South Africa lags behind the rest of the world with regard to the availability of allometric biomass information. There is a complete lack of site specific allometric data for Pinus radiata in the southern Cape region, impeding investment in the renewable energy sector. This shortcoming was addressed by developing up-scalable, single tree biomass models. These models quantify the aboveground biomass of rotation age P. radiata grown in the southern Cape across a range of site conditions. The models use diameter at breast height (DBH) to predict the aboveground component biomass. A nutrient loss risk potential was assigned to each biomass component. Nineteen trees were destructively harvested using a full fresh weight sampling approach. Basic density was determined using a water displacement method, while Newton’s volume equation was used for stemwood volume determination. Log linear models were simultaneously regressed through seemingly unrelated regression (SUR) using the “Systemfit” R statistical package to force component additivity. A categorical variable was applied to the models, grouping the data into two Site Index (SI) based categories, namely “Low” SI and “Medium to High” SI, to account for inter-site variability. The corrected Akaike Information Criteria (AICc) and coefficient of determination (R2) was used to determine the goodness of fit of the models. The McElroy R2 for the SUR system was 0.95. Biomass models were developed that are able to predict various tree component masses at high levels of certainty within site and stand attribute ranges similar to this study. The importance of accurate, site specific wood basic density was demonstrated by its substantial weighting on stem and hence total biomass. Results showed that the stemwood mean basic density range was between 503 kg m-3 and 517 kg m-3 for the “Low” SI sites and 458 kg m-3 for the “Medium to High” SI sites. Site quality can have a major impact on the models, particularly on poorer sites where stemwood production is proportionally less than other tree components. Total aboveground biomass was estimated to range between 58.61 odt ha-1 and 70.85 odt ha-1 for “Low” SI sites, and 185.31 odt ha-1 to 266.58 odt ha-1 for “Medium to High” SI sites. Stemwood biomass accounted for 65% of the total aboveground biomass for “Low” SI sites and 70% for “Medium to High” SI sites.

Thesis (MScFor)--Stellenbosch University, 2013. / ENGLISH ABSTRACT: The study aims at developing models for predicting aboveground biomass for selected drought tolerant Eucalyptus (E) species (E. cladocalyx, E. gomphocephala and E. grandis x camaldulensis) from the dry west coast. Biomass models were fit for each of the species and a cross-species model was parameterised based on pooled data for all the three species. Data was based on destructive sampling of 28 eucalypt trees which were 20 years of age and additional five five-year old E. gomphocephala trees. Preliminary measurements on diameter at breast height (dbh), height (h) and crown height were recorded in the field. The sampled trees were then felled and samples of discs, branches and foliage were collected. Density of the wood discs and the bark was determined by a water displacement method and computer tomography scanning (CT-scanner). Stem biomass was reconstructed using Smalian’s formula for volume determination and the calculated densities. Upscaling of the crown was carried out by regression equations formulated by employing the sampled branches. Further assessment was carried out on a sub-sample by subjecting the samples to different drying temperatures in a series between 60 and 105ºC. Linear models were parameterised by a simultaneous regression approach based on Seemingly Unrelated Regression (SUR) using the “Systemfit” R statistical package. The predictor variables employed in the study were dbh, d2h and h in which the coefficient of determination (R2), Mean Standard Error (MSE) and Root Mean Standard Error (RMSE) were used to determine the goodness of fit for the models. Akaike Information Criteria (AIC) was also used in the selection of the best fitting model. A system of equations consisting of five models was formulated for each Eucalyptus species. The biomass prediction models had degrees of determination (R2) ranging from 0.65 to 0.98 in which dbh and d2h were the main predictor variable while h improved the model fit. The total biomass models were the best fitting models in most cases while foliage biomass had the least good fit when compared to other models. When the samples were subjected to different drying temperatures, stem wood had the largest percentage change of 6% when drying from 60ºC to 105ºC while foliage had the lowest percentage change of less than 2%. / AFRIKAANSE OPSOMMING: Die doel met hierdie studie is om modelle vir die voorspelling van die bogrondse biomassa van drie droogte-bestande Eucalyptus (E) spesies (E. cladocalyx, E. gomphocephala en E. grandis x camaldulensis), gekweek op die droë kusvlakte in Wes-Kaapland, te ontwikkel. Biomassa modelle vir elk van die spesies is gepas en ’n model gegrond op die gekombineerde data van al drie die spesies, is geparameteriseer. Verder is die biomassa variasie onder verskeie droogingstemperature vasgestel. Die data versameling is uitgevoer gegrond op die destruktiewe mostering van 28 Eucalyptus bome wat 20 jaar oud was en ’n bykomende vyf vyfjarige E. gomphocephala bome. Die aanvanklike mates, naamlik deursnee op borshoogte (dbh), boomhoogte (h) en kroonhoogte is in die veld opgemeet. Die gemonsterde bome is afgesaag en monsters van stamhout skywe, takke en die bas is versamel. Die digtheid van die skywe en die bas is deur die waterverplasing metode, en Rekenaar Tomografie skandering (“CT-scanning”) vasgestel. Stam biomassa is rekonstrukteer deur gebruik te maak van Smalian se formule vir die vasstelling van volume en berekende digtheid. Die opskaal van die kroon biomassa is gedoen met behulp van regressie vergelykings van gekose takmonsters. Submonsters is onderwerp aan ’n reeks van verskillende drogingstemperature tussen 60 en 105ºC. Lineêre modelle is deur ’n gelyktydige regressie benadering gegrond op die Seemingly Unrelated Regression (SUR) wat ’n“Systemfit” R statistiese pakket gebruik, parameteriseer. Die voorspeller veranderlikes wat in hierdie studie gebruik is, is dbh, d2h en h waarin die koëffisient van bepaling (R2), gemiddelde standaardfout (MSE) en vierkantswortel van die gemiddelde standaardfout (RMSE) gebruik is om vas te stel hoe goed die model pas. Akaike Inligting Kriteria is gebruik vir die seleksie van die gepaste model. ’n Reeks vergelykings wat bestaan uit vyf modelle is vir elke Eucalyptus spesie geformuleer. Die biomassa voorspelling model het waardes vir die koëffisiente van bepaling (R2) opgelewer wat strek van 0.65 to 0.98% en waarin dbh en d2h die hoof voorspelling veranderlikes is, terwyl h die pas van die model verbeter. Die totale biomassa model het in die meeste gevalle die beste gepas en die blaarbiomassa die swakste as dit met die ander modelle vergelyk word. Tydens droging vind die grootste persentasie verandering van 6% by stamhout plaas tussen temperature van 60ºC tot 105ºC, en die kleinste persentasie verandering van minder as 2% by blare.