AI Disease Risk Forecasts and Fungicide Timing

A septoria risk model will not tell you whether your wheat needs a fungicide. It tells you when the weather has given the disease a chance to infect, and roughly when those infections will show. Used that way, crop disease risk forecasting is one of the more useful digital tools on an arable farm. Used as a reason to stay in the pickup, it will cost you money in exactly the seasons when you can least afford it.

This piece covers how the main weather-driven models for septoria, rust and potato blight have behaved in recent UK seasons, where each one is reliable, and a practical routine for using them to tighten T1 and T2 timing.

What the models actually calculate

Most commercial disease risk tools run a version of the same idea. They take hourly weather (temperature, rainfall, relative humidity, leaf wetness), apply rules about the conditions a pathogen needs, and give you a risk score or an infection date.

For Zymoseptoria tritici (septoria tritici blotch), the key inputs are rain-splash events and long spells of leaf wetness. After an infection event the fungus sits latent in the leaf for roughly 250 to 300 degree-days before lesions appear. In April that can mean three to four weeks of nothing visible, which is the whole problem: the leaf looks clean while it is already infected.

Potato blight runs on fixed criteria. The Hutton Criteria, which replaced the older Smith Period in GB alerts from 2017, flag a risk when two consecutive days each have a minimum temperature of 10°C or above and at least six hours of relative humidity at 90% or higher. BlightWatch sends these alerts by postcode district, free.

Rust is different, and it matters. Yellow rust epidemics depend more on how mild the winter was, which races are about, and the resistance rating of your variety than on any single week of weather. Brown rust likes warm days in late May and June. Weather models for rusts are much weaker predictors than the septoria and blight models, and most agronomists treat them that way.

The tools UK farmers are using

Prices move every year and dealers bundle hardware, so treat that column as a guide for budgeting, not a quote.

The difference between a station on your own farm and modelled weather is bigger than the sales literature suggests. Leaf wetness is very local. A field in a valley bottom by a river can sit wet for six hours longer than a field on a ridge two miles away. If your farm has that kind of variation, one station on the wettest block earns its keep.

How the models performed in recent seasons

Here is how septoria risk models have broadly lined up with what was found in the field, from AHDB disease monitoring and agronomist reports:

2022. A dry spring across most of England. Models showed low septoria risk through April and May, and they were right. Farms that cut T1 rates or held off on lower-risk varieties generally lost nothing. Yellow rust still turned up in susceptible varieties in the east, which no weather model flagged in time. Crop walking caught it.

2023. Wet spells through April and May. The models flagged repeated infection events, and septoria came through on leaf 3 and leaf 2 in susceptible varieties where T1 had been stretched. In the same summer, blight Hutton alerts came thick and fast from late June, and growers who tightened intervals in response did better.

2024. After a very wet autumn and winter, a wet spring drove high septoria pressure almost everywhere. Every model showed high risk for weeks on end. That is a useful lesson in itself: when the risk sits at “high” for six weeks running, the model stops helping you choose timing and only tells you that full protection is needed. Timing discipline still mattered, but the model could not separate one week from the next.

2025. A very dry spring in much of England. Septoria risk ran low for long stretches, and models gave growers enough confidence to use lower T1 doses and, on some resistant varieties, to drop T0. Rust in susceptible varieties remained the thing to watch.

Across these years, the septoria models were good at the easy calls (dry spring, wet spring) and most useful in the middle: a mixed April where one wet week sits between two dry ones. That is where a precise infection date lets you place a spray properly.

A worked example: 180 ha of winter wheat

Take a farm in Shropshire with 180 ha of winter wheat in two varieties: 100 ha with a septoria rating of 6.5 on the AHDB Recommended List, and 80 ha of a more susceptible variety rated 5.

The farm has one Sencrop station on the lower ground. In a mixed spring, the station log for the susceptible block looks something like this:

Date      Rain(mm)  Leaf wet(h)  Septoria risk  Growth stage
12 Apr    0.4       3            Low            GS30
15 Apr    8.2       14           Infection      GS31
16 Apr    3.1       11           Infection      GS31
22 Apr    0.0       1            Low            GS32 (leaf 3 emerging)
27 Apr    0.0       2            Low            GS32 (leaf 3 fully out)
04 May    11.6      19           Infection      GS33

The rain on 15 and 16 April landed before leaf 3 was out, so those infections are mostly on leaves 4 and 5. Leaf 3 is the target for T1, and it was fully emerged by 27 April. The next infection event is forecast for 4 May.

The useful decision: apply T1 between 27 April and 2 May, so it protects leaf 3 before the 4 May rain. Without the station you might have sprayed on a calendar date of 20 April, before leaf 3 was out, and left it unprotected through the May infection.

Now the numbers. A reasonable T1 product and partner costs perhaps £30 to £40/ha, and a spray pass costs around £13 to £15/ha at typical NAAC contractor rates. Getting T1 timing wrong on the susceptible block and needing an extra pass, or losing yield from a poorly protected leaf 3, can easily cost 0.3 to 0.5 t/ha. At £180/t that is £54 to £90/ha. Across 80 ha that comes to £4,300 to £7,200, set against a station that costs a few hundred pounds a year to run.

Then the crop walk. Before the T1 goes on, someone walks the susceptible block, pulls 20 tillers, and checks leaf 4 and below for septoria lesions and pycnidia (the small black dots). They also look over the flag and upper leaves for yellow rust stripes. If they find yellow rust at GS32 in the variety rated 5, the product choice changes. The septoria model said nothing about that, and could not have.

Why the walk is not optional

You do not stop crop walking because you have a model. The model narrows the window. The walk tells you what is in the crop.

There are five things a weather model cannot see:

• Rust. Discussed above. Yellow rust can build in a mild spell that no septoria model scores as risky.
• Actual inoculum. A crop after a clean break with little trash behaves differently from a second wheat sat next to last year’s heavily infected field.
• Your variety’s real performance. Resistance ratings shift as pathogen populations change. A variety rated 6 five years ago may not behave like a 6 now.
• Growth stage errors. Most tools estimate growth stage from drilling date and temperature, and they can be a week out. If leaf 3 is not yet fully emerged when the app says it is, your T1 lands on the wrong leaf. Dissect a few main stems.
• Things that aren’t disease. Manganese deficiency, herbicide scorch and physiological spotting all get mistaken for disease. Spraying fungicide on them wastes money.

For the bigger picture of how satellite imagery and in-field sensing fit alongside these models, including variable-rate fungicide using biomass maps, see our guide to satellite, drone and variable-rate agronomy.

Potatoes: Hutton alerts are good but blunt

BlightWatch alerts by postcode district are a sound early warning. Where they fall short is resolution. One district can cover irrigated fields and dry ones, and irrigation changes the microclimate completely. Growers with on-farm stations running Hutton Criteria in FieldClimate can see which blocks met the threshold rather than relying on the district average.

The practical use: when a Hutton period is forecast, close the spray interval on the most exposed blocks and leave the drier, well-ventilated fields on their normal schedule. Inspect for lesions daily during sustained runs of alerts. Watch especially near dumps and volunteer potatoes, which a forecast cannot account for.

Setting this up for next spring

A simple routine that works on most farms:

1. Before March: Check each variety’s disease ratings on the AHDB Recommended List. Group fields by septoria and yellow rust risk.
2. Place sensors properly: If you buy a station, put it in the wettest block or the one with the most susceptible variety, not by the yard.
3. Log every decision: For each spray, record the model risk, growth stage confirmed by dissection, and what the crop walk found. That paper trail is useful for your agronomist and for any IPM plan you hold under a stewardship agreement.
4. Agree thresholds with your agronomist: For example, “If no infection event since leaf 3 emerged and the variety is rated 6.5 or above, T1 can go at a reduced rate.” Write it down before the season, not in a rush on a wet Tuesday.
5. Review in July: Compare the model’s infection dates with when lesions actually appeared. After two seasons you will know how far to trust your own data.

The farms getting value from these tools are not spraying less across the board. They are spraying at the right time on the right leaf, and they know when a crop walk has caught something the model missed. A good season with a model looks like a field that was walked on 27 April and sprayed on 29 April, ahead of the rain due on 4 May.