The value of probability weather forecasts for UK farms

Many farm operations depend on short-term weather conditions: spraying, drilling, harvesting, and more. When the timing is wrong, the cost can be high. For example, in September 2024, our client Rockscape PLC carried out a pre-emergence herbicide spray on a rye crop:

• The forecast: ~6 mm of rain
• What happened: >20 mm of rain
• Outcome: spray wash-off, an estimated £11,000 loss and risk of pollution

This type of loss is not unusual. Most farmers plan work using weather apps designed for the general public. These typically show a single “best-guess” value for each weather variable. While simple, this approach does not show the chance of disruptive weather, such as heavy rain, that can cause expensive failures.

In reality, the weather models behind these forecasts contain a range of possible outcomes. The most damaging weather is often not the most likely outcome shown, but a lower-probability event that still occurs often enough to matter operationally.

Probcast by CatchmentAI is designed to make this risk visible in its forecasts, and support farm decisions by highlighting lowest-risk windows for specific operations. This summary explains the findings of our study designed to understand the value of giving farmers visibility on weather risk. This study asks a simple question:

How many costly weather-related mistakes can be avoided if farmers can see potential weather outcomes of 10% chance, not just a best-guess forecast?

We focused on the common issue of spraying, where rainfall exceeding 10 mm in 24 hours generally leads to spray wash-off and the need for a full re-spray.

Using published UK rainfall forecast performance results, Rockscape’s real spraying costs and practices, and conservative assumptions about how farmers plan work, we compared two approaches:

• 1. Conventional forecasting – acting on a single best-guess forecast
• 2. Probcast alerts – avoiding windows where there is a 10% risk of > 10 mm in 24 hours

For heavy rainfall events that cause spray failure (a threshold of >10 mm in 24 hours):

• Traditional forecasts warn that weather will exceed the threshold just 22% of the time.
• Probcast-style 10% risk alerts warn of that threshold in advance 77% of the time.

Therefore, based on that difference:

Probability forecasts identify >70% of the heavy rain missed by traditional forecasts.

This does not mean the forecast is “70% more accurate”. It means a farmer using Probcast is able to make around 70% fewer costly weather-related mistakes, because risky windows are visible before operations take place. This does lead to more warnings. But for spraying, changing the timing is usually feasible and cheap, whereas getting it wrong is expensive.

Using Rockscape’s operations and weather station figures:

• 7 spray operations per year
• £80 per hectare per spray
• Damaging rainfall occurs on 10% of days in spraying season

Relative exposure to bad weather leads to different cost scenarios:

• Expected costs if farmer uses no weather info: £56/ha/year (due to spray wash-off)
• Expected costs if farmer uses conventional forecasts: £43/ha/year (22% reduction)
• Expected costs if farmers plans using Probcast alerts: £13/ha/year (77% reduction)

This gives an expected saving of approximately:

~£30/ha/year from mitigated spray wash-off using Probcast alongside traditional forecasts

This is an expected average over time, not a guarantee in any single season, and is consistent with the size of losses seen in real spray failures validated using Rockscape’s operations data.

Spraying is only one weather-sensitive operation. Rockscape estimates that across spraying, drilling, harvesting and other fieldwork, weather-related disruption costs around £125,000 per year.

If Probcast reduces weather-related mistakes by a similar amount across these activities, on average, where ~70% of that cost becomes avoidable:

~£100,000 per year in expected savings (up to £100/ha/year)

• Most rain above a key farming threshold (10 mm) is missed by standard forecasts
• Showing risk, through outcomes at the 10% probability, improves farmer visibility of what’s coming
• High value comes from avoiding expensive mistakes with risk information, not by predicting the weather perfectly

Despite this expected value, there has been a long-standing assumption that farmers are not able to use probability forecasts effectively for daily operations. We have seen first hand on our Probcast platform that farmers find this information intuitive and useful, but farmer engagement and decision quality remains an uncertainty for real-world value. We are testing with early adopter farms in 2026 to understand to what extent improved information and decision support tools generate improved decisions and subsequent reductions in farm costs.

This analysis builds on established research evaluating how well UK rainfall forecasts identify heavy-rain events, combined with Rockscape’s real costs and conservative assumptions about farm operations. A full technical description of the method, assumptions and references is available in the accompanying detailed analysis document.

• Robbins and Titley (2018). Use of ROC curves and AUC for evaluating precipitation forecasts. Meteorological Applications. https://rmets.onlinelibrary.wiley.com/doi/10.1002/met.1720
• RMS (2023). How to interpret an ensemble forecast. Royal Meteorological Society. https://www.rmets.org/metmatters/how-interpret-ensemble-forecast
• Met Office Data Hub. Deterministic forecasting products. https://datahub.metoffice.gov.uk/mo-bpf-deterministic
• Wilks, D. S. (2011). Statistical Methods in the Atmospheric Sciences. Academic Press.
• Zhu, Y., Toth, Z., Wobus, R., Richardson, D., & Mylne, K. (2002). The economic value of ensemble-based weather forecasts. Bulletin of the American Meteorological Society. https://journals.ametsoc.org/view/journals/bams/83/1/1520-0477_2002_083_0073_tevoeb_2_3_co_2.xml