Artificial intelligence in ecological mapping

Artificial Intelligence (AI) and Machine Learning are current buzzwords in the IT sector but what impact will they have on the world of ecology? At Thomson we have been leading the way with the use of AI and Machine Learning to identify habitats and features from aerial mapping and to offer our clients new ways of understanding data.

Using our state of the art GIS tools the Thomson GIS Team has produced high quality mapping and data analysis for a number of years. In doing so we have built up a huge amount of data that has been captured on the ground. As?humans, we can look at a GIS map and identify features or habitat types using our own knowledge and experience. But now we can use this data, coupled with our experience, to?‘teach’ our software too!?

For example, a few years back we undertook a survey to identify Hottentot fig plants in the South West of England. This non-native species is a particular issue in coastal and dune areas as it outcompetes native species. This project was undertaken using traditional survey methods on the ground and led to identification of the species on a traditional GIS map. Recently we pushed this data through our AI tools overlaid with aerial imagery from the same era. The AI/ Machine Learning tools (after a lot of processing time!) ‘learned’ what a Hottentot fig looks like from the air based on the imagery and as identified from the survey.?

We then ran the process against aerial imagery from another location and allowed the AI to attempt to identify the species. We compared the results to survey data we held for this area. It achieved a remarkably high success rate. We could then teach the learning model from this data set too to further improve accuracy – in fact the more it learns the better!

This offers up exciting potential of taking new aerial imagery (either commercial or flown by our own UAV pilots) to quickly identify invasive species over a far wider area than with traditional methods or where access may be limited with a very high degree of accuracy.?

Our next step is to teach our software to identify other invasive species such as Japanese knotweed. We are also trialling its ability to learn what Ash dieback looks like, again from previous ground surveys matched to UAV imagery of the same sites – this could be particularly useful when coupled with our innovative work using near infrared aerial imagery.

This though is just the start of our journey with AI and ecology data. We have many years’ worth of detailed surveys across the UK and are looking to use AI in a number of areas. For example, to build a predictive model based on this data, coupled with imagery, and other data such as temperature, soils, aspect etc to potentially identify habitats, biodiversity and the likelihood of species being present on a site from a current aerial image. Don’t worry - we’re not suggesting this would replace boots on the ground - but it could potentially offer clients an ‘early warning’ system when they are considering sites or allow land owners to understand and manage their land holdings much better.

From our ecologists’ point of view it could help them focus on the most likely places a particular species may be present or better still pre-capture habitat areas before they go on site, speeding up the habitat survey process as a whole.

It’s still early days, but whatever happens it is clear AI is going to have a major impact on the way we work and there are some very exciting times ahead at Thomson as we develop this technology.

Find out more at www.thomsonec.com