moving along with the wind. However, careful inspection reveals that the showers are irregular in both detail and intensity. Forecasted radar is no longer as reliable as it was, as shown in the case of the rain belt. Due to the daily cycle, the radar cannot detect daytime showers before they develop and may forecast showers from the afternoon through to the evening, while in reality they fade.

The MG-MOS forecast is more reliable on typical days where showers are expected. This means it will provide a solid forecast for the likelihood of showers, but it cannot predict the exact time or intensity of the showers. Expanding from a local forecast to county level, it may be that the likelihood of a shower occurring at one or two locations anywhere in a county is 100%, while the likelihood of getting a shower exactly at your home address in this county is only 20%. This clearly illustrates the challenges faced when forecasting rain. Extending a forecast from a 1-hour timeframe to a 6 or 12-hour period also makes the forecast more certain: while you can’t be sure whether a shower will occur at 2pm or 3pm, you may be warned of a shower somewhere between 12pm and 6pm.

To summarize: it is difficult to forecast rain. By using MeteoGroup, you are at least guaranteed to have the most accurate forecast. Depending on your needs, we offer hourly, 6-hourly, 12-hourly, and daily rainfall products (amount and probability, and most likely precipitation type).

Want more information?:
Mail Wim van den Berg, Meteorological Researcher

“It is very hard to predict, especially the future.” (Niels Bohr)