Harvard Forest >

Harvard Forest Symposium Abstract 2018

  • Title: Mapping and monitoring forest insect outbreaks & invasions using Landsat time series
  • Primary Author: Valerie Pasquarella (University of Massachusetts - Amherst )
  • Additional Authors: David Orwig (Harvard Forest)
  • Abstract:

    Introduced insects and pathogens impact millions of acres of forested land in the United States each year, and large-scale monitoring efforts are essential for tracking the spread of outbreaks and quantifying the magnitude and extent of damage. There is a long history of using remote sensing imagery to detect changes in forest condition resulting from pest activity. However, the opening of the Landsat archive for free public use in 2008 has led to new opportunities to use time series of all available Landsat observations to map and monitor forest insect outbreaks and invasions.

    In this project, we explore the use of more than 30 years of Landsat observations for characterizing changes in forest condition resulting from outbreaks of gypsy moth (Lymantria dispar) and hemlock woolly adelgid (Adelges tsugae) in support of the Harvard Forest LTER Program. The forest changes associated with each of these pests are hypothesized to have unique spectral-temporal signatures. Therefore, different approaches will be employed to map impacts on New England’s forest communities.

    Gypsy moth (GM) is a defoliating insect that feeds primarily on deciduous species, particularly oaks (Quercus spp.). Because trees are often exhibit a second flush of foliage following initial defoliation events, GM damage is considered an ephemeral disturbance process. Our previous work focused on the development of a near-real-time monitoring approach that detects changes in forest condition based on deviations from a historical baseline (Figure 1). This approach was used to map the unexpected defoliation that occurred in 2015, 2016, and 2017 for two Landsat scenes covering Southern New England (Figure 2). The total extent of defoliation doubled from each year to the next over this three-year period, though the area of most severe defoliation remained about the same in 2016 and 2017 (Figure 3). Future efforts will focus on improved differentiation of defoliation and recovery processes, as well as an assessment of mortality risk resulting from multiple years of defoliation combined with drought stress and secondary pests and pathogens.

    Hemlock woolly adelgid (HWA) is an aphid-like insect that feeds on storage tissue of hemlock (Tsuga spp), and in the Northeastern US, HWA has become a major threat to eastern hemlock (Tsuga canadensis). Unlike gypsy moth, which cause short-term changes in forest canopies, HWA infestations have resulted in the long-term decline of hemlock trees and shifts in forest composition. Exploratory analysis suggests that changes in hemlock stands caused by HWA infestations are most apparent during the early spring (April-May; Figure 4). Ongoing efforts will focus on a more robust characterization of this spectral-temporal signature, as well as the development of an automated approach for mapping hemlock decline and resulting ecosystem dynamics.

    We anticipate this study will result in improved maps of both distribution and severity of historic and ongoing GM and HWA impacts. We initially expect to produce maps at an annual time step using observations from Landsat 4, Landsat 5, Landsat 7 and Landsat 8 (ca. 1984 through present). There may also be opportunities to map earlier periods as imagery from Landsats 1-3 (ca. 1972 to 1985) are integrated into the USGS Collection 1 data format. Initial pilot studies for both the GM and HWA assessments have been focused on Southern New England, although we intend to scale them to the full New England region. The map products generated from these analyses will be used to parameterize forest landscape models, and will support LTER-VI efforts to simulate scenarios of infestation and quantify regional impacts of invasive species on forest composition and function.

    Figure 1: Example Tasseled Cap (TC) Greenness time series (2013-2017) for a 30m Landsat pixel defoliated by gypsy moth in 2016 and 2017. Harmonic model (solid red line) was fit to all available TCG values for the period 2005-2015 using the Continuous Change Detection and Classification (CCDC) algorithm. Predicted values (dashed red line) are compared to observed values (green points) to assess changes in condition.

    Figure 2: Landsat-based gypsy moth defoliation assessments for 2015, 2016, and 2017 for two scenes (Paths/Rows 12/31 and 13/31). Average difference scores are binned into four severity categories: slight change (deviations 1 to 2 times the model RMSE), moderate change (deviations 2 to 3 times the model RMSE), large change (deviations 3 to 4 times the model RMSE), and very large change (deviations greater than 4 times the model RMSE). Above-average defoliation was first observed in southeastern Massachusetts in 2015. More severe and extensive defoliation was observed in and around Rhode Island in 2016, and defoliation continued to expand into eastern Connecticut and central Massachusetts in 2017

    Figure 3: Total defoliated area by year and severity class. Defoliated forest area in the Southern New England study are (Figure 2) doubled each year from 2015-2017, with the greatest increase in areas exhibiting slight and moderate changes in condition

    Figure 4: Example Tasseled Cap (TC) Wetness time series (1985-2017) for a 30m Landsat pixel in South Springfield, Massachusetts (location of first documented appearance of HWA in the state). This site was initially dominated by hemlock ca. 1990, and significant HWA presence has been documented in field surveys. Prior to HWA invasion, this hemlock stand exhibited a relatively stable seasonal TCW signal, but as time since invasion increased, seasonal variability increased with a notable decrease in TCW during April-May (cyan points).

  • Research Category: Regional Studies; Invasive Plants, Pests & Pathogens; Historical and Retrospective Studies; Ecological Informatics and Modelling

  • Figures:
  • Figure1.png