Estimating and mapping forest structure diversity using airborne laser scanning data

The topic of this doctoral thesis is the investigation of the most effective approaches and techniques that can be used to predict and map indicators of forest structural diversity, in a perspective of a more comprehensive assessment, management and monitoring of biodiversity in forest environments. The thesis is subdivided in two main sections, made up of five different but interdependent and organically connected studies, represented by as many published peer-reviewed original research articles, hereafter reported in Roman numerals as Studies I-V. The first section comprises the studies I-II-III. The contents of this section set the basis of methods and know-how that are subsequently used to estimate and map forest structure diversity in Studies IV and V. Several international cooperation projects has been stipulated in order to cope with the issue of the constantly loss of biodiversity at global scale, and because of the relevant influence that forest structure has on biodiversity, forest structure diversity needs to be to assessed and monitored on large areas. In Study I is demonstrated how this achievement can be efficiently tackled coupling ground data, such as those measured during forest inventory surveys, and remotely sensed data, in particular the ones derived from airborne laser scanning (ALS), which has proved to be a reliable source to characterize forest structure. The specific case of Study I presents how ALS data support the estimates of a common forest parameter, in such case forest above ground biomass (AGB), using field data gathered in a novel two-phase tessellation stratified sampling (TSS) design. In order to be used as a valid source of information for planning conservation strategies, along with the estimation, a detailed map showing the spatial patterns of structural diversity is of great usefulness. Study II presents an extensive meta-analysis carried out during the doctoral time frame where is demonstrated that the non-parametric k-NN is, among the others, the most used and effective technique to spatial predict and map forest attributes, alone or combined together to form synthetic indices. This technique can be further improved implementing an optimization step aimed to set the k-NN parameters in order to achieve the best prediction performance possible. Study III demonstrates that, if an optimization phase is carried out before running the k-NN procedure, the performance in the predictions improved sensibly. In the second and last section, the methods experimented in the first section are applied in two different research studies. Study IV describes the use of ALS data and ground data for the areal estimate of mean values of two forest structural diversity indices in a model-assisted framework. Along with the areal estimates, the study proposes the calculation of the confidence intervals of such estimates and the mapping of the investigated indices. Study V is framed as a methodological paper that takes a step further than Study IV, showing how, using the capability of an optimized k-NN techniques in predict simultaneously different parameters, is possible to map a more comprehensive structural diversity index (SDI) combining different forest structural diversity indices.