Hybrid forest modelling of Pinus Radiata D. Don in Canterbury, New Zealand (2006)
Type of ContentTheses / Dissertations
Degree NameDoctor of Philosophy
PublisherUniversity of Canterbury. New Zealand School of Forestry
AuthorsPinjuv, Guy Lshow all
During this study two models were developed to predict growth of Pinus radiata D.Don plantations in Canterbury, New Zealand. The first, CanSPBL(1.2), is a model for whole rotations of stands owned by Selwyn Plantation Limited in Canterbury. The second model, CanSPBL(water) is a hybrid growth model for the Selwyn estate in Canterbury that incorporates an index of root zone water balance over the simulation period. An existing stand growth and yield model CanSPBL was examined using a validation dataset of PSP measurements that were not used in model fitting. Projection bias was shown for mean top height, basal area per hectare, and residual stand stocking particularly for stands at elevations exceeding 450 metres. The new model, CanSPBL(1.2) showed an increase in precision of 4 - 46% over CanSPBL(1.0) at a stand level. The components of the stand model include mean top height, basal area per hectare, stems per hectare, and diameter distribution. The mortality model was made in conjunction with managers at CanSPBL to exclude catastrophic mortality events from model projections. Data used for model fitting was filtered using a mortality index based on the -3/2 power law. An examination of this model with an independent dataset showed little apparent bias. The new model, CanSPBL(water) was developed to include an index of water balance over the simulation period. Water balance estimates were made using a sub model for root zone water balance included in the hybrid physiological model 3-PG (Landsberg and Waring, 1997). The new model showed an increase in precision of 1 - 4% over CanSPBL(1.2) at a stand level (with the exception of the model for maximum diameter which showed a decrease in precision of 0.78%) using climatic inputs that included yearly variation. However the model showed increases of precision from 0.5 to 8% (with the exception of maximum diameter again, showing a decrease in precision of 0.13%) using long term monthly average climatic inputs. The components of the stand model also include mean top height, basal area per hectare, stems per hectare, and diameter distribution. The mortality model was also fitted with a data set filtered using a mortality severity index based on the -3/2 power law to exclude catastrophic mortality events. An examination of this model with an independent dataset showed little apparent bias. Two models to predict a one sided canopy leaf area index (LAI) of radiata pine stands in the Canterbury Plains of New Zealand were also developed. The models were fitted using non-linear least squares regression of LAI estimates against stem measurements and stand characteristics. LAI estimates were derived from digital analysis of fisheye lens photography. The models were kept simple to avoid computational circularity for physiological modelling applications. This study included an objective comparison and validation of a range of model types. The models CANTY (Goulding, 1995), CanSPBL(1.2) (Pinjuv, 2005), CanSPBL-water (Pinjuv, 2005), and 3-PG (Landsberg and Waring, 1997) were compared and validated with the main criteria for comparison being each model s ability to match actual historical measurements of forest growth in an independent data set. Overall, the models CanSPBL(water), and CanSPBL(1.2) performed the best in terms of basal area and mean top height prediction. Both models CanSPBL(water), and CanSPBL(1.2) showed a slightly worse fit in predictions of stocking than did the model CANTY. The hybrid model 3PG showed a better fit for the prediction of basal area than the statistically based model CANTY, but showed a worse fit for the prediction of final stocking than all other models. In terms of distribution of residuals, CanSPBL(1.2) had overall the lowest skewness, kurtosis, and all model parameters tested significant for normality. 3PG performed the worst on average, in terms of the distribution of residuals, and all models tested positively for the normality of residual distribution.