The British Forestry Commission (FC) provided data for 49 permanent sample plots of Corsican pine (Pinus nigra var. maritima (Ait.) Melville) in Great Britain. They covered various thinning types (low, intermediate, neutral, crown, exploitation), general yield classes (10-22) and initial planting densities (1736-6944 trees per ha). They had been measured at one to six year intervals and thinning was carried out at four to eight year intervals.
The FC follows a detailed procedure for recording sample plot data for various measurements. Computer programmes were written to read these data to do the calculations for the construction of models. Models were initially constructed for separate thinning types by partitioning the data by thinning types (27 sample plots). Later, the possibility of using one set of parameters for each model for all thinning types was tested. However, there were only enough data to construct models for intermediate and neutral thinning types. Each data set was divided into two sets: 75% for constructing the models and 25% for validation.
All models were constructed using regression analysis after determining the basic model structure by examining the scatter distributions and the correlation of selected explanatory variables with the corresponding response variable. All possible combinations of the explanatory variables were tested in order to obtain the best models. It was assumed that there was no natural mortality when thinning was carried out. The performance of the models was tested using statistical tests and standard residual distributions. Two models were constructed initially for each response variable, and after many tests, the best of the two models was selected.
The growth models were constructed to predict the future diameter at breast height (dbh), future total height, current timber height, current total volume and current merchantable volume of individual trees of the main crop trees of Corsican pine growing in Great Britain. The dbh and total height prediction models used the present value of the same variables, a factor to represent the site and the duration of the simulation period. The timber height prediction model used an exponential function developed by multiplying dbh and total height. The total volume prediction model was constructed using basal area and total height of individual trees. The merchantable volume prediction model was a derivation of the selected total volume prediction model. A set of models was also constructed to predict the mean tree basal area, mean dbh and mean total height of the trees removed in thinning. The only explanatory variable of these models was the same
value as the response variable but just before thinning. A general procedure was described to estimate the number of trees removed in each thinning.
Three selected models developed outside Great Britain for other species were recalibrated to Corsican pine in local British conditions without adding new factors or variables to compare the predictability of the new set of models. Bias was highlighted for many re-calibrated models indicating the necessity of new growth functions or variables. Finally the predictions of all the newly constructed and re-calibrated models were tested with the observed values against plantation age.
All the newly constructed models indicated a very low bias and a high modelling efficiency of over 0.9. The signs of the estimated parameters of selected models were corrected to be compatible with the possible biological reality. When compared with the actual data, predictions of the newly constructed models were much closer to the actual values than the predictions from the re-calibrated models.