Monitoring tree vitality with drones

Vitality plays an important role when it comes to assessing the condition of trees. It ultimately determines how successfully an organism can thrive and survive under the given environmental conditions. Leaf green, relative leaf size, and foliage or needle condition are the most important parameters for assessing vitality through simple visual inspection. Photosynthesis takes place in the leaves. Leaf mass is therefore an indicator of a tree’s vitality. Accordingly, leaf discolouration and foliage density can certainly be parameters for assessing vitality. In principle, however, the assessment of vitality is a “relative” assessment, i.e. an assessment that is made in comparison with trees of the same species, at the same stage of development, and on similar sites.

In order to regularly assess the vitality of our forests and detect changes at an early stage, the Forest Condition Survey [Waldzustandserhebung WZE] has been carried out annually in Germany since 1984 in the form of a terrestrial inventory taken at fixed points. In the course of the survey, experts visually assess the crown condition of the forest trees from mid-July to mid-August. The key assessment criterion is the degree of defoliation of the crowns in comparison with a fully leafed or needled crown. The deviation from full foliage is estimated in 5 % steps. 0 % defoliation means a fully leafed crown. 40 % defoliation means that, in comparison with a fully leafed crown, 40 % of the leaf mass is missing, or that only 60 % of the normally expected leaf mass is present.

This approach has proven itself and provides valuable time series data. However, disadvantages include the high annual workload and costs for the federal states, as well as the subjective component of the vitality assessment, which cannot be completely eliminated, despite the constant calibration of the experts on training courses run within the framework of the Forest Condition Survey.

In recent decades, the technology of uncrewed aerial vehicles (UAVs) and data analysis have developed rapidly and offer a wide range of possible applications in forestry. These include UAV-based approaches that have been developed to assess the vitality of forest trees. These mainly use the reflectance behaviour of living, above-ground biomass to calculate spectral vegetation indices, which are intended to allow conclusions to be drawn about vitality and its changes over time. However, there is often no clear correlation between reflected radiation and tree species. Furthermore, numerous studies describe changes in vegetation indices over large areas, without scientifically proving a relationship to the state of vitality of the forest areas or individual trees.

In the joint project “WZE-UAV” carried out by the LWF and the Weihenstephan-Triesdorf University of Applied Sciences (HSWT), the aim was to use high-resolution drone data to overcome these weaknesses and to develop an objective, cost-effective and operable approach for the regular assessment of the vitality of forest trees. To this end, a DJI drone (P4 Multispectral) was used for the first time in 2020 to fly over 263 of the 314 forest condition survey tracts at 100 m above ground level. It took multispectral images with the spectral bands blue, green, red, red edge and near infrared (NIR), with a ground resolution of 5 cm/pixel (Figure 1). Reusable flight plans were created for this, and the flights were carried out by specially trained drone pilots, covering eight to ten Forest Condition Survey tracts per day. The aerial images obtained were batch processed into orthophotos and digital surface models using the software Agisoft Metashape.

In addition, flights were made on an almost weekly basis over the Freising and Ebersberg forest climate stations during the growing seasons of 2020 and 2021. These used the same technology to investigate the phenological aspect and weather influences on the reflection behaviour and derived vegetation indices.

Figure 3 shows the spectral signatures of one healthy, one diseased and one dead tree at the Ebersberg forest climate station. The value spectrum shows a large range of values, and the progression of the signature values differs between vital, infested and dead trees, particularly in the RGB bands in the optically visible wavelength spectrum. The values presented here suggest that time series are very important in order to minimise the noise in the signature values associated with individual images, which also depends on the lighting conditions at the time of image capture (time of day/cloud cover), and to achieve accurate classification.

The statistical analysis of the data of the Forest Condition Survey points was carried out for the main tree species (groups) spruce, pine, fir, beech and oak (Quercus spec). Several vegetation indices (NDVI, NDVRI, GRVI, NDRE) were calculated at individual tree level. These are less sensitive to changing lighting conditions as they utilise relative differences in reflection and are thus better able to compensate for light fluctuations. This proves particularly advantageous when applied to time series datasets, as it allows different recording times to be compared more reliably. Across all of the tree species (groups) considered, it was found that the index values of the vegetation indices correlate closely with the damage levels of the terrestrial inventories and indicate corresponding differences in vitality, with the correlations being more pronounced for spruce, fir and beech than for pine and oak (Figure 4).

Using the high-resolution UAV data with all five spectral bands (R-G-B-RE-NIR), it was possible to automatically classify the species spruce, fir and pine as well as beech and oak with very high accuracy according to the parameters of tree species and level of vitality. When the vitality level was taken into account, F1-scores between 0.51 (pine healthy) and 0.81 (spruce healthy) were achieved (Figure 5). Dead trees could be distinguished very accurately with an F1-score of 0.97. If the vitality parameter was not taken into account, the F1-scores were between 0.77 (fir) and 0.95 (spruce), which are good classification results (Figure 6). The authors refer here to Ecke et al, (2024) for more detailed insights into the results presented here.