Using UAVs to Monitor Transformation to Continuous Cover Forestry

  • Guy Bennett

Traethawd ymchwil myfyriwr: Traethawd Ymchwil DoethurolDoethur mewn Athroniaeth


There is increasing interest in Continuous Cover Forestry (CCF) as a methodof managing forest plantations to ensure they become more resilient to climatechange and to meet changing societal demands. CCF utilises several silvicultural techniques in order to promote and enhance forest structural diversity and favours natural regeneration. A critical component of the transformation of plantations to CCF is the acquisition of up-to-date forest inventory data to direct and influence future management decisions. Recently, the use of Individual Tree Detection (ITD) methodologies derived from uncrewed aerial vehicles (UAVs) have been identified as being a cost effective method for the inventory of forests. This thesis aims to develop novel UAV-based method for monitoring forests in transformation to permanently irregular structures, which is the practice of irregular silviculture, a form of CCF. A novel ITD Bayesian parameter optimisation approach that uses quantile regression and external biophysical tree datasets has been developed to provide a transferable and low cost ITD approach for monitoring forest stands in transformation. This method was applied to a range of stands in transformation in the UK and an independent blind study site to test its transferability. Requiring small amounts of training data (15 reference trees) the novel ITD approach developed had a mean test accuracy (F-score = 0.88) and provided mean tree dbh estimates (RMSE = 5.6 cm) with differences that were not significantly different to the ground data (p < 0.05). This method provides evidence for its use in monitoring forests stands in transformation and thus can also be applied to a wide range of forest structures with limited manual parameterisation between sites. Transformation stages offer the potential to assist in monitoring forests through the process to permanently irregular structures. However, thus far, a lack of consistent and widely adopted terminology is hindering its effectiveness. In order to develop an ITD method which can retrieve a stands transformation stage it was necessary to draw together a consensus on transformation stages to permanently irregular structures. This was achieved by adopting a questionnaire approach based on the visual inspection of simulated forest stands by respondents with a range of CCF experience. The results presented in this thesis, for the first time, demonstrate consensus amongst CCF experts on the classification of irregular silvicultural transformation stages, requiring just a small number of respondents(3) to achieve an absolute agreement of 75%. An opportunity, therefore, presents itself to use these definitions as a guide through the process of transformation to permanently irregular structure. This in turn opens up the potential to integrate other scientific fields, such as remote sensing, and utilise new technologies, such as drones, to assist in monitoring the progression of stands through the transformation stages. The ability to map transformation stages automatically using UAV offers the potential to assist current monitoring strategies at low cost. To achieve this, machine learning which utilised the consensus on transformation stages was used to estimate transformation stages. Combining the classification model and the novel ITD method, a sensitivity analysis was conducted. The objective of the sensitivity analysis was to investigate the error sources of the ITD, and its effects on forest management planning calculations and its ability to retrieve a supervised transformation stage classification. The results from the sensitivity analysis provide evidence that the most important error source for inventory data was tree detection, whilst diameter at breast height (dbh) prediction error was found to be the most important in retrieving structural indices and classifying the transformation stage. At the current level ITD with operational dbh prediction error of 17% rRMSE was found to retrieve transformation stage classifications which could be considered operationally ready. The results also found that current levels of ITD were able to inventory forests more accurately than traditional plot based approaches. This thesis presents a specifically tailored ITD method whilst also developing the first automated transformations stage classification model in the UK which can be derived from ITD or traditional plot data. Development of these methodologies has applications across the UK and Ireland to reduce the perceived complexity and to assist with monitoring transformation to CCF.
Dyddiad Dyfarnu2022
Iaith wreiddiolSaesneg
Sefydliad Dyfarnu
  • Prifysgol Aberystwyth
NoddwyrKnowledge Economy Skills Scholarships
GoruchwyliwrAndy Hardy (Goruchwylydd) & Pete Bunting (Goruchwylydd)

Dyfynnu hyn