Reconstructing past forest composition and abundance by using archived Landsat and national forest inventory data. Issue 10 (18th May 2020)

• Record Type: Journal Article
• Authors: Thapa, Bina; Wolter, Peter T.; Sturtevant, Brian R.; Townsend, Philip A.
• Is Part Of: International journal of remote sensing. Volume 41:Issue 10(2020)
• Page Start: 4022
• Page End: 4056
• Publication Date: 2020-05-18
• Subjects: Remote sensing -- Periodicals; Télédétection -- Périodiques; 621.3678
• DOI: 10.1080/01431161.2019.1711245
• Languages: English
• ISSNs: 0143-1161

Abstract

ABSTRACT: Effective modelling of forest susceptibility to defoliating insect outbreaks requires a better understanding of outbreak dynamics, which includes detailed knowledge of the pre- and post-outbreak forest status as well as subsequent feedback mechanisms. In this paper, we strive to fill the forest status need by combining archived Landsat sensor data (pre- and post-outbreak) with different formats and dates of the U.S. Forest Service's Forest Inventory and Analysis (FIA) data (periodic [1970s, 1990s] and annual [2003–2006]). Specifically, we explore the utility of these FIA ground data for calibrating models of forest species and type abundance for mapping past forest composition in the Border Lakes Ecoregion (BLE) of Upper Midwest of the US. Model calibration results between Landsat reflectance and FIA ground data for both total forest basal area and balsam fir ( Abies balsamea ) relative basal area, a preferred host of the spruce budworm (SBW, Choristoneura fumiferana ), were poor to moderate ( R 2 adj 0.39 and 0.48, respectively). Results for aspen ( Populous tremuloides ) and spruce ( Picea glauca and P. mariana ) abundance yielded substantially better accuracies ( R 2 adj 0.64 and 0.78; RMSE 15.56 and 10.65 m 2 ha −1, respectively). Groupings of tree species into broadleaved and conifers substantially improved model calibration result ( R 2 adj range: 0.72–0.91), except for the SBW host group ( A. balsamea, P. glauca, and P. mariana ). Periodic FIA ground data from the early 1990s generated stronger models compared to other FIA-Landsat date combinations tested. A paired t-test of abundance differences between undisturbed forest from the older 1977 and 1990 periodic inventories was significant ( p -value < 0.0001), suggesting possible effects of variable FIA sampling protocol or ground plot positional accuracy through time. However, a similar paired t-test of abundance difference between periodic FIA (1990) and annual FIA (2003–2006) was not significant ( p -value = 0.249). We posit four potential factors that may have contributed to weak Landsat-FIA calibration results for species abundance: 1) variation in FIA subplot arrangement and sampling protocols through time, 2) variability in species abundance and heterogeneity among FIA sampling across adjacent Landsat orbital paths, 3) understory species (balsam fir) that are largely hidden from remote detection, and 4) cloud cover and orbital phase mismatches preventing capture of key forest phenology aids. While past and present FIA sampling protocols were not specifically designed for integration with 30-meter satellite sensor data, careful pairing of FIA ground data (past or present) with Landsat sensor data can facilitate reasonable estimates, of forest abundance for generalized forest types, and possibly forest species when heterogeneity is low. Nevertheless, we recommend that FIA subplot sampling protocols be augmented to include measurements of forest conditions that are more amenable to integration with 30-meter Landsat sensor data.