Predicting live and dead basal area from LandTrendr variables in beetle-affected forests

Abstract

Recent bark beetle outbreaks in western North America have been widespread and severe. Forest managers need accurate information on beetle-induced tree mortality to make better decisions on how best to remediate beetle-killed forests and restore healthy ecosystem services. We applied the LandTrendr analysis tool, which can detect beetle-induced tree mortality, to Landsat image time series (1984-2010) at study areas affected by bark beetles in Oregon, Alaska, Idaho, Colorado, and Arizona. LandTrendr outputs indicating the timing, magnitude, and duration of disturbance events were generated consistently in all study areas considered and were used to predict Dead Basal Area (BA), Live BA, Total BA, and Percent Dead BA, as summarized from field plot observations geolocated within each study area, using The Random Forest (RF) classification and regression tree algorithm. RF models explained 11-53, 15-55, 11-47, and 17-56% of variance in total, live, dead, and %dead BA, respectively. RF models based on plots from all study areas explained 42, 30, 47, and 51% of variance in total, live, dead, and %dead BA, respectively. This was generally higher than the variance explained within a single study area. These results demonstrate potential for applying LandTrendr for mapping the timing, duration, spatial extent, and magnitude of bark beetle infestations from the historical Landsat image record, which would help forest scientists and managers to better map and monitor beetle effects on coniferous forests and manage their recovery.