If there is one thing Brits love moaning about more than the weather, it is complaining about how wrong the weather forecast was. Just ask BBC weather presenter Michael Fish, who infamously dismissed the possibility of a hurricane immediately before a deadly storm devastated southeast England.
Meteorologists and weather forecasters are incredibly skilled, usually creating impressively accurate forecasts. However, weather forecast inaccuracies can have many unwanted consequences, from unexpected rain showers on what was supposed to be a bluebird day in the hills to ruined crops.
We’ll leave improving weather forecasts to skilled meteorologists. Still, this blog will investigate some of the reasons why weather forecasts, particularly in the mountains, can end up being different to the weather conditions we experience.
The aim of this blog is certainly not to dissuade readers from using the outstanding weather forecasting available to them (MWIS, Met Office, Meteo Blue, Yr). Rather, if anything, the aim is to head into the hills even more prepared for all weather eventualities knowing that sometimes there is a chance of unexpected weather.
Given the importance of weather conditions to almost all human activities, there is a long history of weather forecasting attempts. Attempts to predict weather conditions stretch back thousands of years. Observable weather phenomena can be pre-cursors to future weather events. For example, halos around the sun can indicate an advancing frontal system and the potential for rainfall. More scientific approaches to weather forecasting originated in the 17th century, with the first, often inaccurate, forecasts published in The Times in 1861.
The precursors to modern numerical weather prediction/forecasting developed in the 20th century. Complicated mathematical models and equations are now used to predict future conditions. These mathematical models have become increasingly sophisticated and numerous, accelerated by increases in computing power since the 1950s.
Modern forecasts are largely generated using numerical weather prediction (NWP) techniques.
Starting with observations about the current state of the atmosphere (e.g., at 10h00 on 12/01/2023 the air temperature in Aviemore was 1°C), a supercomputer solves equations describing how atmospheric conditions will change with time (an NWP Model). This produces a forecast for a small period of future time. The forecast is then fed into the original equations and solved to predict conditions further into the future. This is repeated, with key times (e.g., +6hrs, +12hrs) saved until a suitable future time (lead time) is reached.
The results show the projected position of weather systems are manually corrected and passed onto end users (e.g., the public), for example, as a synoptic chart or television broadcast. Weather conditions are extensively observed.
This means weather forecasters have ample opportunities to test and improve equations and models. This has resulted in dramatic improvements in weather forecasting skill.
In short, then, weather forecasting is pretty good and getting better.
A key driver of improving forecasting skills is the ability of meteorologists to continually evaluate forecast accuracy using weather observations.
Weather forecasting models are only as good as the observed data, or starting conditions, fed into them. Poor initial weather data is the most significant cause of weather forecast error.
Weather stations are not generally located according to a representative sampling strategy. Similarly, all measuring instruments are of limited precision and accuracy. Computers also “round off” decimal places. Measurements can be taken on different scales to each other and model grids, meaning they require processing steps and quality control. This can introduce errors, including human error. The 20th century’s worst European windstorm (December 1999’s Cyclone Lothar) was poorly predicted, partially due to quality control errors.
In the British hills, the network of mountain weather stations is thin on the ground. Data is only taken from Cairngorm summit, Cairnwell, Aonach Mor, Bealach na Ba, and Great Dunn Fell.
The readings taken by these stations could be better due to their often-harsh mountain locations. For example, they return 100% relative humidity readings when instruments are rimmed in the winter. Snowfall accumulation measurements are generally considered unreliable due to the precipitation significantly varying across small areas, the impact of wind moving snow about, obstacles deflecting snow, and snow sticking to the side of rain gauges. Rain and snowfall data are only recorded at lowland, valley weather stations rather than mountain stations.
This lack of quality data from mountainous locations limits the ability of meteorologists to tweak and evaluate the success of forecasting models. To get around this, Met Office summit weather forecasts are updated hourly to incorporate the latest observational data. The Mountain Weather Information Service (MWIS) suggest using observations made by Scottish Avalanche Information Service and Fell Top Assessors in the Lake District to supplement weather forecasts with hard knowledge and images from experts on the ground.
Since the atmosphere is a chaotic system, it exhibits seemingly erratic behaviour. Meaning minute changes in one area/data point have magnified consequences in another.
Uncertainty increases with lead time (hence why forecasts ten days out are still far less accurate than those fewer than three days out). If imperfect data is fed into the model, the forecast is affected and potentially subject to a larger error.
Ensemble forecasts run models many times with slightly different starting conditions to rectify this. However, this needs vast computational resources and still relies on imperfect models. It is only effective if the entire ensemble of models that are run fully represent uncertainty in initial conditions.
Standard, deterministic forecasts tend to be preferred by mountain weather forecasters as ensemble models can overlook the complexity of mountain terrain.
Weather forecasting models are also only as good as the coding that makes them. Approximations must be made, often based on incomplete understandings of the incredibly complex atmosphere. This results in small errors in models, which can considerably impact the end result forecast. As it is impossible to represent the climate system mathematically, all models are imperfect.
Expertise at the Mountain Weather Information Service and Met Office allows forecasters to select specific models based on their suitability to different situations. They base this on historical experience with them and how they interact with mountain terrain in the forecast area.
On MeteoBlue, the Multimodel Tool compares different models in a specific location. We can be more certain about the predicted weather conditions where models tend to agree. The greater the difference between models, the greater the uncertainty.
Be careful when comparing different models, as it’s tempting to select the model showing the desired weather forecast, ignoring information that doesn’t suit your plans in other models. Scroll down on the Multimodel page to find information about each model, for example, coverage area and grid cell size.
To localise weather forecasts and for practical computing reasons, the atmosphere is split up into a series of horizontal and vertical grid cells. These are similar (but differently sized and 3D) to those you might see on an Ordnance Survey map.
Weather models are run in each cell to forecast the average weather within that cell. The smaller the cells, the better they will represent the atmosphere and underlying terrain.
Global models encompass the entire atmosphere, requiring large (20-30km) grid cells. Limited-area models cover more localised areas using smaller (up to 1km) cells. Smaller cells can be nestled within larger grids for complex areas (e.g., mountain ranges).
A large grid cell can significantly influence forecast accuracy. For example, three very distinct areas (Valley A; Valley B; Farmland) fall within one grid cell, so they receive a similar broad brush forecast. Each has different, localised weather that will not be fully compensated for during the adjustment phase.
Popular weather forecast Yr.no use high 2.5km resolution grid models for Norway, the Arctic Region and other Nordic countries. It uses the 10km EC-HRES ECMWF model for the rest of the world. This means that in mountainous areas or complex coastlines, many features (e.g, precipitation) are poorly resolved.
The Met Office use a very high 1.5km UK model (UKV) combined with a 10km resolution global model. This is enhanced by the chief forecaster using information from other models, weather observations, satellite images, and their vast experience. Multiple summits (e.g., Am Basteir and Bruach Na Frithe on Skye) can fall within one grid square even with this very fine grid. This means they have an almost identical forecast even though complex topography might make the experienced conditions on each quite different. Models can only partially understand complex topography, smoothing out pointy mountain terrain.
To avoid this problem, the Mountain Weather Information Service offer broad-brush detail across a large area rather than forecasting on highly localised specifics. By highlighting areas where particularly severe conditions are likely, it allows forecasting users to decide what to do and where to go.
If models have worked, wind speed and direction are pretty predictable in lowland terrain. Less predictable is how the wind will interact with complex mountainous terrain. This has knock-on effects on cloud cover and precipitation. Models alone often underestimate terrain, for instance, underestimating wind speeds over the Cairngorm plateau. This is why better mountain forecasts involve experienced human adjustment.
Local topography causes weather anomalies; for example, SSE winds tend to be stronger than forecast in the Northern Cairngorms. Other common highly-localised effects are wind enhancement over ridges, summits and plateaus. Forecast wind speeds, therefore, can only give an indication of likely conditions. This is because complex mountain terrain is (to varying degrees) smoothed out by weather forecasting models.
Cloud cover is tricky to predict on down-wind (lee) slopes due to small-scale variations in terrain and tiny changes in humidity. Too minor to be considered by a forecast model, slight terrain variations (e.g., cold air pooling – or heating up – in hollows, coires, and tight glens) and/or snow on the ground can influence the air temperature.
These small-scale variations in temperature can influence the point at which snow turns to rain. This means the snowline is not necessarily a consistent elevation (e.g., 800m) across a hillside or glen.
As nice as it is to blame a faceless weather forecast for unexpected weather, there is no getting around the fact that the error could be at the user’s end.
A perfect forecast only works if end-users understand it. As a result, significant amounts of thought have gone into communicating forecasts and making them user-friendly. Many Met Office and MWIS forecasters are experienced, regular mountain users themselves. This means they are well aware of the most essential forecast elements to mountain goers and the best language to communicate them. For example, MWIS tries to relate weather conditions to how it will feel on the mountain.
Some weather, such as thunderstorms or the exact location and precipitation volume, are still tricky to predict accurately. Both the Met Office and MWIS communicate via words estimated forecast uncertainty. The MeteoBlue “predictability” score quantifies uncertainty as a percentage. I have certainly interpreted an uncertain forecast to be gospel without fully understanding the levels of doubt shrouding the seemingly authoritative forecast.
There is not a clear-cut threshold beyond which a forecast is perceived to be “wrong” or “inaccurate.” Rather it depends on the perception of forecast users. For instance, a 50m error in cloud height in a very specific point (e.g., a Munro top) may be irrelevant in the overcast glen, while on the summits, the difference between an incredible view and a drizzly, claggy cairn.
Reading forecasts in a rush, misinterpreting pictograms or words used in the forecast or lack of reference points (e.g., actually knowing what the forecast 40mph winds will feel like in different mountain conditions) all might mean we think a highly accurate forecast was wrong. To start building an accurate reference point, there is no shortcut to getting out and experiencing lots of weather conditions (within reason!). This can then be calibrated with weather observations from nearby mountain weather stations or learning from the experience and knowledge of others (e.g., the Mountaineering Scotland article, which converts windspeeds to what they might feel like on the hill.
This article has scratched the surface of a complex and well-studied field. Hopefully, understanding immensely complicated weather forecasting, which is usually over 95% accurate less than three days out, might make readers pause before grumbling about how the weather was not quite as expected.
Hill goers should continue using all the suitable mountain weather forecasts, but be aware that they might be a little wide of the mark periodically. Carry the right equipment to keep you safe in all weather conditions!