Large-scale risk mapping of an eruptive bark beetle - Importance of forest susceptibility and beetle pressure

Bark beetle outbreaks have increased in Europe and North America. To mitigate damage efficiently during outbreaks, robust models predicting where the risk for tree mortality is highest across forest landscapes and better understanding of the underlying mechanisms are required. Using Boosted Regression Trees, we modelled relative risks of infestation by the spruce bark beetle Ips typographus (L.) across a 130,000 ha managed lowland forest landscape in southern Sweden during three years of an outbreak and at a resolution of 100 x 100 m. A second nearby landscape of similar size was used for validation. Both predictors reflecting forest susceptibility and beetle pressure were used. Forest susceptibility predictors included volume per ha of host and non-host trees, tree height and distance to the nearest clear-cut harvested during the last four years, all based on interpretations of satellite images. Bark beetle predictors were based on locations and sizes of previous year infestation spots recorded by helicopter. Model outcomes were similar across years, and there was no major reduction in performance when extrapolating predictions in space or time, indicating the modelled relationships have high reliability. Area under curve (AUC) values varied from 0.729 to 0.818. Including bark beetle predictors increased the AUC value somewhat in one of two years. The most important predictor was volume per ha of the host tree, Norway spruce Picea abies (L.) Karst., which reflects the probability of bark beetles encountering suitable trees. This variable was strongly positively correlated with risk up to 200 m(3) ha(-1). Unexpectedly, the volume of the non-host birch was also positively correlated with infestation risk up to 25 m3 ha-1. Tree height was associated with increased infestation risk above heights of 10 m in 2008 and 15 m in 2009. In 2007 and 2008 there was a weak negative relationship between infestation risk and distance to the nearest clear-cut. Additionally, our study shows that in managed forest landscapes the L typographus-killed trees are distributed in many small infestation spots spread out over the landscape. We demonstrate that high-resolution risk-rating maps can be successfully created for large landscapes using easily accessible satellite data of forest characteristics and aerial surveys of infestation spots. The distribution of killed trees in many small infestation spots, poses a challenge for the forest owners to find and remove colonized trees before the new generation emerge. Our results suggest that mitigation efforts in managed lowland forest should focus on high volume spruce stands. (C) 2014 Elsevier B.V. All rights reserved.