Tree health: Human, machine and science

To explain how new technologies and techniques can be applied to tree health assessment, this article will consider how human, machine and science can be combined for the surveillance and monitoring of our forests.

Image 2 – Left: Aerial photograph (from Image 1), Right: Hyperspectral imagery.

Manual surveying by trained professionals is still one of the best ways to assess the health and conditions of our trees and forests. Nevertheless, with 13 per cent of the country covered by woodland, surveying each tree in this manner becomes a difficult task.

Aerial photography has been used for the last 100 years to provide a bird’s-eye view of the environment below. Over the last century the quality and availability of these images has improved. Viewing the forest canopy from above in this way can be a good means to manually spot disease outbreaks, especially in areas that are difficult to access on the ground. Image 1 demonstrates visible stress in plantation larch trees; subsequent ground surveying revealed this was due to waterlogging.

In addition to standard aerial photography, more sophisticated imaging systems can be carried by aircraft or drones to provide a more detailed snapshot of the forest canopy across the electro-magnetic spectrum. Hyperspectral imagery, for example, proves a detailed picture of the canopy reflectance across the visible and infrared regions of the spectrum. This provides additional detail about canopy condition beyond what is visible to the naked eye.

Image 3 – Far left: Three-band aerial photograph. Centre left: red band. Centre right: green band. Far right: blue band.

Unlike aerial photography which comprises three bands (red, green and blue) (Image 3), hyperspectral imagery can consist of over 400 narrow bands. This provides a more detailed picture of the canopy. It also presents a challenge due to the large amount of data to be stored and analysed.

One way of exploiting this vast amount of data for tree health assessment is to use machine learning. There are a range of machine learning techniques that can be applied but all require ground truth to train the machine to find the object in question. For example, to identify tree crowns suffering from dieback the machine firstly needs some examples of tree canopies in decline (Image 4). The machine then uses this information to identify other canopies in decline.

Image 4 – Trees highlighted with orange circle exemplify training data for a machine learning approach.

The advantages of this approach mean that, subsequently, more and larger adjacent images can be screened to identify canopies in decline. However, the ability of the machine to find these declining tree crowns is highly dependent on the quality of the initial training data. For the best results it is desirable to have training data for a wide range of forest environments and species.

This requirement for training data is the biggest disadvantage of machine learning approaches. Acquiring these datasets can be a time-consuming and costly process. As a result the approach doesn’t lend itself to a rapid surveillance method for new and emerging pests and pathogens.

To overcome this drawback, new approaches exploiting hyperspectral imagery have turned to scientific understanding of the interaction of light with the tree canopy. Due to the structural and chemical composition of leaves and canopies, trees absorb and reflect incoming sunlight in a characteristic manner. Stress and disease affect the chemical and structural properties of tree canopies through discolouration and defoliation, altering the characteristic interaction with incoming solar radiation.

Image 5 – Assessment of canopy condition from hyperspectral imagery overlaying a high-resolution aerial photograph.

These subtle alterations can be detected using hyperspectral cameras (Image 5). The high resolution of the hyperspectral imagery acquired from aircraft also allows these changes in reflectance to be observed at the individual tree crown scale, providing the precise location of affected individuals.

In this approach ground surveying is not used to train the machine; instead it is used to validate the results and provides a means to assess the accuracy. This allows large areas of forest or woodland to be assessed rapidly in a cost-effective manner.

While new technologies cannot replace the educated eye of forestry professionals, the described techniques can provide a useful tool in the monitoring of tree health and efficient deployment of tree health surveyors. When deployed operationally airborne hyperspectral imagery will provide an additional tool for the identification and surveillance of pests and diseases.