Environmentalists Finally Automate Tree Hugging

Ground-based forest inventories are essential tools for managing and conserving the world's forests. These inventories involve measuring tree trunk size to provide important information, like how much carbon is being sequestered or an estimate of the amount of wood that can be harvested sustainably. The insights about overall forest health that can be gleaned from forest inventories also help to assist in planning future conservation efforts.

However, conducting these inventories presents conservationists with many challenges. The task can be time-consuming, requiring teams of field technicians to manually measure every tree within a specific area. It can also be physically demanding, with technicians having to work in difficult terrain and harsh weather conditions. Additionally, inaccurate measurements can occur due to human error or the use of improper measuring tools.

To overcome these challenges, past efforts have sought to automate the process. Some of the more successful methods have used LiDAR sensors to capture the necessary measurements. However, to date these efforts have used costly, special-purpose equipment that is out of reach for many researchers. Moreover, development of these automated systems has been largely conducted in highly-managed forests. Unlike most natural forest settings, these environments consist mostly of trees that are evenly spaced and growing straight, and the undergrowth is regularly cleared.

When it comes to natural forests, where conditions are not ideal, and low-hanging branches and other obstructions are common, a better, more practical solution is needed. A team of researchers at the University of Cambridge have developed a method that makes use of low-cost LiDAR sensors and an algorithm that can provide accurate tree diameter estimates under real-world conditions. With just a smartphone in hand, a field technician can conduct an inventory almost five times faster than they could using manual methods.

The LiDAR sensor on a Huawei P30 Pro smartphone was used for the device, despite the fact that it provides much lower resolution measurements than the types of sensors typically used in forest inventory systems. This rather unimpressive commodity hardware platform also proved to be capable enough to run the analysis algorithm developed by the researchers, proving that a great many smartphones that people already have in their pockets provide all the hardware that is needed.

In the course of developing their algorithm, the team collected a dataset consisting of manual tree trunk measurements paired with images. Using this data, they were able to train an algorithm to distinguish large branches from trunks, and to capture other relevant information, such as the direction a tree is leaning, and the diameter of the trunk.

The smartphone app was tested in three different natural forests (in the US, UK, and Canada), and across three different seasons. A total of 97 sample tree images were captured and analyzed by the phone app. It was discovered that the algorithm was 100% accurate in detecting tree trunks, and when estimating trunk diameter, a mean error rate of 8% was observed. This level of accuracy is in line with manually collected measurements, however, this automated approach was almost five times faster.

The team is planning to make their app publicly available for Android phones later this year.