Reliable weather forecast made easy.
We all use different weather applications (apps) to get the most accurate forecast of the weather in the near future. But how reliable are such programs and where are their limits? And why do the predictions of the apps differ from each other?
The classic weather forecast
Every evening on television, the weather is predicted for the coming days in less than a minute: over the entire area of the country and usually even with parts of the northern Alps. The forecast for such a large area is, of course, roughly rasterized, because small areas cannot be covered in the given time (transmission time: 60 seconds). Therefore, the forecast is limited to the interpretation (!) of the general weather situation (It could be that …, not the look into the crystal ball: it will …).
In fact, each human lives in an extremely small section (grid) of the country. Therefore, a corresponding interpretation of the forecast is required to infer from the rough grid of the forecast to the small grid of the location.
In order to get a feeling for the probability of the forecast, however, it requires a general understanding of the general weather situation (current weather), the desired location (precision), the local topography (small-scale air flows – very interesting in the mountains) and the knowledge of how a forecast is made (calculation of the weather models). It is also important to know that there are commercial and non-commercial forecasts (it’s about reach and advertising revenue – more about this below).
Various weather models
Basically, the current general weather situation is the basis for any forecast. It is confirmed by weather radar. Weather radars can show us the past and present weather situation (actual value), but not the future. Worldwide weather data is collected from different weather stations and flows into complex computer simulations. In addition, there is measurement data from current air traffic (one of the main reasons why the weather forecast is less precise in times of Corona, because there are only a few airplanes flying). These simulations incorporate data from past (real) weather, previous (virtual) calculations and the current (real) measured values of the weather stations. In addition, different (virtual) calculations with different input variables are weighted accordingly. This creates uncertainties and probabilities for weather events that occur (the uncertainty is always part of any serious weather forecast). These simulations reflect a range of weather probabilities. This in turn results in different possibilities for the weather forecast in weather models (published every 6 hours). From these models the predictions (interpretations) are created, which are published daily.
There are numerous weather models from different institutes. Five of the most important models are GFS/NOAA (Global Forecast System/National Oceanic and Atmospheric Administration with a resolution of 22 km) in the USA, ECMWF (European Centre for Medium-Range Weather Forecasts with a resolution of 9 km) in the EU, ICON/DWD (German Meteorological Service with a resolution of 6 km) in Germany, NEMS/Meteoblue (with a resolution of 4 km) in Switzerland and AROME (with a resolution of 1,3 km) in France.
Everybody knows the following situation: While planning a tour, you get the following statement: “My weather app tells me that there should be nice weather at the destination tomorrow (often with the subjective addition: this is the best weather app, because the prediction is mostly correct)”. Another app may come to a different forecast. How comes? And which app is right at the end?
Each weather app has access to a certain weather model and displays its calculation in a prepared form. And because different weather models work with different data bases and input variables, different interpretations are possibly created at the end (same with the soup and the various ingredients). This information is accordingly summarized and displayed as weather symbols (in the best case as meteograms) which represent the forecast for a certain period of time.
» Serious forecasts can be calculated very accurately up to a maximum of three days in advance, but depending on the weather conditions sometimes only for the next six hours and depending on the topography sometimes not at all. Deviations between the different models indicate uncertainties in the weather forecast. It is therefore important to understand that these deviations are not visible in a single weather model – so you need access to different models and their comparisons over several days!
What does a rain probability of 63% mean?
It’s simple: it means that it will rain in 63 out of 100 days with such a weather situation. But how does this exact number come about? Computers read out the data of the weather stations and calculate the weather situation as well as when and where exactly it has rained in the past. In this case, 63 out of 100 days with identical weather situations have experienced rainfall within 24 hours in the past (more precisely: one precipitation event), but on the 37 remaining days it stayed dry – with the same prediction!
In addition, there is a certain uncertainty due to the topography (mountains, hills, rivers). An example: a probability of rainfall of 10% is predicted for the northern part of the Harz (a low mountain range in the middle of Germany). Where is the probability for a rain shower higher – on the Brocken (with 1141 m highest mountain in the Harz) or in Goslar (a town 15 km north of the Brocken in the Harz) on a height of 255 m? And in which wind direction does it rain where?
Due to the spatial conditions it is very likely that it will rain on the windward side (the mountain side facing the wind) and on the mountain (the probability can be 90%), while it will probably stay dry on the leeward side (the mountain side facing away from the wind “always pee in lee”) and in Goslar. For an exact interpretation of this forecast a high resolution (fine grid) of the weather model is very helpful.
Incidentally, the probability of rain does not provide any information about the amount of precipitation, the period of time (shower or continuous rain) or whether it rains, hails or snows. But these are important factors in mountaineering.
Commercial weather apps and its Wet bias (rain distortion)
“Wet bias” refers the process of meteorologists to specify rain probabilities (usually low probabilities of precipitation) in the weather forecast to compensate the interpretation of users (a 5% probability is then predicted as a 20% probability).
» This is about weather psychology. After all, a user tends to rate a weather app as bad if the app does not forecast the precipitation event even though it takes place. Conversely, it is less problematic if the forecast precipitation does not occur (the user is glad that it does not rain in this case). Since weather apps are usually financed by advertising revenue, it is important for their owners to achieve a high user rate through an attractive prediction (so much for the best weather app).
» Nicely calculated weather data is of little help during the planning process in mountain sports, because we need exact data that we can rely on as best as possible. At the end of the day, it's all about our safety.
Different representations and forecasting qualities
» Weather reports (examples: chamonix-meteo.com, meteoschweiz.ch, dwd.de, alpenverein.de):
Daily manual interpretation of the general weather situation (actual value) and weather models (set value) for small areas of professional meteorologists offer a high quality and reliability of the forecast.
The interpretation of the weather models can also be done by yourself with the necessary experience and knowledge of the area.
» Meteograms (examples: meteoblue.com, yr.no, wetterzentrale.de, windy.com):
A meteogram can be used to show trends and temporal progressions. This makes it possible to clearly display many forecasting information on a small space/display. The data is computer generated.
» Weather icons (examples: wetter.com, wetteronline.de):
Simple weather icons in a weather app represent the lowest quality level of forecasting with the greatest inaccuracy and most uncertainties. The data is computer generated.