Mortality predictions of fire-injured large Douglas-fir and ponderosa pine in Oregon and Washington, USA

Abstract Wild and prescribed fire-induced injury to forest treescan produce immediate or delayed treemortality but fire-injured treescan also survive. Land managersuse logistic regression models that incorporate tree-injury variables to discriminate between fatally injured treesand those that will survive. We used data from 4024 ponderosa pine ( Pinus ponderosa Dougl. ex Laws.) and 3804 Douglas-fir ( Pseudotsuga menziesii (Mirb.) Franco) treesfrom 23 fires across Oregon and Washington to assess the discriminatory ability of 21 existing logistic regression models and a polychotomous key (Scott guidelines). We used insights from the validation exercise to build new models for each treespecies and to identify fire-injury variables which consistently produce accurate mortality predictions. Only 8% of Ponderosa pine and 14% of Douglas-fir died within 3 years after fire. The amount of crown volume consumed, the number of bole quadrants with dead cambiumand the presence of beetles were variables that classified most accurately, but surviving treesin our sample displayed a wide range of fire injury making the accurate classification of dead treesdifficult. For ponderosa pine, our new model correctly classified 99% of live treesand 12% of dead treeswhile the Malheur model ( Thies et al., 2006 ) correctly classified 95% of live treesand 24% of dead trees. The Scott guidelines accurately predicted at least 98% of live ponderosa pine treesbut less than 2% of dead ponderosa pine. For Douglas-fir the Scott guidelines accurately predicted at least 80% of live treesand generally less than 10% of dead trees. Misclassification rates can be controlled by the choice of decision criteria used in the models and managers are encouraged to consider costs of the two types of misclassifications when choosing decision criteria for specific land managementdecisions.