Basal area growth and damage prediction models for retained pine seed trees in Sweden

Natural pine regeneration with seed trees is an established and well-used method in Swedish forestry. Retained pines respond to increased exposure and resource availability by increasing their radial stem growth, as observed in the annual rings' width after release. Currently, the basal area growth of seed trees left over after clear felling is simulated (in Heureka) with individual tree models dependent on the tree, competition and site factors. However, a major drawback of those models is that the growth reaction of retained seed trees after liberation is not accounted for leading to potentially biased estimates of the expected growth. This report presents functions for predicting the basal area growth underbark (BAI) for retained pine seed trees. In parallel, models for damage predictions are provided.The basal growth functions were developed in two steps. First, the response after liberation was examined and after that quantified using a logistic growth function. The results showed a significant increase in the BAI, which persisted up to 30 years after liberation and the magnitude of the response depended on the site index. On sites with high site index, the relative increase in growth 10 years before and after liberation was 11%. The corresponding value for sites with low site index was 21%. The maximum relative response culminated at six and seven years after liberation for high and low site indices, respectively. In the second step, the liberation response model was integrated with a BAI prediction model with the predictor variables diameter at breast height, crown ratio, site index and categorical indicators representing stand management class (from barelands to unthinned stands). The new function's simulated outcome demonstrates logical predictions of growth.The regression function for predicting the probability of retained pine trees being damaged at a given time was calibrated with tree height, age and site index as predictor variables. The model accuracy using a cut-off value of 0.5 was moderate (i.e., 66%) and with sensitivity and specificity values of 80% and 40%, respectively. Data from the temporary sample plots of the Swedish NFI distributed throughout the country facilitated this investigation. The limitations and potential developments of the growth and damage models are discussed.