Forests play a significant role in biodiversity-related decision-making as they support approximately 80% of the world’s terrestrial biodiversity. Forest vegetation structure and its complexity affect local biodiversity by modifying microclimatic conditions, providing shelter and breeding sites, and affect the distribution and availability of resources and niches. In assessment of forest vegetation structure, remote sensing data, such as airborne laser scanning (ALS) data and optical data, are widely used. The objective of this dissertation was to examine the potential of ALS data in the assessment of biological and structural diversity of forests.
First, the utilisation of ALS data in the assessment of biological and structural diversity of forests was reviewed. The most studied topics and geographical regions of the study areas were reported, and the most useful and common ALS metrics were listed. Second, the performance of ALS and aerial images for the mapping of ecologically valuable large European aspen (Populus tremula L.) trees was assessed. The ecological importance of European aspen has been highlighted by the large number of Red-listed species that are dependent on it. Remote sensing-based mapping of aspen is known to be difficult as the species is mixed with other deciduous tree species in a forest stand and its occurrence can be sparse. To account for the rarity of large aspen trees and to balance the training data, the synthetic minority oversampling technique (SMOTE) was tested. Third, the performance of ALS and Sentinel-2 data in the prediction of plot-level forest age was analysed. Remote sensing metrics were combined with field data-based categorical variables that describe site conditions. For the prediction of forest age, linear mixed effects modelling (LME) and tree boosting with random effects (GPBoost) were compared. Some of the field plots contained so-called hold over (seed and retention) trees from the previous generation, which hampered age predictions in these plots. This was addressed by testing an alternative prediction method that included a classification step to identify the hold-over plots.
The results showed that most of the research to date with regard to ALS-based assessment of forest biological and structural diversity has been clustered in Europe and North America. Animal ecology, dead trees and tree species diversity measures have been the most frequently studied topics. The ALS data were usually fused with other remote sensing data, especially aerial or satellite images, which was highly advantageous in studies where tree species were considered. There was no single ALS metric that was suitable for all assessments of forest biological or structural diversity. However, the most often utilised and powerful ALS metrics were standard deviation, the mean and the coefficient of variation (COV) of vegetation heights, which were widely utilised across the studied topics.
The classification of large aspen trees, when SMOTE data augmentation was utilised, improved classification accuracy at both the tree- and plot-levels. For the classification of large aspen trees, aerial image metrics were found to be more important than ALS metrics. In particular, the near-infrared band and its ratios with other spectral bands were important. Results suggest that the detection of large aspen trees in genuine populations is still difficult.
The results presented in this dissertation showed that GPBoost was superior to LME in the prediction of plot-level forest age, and that the addition of categorical variables as random group effects led to a clear decrease in the prediction error. Inclusion of categorical variables improved the root mean square error (RMSE) values for LME more than for the GPBoost model. The best modelling strategy was found to include an initial hold-over plot classification before age prediction.
This dissertation demonstrated that ALS data can provide valuable information for the assessment of forest biodiversity at both fine and broad spatial scales. It also showed that it is important to assess the performance of the method with data that provide a realistic picture of the population. Further research on functional diversity, which has received less attention to date, is needed to cover other aspects of forest diversity. Also, the application of the GPBoost model should be further tested for forest attributes other than forest age.