Tim’s take:
A smartphone in the pocket has quietly become the cheapest bit of kit on the farm, and the latest pitch is that it can tell a yellowhammer from a whitethroat before the hedge ends. With farmland bird numbers down 62% since 1970 and an 11% slide in just the last five years, the question is no longer whether there is a problem worth listening to, but whether growers and graziers can afford to ignore it.
Apps like BirdNET, Merlin and ChirpOMatic work well enough on a quiet evening, but the practical bit is what they unlock. Recording even a few minutes in a margin or beetle bank feeds straight into evidence for SFI actions such as AHL1 and IPM2, and that is where the licence trap hides. Most free tiers are fine for personal use, yet anything submitted for payment or published monitoring has to sit under the right licence, and farm advisers do not always flag it. The article is right that the habitat pays, but only if the paperwork holds.
Pick one field margin, run the app on it twice a month through May and June, and log what shows up. That single habit gives a baseline a buyer, an inspector or a future SFI agreement will all recognise, and it costs nothing but ten minutes and a battery.
A bird song identifier app names what is singing in seconds. The tools that work on farm, the licence catch nobody mentions, and what the habitat pays.
The figures that make this worth ten minutes
Defra’s accredited statistics, last updated on 23 September 2025 with data to 2024, put farmland bird populations down 62% since 1970. “Farmland bird populations have declined by 62% since 1970, with turtle dove and tree sparrow showing the steepest declines,” the release says. The number people miss is the short one: an 11% fall in the five years since 2019. That is not a slow historic slide anyone can file under someone else’s problem. It is happening on working ground now.
The indicator covers 19 species, split into 12 farmland specialists and seven generalists. Three quarters of the specialists have declined both since 1970 and in the most recent five-year window.
Whether that decline is mostly about cropping, chemistry, hedgerow management, winter stubble, predation or milder winters is argued over properly by people with datasets. What is not argued over is that most farmers cannot say with any confidence which of those 19 species are actually on their land. You cannot manage what you have never counted.
Why bother naming the bird at all
Three practical reasons, and none of them are sentimental.
First, the money. Several of the actions in the current scheme round pay for habitat aimed squarely at these species, and knowing whether you already have skylarks nesting changes which action is worth taking. Second, the census. The Game and Wildlife Conservation Trust runs the Big Farmland Bird Count each February, and it has functioned as a national farmland bird census since 2014. Thirty minutes on one day over about two hectares is the whole commitment. Third, the baseline. Supply chains have started asking. More on that further down.
The 2026 count ran from 6 to 22 February. GWCT’s results, published on 26 May 2026, logged 275,024 birds across 292,828 hectares from more than 1,200 submissions. Red-listed species accounted for 69,222 of those birds across 34 species, with starling (25,306), lapwing (19,198) and fieldfare (15,816) the most abundant. Suffolk sent the most counts at 139, then Norfolk at 101 and Lincolnshire at 62.
Two figures from that count are worth sitting with: 67% of the farms surveyed were in an agri-environment scheme and 64% were providing supplementary feeding. That is a self-selecting sample of the already-engaged, so it flatters the national picture. It also means the counts are coming from farms that are trying, which is the useful comparison for anyone deciding whether to bother.
Merlin: the free app that names the song
The tool most farmers will actually use is Merlin Bird ID, built by the Cornell Lab of Ornithology and free. The Play Store listing shows over ten million downloads and a 16 July 2026 update. You hold up the phone, it listens, and it names what is singing, in real time, with the species highlighted as each one calls.
It has four identification modes, of which sound is one. The recognition is trained on photos and sound recordings collected through eBird and archived in Cornell’s Macaulay Library. You download a regional bird pack, Europe in our case, rather than the whole world.
Cornell funds it through membership and donations, which is why there is no subscription. The bird-pack download suggests it works away from signal once the pack is on the phone, and that matches how people describe using it in the field, but I could not load Cornell’s own support pages to verify the offline behaviour in writing. Test it yourself in a dead-signal corner before you rely on it for a count.
When you want the recording to run without you
A phone in your hand covers the half hour you are standing in the field. It does not cover dawn, which is when most of the singing happens, and it does not cover the twelve weeks of a breeding season.
That job belongs to BirdNET, a collaboration between Cornell’s K. Lisa Yang Center for Conservation Bioacoustics and Chemnitz University of Technology. It recognises over 6,000 species worldwide, chews through audio in three-second segments, and is the engine behind most of the serious kit. Its current model version is 2.4.
Two tools sit on top of it and are worth knowing apart. birda is a command-line tool written in Rust that runs BirdNET and Google’s Perch models on audio files you have already recorded, entirely offline once the models are installed. Point it at a night’s recording and it returns what it heard. birdnet-go, by the same author, Tomi P. Hakala, is the always-on version: it takes a live soundcard or network audio feed and classifies continuously, around the clock.
Neither is a consumer product. Both assume you are comfortable installing software and reading a config file. If that is a barrier, a cheap recorder plus a friend who likes computers gets you the same result.
The licence catch that matters to a farm business
This part gets skipped in most write-ups and it is the part that could bite.
The BirdNET source code is MIT licensed, which is as permissive as it gets. The trained models are not. They are released under Creative Commons BY-NC-SA 4.0, which is a non-commercial licence, and the project’s own wording is that educational and research use counts as non-commercial and is freely permitted. Commercial use of the models is a different question, and a farm business is a business.
The same split runs through the tools. birda’s own code is MIT. birdnet-go is CC BY-NC-SA 4.0 outright. Merlin is a free consumer app from a non-profit, with its own terms.
For a farmer walking their own fields to see what is there, none of this is likely to trouble anyone. For a contractor selling bird surveys, or a business publishing biodiversity claims to a buyer off the back of these models, it is worth an email to the BirdNET team before you build a service on it. I am flagging the licence, not giving legal advice.
What the habitat is actually worth: three SFI26 actions
Now the money, and this is where reading last year’s article will get you the wrong number.
The Sustainable Farming Incentive stopped accepting new applications on 11 March 2025. “With record numbers of farm businesses in farming schemes and the sustainable farming budget successfully allocated, we will stop accepting new applications for SFI from today,” Defra said at the time. Its successor, SFI26, opened Window 1 on 30 June 2026, restricted to two groups: farms of 50 hectares or under, and farms without an existing ELM revenue agreement. Window 1 carries a £60 million budget, was expected to run about two months, and can close early once that budget is allocated. Window 2 opens in September 2026 for everyone else. There is a cap of one agreement per SBI and £100,000 per agreement year, with a three-hectare minimum.
Defra’s most recent public statement on Window 1 was on 13 July 2026, when 75% of the funding had been allocated. Nothing has been published since, so check the position before you plan around it. Note also that gov.uk’s general funding-for-farmers page still carried the line “SFI is closed for applications” when I read it on 2 August 2026, in the same section that describes Window 1 opening. Trust the scheme pages and the blog over that summary.
The three actions that pay for the birds, all verified on their own gov.uk pages on 2 August 2026:
- CAHL2, winter bird food on arable and horticultural land: £648 per hectare per year, three-year duration.
- AHW2, supplementary winter bird food: £732 per tonne per year, capped at one tonne for every two hectares of winter bird food you are already growing.
- AHW4, skylark plots: £11 per plot per year, minimum two plots per hectare.
SFI26 runs 71 actions and keeps the same action codes as the 2024 version, which is genuinely helpful if you have old notes. There is a 25% cap on how much of your area can go into limited-area actions. Higher Tier Countryside Stewardship is the other route: it opened in September 2025, and from 21 July 2026 you can register an interest in an agreement.
Skylark plots: £11 a plot, and whether they work
Eleven pounds a plot sounds like nothing until you count them up. A plot is an undrilled patch inside a winter cereal field, or one sprayed out after the crop establishes. At the two-per-hectare minimum, a 20-hectare field carries 40 plots, which is £440 a year for ground you were going to drill anyway and now will not.
The reasoning is specific to how skylarks nest. RSPB’s advice for farmers puts the UK skylark population down 63% since 1967 and explains that skylarks nest mid-field and cannot be helped by anything done within ten metres of the boundary. They want vegetation 20 to 50 centimetres high, and they need two or three nesting attempts between April and August. Modern winter wheat grows too tall and too thick to allow more than a single brood. The plots are gaps in that canopy. Chicks eat nothing but insects and spiders for their first week, so the gaps have to be reachable as well as open.
Does it work? The GWCT’s SAFFIE trial, a five-year project across 15 farms that reported in 2007, found skylark densities between 1.3 and 2.8 times higher on fields with skylark plots and wildflower margins than on fields with neither. The same work found 71% fewer skylark food items 32 metres into a standard field than at one metre from the crop edge, which is the mechanism in one number. That trial tested plots and margins together, so it does not isolate the plots on their own, and it is nearly twenty years old now.
One rule in the AHW4 guidance catches people out: you cannot put plots on land bounded by tree lines or next to woods unless the parcel is over ten hectares, because it raises the risk of predator attack. Check your field boundaries before you count up your plots.
Who is already recording, and why buyers are asking
Acoustic monitoring on farms has moved past the hobbyist stage, which is the part that should interest anyone thinking about where supply-chain requirements go next.
The UK Centre for Ecology and Hydrology has run automated monitoring stations combining cameras and acoustic recorders with AI identification, deployed at four pairs of farms in Dorset, Wiltshire, Oxfordshire and Northamptonshire, plus two peatland sites in Cambridgeshire. The pairing is deliberate: each pair compares different levels of agri-environment scheme uptake. Stations run March to October across 2024, 2025 and 2026, funded through NERC and BBSRC. The stated purpose is to demonstrate whether these schemes actually work, which means the evidence base underneath future payment rates is being built with microphones.
On the commercial side, Chirrup.ai, a UK company registered in England and Wales, describes a trial with Pilgrim’s Europe across 40 farms: ten broiler units in Lincolnshire and 30 pig farms in Norfolk, with recorders out for 21 days to establish a biodiversity baseline across the supply chain. That account comes from the company’s own website rather than published research, so treat it as what a vendor says about its own work. The direction of travel is still worth noting.
Whether a recording from a phone app would ever be accepted as evidence for a scheme inspection is a different question, and as far as I can find, nobody has answered it. The apps are for your own knowledge, not your paperwork.
What to do between now and February
If you want to do one thing: put Merlin on your phone, download the Europe pack, and spend twenty minutes at dawn somewhere in the middle of a field rather than along a hedge. Skylarks nest mid-field, out of reach of anything you do to a boundary, and the middle of a field is where you are least likely to have stood and listened.
If you want to do the useful thing: repeat that walk on the same route two or three times before February, write down what you get, then submit a Big Farmland Bird Count when the window opens. Your own repeat counts are worth more to you than the national total, because they tell you whether what you changed made any difference.
If the payments are the point, work out now whether you fall into a Window 2 application in September, read the three action codes above against fields you would actually commit, and check the current budget position before you build a plan on it. The full SFI26 action list and payment rates and our guide to the environmental schemes cover the wider picture, and the Countryside Stewardship guide covers the Higher Tier route.
The 2027 count dates have not been announced. February is the usual month.
Frequently Asked Questions
What is the best bird song identifier app for a UK farm?
Merlin Bird ID from the Cornell Lab of Ornithology is free and identifies birds from their song in real time. Download the Europe bird pack before you go out. For unattended recording across a whole dawn chorus or a full season, BirdNET-based tools such as birda or birdnet-go do the job instead, though both need some technical setup.
Are bird song identification apps accurate enough to rely on?
BirdNET, which powers most of these tools, recognises over 6,000 species and is a research project run by Cornell’s K. Lisa Yang Center for Conservation Bioacoustics with Chemnitz University of Technology. It is good enough for knowing what is on your land. Whether any app recording would be accepted as evidence in a scheme inspection has not been established, so keep them separate from your scheme paperwork.
How much do skylark plots pay in 2026?
Action AHW4 under SFI26 pays £11 per plot per year with a minimum of two plots per hectare, verified on gov.uk on 2 August 2026. You cannot use land bounded by tree lines or adjacent to woods unless the parcel is larger than ten hectares, because of predation risk.
Is SFI open for applications right now?
SFI 2024 closed to new applications on 11 March 2025. SFI26 Window 1 opened on 30 June 2026 for farms of 50 hectares or under and farms without an existing ELM revenue agreement, with a £60 million budget. Defra’s last published update, on 13 July 2026, said 75% of that budget had been allocated. Window 2 is due in September 2026 for everyone else. Check the current position before planning, as gov.uk’s summary pages have carried contradictory wording.
When is the Big Farmland Bird Count 2027?
Dates have not been published. The GWCT has run the count in February every year since 2014, and the 2026 count ran from 6 to 22 February, so expect a similar window.
How far have farmland birds actually declined?
Defra’s accredited statistics, updated on 23 September 2025 with data to 2024, put farmland bird populations down 62% since 1970 and down 11% in the five years since 2019. The indicator covers 19 species, and three quarters of the 12 farmland specialists have declined over both periods.
BritFarmers Weekly
One email a week: what’s moved on schemes, prices, disease control and policy, with links to primary sources. Free. No sponsors. Unsubscribe anytime. Subscribe.
Source: Defra: Wild bird populations in the UK and England, 1970 to 2024
