Using Satellite Imagery to Evaluate Bark Beetle-Caused Tree Mortality Reported in Aerial Surveys in a Mixed Conifer Forest in Northern Idaho, USA

Abstract

Bark beetles cause significant tree mortality in western North America. The United States Forest Service coordinates annual insect and disease surveys (IDS) by observers in airplanes to map and quantify the tree mortality caused by beetles. The subjective nature of these surveys means that accuracy evaluation is important for characterizing uncertainty. Furthermore, the metric reported for quantifying tree mortality recently changed (2012–2018 depending in region) from killed trees per acre to percent tree mortality within damage polygons, posing challenges for linking older and newer records. Here we evaluated IDS severity estimates in a beetle-affected forest in northern Idaho, USA using fine-resolution satellite imagery, which permitted greater areal coverage than field data. We first used well-established methods to map beetle-caused tree mortality in two WorldView-2 (WV2) images with a high accuracy relative to field observations. Trees-per-acre measurements within collocated IDS polygons were then converted to percent mortality using three methods and evaluated with the WV2 maps. The overall accuracies for the three methods ranged from 35–38% (for methods that used five percent-mortality classes) and 49–56% (three classes). When IDS and WV2 estimates of mortality severity that were within ±15% of each other were considered accurate, overall accuracies were 71–78%. Within the aerial survey damage polygons, the total mortality area tended to be overestimated relative to WV2. WV2 imagery identified ~50% more mortality across the study region compared with the IDS methods, with most of the difference occurring where damage was low severity or in wilderness areas. Severity of Douglas-fir beetle-caused tree mortality was estimated the most accurately, whereas severity of mountain pine beetle-caused tree mortality was estimated the least accurately. Future studies that control for temporal ambiguity between IDS and satellite imagery, as well as IDS spatial error, might provide better assessments of IDS severity accuracy. Our study increases the usefulness of the rich aerial survey database by providing estimates of uncertainty in the IDS database of tree mortality severity.

Orographic Factors as a Predictor of the Spread of the Siberian Silk Moth Outbreak in the Mountainous Southern Taiga Forests of Siberia

“…A Landsat-8 30 m Ground Sample Distance (GSD) time-series obtained from the Earth Explorer data portal (https://earthexplorer.usgs.gov/) was used for the analysis. There are well-proven high-precision methods for classifying areas damaged by phytophagous insects using such images in modern studies [13,21,[28][29][30].…”

Supplementing the Forest Health National Aerial Survey Program with Remote Sensing during the COVID-19 Pandemic: Lessons Learned from a Collaborative Approach

“…This scan-and-sketch approach is meant to mimic ADS surveys, where an ADS surveyor scans current-year high-resolution imagery at their desk and digitizes polygons of visible damage, much like they would in an aircraft, with the DMSM tablet. This data is created with the same data standards of a visual survey, which is particularly important when using alternative supplemental data sources, such as remote sensing, when adding to national databases ( Bright et al 2020a , 2020b ). Because data consistency and validation are critical components of ADS data ( Coleman et al 2018 ; Hicke et al 2020 ), additional training efforts went into standardizing the amount of time spent over one area, the amount of zoom used to mimic the distance from an aircraft to damage on the ground, and the best practices for attribution to meet national reporting standards.…”

Mapping Multiple Insect Outbreaks across Large Regions Annually Using Landsat Time Series Data

“…As part of a previous project [37], we stratified the Stanley study area by biomass using pre-outbreak NDVI to sample in a variety of forest conditions. The Elk City study area was stratified by mortality severity and agent using 2017 to 2018 IDS data, to ensure sampling across a range of mortality conditions and agents [38]. In both study areas, we located plot centers following stratified random sampling designs, and navigated to plot centers with professional-grade Trimble GNSS receivers in the field.…”

“…At plot centers, we established circular 400 m 2 fixed-radius plots (radii of 11.3 m) and measured all trees within plot extents with a diameter at breast height (DBH) >7 cm and >10 cm in Stanley and Elk City, respectively. In addition to DBH, we measured tree condition (live or dead), species, canopy dominance, mortality agent, estimated time since death (based on needle, branch, and bole condition, see Bright et al [38], Table A1) or how many needles were remaining on dead trees, and distance and bearing from plot center. We logged several hundred positions at plot centers using the GNSS receivers, and differentially corrected plot center locations so that estimated plot center accuracies were <1 m. Plot center locations along with distance and bearing measurements were used to calculate individual tree locations, which were then used to select pixels for image classification.…”

“…IDS aerial surveyors estimated mortality severity in terms of trees per acre (TPA) prior to 2016, although some USFS Regions adopted the percent mortality metric as early as 2012. To convert legacy TPA estimates to percent mortality so that total mortality area within IDS polygons could be computed, we used two methodologies: Egan et al [56], which provides conservative estimates of mortality severity [38], and the methodology of Meddens et al [57], which provides more liberal mortality severity estimates. The methodology of Egan et al [56], which utilizes the relationship between surveyed field TPA and percent mortality plot data to adjust IDS TPA estimates, assigns percent mortality values of 5%, 20%, and 65% to IDS polygons having TPA estimates from 0 to 9.9, 10 to 30, and >30, respectively.…”