Bridging the divide between remote sensing and forest ecological restoration : applications for effectiveness monitoring

With recent innovations in remote sensing technology allowing the acquisition of ultrahigh-resolution data over large areas, there are now unprecedented opportunities to address key issues in restoration ecology. Among these issues is detecting change in forest restoration plantings through the monitoring of restoration success and structural complexity. With the possibility of individual tree resolution, the utility of such technology to also detect differences among species and provenances is only now starting to be investigated. Such monitoring is becoming increasingly important as the choice of species and provenances to be planted when carrying out new restoration (i.e., seed-sourcing strategies) is currently under debate in the restoration community. Recently, there has been a paradigm shift in the choice of strategies, with climate-adjusted provenancing being integrated into the planning phase of restoration plantings. Indeed, as one of the targets of restoration involves rebuilding functionally and structurally complex communities, the right choice of seed material is key to restoration success. Monitoring restoration plantings is thus key for benchmarking and guiding adaptive management to avoid failure. This thesis uses data from a common garden trial established using seed lots from multiple native provenances of two focal species (Eucalyptus pauciflora and E. tenuiramis), planted in different community treatments (i.e., different co-occurring plant species). The trial was established in 2010 within restoration plantings in the southern Midlands region of Tasmania, Australia. Traditional field survey data were used to examine the performances of species and provenances within species when planted under different community compositions. Subsequent research explored the application of remote sensing of structural and spectral traits at the individual tree-level to provide a timely and cost-effective alternative to traditional field sampling to compare species and provenances. Light Detection And Ranging (LiDAR) data and hyperspectral imagery were obtained from sensors mounted on Unmanned Aerial Systems (UAS). These data were used to compare the structural and spectral traits of the focal species and their provenances. In addition, the application of remote sensing to monitor temporal changes in structural traits was tested over a three-year period using hand-held mobile LiDAR (i.e., ZEB1). Three main findings emerged from the study of community composition effects based on the traditional measurements taken six years after planting. Firstly, the relative performance of a species or provenance was not affected by community composition at this early establishment phase. Secondly, there were significant differences in provenance performance, with the local provenance of both local eucalypts outperforming non-local provenances. Finally, translocation success of provenances differed between species, suggesting that assisted migration/colonisation strategies may be site- and species-specific. The UAS LiDAR data were used to quantify multiple structural traits at the individual tree-level and analysed in a quantitative genetics framework. Several traits reflecting facets of tree architecture were found to differ between species and provenances within species. Of note was the provenance differentiation in crown density, a trait only readily assessed over a large area using this type of technology. This finding reiterates the importance of the choice of provenances for transfers, as the introduction of provenances with sparser crowns may have implications on the dependent organisms (e.g. different trophic levels). The development of tree structural traits was successfully monitored over a three-year period at the individual tree-level using ZEB1 LiDAR data, with growth decreasing and crown sparseness increasing in the last year of monitoring, possibly due to drought. The detection of such trends shows the potential of remote sensing technology for the monitoring of very detailed temporal changes in forest structural traits and to provide base-line measurements to assess plantings along their recovery trajectory. Finally, having found significant differences between species and provenances using LiDAR derived traits, UAS-hyperspectral imagery was used to test whether species and provenances also differed in their spectral reflectance. Several datasets (e.g. pure spectra, vegetation indices, LiDAR-derived traits and their combinations) were used following both an object- and pixel-based approach to classify across three different genetic strata: (i) between two focal eucalypt species (E. pauciflora and E. tenuiramis); (ii) between mainland Australia and Tasmanian provenances of E. pauciflora; (iii) between ten different provenances from within the Tasmanian distribution of E. pauciflora. At the species level, the object-based approach mildly outperformed the results deriving from the pixel-based approach. On the other hand, when classifying the within-species strata (ii-iii), pixel-level results vastly outperformed those based on objects. Lastly, employing an odds-ratio and majority vote on the pixel affinities for each object, we were able to correctly classify more than 95% of trees to the right provenance. These results are discussed in light of the advances made in monitoring tree biodiversity at spatial- and genetic-scale relevant for many land managers. In conclusion, using cutting-edge remote sensing technology, the work contained in this thesis has successfully bridged the fields of forest genetics, restoration ecology and environmental monitoring. Novel results have been achieved on the quantification of tree structural traits, highlighting the important role different genetic provenances play in their development. Moreover, the potential of remote sensing applications for ecosystem conservation has been further showcased. For the first time, several genetic provenances of a eucalypt species have been correctly classified using hyperspectral information collected directly in the field.