Liviu Theodor Ene had his dissertation on the 4. May 2012. The dissertation consists of a trial lecture on the topic “The effects of the accuracy and precision of inventory estimates on decision-making” and a public defense of his thesis entitled “Methods for enhancing forest inventories at different spatial scales using auxiliary information”.
The increasing awareness on the importance of forests and forest management practices to modern society requires accurate and cost-effective methods for monitoring and assessment of forest ecosystems. The four studies included in the thesis are addressing a variety of aspects related to the estimation of forest resources using auxiliary information, considering both the design- and model-based inferential frameworks. More precisely, the case studies described in the present work are focusing on above ground biomass estimation using auxiliary information consisting of airborne laser scanning (ALS) measurements, satellite imagery, and other various cartographical products like elevation data from digital terrain models and land-use maps. The methods discussed in the thesis have a wide applicability, and can be used in relation to forest inventories conducted at national and regional scales, and down to small areas and individual tree level. Large-scale inventories using ALS as sampling tool have the potential to provide timely and reliable estimates of forest characteristics. Such inventories rely on complex sampling designs for acquiring ground measurements and ALS data, thus assessing the validity of the inference cannot always be performed using analytical approaches. A possible solution in such cases is to rely on simulated sampling for assessing the behaviour of various estimators. A sampling simulator was created using empirical datasets, copula modelling and nearest neighbor imputations. The studies demonstrated that ignoring the underlying assumptions required by the estimators can seriously affect the precision of the estimates (e.g., nearly five times overestimation of standard errors). Besides, simulated sampling can provide the means for choosing the appropriate estimator (and even the right sampling strategy) to be used in a real application. Furthermore, the simulation results demonstrated viii that using ALS as sampling tool can be a cost-efficient inventory method for large-area applications. Estimation of forest resources at local scales is often difficult when the sample sizes are very small or missing entirely. Such situations were address within a model-based framework, where the superpopulation model was replaced by a canonical vine (C-vine) copula. Using simulated sampling from the copula function followed by nearest neighbor predictions, the approach demonstrated a higher accuracy compared to bootstrap resampling, the main improvement consisting in a significant bias reduction. To meet the demand for detailed information at tree level required by intensive forest management activities, a novel method for tree top detection and extraction of individual tree attributes was developed. Using the stem number estimates provided by area-based inventories and under mild assumptions regarding the spatial process generating the spatial stem distribution, the algorithm demonstrated a robust behaviour by favourably balancing the omission and commission errors in a heterogeneous boreal forest. An important asset of the method is the potential to be seemingly integrated with area-based, ALS-aided operational forest inventories. The strengths and drawbacks of the methods, as well as further improvements to be considered are also discussed.
Professor Erik Næsset (Main supervisor), Dept. of Ecology and Natural Resource Management
Professor Terje Gobakken (Co-supervisor), Dept. of Ecology and Natural Resource Management
Professor Ron E. McRoberts, University of Minnesota
Professor Annika S. Kangas, University of Helsinki
Professor Tron Eid, Norwegian University of Life Sciences
[bibtex file=SkogRoverAll.bib key=Ene2012a]