Exploring AI’s capabilities to help manage the health of UK forests

According to 2024 Forest Research Woodland Statistics, woodlands are estimated to cover 13% of the UK land area (equating to 3.25 million hectares). These woodlands are overseen by forest or woodland managers who inspect them for pests and diseases as part of their forest maintenance responsibilities. However, this task is time consuming and intrinsically dangerous, from traversing the dense forest floor via foot, to safety risks from tree climbing if a sample is required for lab testing. 

AI-enabled autonomous systems have the potential to transform forest management but only if they can do so safely and efficiently. One application for this technology is identifying infections and infestations in a timely manner for effective treatment that reduces spread and minimises forest destruction.

Introducing ASPEN: Autonomous Systems for Forest Protection

ASPEN: Autonomous Systems for Forest Protection was a multidisciplinary 12-month UKRI Trustworthy Autonomy Systems (TAS) ‘Pump Priming’ programme which ran from 2023 to 2024. The funding from this programme is awarded to multidisciplinary research projects in trustworthy autonomous systems. The aim of ASPEN was to investigate the benefits and practicalities of utilising AI and autonomous systems for effective forest health maintenance. In addition, the project also explored where autonomy can improve safety by removing the need for the physical involvement of forest managers in dangerous scenarios.

The team of Assuring Autonomy International Programme (now the Centre for Assuring Autonomy) researchers Professor Radu Calinescu, Dr Vicky Hodge, Dr Calum Imrie, Dr Paulina Lewinska and Dr Colin Paterson focused on exploring autonomous systems for pest and disease detection, along with researchers from Bangor University, the University of Nottingham, and the Open University.

From lab floor to forest floor

From our initial investigation into this challenge, we discovered that there is a wealth of research and technology available for forest inspections but little consideration for how these findings would actually be realised in application.

One example of this is the use of machine learning (ML) to process hyperspectral sensor data. Hyperspectral sensors capture hundreds of bands of light invisible to the human eye and as such can give an indication if a tree has early stress, which could be caused by a pest or disease. Different bands of light may provide different information depending on the time of year and/or the specific disease, which means it isn’t clear which bands are most useful for the ML to exploit. While research has already indicated that filtering out non-informative bands would improve the ML’s accuracy, even the experts are uncertain how best to do this. 

To advance this important research area, we codified a five-step framework for the forest management process:

• Resist: how can the forest be structured to be naturally resistant to pests and diseases?
• Detect: what are the best methods of identifying forest stressors?
• Test: conducting tests on trees suspected of being infected.
• Treat: removing the pests and diseases, typically by felling and/or burning. 
• Adapt: what can be done to this process as a whole given what we have learnt from the other steps?

Our research provided insight into how autonomous systems can be integrated safely into each of these five steps providing a significant first step into understanding the implications of bringing AI and robotics out of the lab and into forest maintenance. 

Developing an in-house testbed

Within forest maintenance, unmanned aerial vehicles (UAVs) and ground robots can work alongside humans to maximise both capacity and productivity. UAVs in particular are incredibly flexible with their movement and can acquire rich information through a wide range of sensors, whilst ground robots which have significantly longer battery life can acquire 3D point clouds inside the forest. By using both, a series of such data sets collected over time can then be used to identify areas of the forest that might raise concern.

A key goal of ASPEN was to develop an in-house testbed to provide a controlled environment to trial how UAVs and ground robots could operate in this way. Our testbed is based at the Institute for Safe Autonomy (ISA), at the University of York, and is home to a mix of autonomous ground vehicles, such as the Husky robot, and aerial drones including the autonomous PX4-Vision and Phantom Pro 4. In our testbed we can send a drone to initially scan the area and locate any unhealthy trees that can be identified using ML. Used in partnership with a ground vehicle, the drone would give coordinates to the ground vehicle which drives to the location and retrieves a sample.

When the robot is close, we switch to the human operator to manually control the last part of the sampling process. The sample is then returned and collected for lab analysis. This allowed us to showcase a potential end-to-end process of how systems could be deployed that significantly support the woodland managers, who would be provided with updates from the deployed systems throughout their operation. In addition, it would help to diminish the risks that traditionally come with monitoring these woodlands and create safer working environments.

During our experiments we were successful in testing the DeLeaves sampling tool from Outreach Robotics at our facilities for both an indoor and outdoor flight.

What else did we learn and achieve?

During the project we recruited two University of York student interns, Joel Beedle and Eleanor Griffin-Smith to work on additional technical aspects of the project. One output was a lightweight simulator (SALSA) for a swarm of drones tasked with surveying a forest. The simulator enables rapid prototyping of swarm algorithms, and allows users to automate extensive testing for swarm behaviour analysis. You can find out more in the published paper “SALSA: Swarm Algorithm Simulator” which was presented by co-author Joel at Autonomic Computing and Self-Organizing Systems (ACSOS) 2024. 

Eleanor created an exhibit in ISA for showcasing the physical testbed, which has been presented to a variety of visitors. In addition, Eleanor integrated a Large Language Model (LLM) into the platform, as LLMs can provide opportunities for interaction between non-technical experts, such as woodland managers, and the autonomous platforms. 

Governance of autonomous systems in forest health was also an important topic in the ASPEN project. This was investigated by Dr Bev Townsend (University of York), Dr Norman Dandy (the Wildlife Trust for Birmingham and the Black Country), Professor Andy Smith (Bangor University), Dr Seumas Bates (Bangor University), and Rob Taylor (Defra), where they explored additional considerations that had to be taken into account for the UK’s woodland users. These included legal, political and socio-ethical concerns, and non-physical harms such as noise disturbances and intrusions on users’ privacy in addition to physical harm. Exploring these concerns helped to ensure that such autonomous systems’ technologies are deployed in a safe and human-centric manner, and one which protects the privacy and common rights of all those accessing a woodland. 

What’s next for ASPEN

The vision proposed by ASPEN has potential to bring multiple benefits to global forest maintenance. However, further research around data collection with the use of autonomous systems, and better standardisation would greatly improve our understanding. This would then allow better ML models to be trained, and this is something we, along with our industrial partners, have identified as a key issue to be tackled in a future project.

What’s clear is that we have been able to lay the groundwork for understanding how to make the most of AI and autonomous systems’ capabilities to contribute towards forest health through this project. It is a worthwhile cause, and one we are committed to continuing our research into going forward.

ASPEN was a multidisciplinary project carried out by researchers at Bangor University, the University of Nottingham, the Open University, and the University of York. The project was supported by Forest Research, 2Excel Aviation, Defra, UKRI TAS node in resilience, the Assuring Autonomy International Programme and the University of York’s Institute for Safe Autonomy.

This article was adapted from a Trustworthy Autonomous Systems Hub article.