Patterns of Tree Defoliation and Mortality from Insect Damage Using Multi-Scale Remote Sensing

Luke Walter Schefke

doi:10.7273/000006372

With the climate rapidly changing, coniferous trees in North America face many threats, and both native and invasive insects are contributing to their decline and mortality. As insects, particularly bark beetles, successfully attack trees, the foliage of those trees undergoes a color shift from green to red to gray. Attacks from other insects, such as defoliators, can result in defoliation, crown thinning, and loss of needles. These changes may be detected by remote sensing instruments such as satellites and drones. Tree mortality may also come from multiple other variables, such as fire or drought, which then causes tree stress, making plants more susceptible to insect infestation. I analyzed an area with tree disturbance and mortality from three distinct types of insects in Montana, United States to study the detection of forest disturbance by insect outbreaks. This study aims to examine the patterns displayed across a section of forest at different spatial resolutions and scales. Field studies consisted of measuring variables such as diameter, health, and needle color on both trees inside eight-meter fixed-radius plots as well as individual trees not within plots. I analyzed and classified imagery from various sensors, including data from an unmanned aerial vehicle and multiple satellites. Pixels from these data sets are classified using two modeling techniques: maximum likelihood and random forest. This resulted in maps of different tree health classes and other land classes such as bare ground and herbaceous vegetation. I evaluated tree disturbance with classifications of finer spatial resolution pixels (subpixels), which were aggregated to the size of coarser spatial resolution pixels (superpixels) by calculating the percentage of unhealthy trees within, and then comparing them to the classification of the actual classified superpixels. By comparing classification results at different resolution levels, it is possible to extract what information was retained or lost at each step down in spatial resolution, and field measurements provided corroborating evidence of tree disturbance. Random forest models outperformed maximum likelihood models based on accuracy of withheld evaluation points, with overall accuracies ranging from 81.5% to 94.5%. Corroboration of individual trees from the field data was only easily feasible with UAV data, plausible with WorldView-3 data, and not possible with any imagery of 10-m spatial resolution or coarser. Total percent area affected of unhealthy trees was not consistent across resolutions, although coarser imagery tended to underestimate mortality or damage for most intensities of finer imagery disturbance when grouped into distinct disturbance bins but predict more mortality or disturbance across an entire landscape. This study will assist forest managers and natural resource scientists in understanding detection of insect-affected forests, in particular when insect outbreaks are more diffuse and not severe across the entire landscape, giving managers guidelines for where to invest time and resources. This research will also allow for general trends for areas with insect-specific mortality, allowing for potential future comparisons with other causes of tree mortality.

Patterns of Tree Defoliation and Mortality from Insect Damage Using Multi-Scale Remote Sensing

Files and links (1)

Abstract

Metrics

Details