Why Met Office UKV is IMBER’s UK model of choice

IMBER’s main UK forecasts are powered by the Met Office UKV 2km Deterministic Weather Forecast. Our choice of this model is for multiple reasons: it is the highest resolution model covering the British Isles, meaning it accounts for the UK’s complex topography, accurately predicting microclimates like enhanced rainfall over mountain ranges and fog filling steep valleys. It also has convection-permitting scale, explicitly simulating heavy rainfall, sudden snowstorms and convective showers.

Attribute	UKV Model (Met Office)	Global Models (GFS / ECMWF)
Grid Resolution	1.5 km native / 2 km published data — highly detailed regional modelling.	~9–13 km — broader geographical coverage with less local detail.
Terrain Awareness	High; better representation of hills, valleys, coastlines and local terrain.	Moderate; captures major terrain features but smooths smaller geographical variations.
Storm Tracking	Convection-permitting; can simulate individual showers and storm structures in greater detail.	Represents larger-scale weather systems well, but smaller convective processes require more approximation.
Best Used For	Detailed short-range UK forecasting (0–48 hours), with some forecasts extending to 5 days.	Short-to-long-range forecasting, particularly broader weather patterns and trends over 3–10+ days.
Reason Used by IMBER	High-resolution UK forecasts supporting detailed local weather predictions.	Manual forecasts, cross-model comparison.

Every decision made by IMBER has scientific reasoning behind it. Met Office UKV is not perfect in every respect, which is why you can select between models to see which works for you. With time, we will show the confidence of each model in these forecasts. We believe using models like UKV directly provides forecasts that are better in accuracy than commercial blended forecasts.

It is also worth noting that all numerical weather models have limitations, which we understand and account for when writing manual forecasts. These limitations are amplified at certain times of year, like winter, when conditions are often borderline between rain and snow, for example.

Limitation	Explanation	Impact on Forecast Accuracy
Incomplete observations	Models cannot measure the precise atmospheric conditions at every location and altitude.	Small errors in initial conditions can grow over time.
Atmospheric chaos	The atmosphere is a chaotic system, highly sensitive to small changes.	Forecast uncertainty generally increases further into the future.
Limited resolution	Models divide the atmosphere into three-dimensional grid cells rather than simulating every physical process.	Small-scale features such as local showers, fog and wind variations may be missed.
Simplified physics	Processes such as cloud formation, turbulence and radiation must be represented using mathematical approximations.	Can introduce systematic biases in temperature, precipitation and cloud forecasts.
Local geographical effects	Even high-resolution models cannot perfectly represent buildings, vegetation, slopes and coastlines.	Urban temperatures, coastal breezes and local wind conditions may be inaccurate.
Precipitation uncertainty	Rainfall formation and movement depend on complex interactions between temperature, humidity and atmospheric motion.	A model may predict rainfall correctly but misplace its location, timing or intensity.
Extreme weather	Rare or rapidly developing weather events can be particularly difficult to forecast precisely.	Storm intensity, snowfall accumulation and local flooding risk may be underestimated or overestimated.
Computational constraints	Greater resolution and more sophisticated simulations require significant processing power.	Models must balance geographical detail, forecast length, update frequency and computational cost.
Model biases	Every model has assumptions and systematic errors that may vary by season, region and weather situation.	One model may consistently overestimate or underestimate particular variables.
Data latency	Forecasts require processing, distribution and publication after the model’s initialisation time.	Published forecasts may not incorporate the most recent atmospheric developments.

We are particularly interested in how these issues can be overcome. One method of improving numerical weather models could be implementing artificial intelligence, which can analyse past predictions and the subsequent reality, and adjust for this in the forecasts. We will continue to research these improvements and implement them once technically feasable.

Benjamin Raisanen

Founder and Director, IMBER