How Citizen Science Apps Are Changing the Way We Map Old Orchards

I started using iNaturalist in 2019 because I was losing track of my records about which orchard I had photographed which tree in. I had thousands of photos on my phone and a spreadsheet that I dreaded opening.

 

A colleague at the Somerset Wildlife Trust mentioned iNaturalist to me in passing. She used it for butterflies. Out of curiosity, I uploaded my first apple tree that evening. The platform suggested an identification within minutes. That was the beginning of a shift in how I worked.

 

What Problem Was I Trying to Solve?

Heritage orchards in Britain have been disappearing slowly. Many of the varieties were farmed for specific local conditions, flavours, and uses. My duty is to find these trees before they disappear. I identify them, record their locations, and build a lasting database. Doing this alone is impossible.

 

How Technology Changes Mathematics

Citizen science platforms solve the scale problem in a way nothing else does. iNaturalist now hosts over 100 million observations worldwide. Users upload images, tag locations, and the community helps with identification.

 

The community is a mix of naturalists, botanists, and taxonomists logging observations. This is genuinely transformative for orchard work. I can train a volunteer in an afternoon to photograph trees correctly. They learn how to record GPS coordinates, get a usable image of bark, leaf and fruit, and upload observations to the platform. Their work immediately becomes part of a permanent, searchable, geographically mapped database.

 

I have trained twelve volunteers across Somerset and Dorset. Between us, we have logged over 4,000 individual trees in three years. Working alone, that number would have taken me closer to fifteen.

 

Testing Whether the Identification Actually Works

This is where it gets interesting, and where I had to do some proper testing before I trusted the results. iNaturalist uses a computer vision model trained on millions of images. It performs surprisingly well against common varieties.

 

I tested it against my reference collection, which I had confirmed through DNA testing and expert verification. It correctly identified Cox's Orange Pippin, Bramley, and Conference pear over 80% of the time from clear fruit photographs.

 

Heritage varieties like Egremont Russet, one of the oldest documented in Britain, regularly come back as generic "Malus domestica." It gets confused with other russets. The model has not seen sufficient examples to reliably distinguish it.

 

The Workflow I Actually Use

I do not rely on automated identification alone. Every observation I log includes multiple photographs:

• Bark pattern
• Leaf shape
• Fruit profile
• Overall tree form

 

iNaturalist accepts up to 20 images per observation. I typically use four to six, depending on what is useful. The community verification layer is also where a lot of the real value sits. Other users can confirm or correct identifications.

 

Where iNaturalist Has Setbacks

The platform has a hard limitation for this type of work. It is designed for species identification and not cultivar detail. For example, it will tell you "Malus domestica" but not whether you are looking at a Blenheim Orange or a Beauty of Bath. For heritage work, that distinction is everything.

 

For that level of detail, I use a custom database I built in Airtable. The database links to iNaturalist observations via unique identifiers. When I download data from the platform, I import the coordinates and images into my system. Afterwards, I add the cultivar information, rootstock type where known, and condition assessments.

 

The Case That Made It Real

In 2022, a developer submitted planning permission to convert a site near Ilminster into industrial units. The documents referred to "a few old fruit trees." I knew the site from the iNaturalist observations volunteers had logged. Of the seventeen trees logged, three appeared to be Devonshire Quarrenden. This variety of apple was reported to be nearly extinct in commercial cultivation.

 

I captured the coordinates, then visited with members of a local pomology society. We proceeded to collect scion wood for grafting before permission was approved. Without the crowdsourced observations sitting in the database, I would not have known those trees existed until it was far too late.

 

Where This Is Heading

The technology is improving noticeably. I am currently testing whether the platform can flag disease from leaf photographs. Cedar apple rust, canker, and scab all have distinctive visual signatures. Early results suggest the model can identify "unhealthy leaf" with reasonable reliability. What this could eventually mean is large-scale orchard health monitoring via volunteer observations.