Integration of Lidar and Landsat Data to Estimate
Forest Canopy Cover in Coastal British Columbia
Oumer S. Ahmed, Steven E. Franklin, and Michael A. Wulder
Abstract
Airborne Light Detection and Ranging (lidar) data provide
useful measurements of forest canopy structure but are often
limited in spatial coverage. Satellite remote sensing data from
Landsat can provide extensive spatial coverage of generalized
forest information. A forest survey approach that integrates
airborne lidar and satellite data would potentially capitalize
upon these distinctive characteristics. In this study in coastal
forests of British Columbia, the main objective was to deter-
mine the potential of Landsat imagery to accurately estimate
forest canopy cover measured from small-footprint airborne
lidar data in order to expand the lidar measurements to a
larger area. Landsat-derived Tasseled Cap Angle (
TCA
) and
spectral mixture analysis (
SMA
) endmember fractions (i.e.,
sunlit canopy, non-photosynthetic vegetation (
NPV
), shade and
exposed soil) were compared to lidar-derived canopy cover
estimates. Pixel-based analysis and object-based area-weight-
ed error calculations were used to assess regression model
performance. The best canopy cover estimate was obtained
(in the object-based deciduous forest models) with a mean
object size (
MOS
) of 2.5 hectares (adjusted R
2
= 0.86 and RMSE
= 0.28). Overall, lower canopy cover estimation accuracy
was obtained for coniferous forests compared to deciduous
forests in both the pixel and object-based approaches.
Introduction
Accurate information on forest canopy structure is required to
understand and manage forest ecosystems (Wulder and Frank-
lin 2007). Forest canopy cover (
CC
), the area of the ground
covered by a vertical projection of the canopy (Jennings
et al
.,
1999), is a useful metric for several natural resource manage-
ment applications such as: evaluation of wildlife habitat (Koy
et al
., 2005), forest structure classification (Lovell
et al
., 2003;
Fiala
et al
., 2006; Lee and Lucas 2007), characterization of car-
bon sinks (Chopping
et al
., 2008), forest fire behavior and fuel
models (Rollins and Frame, 2006), and estimation of canopy
light transmission (Lieffers
et al
., 1999).
Canopy cover is typically estimated with field instru-
mentation at specific sites, or by remote sensing methods at
increasing spatial scales to support large area monitoring and
modeling applications (e.g., Canadell
et al
., 2008, Coulston
et al
., 2012). Airborne lidar (Light Detection and Ranging)
is an active remote sensing system well suited to measure
canopy structural attributes. Lidar data have proven useful
for estimating
CC
(Hyde
et al
., 2005; Smith
et al
., 2009, Hall
et al
., 2011); while lidar data can provide better estimates of
CC
, wall-to-wall acquisitions of lidar data remain cost prohib-
itive for large forest areas (Coulston
et al
., 2012). Therefore,
lidar-based characterizations of canopy structure are often
restricted in spatial extent. Multispectral remote sensing data-
sets such as the one from Landsat have also been employed to
estimate
CC
. Yet, the lesser sensitivity of these spectral data-
sets to the three dimensional structure of vegetation canopies
(Falkowski
et al
., 2005; Duncanson
et al
., 2010) often degrades
the relationship between
CC
and metrics calculated from the
spectral bands. However, the free availability of large area
multispectral datasets makes them an important data source
for estimating
CC
across large areal extents.
In one recent study, Smith
et al
. (2009) conducted
cross-comparison of multispectral, lidar and coincident field
measurements of
CC
data and they found the relationship
between lidar-derived and field-measured canopy cover was
much stronger and more linear. As forest information is con-
sidered over larger areas, and at higher temporal resolution
(e.g., Goetz
et al
., 2009), such airborne lidar data must be sup-
plemented with other remote sensing datasets, such as those
acquired by the sensors on the Landsat series of platforms
(Wulder
et al
., 2003). The integration of lidar and passive
optical sensors needs to be more thoroughly explored for wall-
to-wall mapping of canopy cover (Hall
et al
., 2011). Recently,
models have been developed to spatially extend airborne lidar
measured forest structural attributes over larger areas using
parametric approaches (Chen
et al
., 2012), which typically
use pixel-based multiple regression to define relationships be-
tween the satellite imagery and airborne lidar-derived canopy
cover (e.g., Smith
et al
., 2009). The integration of such satellite
multispectral remote sensing data with information from
airborne lidar provides opportunities to capitalize upon the
distinctive characteristics of both. This integration could also
serve to make lidar more cost effective over larger areas (e.g.,
Hudak
et al
. 2002; Chen
et al
. 2012). While dealing with meth-
ods to accomplish airborne lidar and Landsat data integration,
among the significant questions that must be addressed are: (a)
the selection of Landsat spectral variables, and (b) the selec-
tion of a pixel-based or object-based sampling approach.
Recent studies have developed empirical models to pro-
duce canopy cover products (e.g., Coulston
et al
., 2013) using
explanatory variables derived from Landsat reflectance values
and derivatives. Various Landsat indices have also been used
Oumer S. Ahmed is with the Geomatics, Remote Sensing and
Land Resources Laboratory, Department of Geography, Trent
University, Ontario, K9J 7B8, Canada (
).
Steven E. Franklin is with the Department of Environmental
and Resource Studies/Science, Department of Geography, and
Office of the President, Trent University, Ontario, K9J 7B8,
Canada.
Michael A. Wulder is with the Canadian Forest Service (Pa-
cific Forestry Centre), Natural Resources Canada, 506 West
Burnside Road, Victoria, British Columbia, V8Z 1M5, Canada.
Photogrammetric Engineering & Remote Sensing
Vol. 80, No. 10, October 2014, pp. 953–961.
0099-1112/14/8010–953
© 2014 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.80.10.953
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
September 2014
953
