ArticlePDF Available

Abstract and Figures

Free and open access to the more than 40 years of data captured in the Landsat archive, combined with improvements in standardized image products and increasing computer processing and storage capabilities, have enabled the production of large-area, cloud-free, surface reflectance pixel-based image composites. Best-available-pixel (BAP) composites represent a new paradigm in remote sensing that is no longer reliant on scene-based analysis. A time series of these BAP image composites affords novel opportunities to generate information products characterizing land cover, land cover change, and forest structural attributes in a manner that is dynamic, transparent, systematic, repeatable, and spatially exhaustive. Herein, we articulate the information needs associated with forest ecosystem science and monitoring in a Canadian context, and indicate how these new image compositing approaches and subsequent derived products can enable us to address these needs. We highlight some of the issues and opportunities associated with an image compositing approach and demonstrate annual composite products at a national-scale for a single year, with more detailed analyses for two prototype areas using 15 years of Landsat data. Recommendations concerning how to best link compositing decisions to the desired use of the composite (and the information need) are presented, along with future research directions.
Content may be subject to copyright.
Canadian Journal of Remote Sensing, 40:192–212, 2014
Published with license by Taylor & Francis
ISSN: 0703-8992 print / 1712-7971 online
DOI: 10.1080/07038992.2014.945827
Pixel-Based Image Compositing for Large-Area Dense
Time Series Applications and Science
J. C. White1,*,M.A.Wulder
1, G. W. Hobart1,J.E.Luther
2, T. Hermosilla3,
P. Griffiths4,N.C.Coops
1, and L. Guindon6
1Canadian Forest Service (Pacific Forestry Centre), Natural Resources Canada, 506 West Burnside
Road, Victoria, British Columbia, V8Z 1M5, Canada
2Canadian Forest Service (Atlantic Forestry Centre), Natural Resources Canada, P.O. Box 960, 20
University Drive, Corner Brook, Newfoundland, A2H 6P9, Canada
3Department of Forest Resource Management, University of British Columbia, 2424 Main Mall,
Vancouver, British Columbia, V6T 1Z4, Canada
4Geography Department, Humboldt-Universit¨
at zu Berlin 10099, Berlin, Germany
5Canadian Forest Service (Northern Forestry Centre), Natural Resources Canada, 5320-122nd Street,
Edmonton, Alberta, T6H 3S5, Canada
6Canadian Forest Service (Laurentian Forestry Centre), Natural Resources Canada, 1055 du P.E.P.S,
succ. Sainte-Foy, Quebec City, Quebec, G1V 4C7, Canada
Abstract. Free and open access to the more than 40 years of data captured in the Landsat archive, combined with improvements
in standardized image products and increasing computer processing and storage capabilities, have enabled the production of
large-area, cloud-free, surface reflectance pixel-based image composites. Best-available-pixel (BAP) composites represent a new
paradigm in remote sensing that is no longer reliant on scene-based analysis. A time series of these BAP image composites
affords novel opportunities to generate information products characterizing land cover, land cover change, and forest structural
attributes in a manner that is dynamic, transparent, systematic, repeatable, and spatially exhaustive. Herein, we articulate the
information needs associated with forest ecosystem science and monitoring in a Canadian context, and indicate how these new
image compositing approaches and subsequent derived products can enable us to address these needs. We highlight some of the
issues and opportunities associated with an image compositing approach and demonstrate annual composite products at a national-
scale for a single year, with more detailed analyses for two prototype areas using 15 years of Landsat data. Recommendations
concerning how to best link compositing decisions to the desired use of the composite (and the information need) are presented,
along with future research directions.
e. L’acc`
es libre et gratuit `
a plus de 40 ans de donn´
ees dans l’archive Landsat combin´
elioration des produits
d’imagerie standardis´
es et l’augmentation des capacit´
es de traitement et de stockage informatiques ont permis la production
d’images composites bas´
ees sur les pixels de r´
eflectance de surface de grande superficie sans nuages. Les composites du « meilleur
pixel disponible » (best-available-pixel;BAP)repr
esentent un nouveau paradigme en mati`
ere de t´
etection qui ne d´
epend plus
de l’analyse par sc`
ene. Une s´
erie chronologique de ces images composites BAP offre de nouvelles occasions de g´
erer des produits
d’information qui caract´
erisent la couverture terrestre, le changement de la couverture terrestre et les attributs structurels de
la forˆ
et d’une mani`
ere dynamique, transparente, syst´
ematique, r´
etable et spatialement exhaustive. Ici, nous articulons les
besoins d’information li´
es `
a la science et `
a la surveillance des ´
emes forestiers dans un contexte canadien, et nous indiquons
comment ces nouvelles approches de composition d’image et les produits qui en d´
ecoulent peuvent nous permettre de r´
a ces besoins. Nous soulignons quelques-uns des probl`
emes et des possibilit´
es associ´
es `
a une approche de composition d’image et
nous d´
emontrons des produits composites annuels `
echelle nationale pour une ann´
ee, avec des analyses plus d´
ees pour deux
zones prototypes utilisant 15 ans de donn´
ees Landsat. Des recommandations concernant la meilleure fac¸on de lier des d´
de composition d’images `
a l’utilisation souhait´
ee du composite (et le besoin d’information) ainsi que les orientations futures de la
recherche sont pr´
Received 30 April 2014; Accepted 9 July 2014.
*Corresponding author e-mail:
©J. C. White, M. A. Wulder, G. W. Hobart, J. E. Luther, T. Hermosilla, P.
Griffiths, N. C. Coops, R. J. Hall, P. Hostert, A. Dyk, and L. Guindon
This is an Open Access article. Non-commercial re-use, distribution, and
reproduction in any medium, provided the original work is properly attributed,
cited, and is not altered, transformed, or built upon in any way, is permitted. The
moral rights of the named author(s) have been asserted.
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
VOL. 40, NO. 3, JUNE/JUIN 2014 193
Free and open access to the Landsat archive enables the
generation of pixel-based composites that can be used to ad-
dress a broad range of information needs associated with na-
tional ecosystem monitoring. For a country the size of Canada
(approaching 10 million km2), with limited access to remote
forests and a multiplicity of jurisdictions responsible for re-
source stewardship, remotely sensed data offers the only viable
means, economic or otherwise, to generate national information
products for ecosystem monitoring in a dynamic, transparent,
systematic, repeatable, and spatially exhaustive manner (Wul-
der et al. 2007a; Falkowski et al. 2009). Landsat data brings
two key elements to ecosystem monitoring: a spatial dimen-
sion that is at a scale appropriate for capturing anthropogenic
impacts (Townshend and Justice 1988), and a temporal dimen-
sion that enables retrospective analyses and characterization of
changes over the more than 40 years of data captured by suc-
cessive Landsat sensors (Wulder et al. 2012). Pixel-based image
compositing with Landsat is a new paradigm in remote sensing
science that applies a suite of user-defined rules to leverage the
extensive Landsat archive for generating cloud-free, radiomet-
rically and phenologically consistent image composites that are
spatially contiguous over large areas (Roy et al. 2010; Hansen
and Loveland 2012; Griffiths et al. 2013).
Pixel-based image compositing of Landsat data has emerged
from a unique confluence of scientific and operational develop-
ments, predicated by free and open access to the Landsat archive
(Woodcock et al. 2008; Wulder et al. 2012), and supported by
the computing and data storage capacity to fully automate ra-
diometric and geometric pre-processing and create increasingly
robust standardized image products (Masek et al. 2006; Roy
et al. 2010). Prior to the era of free access to the Landsat
archive, pixel-based compositing approaches were limited to
low spatial resolution data, with pixels 500 ×500 m or greater,
such as Advanced Very High Resolution Radiometer (AVHRR)
(Holben 1986; Cihlar et al. 1994) and MODerate-resolution
Imaging Spectrometer (MODIS) (Roy 2000; Justice et al. 2002;
Ju et al. 2010). Common to these datasets is their free avail-
ability with near-daily global coverage allowing for selection
of pixels based upon user-defined rules. The compositing ap-
proaches were used primarily to reduce the impact of clouds,
aerosol contamination, and view angle effects, as well as data
volumes (Holben 1986; Cihlar et al. 1994). Due to the large
number of observations available, AVHRR and MODIS com-
positing approaches were relatively simple, using rules such as
the maximum Normalized Difference Vegetation Index (NDVI)
or the minimum view angle to select the “best” observation for a
given pixel within a specified compositing period (e.g., 16 days)
(Wolfe et al. 1998). For Landsat, high data purchase costs prior
to 2008 precluded the application of any such data-intensive
compositing approach. With the opening of the Landsat archive,
compositing approaches using Landsat data became economi-
cally feasible. Such approaches have benefitted and been in-
formed by earlier compositing methods developed for AVHRR
and MODIS data (Roy et al. 2010).
Early efforts at compositing with Landsat serendipitously
took advantage of the overlap between adjacent Landsat scenes.
For example, Du et al. (2001) created a large-area Landsat
mosaic and applied pixel-based compositing in overlap areas
between adjacent scenes, selecting the pixel with the highest
NDVI for the final composite. Post-classification compositing
of Landsat data from scene overlap areas has also been used to
produce large-area land cover products (Guindon and Edmonds
2002; Wijedasa et al. 2012). Epochal global Landsat datasets—
whereby single best observations are selected over a given time
period—were also produced and distributed free of charge to
the science community and the general public via the Global
Land Survey (GLS) project (Townshend et al. 2012; Gutman
et al. 2013). The 1975, 1990, and 2000 GLS datasets were
produced using the best single-date image for each Landsat
path/row (Tucker et al. 2004), whilst the 2005 and 2010 GLS
data were produced via compositing up to three dates of imagery
for each path/row, primarily to fill gaps resulting from Scan Line
Corrector (SLC)-off data (Gutman et al. 2013). Given the lim-
ited availability of cloud-free Landsat data in some areas of the
globe, epochal composites have been used extensively to sup-
port change detection studies (e.g., Hansen et al. 2008; Potapov
et al. 2011). Lindquist et al. (2008) evaluated the potential of
the epochal 2000 and 2005 GLS datasets, relative to more data
intensive per-pixel compositing approaches (e.g., Hansen et al.
2008), for mapping forest cover change in the tropics. The au-
thors concluded that in order to provide sufficient spatial cover-
age to support change detection between epochs, Landsat-based
image compositing approaches should make use of all available
Landsat data for any given path/row (and certainly more than the
three images prescribed for in the GLS products). In a similar
study, Broich et al. (2011) generated epochal Landsat compos-
ites for 2000 and 2005 over Sumatra and Kalimantan, Indonesia
and assessed the efficacy of these composites for quantifying
forest cover change. The authors found that a time series ap-
proach that used “all good land observations” provided more
accurate estimates of forest cover change when compared to
change maps generated from the epochal composites. It should
be noted however that these epochal composites were generated
to serve a broad range of applications and were not tailored
specifically to forest monitoring.
Since the opening of the Landsat archive in 2008, several
Landsat compositing approaches have emerged in the literature
(Table 1). Many of the approaches have relied exclusively on
Landsat Enhanced Thematic Mapper (ETM+data) corrected to
TOA reflectance (Roy et al. 2010; Potapov et al. 2011, 2012).
The per-pixel compositing approach applied by Hansen et al.
(2008) that was later adapted by Potapov et al. (2011), relied on a
MODIS-generated forest/non-forest mask to support radiomet-
ric normalization via a dark object subtraction (DOS) method
(Chavez 1988). Potapov et al. (2012) further refined this ap-
proach, using a 10-year MODIS surface reflectance composite
to facilitate band-wise mean bias adjustments with correspond-
ing Landsat bands. For very large areas, the use of coarser spa-
tial resolution data for normalization increases computational
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
Review of studies applying Landsat pixel-based image compositing approaches
Study Sensor Physical Unit Processing
# of unique
path/row s # of images Compositing period Cloud-masking Rule-base
Hansen et al.
MODIS forest/non-forest mask
used as reference for
normalization via dark object
subtraction (DOS) and a
regression-based surface
anisotropic correction.
20 4–7 per path/row
(98 total)
Two epochs: pre-1996
and post-1996
Custom (supervised
classification using
decision trees)
Selected image date with the
lowest cloud and shadow
likelihood; if two were equal
used the pixel value closest to
the forest reference value
(TOA =100).
Roy et al.
Standard practice for TOA
estimation (Chander et al.
459 6521 Monthly, seasonal,
annual (Dec 2007 to
Nov 2008)
Custom (ACCA and
classification using
decision trees)
Valid surface observations with
minimal cloud, snow, and
atmospheric contamination.
Heritage maximum NDVI for
vegetated pixels; maximum
brightness temperature for
unvegetated pixels; cloud
Potapov et al.
MODIS forest/non-forest mask
used as reference for
normalization via DOS and
regression-based surface
anisotropic correction
406 7227 Two epochs: 2000 and
Custom (supervised
classification using
CART decision trees)
Per-pixel QA assessment based
on criteria for cloud/shadow
and image date (year and
season), then for probability of
water and “no data”. QA
assessment provided a data
pool for selecting “best”
observation. All data pool
observations for each pixel
were then ranked based on NIR
values (band 4). The
observation with a band 4
value closest to the pixel’s
band 4 median was selected for
Potapov et al.
Standard practice for TOA
estimation (Chander et al.
2009). Normalized to MODIS
10-year surface reflectance
using band-wise mean bias.
Regression-based surface
anisotropic correction.
120 8881 Two epochs:
2000–2005 and
Custom (supervised
classification using
CART decision trees)
As per Potapov et al. (2011)
and described above.
Flood 2013 TM, ETM+Surface
1 Not reported.
March 2000 to
November 2012,
minimum of 3
images per season
Seasonal composites
(spring, summer,
autumn, winter)
Fmask (Zhu and
Woodcock 2012)
Medoid NDVI as a seasonally
representative value
Griffiths et al.
TM, ETM+Surface
LEDAPS 42 Y2000 =890;
Y2005 =1478;
Y2010 =1590
Three epochs: 2000,
2005, 2010.
Fmask (Zhu and
Woodcock 2012)
Weighted pixel-based scoring
system based on acquisition
year, acquisition day of year,
and distance of a given pixel to
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
VOL. 40, NO. 3, JUNE/JUIN 2014 195
overhead and imposes temporal limitations related to the op-
erational lifetime of a given satellite or sensor (i.e., precludes
applicability to the pre-MODIS era in this case). A more recent
trend is to combine observations from both TM and ETM+
that have been corrected to surface reflectance (Flood 2012;
Griffiths et al. 2013) using standardized and largely auto-
mated approaches such as LEDAPS (Masek et al. 2006;
Feng et al. 2013). Indeed, the United States Geological Survey
(USGS) now provides Landsat surface reflectance products as a
Climate Data Record (CDR)1, and the provision of these higher
level products could further reduce the amount of pre-processing
required to enable pixel-based compositing in the same way that
the standard Level-1 Terrain Corrected (L1T) products have en-
abled processing efficiencies (Hansen and Loveland 2012). L1T
products are 8-bit with a compressed file size of approximately
250 MB/file (Landsat-8 L1T files are 16-bit with a compressed
file size of 1GB/file). By contrast, CDR products are 16-bit, with
a file size of approximately 500 MB/file. Given the increased
file size of the CDR, there is a trade-off between download-
ing the larger CDR files via the internet, versus downloading
the smaller L1T files and running LEDAPS locally to generate
surface reflectance products.
The majority of compositing approaches detailed in Table 1
have relied on customized cloud detection algorithms, devel-
oped using supervised approaches such as decision tree classi-
fiers (Hansen et al. 2008; Roy et al. 2010; Potapov et al. 2011,
2012). As an alternate, a robust automated cloud detection al-
gorithm is now available that is being included into workflows
for compositing approaches (i.e., Fmask; Zhu and Woodcock
2012). It should be noted that Fmask is also now being used
in the production of CDR products (United States Geological
Survey 2013).
Hansen and Loveland (2012) posit that advances in pixel-
based image compositing signal the end of scene-based analysis
approaches, making way for progressively more novel opportu-
nities for large-area characterization and monitoring. This shift
from a scene-based perspective to a pixel-based perspective for
image understanding and processing is key, and certainly mir-
rors trends and developments in time series analysis approaches
(Kennedy et al. 2010). The full impact of this change from
scene-based to pixel-based analysis has yet to be realized. New
approaches of providing data to users based upon multiple spa-
tial and temporal considerations, rather than solely on image
temporal resolution and programmatic acquisition plans are an-
ticipated (e.g., area of interest driven rather than by path / row).
Further, a best-available pixel philosophy also creates opportu-
nities for fusion of multiple streams of image data from comple-
mentary sensors. For instance, given a sufficient level of similar-
ity between image data and processing levels, the best-available
pixels could come from different sensors. The Landsat series
of sensors, with similar viewing geometry, precise geolocation,
analogous band passes, and—critically—rigorous calibration
1 ECV.php
that enables reflectance conversion, have offered insights into
how measures from multiple satellites can be combined. The
ability to integrate measures from different satellite programs
in a virtual constellation would reduce reliance on individual
satellite programs and offer an increased temporal resolution.
The scheduled 2015 launch of Sentinel-2 will provide measures
that are analogous to those of the Landsat-8 Operational Land
Imager (OLI) and will provide an opportunity to develop and
test a virtual constellation. Sentinel-2 is designed to place two
satellites in orbit. A five day revisit will be realized with the
launch of the second Sentinel-2 satellite (Drusch et al. 2012)
and when combined with Landsat-8, a three day revisit cycle is
anticipated. Areas of persistent cloud, such as some locations
in the humid tropics, will remain a challenge for both Land-
sat and Sentinel, and although the increased revisit frequency
increases the total number of acquisitions, it does not guaran-
tee cloud-free observations (Kovalskyy and Roy 2013). Both
Landsat-8 OLI and Sentinel-2 are calibrated instruments, al-
lowing for conversion of at-satellite digital numbers to surface
reflectance. Data blending techniques (such as those after Gao
et al. 2006) also provide for the production of image values that
could be included in a compositing application. Based upon the
maturity of compositing algorithms and a growing suite of satel-
lites and sources of suitable imagery, operational programs built
upon these data streams are increasingly possible and appear
There are a wide range of information needs that drive na-
tional mapping and change detection efforts in Canada that pro-
vide the motivation for transparent and operational image com-
positing applications. The objective of this paper is to present
the context and current status of data and processing knowl-
edge that allow for these needs to be met in a novel fashion
through pixel-based image composites. We outline various ap-
proaches and considerations when developing pixel-based com-
posites over large areas and define a lexicon that describes the
different types of composites that may be generated and link
these back to the required information needs. To demonstrate
pixel-based compositing over large areas, we present an annual
BAP composite for Canada for the year 2010. Then, to further
demonstrate the potential for products to support times-series
applications, we present 15 years of Landsat data from 1998
to 2012 for the province of Saskatchewan and the island of
Newfoundland. We close with an exploration of potential appli-
cations and product directions and identify persistent issues and
research opportunities that remain.
Forest Ecosystem Monitoring in Canada: Information
In Canada, national forest data are needed to support a range
of science and program information needs, policy development,
and to fulfill national and international reporting obligations.
For example, Canada produces an annual State of the Forest
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
EL ´
ED ´
Key attributes for Canada’s NFI and Carbon Accounting
Land cover
Crown closure
Vo l u m e
Pre-disturbance land cover
Post-disturbance land cover
Disturbance agent
Disturbance year
Disturbance extent (area)
Disturbance intensity
report (Natural Resources Canada 2013), and reports on spe-
cific criteria and indicators for sustainable forest management
(Canadian Council of Forest Ministers 1995) and forest certifi-
cation. Internationally, Canada has reporting commitments as-
sociated with the United Nations FAO Forest Resource Assess-
ment program, The Montreal Process (Criteria and Indicators
for the Conservation and Sustainable Management of Temperate
and Boreal Forests), the United Nations Framework Convention
on Climate Change, and the United Nations Convention on Bi-
ological Diversity. These data must be timely, consistent, and
spatially exhaustive, and must enable assessment of trends over
time, as well as characterization of non-timber resources (Gillis
2001). Canada’s National Forest Inventory (NFI) is designed to
provide the necessary data to support, in part, the aforemen-
tioned information needs; however, the NFI is a sample-based
inventory, designed to survey a minimum of 1% of Canada’s
landmass (Gillis et al. 2005). There is a need to extend this in-
formation beyond the existing sample plots to provide spatially
explicit information that can support increasingly sophisticated
information and modelling requirements for both the NFI and
carbon accounting programs, particularly for Canada’s northern
forest areas where there is currently the greatest paucity of for-
est information (Wulder et al. 2004; Falkowski et al. 2009). The
attributes listed in Table 2 represent a subset of the full suite
of attributes required by the NFI and carbon accounting pro-
grams (Gillis et al. 2005; Wulder et al. 2004; Kurz et al. 2009).
This subset includes those attributes that are fundamental for
meeting programmatic information needs. For example, land
cover, crown closure, and height are necessary for the NFI to
distinguish “forest” from “other wooded land”—an important
distinction for reporting purposes (Gillis 2005). Some of the at-
tributes presented in Table 2 are readily obtained from remotely
sensed data (e.g., land cover, crown closure, and disturbance
related attributes), whilst in other cases, remotely sensed data
can be used to generate a proxy for the attribute of interest
(i.e., time since disturbance as a proxy for age (Helmer et al.
2010)). Other attributes can be modeled (i.e. height, volume,
and biomass), with the aid of an appropriate source of calibra-
tion data (e.g., Chen et al. 2012; Wulder et al. 2012). Of the
items listed in Table 2, species is the most difficult attribute to
determine using medium resolution remotely sensed data such
as Landsat (e.g. van Aardt and Wynne (2001)), although for-
est types (e.g., coniferous, deciduous) can be reliably mapped
(Wulder et al. 2007b).
A lexicon of pixel-based image composites
Composite type Typical compositing period Typical rule-base
Annual BAP Target DOY ±30 days (for a
single year)
1. DOY (relative to a target DOY, i.e., Aug 1)
2. Distance to cloud and cloud shadow
3. Sensor
4. Atmospheric opacity
Multi-year BAP For a given target year and target
DOY ±30 days (±1 or 2 years)
1. Year (relative to a target year)
2. DOY (relative to a target DOY, i.e., Aug 1)
3. Distance to cloud and cloud shadow
4. Sensor
5. Atmospheric opacity
Proxy value
Same as per annual BAP Annual BAP composite used as source (generated using
contextual rule-base as described above. Areas of no data or
anomolous values are assigned a proxy value by examining a
temporal trajectory of pixel values at the same or
neightbouring pixel locations.
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
VOL. 40, NO. 3, JUNE/JUIN 2014 197
FIG. 1. An overview of pixel-based image compositing methods used to generate prototype products for Saskatchewan and
A Lexicon for Pixel-Based Image composites
In overall terms, we propose three unique types of pixel-based
image composites: annual (single-year) composites, multi-year
composites, and proxy-value composites (Table 3). Annual
composites are surface reflectance composites that use the best
available pixel observation (from the target year) for any given
pixel location. Annual composites are produced using a set of
specified rules that are defined according to the information
need. For example, an annual composite may be designed to
capture a specific time period or a limited phenological window
(e.g., August 1 ±30 days). In addition to a day-of-year rule,
rules may also constrain observations according to sensor (e.g.,
preference for Landsat-5 over Landsat-7), distance to cloud and
cloud shadows, and atmospheric opacity (to reduce the impact
of haze). If there are no observations that satisfy the composit-
ing rules for a given pixel location, then the pixel is coded as
“no data” and as a result, annual composites may have areas of
missing data. Multi-year composites are generated according to
a set of user-specified rules; however, pixel observations from
previous or subsequent years may be used when no suitable
observation is found within a desired target year. Proxy value
composites are annual BAP composites where “no data” pixels
are populated using a time series of annual BAP composites
to determine proxy values. Likewise, pixels with anomalous
values—those that exceed a pre-defined range of expectation
or which have opacity values indicative of hazy imagery—may
also be assigned a proxy value. In essence, the objective of
the proxy value composite is to assign the most similar value
in time and space to a pixel that either has “no data” or has
an anomalous value. Ideally, a pixel’s full temporal trajectory
is used to determine whether the pixel is stable over time (i.e.,
within an expected range of spectral values), or whether spectral
change has occurred beyond a specified range of expectation.
If a pixel has a stable trajectory, the infilling of missing val-
ues or resetting of anomalous values through the averaging (or
some other method) of pixel values in the pixel’s trajectory is
possible. If the pixel’s value is not stable through time, proxy
values will have to be determined in accordance with spectral
change events and (possibly) by examining the pixel’s immedi-
ate spatial neighbourhood in the year of missing data. Detection
of change events in the pixel’s spectral trajectory is a necessary
precursor to the generation of proxy value composites.
An overview of the pixel-based imaging composite methods
used to develop annual BAP composites is provided in Figure 1;
details follow in the subsequent sections.
Study Area
Approaching 10 million km2in area, Canada is a large
nation with a gradient in ecosystem productivity that is in-
fluenced by latitude and precipitation (Hofgaard et al. 1999)
(Figure 2). Forested ecosystems represent over 60% of Canada’s
land area (Wulder et al. 2008) and ninety-three percent of
Canada’s forests are publicly owned (Wulder et al. 2007a), with
the capacity to harvest and process timber allocated through
tenure agreements to the private sector (Haley and Luckert
1990). As a complex mosaic of trees, wetlands, and lakes, the
forests of Canada represent 10% of global forests, and offer
a range of ecosystem services that are economic, ecological,
and sociocultural in nature. Canada’s terrestrial area is repre-
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
EL ´
ED ´
FIG. 2. Study area: Canada, Saskatchewan, and Newfoundland.
sented by 1285 unique WRS-2 path/rows and approximately
10.7 billion 30 x 30 m pixels. Saskatchewan is the seventh
largest province or territory in Canada with an area of approx-
imately 651,900 km2. More than half of this area is forest,
with wildfire as the dominant disturbance agent (Saskatchewan
Ministry of Environment 2012). Saskatchewan is comprised of
approximately 702 million 30 ×30 m pixels (with 71 WRS-2
path/row locations), and has a diverse forest assemblage and a
latitudinal gradient. Newfoundland has an area of approximately
108,860 km2with 46% of this area considered forest (Depart-
ment of Forest Resources and Agrifoods 2003). Newfoundland
is the most eastern extent of the Boreal Shield Ecozone and its
forests are typically composed of small trees that are primar-
ily coniferous species, with the dominant species being Black
spruce (Picea mariana [Mill.] B.S.P.). Newfoundland is an is-
land and is comprised of approximately 122 million 30 ×30 m
pixels (with 21 WRS-2 path/row locations).
Candidate images for producing an annual composite for a
given year were selected from the Landsat metadata archive
based on distance to a specified target day of year (DOY) and
percent cloud cover. Images acquired within ±62 days of the
target DOY with less than 70% cloud cover were considered
viable candidates for compositing. A target DOY of August
1 (Julian day 213) was selected for its likelihood of being
within the growing season for the majority of Canada’s ter-
restrial landscape (McKenney et al. 2006). In Canada’s north
(that is, >60N) the date range for candidate images was fur-
ther restricted to ±30 days, to minimize the presence of snow
cover. Image overlap is significant in the north (i.e. 85% at
80N; Wulder and Seemann 2001) increasing the likelihood
of cloud-free observations despite the more limited date range
for candidate images. When cloud cover exceeds 70%, images
are difficult to geometrically correct due to obscured ground
control points (White and Wulder 2013). Identified candidate
images were downloaded from the USGS Landsat archive as
L1T products; L1T products are systematically corrected for ra-
diometric, geometric, and terrain distortions (Irons et al. 2012).
A master list of all candidate images for the area of interest is
compiled that contains a user-generated unique numeric identi-
fier for each image, as well as the unique Landsat scene identifier
(i.e., LT50350232001211PC00). This list is used as a look-up
table (LUT) to track the source image that is used for each pixel
in order to support the development of image composites.
The L1T candidate images are pre-processed using Fmask
(version 2.1), an object-based algorithm designed to identify
clouds and cloud shadows, as well as clear land and water pixels,
snow, and areas of no data (Zhu and Woodcock 2012). The
Landsat Ecosystem Disturbance Adaptive System (LEDAPS;
version 1.3.0; Schmidt et al. 2013) is used to generate surface
reflectance values for use in the final image composite. LEDAPS
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
VOL. 40, NO. 3, JUNE/JUIN 2014 199
produces top-of-atmosphere (TOA) reflectance from Landsat
TM and ETM+digital numbers (DN) and applies atmospheric
corrections to generate a surface reflectance product (Masek
et al. 2006). LEDAPS atmospheric corrections are based on the
Second Simulation of a Satellite Signal in the Solar Spectrum
(6S) radiative transfer model (Vermote et al. 1997).
Calculate Pixel Scores
Four scores were calculated for each pixel: sensor score, day
of year score, distance to cloud or cloud shadow score, and
opacity score. The sensor and DOY score were calculated at
the image level (i.e., all pixels within the image receive the
same score), whilst the cloud/cloud shadow and opacity scores
were unique to each pixel. All scores were then summed to
provide a total score for each pixel, and the pixel with the largest
score (i.e., the BAP) was used in the image composite.
Sensor Score
In order to give preference to images captured by the Landsat
TM sensor and mitigate the impact of spatial gaps associated
with Landsat 7 ETM+SLC-off data, a sensor score is assigned
to the pixels in each image acquired after May 31, 2003 as
follows: pixels from Landsat TM images were assigned a score
of 1; pixels in Landsat ETM+images were assigned a score of
0.5. Pre-2003, both sensors received a score of 1.
Day of Year Score
A score was assigned to all pixels in an image according to
the DOY the image was acquired relative to the target DOY. A
DOY score was assigned to all pixels in an image as per Griffiths
et al. (2013):
ScoreDOY =1
where μand σdenote the mean and standard deviation respec-
tively of all the image DOYs, and xiis the DOY for the image
being assessed. In our processing, μwas forced to our target
DOY (August 1 or Julian day 213) and the standard deviation
was set to 38 by examining the distribution of DOYs for all
candidate images in the two prototype regions of Saskatchewan
and Newfoundland. The DOY score is scaled to a value between
0 and 1 by dividing by the maximum score.
Distance to Cloud or Cloud Shadow Score
Using the outputs from Fmask, a distance to cloud or cloud
shadow score is assigned, whereby pixels identified as clouds
or cloud shadows are assigned “no data” value and any pixel
located at a distance greater than 50 pixels from an identified
cloud or cloud shadow pixel is assigned a score of 1. Pixels that
are not identified as clouds or cloud shadows and that are less
than 50 pixels away from clouds and cloud shadows are assigned
a score between 0 and 1 using Equation (2) as per Griffiths et al.
ScoreCloudDistance =1
1+e0.2min(Di,Dreq)DreqDmi n
2 [2]
where Diis the pixel’s distance to cloud or cloud shadow, Dreq
is the minimum required distance (i.e., 50 pixels), and Dmin is
the minimum distance of the given pixel observations (i.e., 0
Opacity Score
LEDAPS applies the dark dense vegetation method of Kauf-
man et al. (1997) to estimate aerosol optical thickness (AOT)
directly from the imagery and the AOT is one of the inputs used
in the 6S radiative transfer model (Ju et al. 2012). The LEDAPS
surface reflectance output includes an AOT map derived from
the Landsat TM or ETM+blue band (Masek et al. 2006), here-
after referred to as opacity. In general, opacity values less than
0.1 are considered clear, opacity values between 0.1 and 0.3
are average, and opacity values greater than 0.3 are considered
hazy2. Since hazy images can confound the generation of qual-
ity image composites, an opacity score was calculated using the
atmospheric opacity band output by LEDAPS. Pixels with an
opacity value <0.2 were assigned a score of 1 and pixels with
an opacity value >0.3 were labelled as “no data”. Pixels with
opacity values 0.2 and <0.3 were assigned a score between
0 and 1 using Equation (3).
ScoreOpaci ty =11
1+e0.2min(Oi,Omax )Omax Omi n
where Oiis the pixel’s opacity value, Omax is the maximum
opacity value (i.e., 0.3), and Omin is the minimum opacity value
(i.e., 0.2).
Pixel-Based Image Compositing
The objective of the composite processing is to populate
the final image composite with the surface reflectance value
from the BAP for each pixel in the composite. To develop the
pixel-based image composites, the summed score rasters for all
available L1T candidate images are evaluated to determine the
image with the maximum score for each pixel location. This is
accomplished in two stages: first, the BAP for each pixel in each
WRS-2 path/row is identified, and then the overall area of inter-
est is considered, at which time overlap with adjacent path/row
locations also figures into the determination of the BAP. Once
the final BAPs are identified, the surface reflectance data is pro-
jected from its source UTM projection to a standard national
Lambert Conformal Conic projection using cubic convolution,
to populate the final composite.
2 Tables.php
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
EL ´
ED ´
Annual BAP Composites
A national, annual BAP composite was generated for the tar-
get year 2010. This composite used Landsat TM and ETM+
imagery from 2010. For this national annual BAP composite,
we assessed the observation yield (the number of BAP obser-
vations from within the target range of ±30 days of August 1),
distance to target DOY, and prevalence and spatial distribution
of missing data (i.e., data >±30 days from August 1). In addi-
tion, annual composites were generated for Saskatchewan and
Newfoundland to demonstrate the image compositing approach
for time series applications. We assessed the observation yield
and evaluated the number of consecutive years where the an-
nual composites had missing data. Newfoundland is, by nature
of its geography, prone to clouds and our expectation was that
the production of annual composites in this area would result in
more areas of “no data”.
The radiometric consistency of the annual BAP composites
for Saskatchewan and Newfoundland were assessed by select-
ing a single path/row in each area, identifying a target year, and
withholding the most cloud-free image in each case as a refer-
ence image. For Saskatchewan we used WRS-2 Path 38, Row
22 and generated an annual BAP composite for 2001 using 18
unique images with DOY ranging from July 10 to August 27.
The reference image was a Landsat 7 ETM+image from Au-
gust 3, 2001. For Newfoundland, we used Path 3, Row 26 and
generated an annual BAP composite for 2000 using 8 unique
images with DOY ranging from July 1 to August 27. The ref-
erence image was a Landsat 7 ETM+image from August 27,
2000. The remaining set of images was used to build an an-
nual BAP composite for the same path/row locations using the
aforementioned scoring system. Samples of pixels (n =500)
were then randomly selected from areas of dense forest for
each of these two scenes, constrained by a DOY of August 1
±30 days. Dense forest areas were identified using a combina-
tion of circa 2000 land cover classification (Wulder et al. 2008)
and an NDVI threshold of 0.9. Values for surface reflectance
were extracted at the selected sample locations from both
the annual BAP composites and the reference images and the
strength of the band-wise correspondence between reflectance
values was evaluated using the coefficient of determination
National 2010 Annual BAP Composite
We generated a national 2010 annual BAP composite for
Canada (Figure 3a) using the aforementioned rule-base and as-
sessed the observation yield relative to our target DOY, August
1 (Figure 3b). Approximately 17% of pixels do not have BAP
observations within ±30 days of our target DOY; however, if
the date range is expanded to ±45 days, then only 4% of pixels
have no BAP observations (Figure 4). There are some areas in
the country (i.e., Rocky Mountains, central Saskatchewan, east-
ern Newfoundland), where pixels were obscured with persistent
cloud in 2010 and yielded no cloud-free observations within
a±62 day window of August 1; however, this accounts for
less than 1% of pixels. To fill in these areas would require the
development of either a proxy BAP composite or a multi-year
BAP composite (Table 3). We generated a multi-year composite
(Figure 5) wherein pixels with no observations within the ±
30 day window for 2010 were populated with BAP observa-
tions from 2009 and 2011.
Annual BAP composites for Saskatchewan
and Newfoundland: 1998–2012
BAP Observation Yield
For the annual image composites in Saskatchewan, 29% of
pixels had observations from within ±30 days of August 1 for
all 15 years considered, whilst 74% of pixels have 13 or more
years of observations within the target date range. All pixels in
Saskatchewan had at least seven years of observations within
the target date range (Figure 6a). Approximately 86% of pixels
in Saskatchewan have no consecutive years of missing data (i.e.,
they are either missing only 1 or no years of data) (Figure 7a).
Of the 14% of pixels with consecutive years of missing data,
approximately 11% have two consecutive years of missing data,
whilst 3% of pixels have 3 or more consecutive years of missing
data. The geometric fidelity of Landsat imagery and the SLC-
off failure are major contributors to this 14% of pixels that have
several consecutive years of missing data, since this situation is
more common after 2003 (the year of the SLC failure), as the
SLC-off no-data gaps recur spatially.
The observation yield from the annual composites for the
island of Newfoundland contrasts with that in Saskatchewan:
less than 1% of pixels had observations for 14 or more years
within ±30 days of August 1, whilst 52% of pixels have 10
or more years of observations within the target date range (Fig-
ure 6b). All pixels in Newfoundland had at least three years of
observations within the target date range. Only about 18% of
pixels in Newfoundland have no consecutive years of missing
data, whilst 82% are missing two or more consecutive years of
observations (Figure 7b). Similar to the results of our analysis
in Saskatchewan, more pixels in Newfoundland had fewer con-
secutive years of missing data before 2003: 62% of pixels had
no consecutive years of missing data before 2003, compared to
only 24% of pixels after 2003.
Assessment of the Radiometric Consistency of Annual BAP
Our assessment of the radiometric consistency of our an-
nual BAP composites is summarized in Figure 8. The sur-
face reflectance values in the annual BAP composite for both
Saskatchewan and Newfoundland had strong correspondence
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
VOL. 40, NO. 3, JUNE/JUIN 2014 201
FIG. 3. 2010 annual best available pixel (BAP) composite (A) using August 1 ±30 days. Approximately 17% of pixels have no
observations within that date range. The distance to target day of year (DOY; B) indicates the DOY distribution; less than 1% of
pixels have no observations (no data).
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
EL ´
ED ´
FIG. 4. Observation yield for 2010 annual BAP composite for Canada.
with the single-date reference images in the near- and mid-
infrared regions (bands 4, 5, 7), with R2>0.79. Correspondence
was lower in the visible wavelengths (bands 1, 2, and 3), with
R2ranging from 0.60 to 0.75—likely as a result of atmospheric
effects, which are more prevalent in these wavelengths. More-
over, despite the greater likelihood of clouds and cloud shadows
in Newfoundland, which further limits the selection of the BAP
and reduces the observation yield (Figures 6 and 7) compared to
Saskatchewan, the Newfoundland composites had greater corre-
spondence in the visible bands. Using a stable target (i.e., dense
forest), the assessment of the annual BAP composites relative to
an independent, single-date image indicates that no systematic
artifacts are being introduced in the compositing process.
Pixel-Based Compositing Over Large Areas
A pixel-based image compositing approach has been pro-
posed to address information needs associated with forest
ecosystem science and monitoring in the Canadian context.
Herein, we demonstrate the approach over large areas by pre-
senting an annual BAP composite for Canada generated with
Landsat data for the year 2010 and annual BAP time-series
from 1998 to 2012 for the province of Saskatchewan and the
island of Newfoundland. The compositing approach we present
is inherently flexible, and the rules used for compositing can,
in theory, be adjusted to accommodate a range of information
needs (Table 4). The parametric scoring mechanisms for se-
lecting BAP observations can be adjusted to give more or less
weight to particular sensors, DOY, or the influence of clouds
and haze according to specific information needs or application
requirements, and this flexibility is one of the key advantages of
the scoring mechanism used (Griffiths et al. 2013).
As indicated in our methods, we consider images acquired
within ±62 days of the target DOY with less than 70% cloud
cover as viable candidates for compositing. However, given our
knowledge of phenology and growing season length for much
of the forested area of Canada, we limit our final composites
to BAP observations that are within ±30 days of our target
DOY. We assign all BAPs from outside this temporal window
as “no data”. We would then attempt to fill these “no data” pixels
with proxy values derived using a time series approach or by
generating a multi-year composite (Figure 5). The majority of
BAP observations in the example presented in Figure 9 came
from a Landsat-5 TM image acquired on day 210 (July 29), 2010
(LT50440202010210EDC00; Figure 9a). Pixels with clouds and
cloud shadows in this image were identified with Fmask, and a
distance to cloud score was assigned. In the final BAP composite
(Figure 9b), these pixels were filled using observations from a
number of different Landsat scenes (Figure 9c). Observations
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
VOL. 40, NO. 3, JUNE/JUIN 2014 203
FIG. 5. Multi-year BAP composite for Canada incorporating pixels from years 2009, 2010, 2011.
from images acquired outside the August 1 ±30 day window
(Figure 9d) are visibly different from the surrounding pixels,
clearly demonstrating the need to constrain the DOY range.
Characteristics relevant to particular geographic regions,
such as persistence of cloud cover, topography, dynamism of
landscape processes, phenology, and Landsat data availability,
are important considerations when applying a compositing ap-
proach. Likewise, different information needs (e.g., disturbance
mapping, estimation of biophysical parameters) may dictate
different compositing strategies, target dates, and compositing
rules that are specific to the application (Table 4). Usually, there
are trade-offs to be made in producing composite products. In
general, it is desirable to generate composites that have consis-
tent phenology with minimal “no data” pixels that best represent
the phenomenon of interest (e.g., land cover classes, forest struc-
tural attributes, or specific disturbance events). These trade-offs
may be location or application specific. For example, in north-
ern Canada, a narrower DOY range for composite development
is required to capture the shorter growing season in this area.
Although a narrow date range in theory would reduce the proba-
bility of obtaining a cloud-free observation, the 80% overlap be-
tween Landsat scenes in the north (Wulder and Seemann 2001)
ensures sufficient observations for annual BAP composites at
this latitude. To enable the characterization of discrete distur-
bance events such as wildfires, it may be preferable to have a
DOY range that is referenced to the end of the fire season (Ta-
ble 4). Similarly, to characterize disturbances related to insects,
the DOY range would very much depend on the type of dam-
age requiring detection (i.e., a defoliator versus a bark beetle).
An example of a program-specific information need is that of
the National Deforestation Monitoring System (NDMS), which
uses remotely sensed data to help create an estimate of anthro-
pogenic change from a forest to a non-forest land use. NDMS
is a sample-based program that produces national estimates of
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
EL ´
ED ´
FIG. 6. The observation yield of BAP pixels that provide the basis for composite development for Saskatchewan (A) and
Newfoundland (B).
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
VOL. 40, NO. 3, JUNE/JUIN 2014 205
FIG. 7. The number and spatial distribution for consecutive years of missing data for Saskatchewan (A) and Newfoundland (B).
deforestation for a variety of clients; however, the mapping pro-
cess could benefit from increased automation. The detection of
possible deforestation events requires two core dates of imagery,
typically bracketing a five year time period. The mapping pro-
cess could be performed more efficiently by having access to
two national annual BAP composites, which would allow for
the display of various band combinations and the generation of
a change enhanced image between the two composite years.
Fire is the primary natural disturbance agent in most boreal
forests (Bond-Lamberty et al. 2007), and high inter-annual vari-
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
EL ´
ED ´
FIG. 8. Quality assessment of annual BAP composite for a
sample path/row in Saskatchewan (p38r22) and Newfoundland
(p3r26). For a randomly selected set of 500 pixels, surface
reflectance values from a reference image are plotted against
BAP composite surface reflectance values (reflectance values
are scaled by 10000).
ability in the area burned, combined with large insect outbreaks,
can have a profound influence on the greenhouse gas balance of
Canada’s forests (Kurz et al. 2008). Monitoring these types of
natural disturbances requires acquisition of imagery before and
after the disturbance to best detect spectral response changes
that are in turn used to derive the spatial location, area, and
magnitude of the disturbance. Fire is a random event that can
occur over a broad temporal window from spring to late fall.
Presently, a national annual map of areas burned in Canada
is generated by compiling the best available information on
burned areas from a variety of data sources such as Landsat TM
or ETM+imagery, SPOT VEGETATION, as well as data from
provincial/territorial government fire management agencies (de
Groot et al. 2007; Natural Resources Canada 2014). When the
presence of cloud or smoke prevents the selection of a Landsat
image, the burn polygon is extracted from fire agency data which
may or may not be as accurate as what could be mapped from
Landsat. Implementation of the BAP concept could potentially
increase the likelihood of having a suitable cloud-free image for
mapping burned areas.
The mapping of severe insect disturbances using remotely
sensed data is an application that is highly time sensitive since
the manifestation and detectability of damage varies by causal
agent and time of year (Hall et al. 2006; Hicke et al. 2012).
Particularly with insect defoliation, the opportunity to acquire
cloud-free Landsat images may be confined to two or three satel-
lite passes in the growing season (Hall et al. 2006). This very
narrow temporal window can limit the use of optical satellite
imagery for operational mapping of defoliation events. The po-
tential to create cloud-free BAP composites from Landsat and
Sentinel-2, with a three and five day revisit cycle could greatly
increase the prospects of having suitable pre- and post-outbreak
images for mapping of insect disturbances.
The most versatile compositing approach would have the en-
tire Landsat image archive of Canada accessible on demand.
Then, for any given information need, a customized rule-base
could be constructed and implemented to create a suitable BAP
composite. In this way, users bring their algorithms (and their
information needs) to the data, rather than the opposite, where
multiple users are downloading and replicating the Landsat
archive of Canada in order to generate data products inde-
pendently to support their information needs. Examples of this
capacity include the NASA Earth Exchange (NEX) (Nemani
et al. 2011), Google Earth Engine (Hansen et al. 2013), and
Geoscience Australia (Geoscience Australia 2014). While such
versatility is the ultimate goal, initial efforts are required to
demonstrate pixel-based compositing over large areas, evaluate
annual pixel-based composite products, and assess the potential
for annual pixel-based composite products to support time-series
Issues and Opportunities
Canada’s large size and geographic complexity pose signifi-
cant challenges for a pixel-based compositing approach. Primar-
ily, the probability of cloud-free observations is a complex com-
bination of regional weather conditions, acquisition scheduling,
and historic downlink activity, among other issues (White and
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
VOL. 40, NO. 3, JUNE/JUIN 2014 207
Examples of different information requirements and associated compositing criteria
Application area Date range Rule-base
Land cover Aug 1 ±30 days Sensor
Day of year
Distance to cloud/cloud shadow
Forest structural attribution
(height, volume, biomass)
Aug 1 ±30 days Sensor
Day of year
Distance to cloud/cloud shadow
Deforestation monitoring Sep 1 +30, 60 days Sensor
Day of year
Distance to cloud/cloud shadow
Pixel temporal grouping
Wildfires Nominally Sensor
Aug 20 to Oct 10 Day of year
Pre- and post-burn image dates are
influenced by early or late season
Distance to cloud/cloud shadow
Date of end fire date
Insects Variable depending on insect Sensor
Day of year
Distance to cloud/cloud shadow
Type of damage: eg., defoliator vs bark
beetle based on agency aerial survey
Wulder 2013). When it comes to scheduling, the conterminous
United States (CONUS) has preferential data acquisition under
the Landsat program (Kovalskyy and Roy 2013), with approxi-
mately 831,318 images for 459 unique path/rows for the period
1972–2012. For that same time period, Canada, as a strong In-
ternational Cooperator in the Landsat program and with close
proximity to CONUS, has approximately 605,000 images in the
archive for 1224 unique path/rows (White and Wulder 2013).
With this lower total number of images also comes a lower prob-
ability of cloud-free imagery. Given the large size of Canada and
the need to monitor terrestrial ecosystems in a way that is sys-
tematic, consistent, transparent, and repeatable, BAP composit-
ing offers the potential to increase the availability of cloud-free
EO observations at a spatial resolution that can support monitor-
ing requirements, however, the challenge of acquiring complete
spatial and temporal coverage over large and geographically
complex areas remains. With Landsat-8, a more globally inclu-
sive data acquisition plan has been implemented that captures
the majority of terrestrial WRS-2 path/rows within the satellite’s
orbit each day (Roy et al. 2014).
Through our product development, it is clear that the ob-
servation yield for our standard rule-base varies markedly in
different regions of the country. Newfoundland is more prone to
cloud cover, resulting in lower annual observation yield and the
BAP compositing approach is most challenging in environments
that are prone to cloud cover and haze. In such circumstances,
leveraging the time series of Landsat is an important consid-
eration. Saskatchewan may have fewer issues associated with
cloud cover, but has a very dynamic natural disturbance regime,
which is also challenging if the information need is to charac-
terize discrete disturbance events.
The issue of “missing data” (i.e., pixels that have no BAP
observations within the specified DOY range) requires spe-
cific attention. This is especially true for areas where cloud
and haze conditions limit opportunities for pixels to meet the
criteria for consideration as BAPs. These data gaps limit ap-
plications that require spatially exhaustive and location-specific
coverage. Therefore, alternate methods are required to populate
missing data. As presented in the lexicon for pixel-based im-
age composites, the BAP approach could produce multi-year
composite products or proxy-value composites depending on
the information needs of the application. Multi-year compos-
ite methods, while simpler to develop, might be suitable for
generating products for national reporting for specific time win-
dows, while more complex methods for developing proxy-value
products are needed for annual monitoring and modeling appli-
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
EL ´
ED ´
FIG. 9. Areas of cloud in a single-date image (LT50440202010210EDC00) (A) are replaced with cloud-free observations in the
annual BAP composite for 2010 (B). The composite uses observations from several unique images (C), with the greatest difference
visible for those images acquired outside the August 1 ±30 day window (D).
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
VOL. 40, NO. 3, JUNE/JUIN 2014 209
cations. In such cases, detection of change events in the pixel’s
spectral trajectory is a necessary precursor to the generation
of proxy-value composites. For the development of proxy
composites, the distribution of BAP observations within the
15-year period is also important: theoretically it is simpler to
assign proxy values to single years of missing data interspersed
in a time series of observations, than it is to assign proxy values
to five consecutive years of missing data. Further development
is needed in this area for operational implementation over large
Herein, we focused on forest ecosystem science and moni-
toring needs of the Canadian Forest Service. The goal was to
create satellite-based image composite products that could ex-
pand the current sample-based NFI monitoring approach and
provide spatially explicit information on forest dynamics for
carbon accounting as well as a myriad of other information
needs in Canada that would benefit from wall-to-wall image
and data products with an annual time stamp. In the case of
the NFI, satellite-based analyses are needed when plot density
or temporal coverage is insufficient to meet the monitoring and
reporting requirements at regional and national levels, such as
can occur in northern Canada (Falkowski et al. 2009). For the
products presented, we selected compositing parameters to sup-
port key forest ecosystem monitoring needs— characterization
of land cover and land cover change, and attribution of forest
structural parameters. Further research is needed to optimize
the parameters for more specific applications. For example, an
alternate DOY or narrower phenological window for identifying
BAPs may result in improved land cover classification accuracy
or improved predictions of structural parameters at the expense
of areal coverage (i.e. more missing data). For time sensitive
applications such as monitoring disturbances (e.g. fire, insect
defoliation), imagery is required following the disturbance. Of-
ten, there is a very small window for detection and the optimal
timing can vary by disturbance type. Capacity to generate prod-
ucts for such applications is currently limited by the frequency
with which images are collected, but this limitation may be
addressed with future sensors, with especial interest in the ca-
pacity of combined Sentinel-2 and Landsat 8 applications. Both
Sentinel 2’s remain to be launched and the complementarity of
the measurements yet to be determined, as well as the nature of
the data access. Opportunities for virtual constellations notwith-
standing, continuity of Landsat measures remains of the highest
priority, and the integration of Landsat-8 and Sentinel-2 obser-
vations into the BAP compositing process is an issue requiring
further research.
Herein we have made some recommendations concerning
how to best link compositing decisions to the desired use of
the composite; however, further research may help to optimize
these decisions across a broader range of applications. While
our efforts to date have focused on information needs of the
Canadian Forest Service (specifically, NFI and carbon account-
ing), monitoring, reporting, and science-based policy support
are core activities of several other federal agencies in Canada.
Each organization has sector-specific objectives associated with
terrestrial monitoring, and while these objectives can often be
met using the same or similar data, in some cases, optimization
may improve product suitability for more specific information
requirements (e.g. wetlands and agricultural monitoring).
The approach and products presented in this paper demon-
strate that annual BAP compositing is possible for large areas
and for time series application in the Canadian context. This is
the first time that an annual BAP composite has been generated
from Landsat imagery for all of Canada. In addition, annual
BAP time-series extending from 1998 to 2012 were produced
for Saskatchewan and the island of Newfoundland to support
the further development of ecosystem monitoring applications.
An important limitation for the application of annual BAP prod-
ucts to derive forest ecosystem information products (i.e., land
cover, land cover change, and structural estimation) is missing
data—pixels with no observations due to clouds, cloud shad-
ows, sensor issues (SLC-off), or restricted acquisition period. In
the future, it is expected that these limitations will be mitigated
with increased coverage from future satellite sensors. However,
to address historical and current information needs, further re-
search is necessary to develop multi-year and BAP proxy-value
composites targeted to specific monitoring applications. Future
research will focus on developing approaches for using a time
series of BAP proxy-value composites to generate information
products for land cover, land cover change, and forest structure.
This research was undertaken as part of the “National Ter-
restrial Ecosystem Monitoring System (NTEMS): Timely and
detailed national cross-sector monitoring for Canada” project
jointly funded by the Canadian Space Agency (CSA) Govern-
ment Related Initiatives Program (GRIP) and the Canadian For-
est Service (CFS) of Natural Resources Canada.
Bond-Lamberty, B., Peckham, S. D., Ahl, D. E., and Gower, S. T. 2007.
Fire as a dominant driver of central Canadian boreal forest carbon
balance. Nature Letters, Vol. 450: pp. 89–92.
Broich, M., Hansen, M. C., Potapov, P., Adusei, B., Lindquist, E., and
Stehman, S. V. 2011. Time series analysis of multi-resolution optical
imagery for quantifying forest cover loss in Sumatra and Kalimantan,
Indonesia. International Journal of Applied Earth Observation and
Geoinformation, Vol. 13: pp. 277–291.
Canadian Council of Forest Ministers. 1995. Defining sustainable for-
est management: A Canadian approach to criteria and indicators.
Available online:
Chander, G., Markham, B. L., and Helder, D. L. 2009. Summary of
current radiometric calibration coefficients for Landsat MSS, TM,
ETM+and EO-1 ALI sensors. Remote Sensing of Environment,Vol.
113: pp. 893–903.
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
EL ´
ED ´
Chavez, P. S. 1988. An improved dark-object subtraction technique
for atmospheric scattering correction of multispectral data. Remote
Sensing of Environment, Vol. 24(No. 3): pp. 459–479.
Chen, G., Wulder, M. A., White, J. C., Hilker, T. H., and Coops,
N. C. 2012. Lidar calibration and validation for geometric-optical
modeling with Landsat. Remote Sensing of Environment, Vol. 124:
pp. 384–393.
Cihlar, J., Manak, D., and D’Iorio, M. 1994. Evaluation of composit-
ing algorithms for AVHRR data over land. IEEE Transactions on
Geoscience and Remote Sensing, Vol. 32(No. 2): pp. 427–436.
de Groot, W. J., Landry, R., Kurz, W. A., Anderson, K. R., Englefield,
P., Fraser, R., Hall, R. J., Raymond, D., Decker, V., Lynham, T. J.,
Banfield, E., and Pritchard, J. M. 2007. Estimating direct carbon
emissions from Canadian wildland fires. International Journal of
Wildland Fire, Vol. 16(No. 5): pp. 593–606.
Department of Forest Resources and Agrifoods. 2003. Provincial
sustainable forest management strategy. Government of New-
foundland and Labrador. Available online:
Drusch, M., Del Bello, U., Carlier, S., Colin, O., Fernandez, V., Gascon,
F., Hoersch, B., Isola, C., Laberinti, P., Martimort, P., Meygret, A.,
Spoto, F., Sy, O., Marchese, F., and Bargellini, P. 2012. Sentinel-2:
ESA’s optical high-resolution mission for GMES operational ser-
vices. Remote Sensing of Environment, Vol. 120: pp. 25–36.
Du, Y., Cihlar, J., Beaubien, J., and Latifovic, R. 2001. Ra-
diometric normalization, compositing, and quality control for
satellite high resolution image mosaics over large areas. IEEE
Transactions on Geoscience and Remote Sensing, Vol. 39:
pp. 623–634.
Falkowski, M. J., Wulder, M. A., White, J. C., and Gillis, M. D. 2009.
Supporting large-area, sample-based forest inventories with very
high spatial resolution satellite imagery. Progress in Physical Geog-
raphy, Vol. 33(No. 3): pp. 403–423.
Feng, M., Sexton, J. O., Huang, C., Masek, J. G., Vermote, E. F., Gao,
F., Narasimhan, R., Channan, S., Wolfe, R. E., and Townshend, J. R.
2013. Global surface reflectance products from Landsat: Assessment
using coincident MODIS observations. Remote Sensing of Environ-
ment, Vol. 134: pp. 276–293.
Flood, N. 2013. Seasonal composite Landsat TM/ETM +images using
the Medoid (a multi-dimensional median). Remote Sensing,Vol.
5(No. 12): pp. 6481–6500.
Flood, N., Danaher, T., Gill, T., Gillingham, S. 2013. An operational
scheme for deriving standardised surface reflectance from Landsat
TM/ETM+and SPOT HRG Imagery for Eastern Australia. Remote
Sensing, Vol. 5: pp. 83–109.
Gao, F., Masek, J., Schwaller, M., Hall, F. 2006. On the blending of the
Landsat and MODIS surface reflectance: predicting daily Landsat
surface reflectance. IEEE Transactions on Geoscience and Remote
Sensing, Vol. 44: pp. 2207–2218.
Geoscience Australia. 2014. The future of the Landsat archive. Avail-
able online:
Gillis, M. D. 2001. Canada’s National Forest Inventory (responding to
current information needs). Environmental Monitoring and Assess-
ment, Vol. 67: pp. 121–129.
Gillis, M. D., Omule, A. Y., and Brierley, T. 2005. Monitoring Canada’s
forests: The National Forest Inventory. The Forestry Chronicle,Vol.
81: pp. 214–221.
Griffiths, P., van der Linden, S., Kuemmerle, T., and Hostert, P. 2013. A
pixel-based Landsat compositing algorithm for large area land cover
mapping. Journal of Selected Topics in Applied Earth Observations
and Remote Sensing, Vol. 6(No. 5): pp. 2088–2101.
Guindon, B., and Edmonds, C. M. 2002. Large-area land cover
mapping through scene-based classification compositing. Pho-
togrammetric Engineering and Remote Sensing, Vol. 68: pp. 589–
Gutman, G., Huang, C., Chander, G., Noojipady, P., and Masek, J.
G. 2013. Assessment of the NASA-USGS Global Land Survey
(GLS) datasets. Remote Sensing of Environment, Vol. 134: pp. 249–
Haley, D. and Luckert, M. K. 1990. Forest tenures in Canada: A frame-
work for policy analysis. Ottawa, Forestry Canada: Information Re-
port E-X-43, 104 p.
Hall, R. J., Skakun, R. S. Skakun, and Arsenault, E. J. 2006. Remotely
sensed data for mapping insect defoliation. pp. 85–111 in M.A.
Wulder and S.E. Franklin (editors). Forest Disturbance and Spatial
Pattern: Remote Sensing and GIS Approaches. Taylor and Francis.
CRC Press.
Hansen, M. C., and Loveland, T. R. 2012. A review of large area
monitoring of land cover change using Landsat data. Remote Sensing
of Environment, Vol. 122: pp. 66–74.
Hansen, M. C., Potapov, P. V., Moore, R., Hancher, M., Turubanova, S.
A., Tyukavina, A., Thau, D., Stehman, S. V., Goetz, S. J., Loveland,
T. R., Kommareddy, A., Egorov, A., Chini, L., Justice, C. O., and
Townshend, J. R. G. 2013. High resolution global maps of 21st-
century forest cover change. Science, Vol. 342: pp. 850–853.
Hansen, M. C., Roy, D. P., Lindquist, E., Adusei, B., Justice, C. O.,
Alstatt, A. 2008. A method for integrating MODIS and Landsat
data for systematic monitoring of forest cover and change in the
Congo Basin. Remote Sensing of Environment, Vol. 112(No. 5): pp.
Helmer, E. H., Ruzycki, T. S., Wunderle, J. M., Vogesser, S., Ruefe-
nacht, C., Kwit, C., Brandeis, T. J., and Ewert, D. N. 2010. Mapping
tropical dry forest height, foliage height profiles and disturbance
type and age with a time series of cloud-cleared Landsat and ALI
image mosaics to characterize avian habitat. Remote Sensing of En-
vironment, Vol. 114: pp. 2457–2473.
Hicke, J. A., Allen, C. D., Desai, A. R., Dietze, M. C., Hall, R. J.,
Hogg, E. H., Kashian, D. M., Moore, D., Raffa, K., Sturrock, R., and
Vogelmann, J. 2012. Effects of biotic disturbances on forest carbon
cycling in the United States and Canada. Global Change Biology,
Vol. 18(No. 1): pp. 7–34.
Hofgaard, A., Tardif, J., and Bergeron, Y. 1999. Dendroclimatic re-
sponse of Picea mariana and Pinus banksiana along a latitudinal
gradient in the eastern Canadian boreal forest. Canadian Journal of
Forest Research, Vol. 29: pp. 1333–1346.
Holben, B. 1986. Characteristics of maximum-value composite im-
ages from temporal AVHRR data. International Journal of Remote
Sensing, Vol. 7: pp. 1417–1434.
Irons, J. R., Dwyer, J. L., and Barsi, J. A. 2012. The next Landsat
satellite: the Landsat Data Continuity Mission. Remote Sensing of
Environment, Vol. 122: pp. 11–21.
Ju, J., Roy, D. P., Shuai, Y., and Schaaf, C. 2010. Development of
an approach for generation of temporally complete daily nadir
MODIS reflectance time series. Remote Sensing of Environment,Vol.
114(No. 1): pp. 1–20.
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
VOL. 40, NO. 3, JUNE/JUIN 2014 211
Justice, C. O., Townshend, J. R. G., Vermote, E. F., Masuoka, E., Wolfe,
R. E., Saleous, N., Roy, D. P., and Morisette, J. T. 2002. An overview
of MODIS land data processing and product status. Remote Sensing
of Environment, Vol. 83: pp. 3–15.
Kaufman, Y. J., Tanre, D., Remer, L. A., Vermote, E. F., Chu, A.,
and Holben, B. N. 1997. Operational remote sensing of tropospheric
aerosol over the land from EOS-MODIS. Journal of Geophysical
Research-Atmosphere, Vol. 102: pp. 17051–17068.
Kennedy, R., Yang, Z., and Cohen, W. B. 2010. Detecting trends in
forest disturbance and recovery using yearly Landsat time series: 1.
LandTrendr—temporal segmentation algorithms. Remote Sensing of
Environment, Vol. 114: pp. 2897–2910.
Kovalskyy, V., and Roy, D. P. 2013. The global availability of Landsat 5
TM and Landsat 7 ETM+land surface observations and implications
for global 30 m Landsat data product generation. Remote Sensing of
Environment, Vol. 130: pp. 280–293.
Kurz, W. A., Stinson, G., Rampley, G. J., Dymond, C. C., and Neilson,
E. T. 2008. Risk of natural disturbances makes future contribution
of Canada’s forests to the global carbon cycle highly uncertain.
Proceedings of the National Academy of Sciences of the United
States of America, Vol. 105: pp. 1551–1555.
Kurz, W. A., Dymond, C. C., White, T. M., Stinson, G., Shaw, C.
H., Rampley, G. J., Smyth, C., Simpson, B. N., Neilson, E. T.,
Trofymow, J. A., Metsaranta, J., and Apps, M. J. 2009. CBM-
CFS3: A model of carbon dynamics in forestry and land-use change
implementing IPCC standards. Ecological Modelling, Vol. 220:
pp. 480–504.
Lindquist, E. J., Hansen, M. C., Roy, D. P., and Justice, C. O. 2008.
The suitability of decadal image data sets for mapping tropical forest
cover change in the Democratic Republic of Congo: implications for
the global land survey. International Journal of Remote Sensing,
Vol. 29: pp. 7269–7275.
Masek, J. G., Vermote, E. F., Saleous, N., Wolfe, R., Hall, F. G.,
Huemmrich, F., Gao, F., Kutler, J., and Lim, T. K. 2006. A Landsat
surface reflectance data set for North America, 1990–2000. IEEE
Geoscience and Remote Sensing Letters, Vol. 3(No. 1): pp. 68–72.
McKenney, D., Pedlar, J. H., Papadopol, P., and Hutchinson, M. F.
2006. The development of 1901-2000 historical monthly climate
models for Canada and the United States. Agricultural and Forest
Meterology, Vol. 138(Nos. 1–4): pp. 69–81.
Natural Resources Canada. 2013. The State of Canada’s Forests Annual
Report 2013. Canadian Forest Service, Headquarters, Ottawa. 56 p.
Available online:
Natural Resources Canada. 2014. FireMARS [Internet]. Ottawa, ON:
Nat. Resour. Can. Available online:
Nemani, R., Votava, P., Michaelis, A., Melton, F., and Milesi, C.
2011. Collaborative supercomputing for global change science. EOS
Transactions, Vol. 92(No. 13): pp. 109–110.
Potapov, P., Turubanova, S., and Hansen, M. C. 2011. Regional-scale
boreal forest cover and change mapping using Landsat data compos-
ites for European Russia. Remote Sensing of Environment, Vol. 115:
pp. 548–561.
Potapov, P., Turubanova, S., Hansen, M. C., Adusei, B., Broich, M.,
Alstatt, A., Mane, L., and Justice, C. O. 2012. Quantifying forest
cover loss in Democratic Republic of the Congo, 2000-2010, with
Landsat ETM+data. Remote Sensing of Environment, Vol. 122: pp.
Roy, D. P. 2000. The impact of misregistration upon composited wide
field of view satellite data and implications for change detetcion.
IEEE Transactions on Geoscience and Remote Sensing, Vol. 38: pp.
Roy, D. P., Ju, J., Kline, K., Scaramuzza, P. L., Kovalskyy, V., Hansen,
M., Loveland, T. R., Vermote, E., and Zhang, C. 2010. Web-enabled
Landsat Data (WELD): Landsat ETM+composited mosaics of the
conterminous United States. Remote Sensing of Environment,Vol.
114: pp. 35–49.
Roy, D. P., Wulder, M. A., Loveland, T. R., Woodcock, C. E., Allen,
R. G., Anderson, M. C., Helder, D., Irons, J. R., Johnson, D. M.,
Kennedy, R., Scambos, T. A., Schaaf, C. B., Schott, J. R., Sheng,
Y., Vermote, E. F., Belward, A. S., Bindschadler, R., Cohen, W. B.,
Gao, F., Hipple, J. D., Hostert, P., Huntington, J., Justice, C. O.,
Kilic, A., Kovalskyy, V., Lee, Z. P., Lymburner, L., Masek, J. G.,
McCorkel, J., Shuai, Y., Trezza, R., Vogelmann, J., Wynne, R. H., and
Zhu, Z. 2014. Landsat-8: Science and product vision for terrestrial
global change research. Remote Sensing of Environment. Vol. 145:
pp. 154–172.
Saskatchewan Ministry of Environment. 2012. 2012 Report on
Saskatchewan Forests. Government of Saskatchewan. Available on-
Schmidt, G. L., Jenkerson, C. B., Masek, J., Vermote, E., and
Gao, F. 2013. Landsat ecosystem disturbance adaptive process-
ing system (LEDAPS) algorithm description. U.S. Geological
Survey Open-File Report 2013–1057, 17 p. Available online:
Townshend, J. R. G., and Justice, C. O. 1988. Selecting the spatial reso-
lution of satellite sensors required for global monitoring of land trans-
formations. International Journal of Remote Sensing, Vol. 9(No. 2):
pp. 187–236.
Townshend, J. R., Masek, J. G., Huang, C., Vermote, E. F., Gao, F.,
Channan, S., Sexton, J. O., Feng, M., Narasimhan, R., Kim, D., Song,
K., Song, D., Song, X., Noojipady, P., Tan, B., Hansen, M. C., Li,
M., and Wolfe, R. E. 2012. Global characterization and monitoring
of forest cover using Landsat data: opportunities and challenges.
International Journal of Digital Earth. Vol 5(No. 5): pp. 373–397.
Tucker, C. J., Grant, D. M., and Dykstra, J. D. 2004. NASA’s global
orthorectified Landsat data set. Photogrammetric Engineering and
Remote Sensing, Vol. 70: pp. 313–322.
United States Geological Survey. 2013. Product Guide Landsat Cli-
mate Data Record (CDR) Surface Reflectance. Version 3.4. De-
cember, 2013. Available online:
cdr sr product guide.pdf
van Aardt, J. A. N., and Wynne, R. H. 2001. Spectral separability
among six southern tree species. Photogrammetric Engineering and
Remote Sensing, Vol. 67(No. 12): pp. 1367–1375.
Vermote, E. F., Tanre, D., Deuze, J. L., Herman, M., and Morcrette,
J.-J., 1997, Second simulation of the satellite signal in the solar
spectrum, 6S—An overview. IEEE Transactions on Geoscience and
Remote Sensing, Vol. 35(No. 3): pp. 675–686.
White, J. C., and Wulder, M. A. 2013. The Landsat observation record
of Canada: 1972-2012. Canadian Journal of Remote Sensing,Vol.
39: pp. 455–467.
Wijedesa, L. S., Sloan, S., Michelakis, D. G., and Clements, G. R.
2012. Overcoming limitation with Landsat imagery for mapping of
peat swamp forests in Sundaland. Remote Sensing, Vol. 4(No. 9):
pp. 2595–2618.
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
EL ´
ED ´
Wolfe, R. E., Roy, D. P., and Vermote, E. 1998. MODIS land data
storage, gridding, and compositing methodology: Level 2 grid.
IEEE Transactions on Geoscience and Remote Sensing, Vol. 36:
pp. 1324–1338.
Woodcock, C. E., Allen, R., Anderson, M., Belward, A., Bindschadler,
R., Cohen, W., Gao, F., Goward, S. N., Helder, D., Helmer, E.,
Nemani, R., Oreopoulos, L., Schott, J., Thenkabail, P. S., Ver-
mote, E. F., Vogelmann, J., Wulder, M. A., and Wynne, R. 2008.
Free access to Landsat imagery. Science, Vol. 320(No. 5874):
p. 1011.
Wulder, M. A., and Seemann, D. 2001. Spatially partitioning
Canada with the Landsat Worldwide Referencing System. Canadian
Journal of Remote Sensing, Vol. 27(No. 3): pp. 225–231.
Wulder, M. A., Kurz, W., and Gillis, M. D. 2004. National level forest
monitoring and modelling in Canada. Progress in Planning, Vol. 61:
pp. 365–381.
Wulder, M. A., Campbell, C., White, J. C., Flannigan, M., and Camp-
bell, I. D. 2007a. National circumstances in the international cir-
cumboreal community. The Forestry Chronicle, Vol. 83(No. 4): pp.
Wulder, M. A., White, J. C., Magnussen, S., and McDonald, S. 2007b.
Validation of a large-area land cover product using purpose ac-
quired airborne video. Remote Sensing of Environment, Vol. 106:
pp. 480–491.
Beaudoin, A., Goodenough, D. G., and Dechka, J. A. 2008. Moni-
toring Canada’s forests. Part 1: Completion of the EOSD land cover
project. Canadian Journal of Remote Sensing Vol. 34(No. 6m): pp.
Wulder, M. A., Masek, J. G., Cohen, W. B., Loveland, T. R.,
and Woodcock, C. E. 2012. Opening the archive: How free data
has enabled the science and monitoring promise of Landsat. Remote
Sensing of Environment, Vol. 122: pp. 2–10.
Zhu, Z., and Woodcock, C. E. 2012. Object-based cloud and cloud
shadow detection in Landsat imagery. Remote Sensing of Environ-
ment, Vol. 118: pp. 83–94.
Downloaded by [Canadian Forest Service] at 15:53 08 April 2016
... The input data used were based on pixel-based image compositing [32,33]. Such image composites on a national basis support automation of a wide range of image classification methods and solve the problem of selection and access to cloud-free imagery [34,35]. ...
... Such image composites on a national basis support automation of a wide range of image classification methods and solve the problem of selection and access to cloud-free imagery [34,35]. Pixel-based compositing methods are increasingly being recognised as a valuable remote sensing technique to leverage the large volumes of available data by creation of representative images for annual or seasonal time periods [33] as input to model development. Annual geomedian data [36] contain pixel composites of Landsat images using a high dimensional statistic called the 'geometric median', providing a method of combining images that produces a representative cloud-free annual composite image maintaining the 'spectral relationship between bands, reduced spatial noise, and consistency across scene boundaries' [37]. ...
Full-text available
Land cover mapping from satellite images has progressed from visual and statistical approaches to Random Forests (RFs) and, more recently, advanced image recognition techniques such as convolutional neural networks (CNNs). CNNs have a conceptual benefit over RFs in recognising spatial feature context, but potentially at the cost of reduced spatial detail. We tested the use of CNNs for improved land cover mapping based on Landsat data, compared with RFs, for a study area of approximately 500 km × 500 km in southeastern Australia. Landsat 8 geomedian composite surface reflectances were available for 2018. Label data were a simple nine-member land cover classification derived from reference land use mapping (Catchment Scale Land Use of Australia—CLUM), and further enhanced by using custom forest extent mapping (Forests of Australia). Experiments were undertaken testing U-Net CNN for segmentation of Landsat 8 geomedian imagery to determine the optimal combination of input Landsat 8 bands. The results were compared with those from a simple autoencoder as well as an RF model. Segmentation test results for the best performing U-Net CNN models produced an overall accuracy of 79% and weighted-mean F1 score of 77% (9 band input) or 76% (6 band input) for a simple nine-member land cover classification, compared with 73% and 68% (6 band input), respectively, for the best RF model. We conclude that U-Net CNN models can generate annual land cover maps with good accuracy from proxy training data, and can also be used for quality control or improvement of existing land cover products.
... First, we characterized each sector concerning on-site species richness and alphadiversity and assessed differences between sectors to compare biodiversity patterns within the beech forest (Hsieh et al., 2016). Second, for a 37-year study period (1984-2020), we constructed Landsat Best Available Pixel composites (White et al., 2014) from which an NDVI time series was obtained and used to calculate eight Temporal Metrics (TMs) that summarise the NDVI trend over time. Then, we investigated the relationship -in terms of Pearson's product-moment correlation -between Landsat TMs and the abundance of saproxylic key species at the family and trophic category level. ...
... For each year in the study period (1984-2020), we calculated cloudfree composites of the study area. When more than one cloud-free observation was available for a specific year, the "best" pixel was selected using the Best Available Pixel procedure (BAP), for which a comprehensive description was given by Griffiths et al. (2013), White et al. (2014) and Hermosilla et al. (2015a, b). When any observation was available for a specific year (data gaps), we obtained a valid synthetic observation by linear interpolating the first two valid observations in previous and subsequent years. ...
Full-text available
Background Rapid climate changes lead to an increase in forest disturbance, which in turn lead to growing concerns for biodiversity. While saproxylic beetles are relevant indicators for studying different aspects of biodiversity, most are smaller than 2 mm and difficult to sample. This, together with a high number of species and trophic roles, make their study remarkably challenging, time-consuming, and expensive. The Landsat mission provides data since 1984 and represents a powerful tool in this scenario. While we believe that remote sensing data cannot replace on-site sampling and analysis, in this study we aim to prove that the Landsat Time Series (TS) may support the identification of insects’ hotspots and consequently guide the selection of areas where to concentrate field analysis. Methods With this aim, we constructed a Landsat-derived NDVI TS (1984–2020) and we summarised the NDVI trend over time by calculating eight Temporal Metrics (TMs) among which four resulted particularly successful in predicting the amount of saproxylic insects: (i) the slope of the regression line obtained by linear interpolating the NDVI values over time; (ii) the Root Mean Square Error (RMSE) between the regression line and the NDVI TS; (iii) the median, and the (iv) minimum values of the NDVI TS. The study area consists of four monitoring sectors in a Mediterranean-managed beech forest located in the Apennines (Molise, Italy), where 60 window flight traps for flying beetles were installed. First, the saproxylic beetle's biodiversities of monitoring sectors were quantified in terms of species richness and alpha-diversity. Second, the capability of TMs in predicting the richness of saproxylic beetles family and trophic categories was assessed in terms of Pearson's product-moment correlation. Results The alpha diversity and species richness analysis indicate dissimilarities across the four monitored sectors (Shannon and Simpson's index ranging between 0.67 to 2.31 and 0.69 to 0.88, respectively), with Landsat TS resulting in effective predictors for estimating saproxylic beetle richness. The strongest correlation was reached between the Monotomidae family and the RMSE temporal metric (R = 0.66). The mean absolute correlation (r) between the NDVI TMs and the saproxylic community was 0.46 for Monotomidae, 0.31 for Cerambycidae, and 0.25 for Curculionidae. Conclusions Our results suggest that Landsat TS has important implications for studying saproxylic beetle distribution and, by helping the selection of monitoring areas, increasing the amount of information acquired while decreasing the effort required for field analysis.
... More specifically, imagery acquired by Landsat 5, 7, and 8 with a cloud cover threshold <70% is used. Images with cloud cover >70% are excluded because they are more prone to geographical location errors, due to the challenges of performing geometrical corrections when ground control points are obscured (White et al., 2014). The remaining clouds in Landsat imagery were then masked by the CFMASK algorithm (Foga et al., 2017). ...
Snow cover is a key hydrological variable, critical to understanding water cycles and informing management decisions around resource extraction and recreational activities. Remote sensing open-access data and cloud-based computing platforms are two innovative tools for snow cover estimation. In this paper, we present SnowWarp, a processing framework that uses Google Earth Engine and the R programming languages to combine Landsat 30 m with MODIS 500 m satellite imagery and produce daily-30-m spatial resolution snow cover data anywhere globally. SnowWarp was applied in an alpine catchment in Northern Italy from 2000-2019 and validated using hydrometeorological datasets. Strong correlations between snow cover and ground data were found with correlations in terms of R up to −0.84 for temperature, −0.17 for precipitation, 0.74 for snow depth, and −0.43 for streamflow. The SnowWarp tool is an open-source framework enabling users to map fine spatial and temporal dynamics of snow cover to the ecosystem and hydrological monitoring.
... It will be more challenging to examine gap-filling effects for a longer period (Brandt et al., 2018) as few gap-filling methods have been suggested to deliver good performance in a long time series. Further studies can focus on examining other types of gap-filling methods, other predictors, such as best-pixel composites (White et al., 2014), and other downstream tasks, such as land surface temperature monitoring (Li et al., 2013). ...
Full-text available
Preprocessing of Landsat images is a double-edged sword, transforming the raw data into a useful format but potentially introducing unwanted values with unnecessary steps. Through recovering missing data of satellite images in time series analysis, gap-filling is an important, highly developed, preprocessing procedure, but its necessity and effects in numerous Landsat applications, such as tree canopy cover (TCC) modelling, are rarely examined. We address this barrier by providing a quantitative comparison of TCC modelling using predictor variables derived from Landsat time series that included gap-filling versus those that did not include gap-filling and evaluating the effects that gap-filling has on modelling TCC. With 1-year Landsat time series from a tropical region located in Taita Hills, Kenya, and a reference TCC map in 0–100 scales derived from airborne laser scanning data, we designed comparable random forest modelling experiments to address the following questions: 1) Does gap-filling improve TCC modelling based on time series predictor variables including the seasonal composites (SC), spectral-temporal metrics (STMs), and harmonic regression (HR) coefficients? 2) What is the difference in TCC modelling between using gap-filled pixels and using valid (actual or cloud-free) pixels? Two gap-filling methods, one temporal-based method (Steffen spline interpolation) and one hybrid method (MOPSTM) have been examined. We show that gap-filled predictors derived from the Landsat time series delivered better performance on average than non-gap-filled predictors with the average of median RMSE values for Steffen-filled and MOPSTM-filled SC’s being 17.09 and 16.57 respectively, while for non-gap-filled predictors, it was 17.21. MOPSTM-filled SC is 3.7% better than non-gap-filled SC on RMSE, and Steffen-filled SC is 0.7% better than non-gap-filled SC on RMSE. The positive effects of gap-filling may be reduced when there are sufficient high-quality valid observations to generate a seasonal composite. The single-date experiment suggests that gap-filled data (e.g. RMSE of 16.99, 17.71, 16.24, and 17.85 with 100% gap-filled pixels as training and test datasets for four seasons) may deliver no worse performance than valid data (e.g. RMSE of 15.46, 17.07, 16.31, and 18.14 with 100% valid pixels as training and test datasets for four seasons). Thus, we conclude that gap-filling has a positive effect on the accuracy of TCC modelling, which justifies its inclusion in image preprocessing workflows.
... Therefore, these methods are not well suitable for our research. Recently, White et al. (2014) proposed a BAP method to generate cloudless composites for a large area. Since BAP compiles cloud-free images by selecting the best available observation based on user-defined criteria (Gomez et al., 2016;Griffiths et al., 2013), the BAP composites can retain the source image information from which they came. ...
Full-text available
Monitoring the water clarity of lakes is essential for the sustainable development of human society. However, existing water clarity assessments in China have mostly focused on lakes with areas > 1 km2, and the monitoring periods were mainly in the 21st century. In order to improve the understanding of spatiotemporal variations in lake clarity across China, based on the Google Earth Engine cloud platform, a 30 m long-term LAke Water Secchi depth (SD) dataset (LAWSD30) of China (1985–2020) was first developed using Landsat series imagery and a robust water-color parameter-based SD model. The LAWSD30 dataset exhibited a good performance compared to concurrent in situ SD datasets, with an R2 of 0.86 and a root mean square error of 0.225 m. Then, based on our LAWSD30 dataset, long-term spatiotemporal variations in SD for lakes > 0.01 km2 (N = 40 973) across China were evaluated. The results show that the SD of lakes with areas ≤ 1 km2 exhibited a significant downward trend in the period of 1985–2020, but the decline rate began to slow down and stabilized after 2001. In addition, the SD of lakes with an area > 1 km2 showed a significant downward trend before 2001, and began to increase significantly afterwards. Moreover, in terms of the spatial patterns, the proportion of small lakes (area ≤ 1 km2) showing a decreasing SD trend was the largest in the Mongolian–Xinjiang Plateau Region (MXR) (about 30.0 %), and the smallest in the Eastern Plain Region (EPR) (2.6 %). In contrast, for lakes > 1 km2, this proportion was the highest in MXR (about 23.0 %), and the lowest in the Northeast Mountain Plain Region (NER) (16.1 %). The LAWSD30 dataset and the spatiotemporal patterns of lake water clarity in our research can provide effective guidance for the protection and management of lake environment in China.
... Its aim is to simplify the work of users, by providing almost cloud free images. Most Level-3A products are based on a best available pixel method [23], which selects, for each pixel, the best date in a surface reflectance time series. The most classical selection criterion comprises selecting the date for which a vegetation index is maximal [24], knowing that clouds have a very low vegetation index. ...
Full-text available
VENμS (Vegetation and Environment New micro (μ) Satellite) is a micro satellite launched in 2017 by the Israeli Space Agency (ISA) and the French Centre National d’Etudes Spatiales (CNES). VENμS is a research satellite containing two very different devices: an electric Hall effect thruster and a multispectral optical camera. This paper focuses on the multispectral camera. The camera provides images at a resolution of 5 m, with a field of view of 27 km, and the orbit of the satellite was chosen to allow us to revisit of each observed site with constant angles every second day. In November 2020, VENμS ended the first phase of its mission. This phase, called VM01, allowed us to provide about 150 accurate time series over selected scientific sites over almost three years. Extensive work was conducted to calibrate the camera and assess the quality of the products. Not everything worked as planned before launch and a large amount of work was necessary to correct some defects of the camera or to improve the geometric registration of images. This article establishes the image quality VM01 final assessment including the presentation of radiometric and geometric calibration methods, the estimation of instrument performances and their associated temporal stabilities and the monitoring activities. In addition, it highlights the whole mechanism of data programming, reception and production. The end of VM01 phase is not the end of the VENμS mission, and a new phase started on a one-day repeat orbit.
... Pixel-based compositing methods can provide a new analysis paradigm instead of depending on a single scene [40], resulting in an unprecedented opportunity for standardizing and automating fire severity and burned area evaluations [27,[41][42][43]. Nevertheless, the image composite techniques analyzed in the literature have largely varied, depending on the study goal, location, and period analyzed. ...
Full-text available
The remote sensing of fire severity and burned area is fundamental in the evaluation of fire impacts. The current study aimed to: (i) compare Sentinel-2 (S2) spectral indices to predict field-observed fire severity in Durango, Mexico; (ii) evaluate the effect of the compositing period (1 or 3 months), techniques (average or minimum), and phenological correction (constant offset, c, against a novel relative phenological correction, rc) on fire severity mapping, and (iii) determine fire perimeter accuracy. The Relative Burn Ratio (RBR), using S2 bands 8a and 12, provided the best correspondence with field-based fire severity (FBS). One-month rc minimum composites showed the highest correspondence with FBS (R2 = 0.83). The decrease in R2 using 3 months rather than 1 month was ≥0.05 (0.05–0.15) for c composites and <0.05 (0.02–0.03) for rc composites. Furthermore, using rc increased the R2 by 0.05–0.09 and 0.10–0.15 for the 3-month RBR and dNBR compared to the corresponding c composites. Rc composites also showed increases of up to 0.16–0.22 and 0.08–0.11 in kappa values and overall accuracy, respectively, in mapping fire perimeters against c composites. These results suggest a promising potential of the novel relative phenological correction to be systematically applied with automated algorithms to improve the accuracy and robustness of fire severity and perimeter evaluations.
... Generally, canopy cover loss alerts indicate likely changes in near-real-time, often with some measure of confidence [30,31], whereas annual change maps, such as the global forest change product of the University of Maryland [37], show high confidence with Remote Sens. 2022, 14, 562 3 of 19 regard to the detected changes actually being real changes on the ground. However, the latter typically use annual best-pixel composites [38][39][40] or temporal metrics [41][42][43] that lead to some ambiguity in correctly assigning a change to a particular year, as images taken under optimal conditions are typically weighted more heavily than winter acquisitions. Hence, changes happening late in a particular year are likely to be attributed to the next year. ...
Conference Paper
Vast areas of Central and Northern Europe experienced a pronounced drought in 2018. Germany, among other countries, was heavily affected. In some parts of the country, exceptionally dry conditions continued into spring 2021. The effects of the 2018 drought had a strong impact on Central European forests, particularly in the Czech Republic and Germany. Extensive droughts cause severe stress to trees, which is amplified by the specific situation in Germany, where forests are often located in hilly regions or on poor soils, and many trees are planted at the margins of their climatic niche. Once stressed by drought, trees are generally more susceptible to insect damage. While deciduous trees often have the potential to recover from insect infestations, the situation is different for coniferous trees. The European spruce bark beetle (Ips typographus [L.]) is one of the most damaging pest insects of spruce forests in Europe: successful infestation is typically fatal to trees. During the 2018-2020 drought, bark beetle management in Germany had a strong focus on the prevention of outbreak expansion by massive salvage and sanitation logging in outbreak areas and their surroundings. Actual numbers of the associated forest loss are provided based on statistical sampling and are not spatially explicit. Besides, the temporal development can only be traced at annual intervals. Remote sensing has proven to be valuable in detecting forest changes, particularly stand replacing changes. However, annual change maps typically use annual best-pixel composites or temporal metrics. These can lead to some ambiguity in correctly assigning a change to a particular year, as images taken under optimal conditions are typically weighted more heavily than winter acquisitions. Hence, changes happening late in a year are likely attributed to the next year. Common silvicultural practice in Germany avoids large-scale clear-cuts. This has changed in response to the recent drought. Clear-cuts are common practice to implement salvage logging. To our knowledge there is currently no comprehensive, spatially-explicit assessment of clear-cuts and tree loss in Germany. We demonstrate an efficient method to map clear-cuts in temperate Central European forests with high spatial (10 m) and temporal (monthly) resolution. We present a first spatially-explicit assessment of the tree-loss areas in response to the 2018-2020 drought in Germany. To achieve this goal, we used time series of Sentinel-2 and Landsat 8 data and a spectral index largely insensitive to illumination conditions, the disturbance index (DI, Healey et al., 2005). The dense time series was aggregated to monthly composites, thereby removing outliers. From the monthly time series (January 2018-April 2021), we computed anomalies with respect to a reference period (2017) and applied simple thresholding to separate clear-cuts and dead trees from healthy and stressed forest stands. We identified changes (i.e. tree loss) persisting over the monitoring period, determined tree loss dates at per-pixel scale and aggregated the results to different administrative levels. Our results reveal that about 588,489 ha of forest were lost in Germany between January 2018 and April 2021, corresponding to more than 5 % of the total forest area. This figure contains also dead trees that were not yet logged, but mainly refers to cleared forests. In 2018, the tree loss area was still rather low as it took some time for the trees to die in response to the heavy 2018 drought. Most of the cleared areas of 2018 are likely the result of the removal of windthrown trees in the aftermath of 2017 summer storms (e.g. near Passau in Bavaria, South-East Germany) and 2018 winter storms such as “Friederike” (e.g. in Northern and Eastern Germany). Drought induced mortality in beech and spruce trees started in 2018, and was accelerated by bark beetle infestation in spruce trees, which started in 2018 and continued in several outbreak phases until 2021. Salvage logging as radical management strategy started already in 2018 in some federal states such as Saxony-Anhalt, but accelerated through 2019 and 2020, particularly in Hesse and North Rhine-Westphalia. Consequently, the spatial pattern of tree loss changed from larger areas in Eastern and South-Eastern Germany in 2018 to dominant changes in Central and Western Germany in 2019, 2020 and 2021. Considering all forest types, tree loss was evident throughout Germany, even though Northern and Southern Germany were less affected than Central Germany. Central Western and Eastern Germany were most heavily affected with regard to forest loss in coniferous forests. In a belt ranging from the western to the eastern borders of the country, a large share of the coniferous forests was cleared, in some areas more than three quarters. At district level (Landkreis), the pattern becomes clearer than on federal state level. The district of Soest in North Rhine-Westphalia, for example, lost two thirds of its coniferous forests. While existing annual crown condition assessments are a valuable source to identify general (long-term) forest health developments, spatially-explicit mapping of tree loss is still missing in Germany. We aim to support forest management and scientific understanding with this first assessment of tree loss after the 2018-2020 drought years.
... Frontiers in Remote Sensing | June 2022 | Volume 3 | Article 894618 White et al., 2014) and maximum NDVI compositing schemes to perform better in most situations. Still, even if compositing is possible in areas of low data density, it is unlikely that the composites would be dense enough to enable monitoring algorithms such as LandTrendr and CCDC. ...
Full-text available
The Landsat program has the longest collection of moderate-resolution satellite imagery, and the data are free to everyone. With the improvements of standardized image products, the flexibility of cloud computing platforms, and the development of time series approaches, it is now possible to conduct global-scale analyses of time series using Landsat data over multiple decades. Efforts in this regard are limited by the density of usable observations. The availability of usable Landsat Tier 1 observations at the scale of individual pixels from the perspective of time series analysis for land change monitoring is remarkably variable both in space (globally) and time (1985–2020), depending most immediately on which sensors were in operation, the technical capabilities of the mission, and the acquisition strategies and objectives of the satellite operators (e.g., USGS, commercial company) and the international ground receiving stations. Additionally, analysis of data density at the pixel scale allows for the integration of quality control data on clouds, cloud shadows, and snow as well as other properties returned from the atmospheric correction process. Maps for different time periods show the effect of excluding observations based on the presence of clouds, cloud shadows, snow, sensor saturation, hazy observations (based on atmospheric opacity), and lack of aerosol optical depth information. Two major discoveries are: 1) that filtering saturated and hazy pixels is helpful to reduce noise in the time series, although the impact may vary across different continents; 2) the atmospheric opacity band needs to be used with caution because many images are removed when no value is given in this band, when many of those observations are usable. The results provide guidance on when and where time series analysis is feasible, which will benefit many users of Landsat data.
The United Nations 2030 Agenda for Sustainable Development provides an important framework for economic, social, and environmental action. A comprehensive indicator system to aid in the systematic implementation and monitoring of progress toward the Sustainable Development Goals (SDGs) is unfortunately limited in many countries due to lack of data. The availability of a growing amount of multi-source data and rapid advancements in big data methods and infrastructure provide unique opportunities to mitigate these data shortages and develop innovative methodologies for comparatively monitoring SDGs. Big Earth Data, a special class of big data with spatial attributes, holds tremendous potential to facilitate science, technology, and innovation toward implementing SDGs around the world. Several programs and initiatives in China have invested in Big Earth Data infrastructure and capabilities, and have successfully carried out case studies to demonstrate their utility in sustainability science. This paper presents implementations of Big Earth Data in evaluating SDG indicators, including the development of new algorithms, indicator expansion (for SDG 11.4.1) and indicator extension (for SDG 11.3.1), introduction of a biodiversity risk index as a more effective analysis method for SDG 15.5.1, and several new high-quality data products, such as global net ecosystem productivity, high-resolution global mountain green cover index, and endangered species richness. These innovations are used to present a comprehensive analysis of SDGs 2, 6, 11, 13, 14, and 15 from 2010 to 2020 in China utilizing Big Earth Data, concluding that all six SDGs are on schedule to be achieved by 2030.
Full-text available
With the advent of the free U.S. Landsat data policy it is now feasible to consider the generation of global coverage 30 m Landsat data sets with temporal reporting frequency similar to that provided by the monthly Web Enabled Landsat (WELD) products. A statistical Landsat metadata analysis is reported considering more than 800,000 Landsat 5 TM and Landsat 7 ETM + acquisitions obtained from the U.S. Geological Survey (USGS) Earth Resources Observation and Science (EROS) Center archive. The global monthly probabilities of acquiring a cloud-free land surface observation for December 1998 to November 2001 (2000 epoch) and from December 2008 to November 2011 (2010 epoch) are reported to assess the availability of the Landsat data in the USGS Landsat archive for global multi-temporal land remote sensing applications. The global probabilities of acquiring a cloud-free land surface observation in each of three different seasons with the highest seasonal probabilities of cloud-free land surface observation are reported, considering one, two and three years of Landsat data, to assess the availability of Landsat data for global land cover mapping. The probabilities are derived considering Landsat 5 TM only, Landsat 7 ETM + only, and both sensors combined, to examine the relative benefits of using one or both Landsat sensors. The results demonstrate the utility of combing both Landsat 5 TM and Landsat 7 ETM + data streams to take advantage of their different acquisition patterns and to mitigate the deleterious impact of the Landsat 7 ETM + 2003 scan line failure. Sensor combination provided a greater global acquisition coverage with a 1.7% to 14.4% higher percentage of land locations acquired monthly compared to considering Landsat 7 ETM + data alone. The mean global monthly probability of a cloud-free land surface observation for the combined sensors was up to nearly 1.4 and 6.7 times greater than for ETM + and TM alone respectively. The probability of acquiring a cloud-free Landsat land surface observation in different seasons was greater when more years of data were considered and when both Landsat sensor data were combined. Considering combined sensors and 36 months of data, 86.4% and 84.2% of the global land locations had probabilities >= 0.95 for the 2000 and 2010 epochs respectively, with a global mean probability of 0.92 (sigma 0.24) for the 2000 epoch and 0.90 (sigma 0.28) for the 2010 epoch. These results indicate that 36 months of combined Landsat sensor data will provide sufficient land surface observations for 30 m global land cover mapping using a multi-temporal supervised classification scheme.
Full-text available
Multi-temporal satellite imagery can be composited over a season (or other time period) to produce imagery which is representative of that period, using techniques which will reduce contamination by cloud and other problems. For the purposes of vegetation monitoring, a commonly used technique is the Maximum NDVI Composite, used in conjunction with variety of other constraints. The current paper proposes an alternative based on the medoid (in reflectance space) over the time period (the medoid is a multi-dimensional analogue of the median), which is robust against extreme values, and appears to be better at producing imagery which is representative of the time period. For each pixel, the medoid is always selected from the available dates, so the result is always a single observation for that pixel, thus preserving relationships between bands. The method is applied to Landsat TM/ETM+ imagery to create seasonal reflectance images (four per year), with the aim being a regular time series of reflectance values which captures the variability at seasonal time scales. Analysis of the seasonal reflectance values suggests that resulting temporal image composites are more representative of the time series than the maximum NDVI seasonal composite.
Full-text available
The Landsat data archive represents more than 40 years of Earth observation, providing a valuable information source for monitoring ecosystem dynamics. In excess of 605 000 images of Canada have been acquired by the Landsat program since 1972. Herein we report several spatial and temporal characteristics of the Landsat observation record for Canada (1972_2012), including image availability by year, growing season, sensor, ecozone, and provincial or territorial jurisdiction. In contrast to the global Landsat archive, which is dominated by Enhanced Thematic Mapper Plus (ETM_) data, the majority of archived Landsat images of Canada were acquired by the Thematic Mapper (TM) sensor (57%). Approximately 55% of archived Landsat images were acquired within ± 30 days of 1 August, and 74% of Worldwide Reference System_2 path_row locations in Canada have more than 200 images acquired between 1 June and 30 September. Issues such as cloud cover and the availability of imagery to support pixel-based image compositing and time series analyses are explored and documented. For a pixel-based image compositing scenario whereby images (TM and ETM_) acquired after 1981 with less than 70% cloud cover and a target date of 1 August 930 days are considered, 60% of the path_row locations have five or fewer years of missing data in the archive. For time series analyses (i.e., ecosystem monitoring scenario) with the same temporal constraint but with less than 10% cloud cover, only 2% of path_row locations are missing five or fewer years of data, with a mean and median of 17 years of missing data. However, if a broader temporal window (1 June to 30 September) is considered for this scenario, 18% of path_row locations have five or fewer years of missing data. Free and open-access to the Landsat data archive, combined with the continuity of new data collections provided by the recently launched Landsat 8 satellite, offer many opportunities for scientific inquiry concerning the status and trends of Canada’s terrestrial ecosystems.
Full-text available
Landsat 8, a NASA and USGS collaboration, acquires global moderate-resolution measurements of the Earth's terrestrial and polar regions in the visible, near-infrared, short wave, and thermal infrared. Landsat 8 extends the remarkable 40 year Landsat record and has enhanced capabilities including new spectral bands in the blue and cirrus cloud-detection portion of the spectrum, two thermal bands, improved sensor signal-to-noise performance and associated improvements in radiometric resolution, and an improved duty cycle that allows collection of a significantly greater number of images per day. This paper introduces the current (2012–2017) Landsat Science Team's efforts to establish an initial understanding of Landsat 8 capabilities and the steps ahead in support of priorities identified by the team. Preliminary evaluation of Landsat 8 capabilities and identification of new science and applications opportunities are described with respect to calibration and radiometric characterization; surface reflectance; surface albedo; surface temperature, evapotranspiration and drought; agriculture; land cover, condition, disturbance and change; fresh and coastal water; and snow and ice. Insights into the development of derived ‘higher-level’ Landsat products are provided in recognition of the growing need for consistently processed, moderate spatial resolution, large area, long-term terrestrial data records for resource management and for climate and global change studies. The paper concludes with future prospects, emphasizing the opportunities for land imaging constellations by combining Landsat data with data collected from other international sensing systems, and consideration of successor Landsat mission requirements.
A new national forest inventory is being installed in Canada. For the last 20 years, Canada's forest inventory has been a compilation of inventory data from across the country. Although this method has a number of advantages, it lacks information about the nature and rate of changes to the resource, and does not permit projections or forecasts. To address these limitations a new National Forest Inventory (NFI) was developed to monitor Canada's progress in meeting a commitment towards sustainable forest management, and to satisfy requirements for national and international reporting. The purpose of the new inventory is to "assess and monitor the extent, state and sustainable development of Canada's forests in a timely and accurate manner." The NFI consists of a plot-based system of permanent observational units located on a national grid. A combination of ground plot, photo plot and remote sensing data are used to capture a set of basic attributes that are used to derive indicators of sustainability. To meet the monitoring needs a re-measurement strategy and framework to guide the development of change estimation procedures has been worked out. A plan for implementation has been drafted. The proposed plan is presented and discussed in this paper.
Over the past decade, a number of initiatives have been undertaken to create definitive national and global data sets consisting of precision corrected Landsat Multispectral Scanner (MSS) and Thematic Mapper (TM) scenes. One important application of these data is the derivation of large area land-cover products spanning multiple satellite scenes. A popular approach to land-cover mapping on this scale involves merging constituent scenes into image mosaics prior to image clustering and cluster labeling, thereby eliminating redundant geographic coverage arising from overlapping imaging swaths of adjacent orbital tracks. In this paper, arguments are presented to support the view that areas of overlapping coverage contain important information that can be used to assess and improve classification performance. A methodology is presented for the creation of large area land-cover products through the compositing of independently classified scenes. Statistical analyses of classification consistency between scenes in overlapping regions are employed both to identify mislabeled clusters and to provide a measure of classification confidence for each scene at the cluster level. During classification compositing, confidence measures are used to rationalize conflicting classifications in overlap regions and to create a relative confidence layer, sampled at the pixel level, which characterizes the spatial variation in classification quality over the final product. The procedure is illustrated with results from a synoptic mapping project of the Great Lakes watershed that involved the classification and compositing of 46 Landsat MSS scenes.
NASA has sponsored the creation of an orthorectified and geodetically accurate global land data set of Landsat Multispectral Scanner, Thematic Mapper, and Enhanced Thematic Mapper data, from the 1970s, circa 1990, and circa 2000, respectively, to support a variety of scientific studies and educational purposes. This is the first time a geodetically accurate global compendium of orthorectified multi-epoch digital satellite data at the 30- to 80-m spatial scale spanning 30 years has been produced for use by the international scientific and educational communities. We describe data selection, orthorectification, accuracy, access, and other aspects of these data.