PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
November 2016
853
Quantifying Early-Seral Forest Composition with
Remote Sensing
Rayma A. Cooley, Peter T. Wolter, and Brian R. Sturtevant
Abstract
Spatially explicit modeling of recovering forest structure within
two years following wildfire disturbance has not been attempt-
ed, yet such knowledge is critical for determining successional
pathways. We used remote sensing and field data, along with
digital climate and terrain data, to model and map early-ser-
al aspen structure and vegetation species richness following
wildfire. Richness was the strongest model (
RMSE
= 2.47 species,
Adj. R
2
= 0.60), followed by aspen stem diameter, basal area
(
BA
), height, density, and percent cover (Adj. R
2
range = 0.22 to
0.53). Effects of pre-fire aspen
BA
and fire severity on post-fire
aspen structure and richness were analyzed. Post-fire recovery
attributes were not significantly related to fire severity, while all
but percent cover and richness were sensitive to pre-fire aspen
BA
(Adj. R
2
range = 0.12 to 0.33, p <0.001). This remote map-
ping capability will enable improved prediction of future forest
composition and structure, and associated carbon stocks.
Introduction
Individual wildfires often include a range of impacts (i.e.,
severity), producing a mosaic of biological legacies that can
have persistent influence on post-fire composition and succes-
sional pathways (Franklin
et al.
, 2007). Patterns of early-seral
forest composition and density established within a few
years post-fire are strong predictors of the initial successional
trajectory of a forest (Johnstone
et al.
, 2004). In regions with
rich fire legacies, the consequences for future dynamics of
subsequent forests are dramatic (Johnstone
et al.
, 2010). For
example, in boreal and sub-boreal systems of North America,
the relative dominance of conifer versus deciduous tree spe-
cies will strongly affect nutrient dynamics (Frelich and Reich,
1995), fire behavior (DeByle and Winokur, 1985; Carlson
et
al.
, 2011), susceptibility to insect disturbance (Charbonneau
et al.
, 2012), wildlife habitat (Pastor
et al.
, 1988), and regen-
eration capacity in response to future disturbances (Frelich,
2002). A growing body of literature suggests that anthropo-
genic activities and effects, including climate change, land
use, and fire suppression, are modifying regional patterns in
fire severity (Stephens
et al.
, 2014). Yet current understand-
ing of the effects of fire severity on forest development lags
behind (Keeley, 2009), in part because early-seral regeneration
patterns are difficult to quantify at the scale of large burns.
Hence, the ability to accurately characterize early-seral forest
structure soon after disturbance is of critical importance for
understanding successional trajectories.
Remote sensing has been used in boreal and sub-boreal for-
ests of North America to reliably map mature forest composi-
tion and structure (Wolter
et al.,
2009; Wolter and Townsend,
2011). However, the spectral signal of early-seral forest
regeneration (one to two years) following disturbance is often
indistinct and may be confused with other vegetation life
forms, coarse woody debris, and soil prior to canopy closure,
which complicates the composition and structure mapping
process using medium spatial resolution sensors such as
Landsat (Veraverbeke
et al.,
2012). Nevertheless, the ability
to map tree species recruitment and structure information so
early in a forest’s successional state (over large landscapes)
would be a valuable asset for forest managers and scientists in
providing insight into future forest development patterns in
time and space (Veraverbeke
et al.,
2012).
The Pagami Creek Fire (
PCF
, 18 August to 12 October 2011)
burned over 38,000 hectares (ha) of the Superior National Forest
(
SNF
), most of which (90.2 percent) occurred in the Boundary
Waters Canoe Area Wilderness (
BWCAW
) (Figure 1). The
PCF
is
the largest and most recent in a series of major fires affecting
the
BWCAW
. The initial behavior of the fire was characterized by
low-intensity surface fires, until mid-September when the fire
transitioned to a high-intensity crown-fire, creating a range of
fire severity patterns. Prior to the burn, this area served as a hot-
spot for remote sensing research mapping forest composition
and structure (Wolter
et al.,
2008; Wolter
et al.,
2009; Wolter and
Townsend, 2011). The
PCF
therefore provided a rare opportunity
to investigate interactions between pre-fire forest conditions
and fire severity as they affected post-fire regeneration patterns.
Among the first tree species to emerge from the fire disturbance
were quaking and bigtooth aspen (
Populus tremuloides
and
P.
grandidentata
, respectively), which can sprout vigorously from
clonal root networks (Frelich and Reich, 1995).
The objectives of our study were twofold. First, we as-
sessed the degree to which early regeneration structure of
aspen and vegetation species richness could be reliably esti-
mated and mapped using field-collected data measurements,
in conjunction with image-based remote sensing variables
and other spatially-explicit biophysical information. To our
knowledge, two years post-disturbance is the earliest that
detailed mapping of aspen regeneration structure (i.e., height,
basal area, stem density, stem diameter, and percent cover) has
been attempted using image-based remote sensing techniques.
Indeed, two-year aspen regeneration is expected to be short,
mixed with both herbaceous and woody growth, with either
sparse or heterogeneous leaf area that may be well below the
spatial resolution of commonly-used satellite sensors such as
Landsat (30-meter) to detect. We therefore integrated Landsat
imagery with one meter spatial resolution National Agricul-
ture Imagery Program (
NAIP
) color-infrared aerial image data to
develop image-based models of aspen regeneration abundance
and structure, as well as vegetation species richness.
Rayma A. Cooley is with the US Forest Service, Six Rivers
National Forest, 741 State Hwy 36, Bridgeville, CA 95526,
and formerly with the Iowa State University (raymacooley@
fs.fed.us).
Peter T. Wolter is with the Iowa State University, 339 Science
Hall II, Ames, IA 50011.
Brian R. Sturtevant is with the US Forest Service, Northern
Research Station,5985 Highway K, Rhinelander, WI 54501.
Photogrammetric Engineering & Remote Sensing
Vol. 82, No. 11, November 2016, pp. 853–863.
0099-1112/16/853–863
© 2016 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.82.11.853
