See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/312202874

A survival guide to Landsat preprocessing

Article  in  Ecology · March 2017

DOI: 10.1002/ecy.1730

CITATIONS READS

43 3,566

6 authors, including:

Nicholas E. Young Ryan Anderson

Colorado State University Colorado State University
41 PUBLICATIONS   333 CITATIONS    12 PUBLICATIONS   109 CITATIONS   

SEE PROFILE SEE PROFILE

Stephen Chignell Anthony Vorster

University of British Columbia - Vancouver Colorado State University
15 PUBLICATIONS   100 CITATIONS    11 PUBLICATIONS   70 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Ethnobotany and livelihoods in Madagascar and Eastern Africa View project

Secondary Cities View project

All content following this page was uploaded by Nicholas E. Young on 27 October 2017.

The user has requested enhancement of the downloaded file.

 CONCEPTS & SYNTHESIS
EMPHASIZING NEW IDEAS TO STIMULATE RESEARCH IN ECOLOGY

Ecology, 98(4), 2017, pp. 920–932

© 2017 The Authors. Ecology, published by Wiley Periodicals, Inc., on behalf of the Ecological Society of America.
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any
medium, provided the original work is properly cited.

A survival guide to Landsat preprocessing

Nicholas E. Young,1,4 Ryan S. Anderson,1 Stephen M. Chignell,1,2 Anthony G. Vorster,1,2
Rick Lawrence,3 and Paul H. Evangelista1,2
1Natural Resource Ecology Laboratory, Colorado State University, 1499 Campus Delivery, Fort Collins, Colorado 80523 USA
2Department of Ecosystem Science and Sustainability, Colorado State University, Fort Collins, Colorado 80523 USA
3Land Resources and Environmental Sciences Department, Spatial Sciences Center, 
Montana State University, Bozeman, Montana 59717 USA

Abstract. Landsat data are increasingly used for ecological monitoring and research. These
data often require preprocessing prior to analysis to account for sensor, solar, atmospheric,
and topographic effects. However, ecologists using these data are faced with a literature con-
taining inconsistent terminology, outdated methods, and a vast number of approaches with
contradictory recommendations. These issues can, at best, make determining the correct pre-
processing workflow a difficult and time-­consuming task and, at worst, lead to erroneous
­results. We address these problems by providing a concise overview of the Landsat missions
and sensors and by clarifying frequently conflated terms and methods. Preprocessing steps
commonly applied to Landsat data are differentiated and explained, including georeferencing
and co-­registration, conversion to radiance, solar correction, atmospheric correction, topo-
graphic correction, and relative correction. We then synthesize this information by presenting
workflows and a decision tree for determining the appropriate level of imagery preprocessing
given an ecological research question, while emphasizing the need to tailor each workflow to
the study site and question at hand. We recommend a parsimonious approach to Landsat pre-
processing that avoids unnecessary steps and recommend approaches and data products that
are well tested, easily available, and sufficiently documented. Our focus is specific to ecological
applications of Landsat data, yet many of the concepts and recommendations discussed are
also appropriate for other disciplines and remote sensing platforms.
Key words: atmospheric correction; change detection; decision tree; ecology; image; normalization;
­radiometric correction; remote sensing; review; satellite; topographic correction; workflow.

Introduction they must be preprocessed to account for sensor, solar,

atmospheric, and topographic effects. However, each
Landsat data have become exceedingly integrated into
preparation step further alters the data from their original
Earth observation and monitoring applications, particu-
values, increasing the potential to introduce error
larly within the last decade (Wulder et al. 2012, Turner
(Kennedy et al. 2009). Determining the appropriate level
et al. 2015). This recent increase is due in part to Landsat’s
of preprocessing is a significant barrier to non-­remote
free and global coverage; when Landsat data became
sensing scientists who lack expertise in the numerous and
freely available in 2009, the USGS saw a 50-­fold annual
constantly changing techniques necessary to preprocess
increase in image downloads (Miller et al. 2011). The
these data (Pettorelli et al. 2014). This difficulty is exac-
Landsat program’s ever-­expanding image archive is an
erbated by preprocessing approaches that are similar but
invaluable data set for ecological monitoring, change
distinct, each with numerous possible workflows that
detection, and biodiversity conservation (Cohen and
analysts must navigate. Furthermore, explanations of the
Goward 2004, Loveland and Dwyer 2012, Kennedy et al.
specific preprocessing steps taken in ecological studies
2014, Turner et al. 2015, Vogelmann et al. 2016). Before
vary; some offer detailed technical accounts, while others
these data can be used for certain ecological analyses,
provide just a few broad summary sentences. These
inconsistencies result in confusion and ambiguity, par-
Manuscript received 28 July 2016; revised 23 December
2016; accepted 3 January 2017. Corresponding Editor: David ticularly for researchers and managers who would like
S. Schimel. to use Landsat data in their work but do not have a clear
4E-mail: [email protected] roadmap for how to do so.

920
April 2017 SURVIVAL GUIDE TO LANDSAT PREPROCESSING 921

Terminology adds to the challenge of determining the example, the spectral range of many bands have changed,
most appropriate preprocessing workflow. The remote which can create issues for time series analyses (e.g.,
sensing community has made efforts to standardize its Holden and Woodcock 2016). The Landsat satellites have
terminology (Schaepman-­Strub et al. 2006), but terms are been organized into three groups based on their sensor
often conflated in more applied studies published in eco- and platform characteristics (Chander et al. 2009) and in
logical journals. This contributes to the continued use of the following section, we modify this framework and
outdated or redundant terms, which creates confusion include recent and upcoming Landsat missions.
for researchers attempting to reproduce or expand on The first group, Landsat 1–3, were equipped with the
published research methods. Multispectral Scanner (MSS), which recorded data in four
Our purpose is to clarify and synthesize the preparation spectral bands; two visible and two near-­infrared (NIR;
of Landsat imagery for ecological applications. We begin Table 1). The next group, Landsat 4–7, carried either the
with a concise overview of the Landsat missions. We then Thematic Mapper (TM) or Enhanced Thematic Mapper
describe the different levels of preprocessing and the (ETM+) sensors, which featured finer spatial resolution
sequence in which preprocessing steps should be applied, (i.e., pixel size), and increased radiometric resolution (i.e.,
while clarifying inconsistent remote sensing terminology bit depth) than the MSS. This group also had expanded
(Appendix S1: Table S1). Finally, we present a decision tree spectral coverage, adding bands in the middle-­infrared and
to help analysts determine the level of preprocessing nec- thermal-­infrared wavelengths. It is worth noting that the
essary for their study, and provide examples of its imple- middle-­infrared is now often referred to as the shortwave
mentation in common ecological analyses. We stress that infrared (SWIR). While middle-­infrared and SWIR gen-
this paper does not contain the only available or appro- erally cover different, but largely overlapping, spectral
priate approaches for all circumstances, nor is it meant to ranges, the relevant Landsat bands fall within this overlap.
serve as an introduction to the field of remote sensing. Our For these reasons, we use the term SWIR.
aim is to address common questions that have repeatedly The satellites in the TM/ETM+ group have slightly dif-
been raised during our collective experience as scientists ferent characteristics from one another. In addition to the

Concepts & Synthesis

working with Landsat data. As such, this paper is intended TM instrument, Landsat 4–5 also had an MSS scanner
as a guide for novice users of Landsat imagery as well as a onboard. Landsat 7 included a panchromatic band with
concise reference for those with more experience. increased spatial resolution. The third group currently
consists of Landsat 8, equipped with its Operational Land
Imager (OLI) and Thermal Infrared (TIRS) sensors. The
Overview and synthesis of the Landsat missions
OLI augments the spectral resolution of the TM group
The National Aeronautics and Space Administration with the addition of a deep blue and a cirrus band, while
(NASA) launched the first Landsat satellite in 1972 and TIRS adds a second thermal band. These instruments also
has since followed with six successfully launched satellites offer a number of sensor and calibration improvements
(Fig. 1). While NASA builds and launches the satellites, over the previous Landsat missions (Roy et al. 2014). The
the U.S. Geological Survey (USGS) has operated the mis- other member of the third group, Landsat 9, is currently
sions since the 1990s. The Landsat program offers a near-­ planned as an upgraded rebuild of Landsat 8 and is
continuous record of imagery, but there are a number of scheduled to be launched in 2020.
differences among the satellites and sensors that pose The data generated from the Landsat program offer a
challenges for experts and non-­experts alike (Table 1). For number of characteristics that can support ecological

Fig. 1.  A timeline of Landsat satellites and sensors. Landsat 9 launch date is based on recent congressional appropriations
language.
 922 NICHOLAS E. YOUNG ET AL. Ecology, Vol. 98, No. 4

Table 1.  Summary of band designations and pixel size (m) for all Landsat satellites (LS) and sensors.

Landsat sensor LS 1–5 MSS LS 4–5 TM LS 7 ETM+ LS 8 OLI/TIRS Pixel size (m)

Coastal aerosol B1 (0.43–0.45) 30

Blue B1 (0.45–0.52) B1 (0.45–0.52) B2 (0.45–0.51) 30
Green B1 (0.5–0.6) B2 (0.52–0.60) B2 (0.52–0.60) B3 (0.53–0.59) 30 (60† for MSS)
Red B2 (0.6–0.7) B3 (0.63–0.69) B3 (0.63–0.69) B4 (0.64–0.67) 30 (60† for MSS)
NIR 1 B3 (0.7–0.8) 60
NIR B4 (0.8–1.1) B4 (0.76–0.90) B4 (0.77–0.90) B5 (0.85–0.88) 30 (60† for MSS)
SWIR 1 B5 (1.55–1.75) B5 (1.55–1.75) B6 (1.57–1.65) 30
SWIR 2 B7 (2.08–2.35) B7 (2.09–2.35) B7 (2.11–2.29) 30
Thermal B6 (10.40–12.50) B6‡ (10.40–12.50) B10 (10.60–11.19) 30†
B11 (11.50–12.51)
Pan-­Chromatic B8 (0.52–0.90) B8 (0.50–0.68) 15
Cirrus B9 (1.36–1.38) 30

Notes: The table shows each band number and the corresponding wavelength range (in parentheses, micrometers). The exact spec-
tral range of each band varies among sensors, but are comparable for many applications. The MSS bands were originally numbered
4–7, but were relabeled 1–4 with the launch of Landsat 4. Empty cells occur where a particular sensor was not present in a satellite.
This table was adapted from USGS (2015a).
† Data are recorded at a coarser pixel size, but upsampled and delivered at the size listed in table. Native pixel size of MSS data was
79 × 57 m. Native pixel size of thermal data has changed with the sensors: 60 m (TM), 120 m (ETM+), 100 m (TIRS).
‡ Landsat 7 thermal data are acquired at both high and low gain settings. The data are therefore delivered with two Band 6 files
(USGS 2015a).
Concepts & Synthesis

studies (Pasquarella et al. 2016). The multispectral bands that exists on the ground, while the image is the collection
enable local to regional-­scale mapping of vegetation type of spatially arranged measurements (i.e., bands) captured
and condition (Jones and Vaughan 2010). The NIR and in the scene at a single time. Each scene has an assigned
SWIR bands are particularly useful for mapping plant location that is defined by the Worldwide Reference
and soil moisture characteristics, as well as water quality System (WRS). This system assigns a path determined by
in wetlands, rivers, and coastal environments (Roy et al. the s­atellite’s orbit (i.e., vertical, latitudinal) and a row
2014). The thermal infrared bands play an important role (i.e., ­horizontal, longitudinal) for each scene, providing a
in mapping and understanding wildfire ecology (Wang global index for database cataloging, querying, and dissem-
et al. 2010), managing of water resources and monitoring ination. Landsat data are distributed in two WRSs; Landsat
of evapotranspiration (Anderson et al. 2012), and land 1–3 follow WRS1, while Landsat 4–8 follow WRS2 due to
cover classification (Sun and Schulz 2015). The majority the differing orbit altitudes of the satellites.
of Landsat data are delivered at a pixel size of 30 m The USGS manages Landsat data and disseminates
(Table 1). This pixel size prevents fine-­scale mapping of them via a number of online portals (e.g., EarthExplorer;
surface features; however, it is often beneficial in ecology, USGS 2015b). Although a number of Landsat-­ based
as it accurately captures landscape-­scale characteristics data products are available at various levels of preproc-
while avoiding the significant computational require- essing (e.g., WELD; Roy et al. 2010), we focus on the
ments associated with hyper-­spatial and hyper-­spectral current and primary Landsat archive processed by the
sensors. This pixel size also tends to correspond well with USGS, with particular emphasis on the Level-­ 1 and
many management-­level activities. Landsat 4–8 have a higher-­ level Climate Data Records (CDR) products.
revisit interval of 16 d (Schowengerdt 2007), which facil- Level-­1 products are a part of the Collection 1 data sys-
itates studies of landscapes through time (for a review, see tematically processed by the USGS to standardized tiers
Hansen and Loveland 2012, Willis 2015). Additionally, based on data quality and processing level, while higher-­
ecologists have increasingly incorporated Landsat level CDR products provide additional levels of preproc-
imagery, Global Positioning System (GPS) field data, essing. These products are explained further in the
topographic data, and other ancillary variables in correl- following sections. At this time, former Landsat products
ative spatial models over the last decade. (e.g., L1T) are being reprocessed to be included in the
Collection 1 Level-­1 data archive.
The USGS is working on a project to provide data in an
Data coverage and dissemination
alternative format called Analysis Ready Data (ARD) as
Landsat satellites record data as they orbit the Earth; an addition to the products already available. The initial
these data are then systematically partitioned into images purpose for these data is to provide standardized products
based on scene location and date. The terms scene and for users of large data amounts over space and/ or time
image are often confused in the literature when referring to and to reduce or eliminate the need for preprocessing.
these data. We use the framework provided by Strahler These efforts may provide a reasonable alternative to
et al. (1986) who define the scene as the extent, or footprint, some or all of the preprocessing steps described below for
April 2017 SURVIVAL GUIDE TO LANDSAT PREPROCESSING 923

certain users, but understanding the motivation and steps (Chander et al. 2009). Landsat Level-­ 1 products are
of preprocessing will still be valuable in determining delivered as digital numbers, which can be converted to
whether Level-­1 or ARD is appropriate. absolute units of radiance or reflectance. Reflectance is a
unitless measure of the ratio of radiation reflected by an
object relative to the radiation incident upon the object.
Preprocessing
Ecological studies most commonly use DN and reflec-
Images acquired by Landsat sensors are subject to dis- tance. For the thermal bands, studies often use DN or
tortion as a result of sensor, solar, atmospheric, and top- temperature (degrees Kelvin).
ographic effects. Preprocessing attempts to minimize Preprocessing to these specific units, along with cor-
these effects to the extent desired for a particular appli- recting for radiometric artifacts, typically follows a
cation. However, preprocessing steps are time-­consuming, general workflow (Fig. 3). Most of the steps can be clas-
imperfectly address the artifacts to be removed, and have sified into three broad groups: geometric, absolute, and
the potential to introduce additional sources of error. relative. Some steps convert the imagery from one unit to
Many ecological applications require further preproc- another (e.g., at-­sensor radiance to top-­of-­atmosphere
essing than that provided by Level-­1 products or even reflectance via solar correction), while others specifically
CDR products before performing an analysis. These pre- address potential artifacts (e.g., topographic correction).
processing steps can significantly impact analysis results We present the workflow assuming the analyst is starting
(Sundaresan et al. 2007) and have a general order in with either Level-­1 or high-­level surface reflectance CDR
which they should be performed. While methods sections products. Although other products (e.g., Level 0 raw
in the literature commonly mention these steps, the justi- imagery) may be desired for a particular analysis, they are
fication or reasoning for performing them are often vague rarely used in ecological studies and require additional
or omitted, creating confusion about which steps should expertise and understanding to process. Note that Level-­1
be considered and what they accomplish for a particular products have assumptions (e.g., the use of cubic convo-
application. In this section, we provide a description of lution resampling vs. alternative resampling methods)

Concepts & Synthesis

the most common preprocessing steps applied to Landsat that may need to be considered for some applications.
products and their importance. The preprocessing steps presented in the workflow can be
Before preprocessing Landsat imagery, it is important implemented using a variety of software packages that
to understand the units commonly associated with these continue to develop and improve. Some free options
data: digital number (DN), radiance, and reflectance for include R packages ‘RStoolbox’ and ‘landsat’, as well as
the visible to SWIR (vis-­ SWIR) bands and, for the QGIS, GRASS GIS, Google Earth Engine, and
thermal bands, DN, radiance, and temperature (Fig. 2). MultiSpec. Proprietary software includes ENVI, ArcGIS,
Initial sensor-­recorded signals are calibrated to radiance ERDAS IMAGINE, Geomatica, and MATLAB. Some
values using gains and offsets that differ among sensors of these software options provide a graphical user
and over time due to sensor degradation. Radiance interface while others use command-­ line processing,
(watts·steradian−1·m−2·μm−1) is the measure of energy which requires code input in various programming lan-
flux recorded by the sensor. These values are then res- guages. Therefore, software selection is largely driven by
caled to digital number as 6-­bit or 7-­bit (MSS), 8-­bit the familiarity of the analyst and their collaborators with
(TM, ETM+), or 12-­bit (OLI, TIRS) unsigned integers the software as well as the demands of a given analysis.

Fig. 2.  The common units of Landsat imagery used in ecological analysis. The units change as each step of absolute correction
is performed: conversion to radiance, solar correction/thermal calibration, and atmospheric correction.
 924 NICHOLAS E. YOUNG ET AL. Ecology, Vol. 98, No. 4
Concepts & Synthesis

Fig. 3.  Sequential workflow of potential preprocessing steps. Steps are presented in the recommended sequence, but not all
steps are necessary for all analyses. Boldface type indicates preprocessing level. The two flowcharts show (a) general preprocessing
steps for visual and shortwave infrared (vis-­SWIR) data from both high-­level data products (dashed line; e.g., Landsat surface
reflectance CDR) and Level-­1 Landsat products (solid line) and (b) the general preprocessing steps for thermal data. Note that a
high-­level land surface temperature product is not currently available, but is expected to be available soon. Refer to the section
Preprocessing in the text for more detail and discussion of each step.

Geometric correction model (DEM). The Landsat Level-­1 products are termed
“terrain-­corrected” and the majority of images can be
The processes of georeferencing (alignment of imagery used as delivered by the USGS (2015c). Collection 1
to its correct geographic location) and orthorectifying images are classified into tiers based on quality and pro-
(correction for the effects of relief and view direction on cessing level (USGS 2016c). Tier 1 products are consist-
pixel location) are components of geometric correction ently georegistered within ≤12 m radial root mean square
necessary to ensure the exact positioning of an image. error, making them suitable for time-­ series pixel-­
level
Imagery can be positioned relative to the datum, topog- analysis. The other tiers (currently Tier 2 and Real-­Time)
raphy, or other data types, including reference data and may not necessarily be as accurate and should be eval-
additional geospatial layers that might be used in the uated on a case-­ by-­
case basis. Comparatively, the
analyses. Landsat Level-­1 products are precision regis- geometric accuracy of Pre-­Collection L1T products is
tered and orthorectified through a systematic process that approximately 30 m within the United States and about
involves ground control points and a digital elevation 50 m globally but the relative geolocation within an image
April 2017 SURVIVAL GUIDE TO LANDSAT PREPROCESSING 925

is much better (Loveland and Dwyer 2012). A small acquisition dates to a common scale for consistent Earth
fraction of Tier 2 and Pre-­Collection L1T images contain monitoring through time (Chander et al. 2009, Markham
errors and there may be cases where specific Landsat and Helder 2012). Digital numbers cannot be used to
images require additional georeferencing. For example, compare spectral values across time due to sensor degra-
additional georeferencing is more likely to be needed dation and differences between sensors. While DNs can
when working across large spatial extents or timeframes, be used effectively for many single-­ image analyses,
working with older imagery in the Landsat archive, or absolute correction is needed to bring the values to a com-
working in areas of the world where the USGS has not parable scale. As described above, the DNs provided in
been able to obtain sufficient ground control points Level-­1 products are calibrated radiance values that have
(although the USGS is continually improving its ground been scaled to varying bit depths. Conversion to radiance
control; Bodart et al. 2011, Avitabile et al. 2012). When a is the preprocessing step whereby the DNs are converted
workflow involves stacking multiple images, mosaicking back to radiance (often termed at-­sensor radiance) by
adjacent images, compositing ancillary spatial layers, or using rescaling factors (i.e., calibration coefficients) asso-
using georeferenced ground data, the alignment of fea- ciated with each band for a given sensor. The rescaling
tures in the images should be checked quantitatively with factors are stored in the metadata file associated with each
ground control points or, at the very least, through visual image. Often this preprocessing step is described as sensor
assessment. Discrepancies should be corrected prior to calibration, which specifically refers to the determination
analysis using a process known as co-­registration (often of the coefficients used to conduct the conversion, not the
referred to as just registration). Registration involves conversion itself. The conversion to radiance step is nec-
aligning data layers relative to one another, while georef- essary before additional absolute correction steps;
erencing involves aligning layers to the correct geographic however, this level of preprocessing alone should rarely be
location. Registration is a critical step in preprocessing used for analysis because the conversion is linear and
Landsat imagery for ecological analysis, since a misregis- therefore little additional information is gained relative to
tration can result in significant errors, especially in change using DNs. Many of the available software programs will

Concepts & Synthesis

detection analyses (Sundaresan et al. 2007). When relating automatically perform this conversion, as well as an addi-
Landsat data to ancillary georeferenced data, such as tional solar correction (see Solar correction), when sup-
GPS-­marked plot data, images should be georeferenced plied with the metadata file associated with the image (a
rather than registered to maintain alignment between text file with “MTL” in the name).
data. There are numerous approaches for both georefer-
encing and registering Landsat data, and the process
Solar correction
might involve a simple pixel shift or a more complex auto-
mated feature detection and matching between images The next preprocessing step, solar correction, accounts
(for review, see Brown 1992, Zitová and Flusser 2003). for solar influences on pixel values. Solar correction con-
verts at-­sensor radiance to top-­ of-­
atmosphere (TOA)
reflectance by incorporating exoatmospheric solar irra-
Absolute radiometric correction
diance (power of the sun), Earth-­Sun distance, and solar
Absolute radiometric correction (sometimes referred elevation angle. These vary with date, time, and latitude
to as just absolute correction) can refer to a single pre- so their effects must be accounted for when working
processing step or a collection of preprocessing steps that across multiple images, even within a single scene. Top-­
account for sensor, solar, atmospheric, and topographic of-­atmosphere reflectance is a measure of the proportion
effects. The term absolute is used to describe the process of incoming radiation reflected from a surface as detected
of obtaining “true” and comparable values, although the from above the atmosphere. The solar correction step is
corrected values are still approximations. Values obtained often grouped with conversion to radiance in the liter-
from absolute correction can be compared (across time, ature. Similar to conversion to radiance, the solar cor-
space, or sensor) to images that have undergone the same rection values for Landsat data can be retrieved from the
level of correction. Some levels of absolute correction are metadata files associated with each image or, in some
better suited for comparing across images (e.g., surface cases, found in lookup tables. Landsat 8 provides coeffi-
reflectance/land surface temperature) than others (e.g., cients to convert directly to a TOA reflectance product
at-­
sensor radiance). The term absolute correction is from DNs, but this should not be considered true TOA
heavily used in the Landsat preprocessing literature to reflectance since the process does not provide a correction
clarify the interpretation of the resulting values, and is for the solar elevation angle (USGS 2014).
often contrasted with relative radiometric correction (see
Relative radiometric correction).
Atmospheric correction
The energy that is captured by Landsat sensors is influ-
Conversion to radiance
enced by the Earth’s atmosphere. These effects include
Significant efforts have been made to bring the data col- scattering and absorption due to interactions of the elec-
lected across the multiple Landsat missions and tromagnetic radiation with atmospheric particles (i.e.,
 926 NICHOLAS E. YOUNG ET AL. Ecology, Vol. 98, No. 4
Concepts & Synthesis

Fig. 4.  Dichotomous decision tree for determining the level of preprocessing necessary for most ecological applications, guided
by the spatial and temporal extent of the study. The graphics on the tree branches represent images (gray squares) and the project
study area (dashed lines). “Single time” refers to images captured on the same day (or closest day if adjacent paths) to represent one
period in time, or images made into a mosaic to represent a single date. Normalization for multi-­scene and multi-­temporal studies
should be done in two steps: horizontal (images from across multiple with same/similar dates) relative correction then vertical
(images from across different times) relative correction.

gases, water vapor, and aerosols). Atmospheric correction such, we recommend not performing atmospheric cor-
attempts to account for these effects. However, some rection unless necessary (discussed in Determining the
atmospheric effects are highly variable over the Earth’s appropriate level of preprocessing and in Fig. 4) and to use
surface and can be difficult to correct in Landsat imagery. freely available high-­ level products when needed
While it is not always necessary to atmospherically correct (e.g., Landsat CDR; discussed in High-level products).
Landsat data to surface values, there are instances where
this level of correction is needed. In general, absolute
Topographic correction
atmospheric corrections are needed when (1) an empirical
model is being created for application beyond the data Solar correction does not account for illumination
used to develop it, (2) there is a comparison being made to effects from slope, aspect, and elevation that can cause
ground reflectance data such as a field-­based spectroradi- variations in reflectance values for similar features with
ometer, or (3) as an alternative to relative correction when different terrain positions (Riaño et al. 2003). Topographic
comparisons are being made across multiple images. All correction is the process used to account for these effects.
atmospheric correction methods have associated assump- While this correction is not always required, it can be
tions about the target and the nature of the atmospheric especially important for applications in mountain systems
particles or emissivity (for land surface temperature). or rugged terrain (Colby 1991, Riaño et al. 2003, Shepherd
There are numerous atmospheric correction methods and Dymond 2003), which are common settings for sat-
available, ranging from simple approaches that use only ellite monitoring due to the difficulty of accessing these
within-­image information such as dark object subtraction environments for field measurements.
(Chavez 1988), to more complex and data-­ intensive An important distinction should be made between top-
approaches such as the method used for the Landsat ographic and terrain correction. Topographic correction
Ecosystem Disturbance Adaptive Processing System is a radiometric process while terrain correction is
(LEDAPS) products (Masek et al. 2006). The complex geometric in nature. Although Landsat Level-­1 products
methods are generally more accurate than simpler are terrain corrected, this does not account for the same
methods; however, they often require ancillary data about effects as a topographic correction. Terrain correction
atmospheric conditions at the time of image collection ensures each pixel is displayed as viewed from directly
and can be difficult to implement, especially for non-­ above regardless of topography or view angle, and, while
remote sensing experts. Further, these corrections can important, does not account for the same effects as top-
introduce additional errors (Schroeder et al. 2006). As ographic correction.
April 2017 SURVIVAL GUIDE TO LANDSAT PREPROCESSING 927

Numerous methods exist for performing topographic transformed. Two general approaches are used to
correction. Band ratios are a simple way to partially compare these overlapping regions. The first is histogram
reduce impacts from illumination effects without the use matching (also referred to as equalizing), which uses
of a digital elevation model (DEM; Holben and Justice information from all pixels in the overlapping regions.
1981, Colby 1991, Hale and Rock 2003). This method We do not recommend histogram matching Landsat
assumes reflectance will increase or decrease propor- data, as this approach is generally used for relatively cor-
tionally in both bands being used, however this recting multiple images from the same day when there is
assumption is demonstrably false for indirect radiation little variation in the solar and atmospheric conditions.
(i.e., diffuse irradiance; Leprieur et al. 1988, Conese et al. Adjoining Landsat images collected on the same day (i.e.,
1993). Conversely, more complex approaches can be used along the same path) can be mosaicked without a relative
with a DEM to model the illumination effects by taking correction.
into account the specific slope and aspect of the terrain. The second approach uses pseudo invariant features
Several methods are commonly used in the literature (PIF; generally consisting of one or more pixels) or pseudo
(Riaño et al. 2003, Vanonckelen et al. 2013), and invariant targets (PIT; generally a single pixel), where non-­
numerous studies comparing these methods have been changing features or targets in the overlapping area are
published with varying results depending on study used to bring the images to a common scale (Schott et al.
location and application (e.g., Meyer et al. 1993, Lu et al. 1988). These pseudo invariant features/targets can be
2002, Hale and Rock 2003, Riaño et al. 2003, Richter selected manually or automatically through statistical
et al. 2009, Hantson and Chuvieco 2011, Vanonckelen algorithms (Du et al. 2002, Bao et al. 2012). Comparisons
et al. 2013, Adhikari et al. 2015). While no specific topo- have shown that performing a relative correction might
graphic correction method is superior in all cases, the provide more accurate results than an absolute atmos-
Minnaert Correction (Minnaert 1941) and C-­correction pheric correction, especially when comparability is more
(Teillet et al. 1982) methods have shown consistently important than the pixel value (Schroeder et al. 2006).
good performance for removing topographic effects Furthermore, some studies have performed relative cor-

Concepts & Synthesis

within Landsat imagery (Riaño et al. 2003, Gao and rection of atmospherically corrected imagery to account
Zhang 2009, Richter et al. 2009, Hantson and Chuvieco for residual differences between images arising from
2011, Vanonckelen et al. 2013). Sola et al. (2016) provide imperfect preprocessing when surface reflectance is desired
a recent and thorough evaluation of many of these (Li et al. 2014). We suggest this approach only if both com-
methods, which can be implemented in several open-­ parability between images and surface reflectance units are
source and commercial software products. Given the needed for a particular application. If the objective is to
number of available approaches and the difficulty of their obtain images that are spatially or temporally consistent
application by non-­ remote sensing experts, analysts (and that is the only goal of the relative correction), then
should carefully consider whether a topographic cor- relative correction from a more basic preprocessing level,
rection is necessary for their study. While this preproc- such as DN or TOA, would be the most appropriate
essing step can be more important than atmospheric option, as correcting to surface reflectance serves no
correction for some applications in topographically purpose and injects a potential source of error.
complex regions (Vanonckelen et al. 2013), this step is not
needed for every scenario.
Preprocessing for spectral indices
Spectral indices are transformations and reductions of
Relative radiometric correction
spectral data that are used to highlight specific phe-
Ecologists often conduct analyses that require con- nomena on the landscape. Indices facilitate data interpre-
sistent spectral values across space and time. However, tation and comparison, both qualitative (e.g., pattern
such studies do not necessarily require the use of true visualization) and quantitative (e.g., monitoring plant
surface reflectance values. A common preprocessing step health). A plethora of spectral indices have been designed
when using multiple images is a relative radiometric cor- to highlight vegetation, hydrology, geology, burned
rection or normalization (hereafter referred to as relative areas, and snow, to name a few (for a review, see Jackson
correction). Relative correction brings each band of an and Huete 1991, Bannari et al. 1995, Lozano et al. 2007).
image to the same radiometric scale as the corresponding They can be particularly useful as relative measures
band of a reference image. This presumably accounts for where reference data do not exist to model such variables
sensor, solar, and atmospheric differences, but does not quantitatively.
expressly account for vegetation phenological differences. The approach to preparing multiple bands for use in
Similar to atmospheric and topographic correction, spectral indices should be the same as the approach for
there are many approaches to relative correction, and preparing individual bands for any analysis; the analyst
there is no clear preferred method that consistently per- should determine what level of preprocessing is necessary
forms best across all applications (Chen et al. 2005). based on the characteristics and objectives of the study.
Relative correction typically uses the overlapping regions Ratio-­based indices account for multiplicative atmos-
between a reference image and another image to be pheric artifacts, but do not correct additive effects and
 928 NICHOLAS E. YOUNG ET AL. Ecology, Vol. 98, No. 4

atmospheric differences between bands. Therefore, when factors; USGS 2016b). These are available through the
comparing spectral indices across multiple images (time USGS EROS Center Science Processing Architecture
or space), we recommend a correction to eliminate such (ESPA) on demand interface (USGS 2016d). Fur­
effects. There are potential exceptions; notably, some thermore, at the time of this writing, the USGS has
indices have correction factors that are dependent on the expressed that they will deliver essential climate variables
data being in a specific unit, such as the soil adjusted veg- (ECVs) that will include land surface temperature,
etation index (reflectance) or the tasseled cap transfor- burned area extent, dynamic surface water, and snow
mation (varies by sensor/method). Further, if image cover area as additional high-­level products in the near
correction is not feasible, atmospherically resistant future. While these products will provide the data
indices may be used to reduce some atmospheric effects required to perform large extent and accurate ecological
(Pinty and Verstraete 1992, Huete et al. 1997). analyses with only a fraction of the preprocessing time,
users will benefit from understanding the concepts and
assumptions that are inherent in their creation.
High-­Level Products
High-­level Landsat products are data that have been
Determining the Appropriate Level of Preprocessing
preprocessed to a level beyond the specifications of
Level-­1 products and are directly available for download. While understanding the levels and appropriate order
These products eliminate the need for the user to com- of preprocessing steps provides a necessary foundation,
plete certain preprocessing steps, a welcome advancement the analyst is ultimately left with the question, “What
that simplifies the preprocessing workflow. level of correction is ‘good enough’ for my study?” To
The USGS Landsat Surface Reflectance Climate Data help answer this question, we start with the guiding prin-
Records (CDR) are the most notable of the current high-­ ciple that only the preprocessing steps necessary for a
level data products. These are surface reflectance products given analysis should be applied because each step may
Concepts & Synthesis

that can be downloaded from multiple USGS sources risk introducing artifacts and/or error into the data (Song
(USGS 2015b, 2016a). The Landsat surface reflectance et al. 2001, Riaño et al. 2003, Kennedy et al. 2009). Even
CDRs are processed to calibrate raw DNs to TOA reflec- with this core tenet, it can still be challenging to identify
tance (also available for download), and then corrected the appropriate level of preprocessing.
to surface reflectance using atmospheric parameters and We designed a flexible decision tree to help navigate
a DEM. Furthermore, TOA brightness temperature, as these choices (Fig. 4). The first row of the tree refers to
well as masks for clouds, cloud shadows, adjacent clouds, analyses that require the data be converted to a specific
land, and water are also available as high-­level products value type, such as a parameter for an existing model.
from the USGS. Land surface temperature and MSS Beyond these scenarios, we recommend the analyst guide
data are not currently included with the Landsat surface their decision by identifying how the spatial and temporal
reflectance CDRs, however, the USGS has publicly extent of their analysis relates to the existing scene bound-
stated its intent to release a land surface temperature aries. The questions in the decision tree specifically refer
product (Laraby 2016). The quality of the surface reflec- to when and where the analysis of spectral values is taking
tance CDR products can be adversely affected by low sun place within the broader workflow. Row two of the
angle, excessive clouds, high latitudes (>65 degrees north decision tree focuses on analyses taking place within a
or south), or locations that are very arid or snow/water single image, and thus calls for the use of DNs to avoid
dominated (USGS 2015d). Moreover, the process to unnecessary preprocessing. Row three considers analyses
develop surface reflectance CDR products contains non- using multiple images. This could be multi-­temporal (i.e.,
linear elements, which can introduce error and unwanted multiple images from different dates analyzed as separate
data artifacts. Nevertheless, these data are the products time steps, also referred to as multi-­date), multi-­scene
of high-­quality, well-­documented approaches that are (i.e., images with different path/row designations), or a
likely as reliable as any approach readily available to combination of the two. This row calls for either surface
ecologists. Quality assurance layers are available as an or relative units to account for differences between
additional USGS high-­level product for pixel-­level con- images resulting from sensor, solar, atmospheric, and
dition and validity flags. topographic effects. In such cases, we recommend per-
The USGS continues to release additional high-­level forming a relative correction to normalize values.
data and make improvements to the existing products. In Nevertheless, because relative corrections typically
addition to surface reflectance, the USGS has made require more expertise than obtaining surface reflectance
available some of the most widely used spectral indices CDR products, these products can be a valid and con-
derived from Landsat Surface Reflectance high-­ level venient alternative with high accuracy (Vuolo et al. 2015),
products (Masek et al. 2006). These products are gen- especially if the images are similar in time, phenology, or
erated from the surface reflectance data and therefore growing degree days. Caution should be exercised with
have the same caveats. However, they also carry addi- this approach when the analysis spans different sensors
tional assumptions inherent to the derivation of the (e.g., Landsat 7 ETM+ and Landsat 8 OLI) since sys-
indices themselves (e.g., coefficient values and correction tematic biases appear among their surface reflectance
April 2017 SURVIVAL GUIDE TO LANDSAT PREPROCESSING 929

values, and these biases can be magnified in products Although multiple times are used, the spectral values are
such as the normalized difference vegetation index being analyzed independently to create two categorical
(NDVI; Roy et al. 2016). maps of forest cover because only one image is used for
We illustrate these concepts in the following section by each classification. The analysis is detecting changes in
walking through three examples commonly encountered the cover classes from each image, not the spectral val-
by ecologists working with Landsat data: specific value ues. It is therefore appropriate in this case to classify each
type, change detection, and correlative modeling. These image using its respective DN values (Fig. 4, row 2) and
examples are not intended to be exhaustive, but demon- subsequently perform the change detection. If the anal-
strate the use and flexibility of the decision tree. ysis involved multiple scenes for each time period, then
this would bring the user to the third row of the ­decision
tree and preprocessing to relative values or u ­ sing surface
Example scenario 1: Specific value type
reflectance would be required (Fig. 4, row 3).
Landsat data are often used to measure specific
physical properties of the Earth’s surface. For example, a
Example scenario 3: Mapping with correlative models
researcher may want to use Landsat data to measure
surface albedo to evaluate post fire dynamics in a forest Landsat bands and derived spectral indices are
environment. Because surface reflectance is used to derive increasingly used as variables in correlative models (e.g.,
surface albedo, the researcher would require a specific regression tree analysis, random forests, MaxEnt). These
value type (Fig. 4, row 1) and preprocess the data to are used to map continuous responses (e.g., percent
surface reflectance. Another example is the tasseled cap cover, biomass, habitat suitability) as well as discrete
transformation. The coefficients used for the tasseled cap classes (e.g., land cover type; He et al. 2015, Lawrence
transformation vary between Landsat sensors; some are and Moran 2015). In such cases, if the study area is
intended to be used with reflectance data, while others encompassed within a single image, DN is an appro-
require DNs. In this case, the researcher would again priate preprocessing level to use (Fig. 4, section 2).

Concepts & Synthesis

require a specific value type and identify the specific levelHowever, including multi-­temporal imagery in correl-
of preprocessing required for their sensor. ative models can significantly improve mapping of tem-
porally dynamic features such as wetlands, vegetation,
or deciduous forests (Baker et al. 2006, Evangelista et al.
Example scenario 2: Change detection
2009, Savage et al. 2015). There are a variety of ways to
The Landsat program’s historical archive makes it par- include multi-­temporal imagery in such models. Digital
ticularly useful for mapping environmental change. number is appropriate when bands or indices from
There are many approaches to change detection with multi-­temporal images are included as individual model
remote sensing data, each with their own advantages and variables and the data are not combined in any way
limitations. Most ecological studies addressing thematic (e.g., differencing, ratioing). This is because the images
change take either a pre-­ classification or post-­ are being treated as individual variables and no analysis
classification approach. is taking place between the spectral values prior to model
development. However, if the bands or indices are
Pre-­classification change detection.—Pre-­classification planned to be combined to produce a new variable, the
change detection involves comparing the spectral values spectral analysis is taking place across time and the data
from multiple dates and subsequently classifying the pix- should be corrected to surface or relative values before
els by their change in values. Examples include image dif- combination and subsequent model development (Fig. 4,
ferencing (i.e., subtracting pixel values between each pair row 3). If the study area spans multiple scenes for any of
of bands) and multi-­temporal principal component anal- the scenarios above, surface reflectance or relative values
ysis, in which the bands of two images are combined into are necessary (Fig. 4, row 3).
a single composite raster and the data are transformed to
identify change classes. In these cases, the spectral values
Conclusion
are being compared across images from multiple times,
and the imagery should be preprocessed to either surface Landsat data are a rich resource for ecological analysis,
reflectance or relative values to ensure comparability but the literature and available resources to appropri-
(Fig. 4, row 3). ately make use of these data are inconsistent, disjointed,
and challenging to interpret for non-­ experts. Our
Post-­classification change detection.—Post-­classification overview and synthesis summarizes the Landsat sensor
change detection involves classifying images from characteristics and clarifies key preprocessing steps nec-
­multiple dates and subsequently comparing one cate- essary to work with the data. The workflow and decision
gorical map to the other to identify changes. Consider a tree can be used as guides for determining the necessary
study designed to map forest loss by classifying the pixels sequence of steps to achieve the appropriate level of
of a 1980 Landsat image and a 2010 image of the same image preprocessing for common ecological applica-
scene and subsequently calculating the changes in ­classes. tions. Equipped with this knowledge, readers will be able
 930 NICHOLAS E. YOUNG ET AL. Ecology, Vol. 98, No. 4

to critically evaluate the approaches described in the Baker, C., R. Lawrence, C. Montagne, and D. Patten. 2006.
­literature and adapt them to their own application. Mapping wetlands and riparian areas using landsat etm+ im-
We recommend taking a parsimonious approach to agery and decision-­tree-­based models. Wetlands 26:465–474.
Bannari, A., D. Morin, F. Bonn, and A. R. Huete. 1995. A
preprocessing; correct the artifacts necessary for a par- ­review of vegetation indices. Remote Sensing Reviews 13:
ticular application, but avoid unnecessary steps that may 95–120.
introduce additional artifacts without gaining additional Bao, N. S., A. M. Lechner, A. Fletcher, A. Mellor, D. Mulligan,
value (Song et al. 2001, Riaño et al. 2003, Kennedy et al. and Z. K. Bai. 2012. Comparison of relative radiometric nor-
2009). When analyzing across multiple images, we rec- malization methods using pseudo-­ invariant features for
ommend performing a relative correction to normalize change detection studies in rural and urban landscapes.
Journal of Applied Remote Sensing 6:63571–63578.
values. However, the quality and availability of high-­
Bodart, C., H. Eva, R. Beuchle, R. Raši, D. Simonetti, H.-J.
level surface reflectance products continue to improve Stibig, A. Brink, E. Lindquist, and F. Achard. 2011. Pre-­
(Vuolo et al. 2015) and we encourage readers to consider processing of a sample of multi-­scene and multi-­date Landsat
using these products while taking into account the imagery used to monitor forest cover changes over the trop-
assumptions inherent in their creation. ics. ISPRS Journal of Photogrammetry and Remote Sensing
This survival guide reduces the potential for confusion 66:555–563.
Brown, L. G. 1992. A survey of image registration techniques.
and ambiguity that many scientists face when determining
ACM Computing Surveys (CSUR) 24:325–376.
how to preprocess their imagery for analysis. A better Chander, G., B. L. Markham, and D. L. Helder. 2009. Summary
understanding of Landsat preprocessing can improve of current radiometric calibration coefficients for Landsat
reproducibility and accuracy as ecologists and others con- MSS, TM, ETM+, and EO-­1 ALI sensors. Remote Sensing
tinue to develop new applications for remote sensing data. of Environment 113:893–903.
This will facilitate integration with other remote sensing Chavez, P. S. 1988. An improved dark-­object subtraction tech-
programs, such as the European Space Agency’s Sentinel-­2 nique for atmospheric scattering correction of multispectral
data. Remote Sensing of Environment 24:459–479.
mission, which has similar spectral bands and is designed
Concepts & Synthesis

Chen, X., L. Vierling, and D. Deering. 2005. A simple and

to integrate with the Landsat Program. Although there ­effective radiometric correction method to improve landscape
are still some issues when using these two data sources change detection across sensors and across time. Remote
together (e.g., Storey et al. 2016), there are significant Sensing of Environment 98:63–79.
efforts underway (MuSLI, Multi-­Source Land Imaging) Cohen, W. B., and S. N. Goward. 2004. Landsat’s role in ecologi-
to resolve them. This cross-­platform consistency makes cal applications of remote sensing. BioScience 54:535–545.
Colby, J. D. 1991. Topographic normalization in rugged ­terrain.
our conceptual workflow easily adaptable to current and
Photogrammetric Engineering and Remote Sensing 57:
future multispectral remote sensing missions, enabling 531–537.
unprecedented opportunities for comparative research Conese, C., M. A. Gilabert, F. Maselli, and L. Bottai. 1993.
and monitoring of ecological processes. Topographic normalization of TM scenes through the use of
an atmospheric correction method and digital terrain models.
Acknowledgments Photogrammetric Engineering and Remote Sensing 59:
1745–1753.
This research was supported by the Agriculture and Food Du, Y., P. M. Teillet, and J. Cihlar. 2002. Radiometric normal-
Research Initiative Competitive Grant (grant number ization of multitemporal high-­resolution satellite images with
2013-­68005-­21298; Bioenergy Alliance Network of the Rockies) quality control for land cover change detection. Remote
from the USDA-­NIFA, the U.S. Geological Survey, and the Sensing of Environment 82:123–134.
National Science Foundation (no. 1313728). We would like to Evangelista, P. H., T. J. Stohlgren, J. T. Morisette, and
thank the undergraduate and graduate students who provided S. Kumar. 2009. Mapping invasive tamarisk (Tamarix): a
invaluable feedback on early versions of the concept, workflow comparison of single-­ scene and time-­ series analyses of
and decision tree. In addition, we are grateful to Warren Cohen ­remotely sensed data. Remote Sensing 1:519–533.
for a thoughtful, collegial review of earlier versions of this Gao, Y., and W. Zhang. 2009. LULC classification and topo-
manuscript. graphic correction of Landsat-­ 7 ETM+ imagery in the
Yangjia River Watershed: the influence of DEM resolution.
Sensors 9:1980–1995.
Literature Cited
Hale, S. R., and B. N. Rock. 2003. Impact of topographic nor-
Adhikari, H., J. Heiskanen, E. E. Maeda, and P. K. E. Pellikka. malization on land-­ cover classification accuracy. Photo­
2015. Does topographic normalization of landsat images grammetric Engineering & Remote Sensing 69:785–791.
­improve fractional tree cover mapping in tropical mountains? Hansen, M. C., and T. R. Loveland. 2012. A review of large
ISPRS – International Archives of the Photogrammetry, area monitoring of land cover change using Landsat data.
Remote Sensing and Spatial Information Sciences XL-7/ W3: Remote Sensing of Environment 122:66–74.
261–267. Hantson, S., and E. Chuvieco. 2011. Evaluation of different
Anderson, M. C., R. G. Allen, A. Morse, and W. P. Kustas. topographic correction methods for Landsat imagery.
2012. Use of Landsat thermal imagery in monitoring evapo- International Journal of Applied Earth Observation and
transpiration and managing water resources. Remote Sensing Geoinformation 13:691–700.
of Environment 122:50–65. He, K. S., B. A. Bradley, A. F. Cord, D. Rocchini, M.-N.
Avitabile, V., A. Baccini, M. A. Friedl, and C. Schmullius. 2012. Tuanmu, S. Schmidtlein, W. Turner, M. Wegmann, and
Capabilities and limitations of Landsat and land cover data N. Pettorelli. 2015. Will remote sensing shape the next gener-
for aboveground woody biomass estimation of Uganda. ation of species distribution models? Remote Sensing in
Remote Sensing of Environment 117:366–380. Ecology and Conservation 1:4–18.
April 2017 SURVIVAL GUIDE TO LANDSAT PREPROCESSING 931

Holben, B., and C. Justice. 1981. An examination of spectral in the United States. Executive report. U.S. Geological
band ratioing to reduce the topographic effect on remotely Survey, Reston, Virigina, USA.
sensed data. International Journal of Remote Sensing 2: Minnaert, M. 1941. The reciprocity principle in lunar photome-
115–133. try. Astrophysical Journal 93:403–410.
Holden, C. E., and C. E. Woodcock. 2016. An analysis of Pasquarella, V. J., C. E. Holden, L. Kaufman, and C. E.
Landsat 7 and Landsat 8 underflight data and the implications Woodcock. 2016. From imagery to ecology: leveraging time
for time series investigations. Remote Sensing of Environment series of all available Landsat observations to map and mon-
185:16–36. https://doi.org/10.1016/j.rse.2016.02.052 itor ecosystem state and dynamics. Remote Sensing in
Huete, A., H. Liu, W. Batchily, and V. Leeuwen. 1997. A com- Ecology and Conservation 2:152–170.
parison of vegetation indices over a global set of TM images Pettorelli, N., W. F. Laurance, T. G. O’Brien, M. Wegmann,
for EOS-­ MODIS. Remote Sensing of Environment 59: H.  Nagendra, and W. Turner. 2014. Satellite remote sensing
440–451. for applied ecologists: opportunities and challenges. Journal
Jackson, R. D., and A. R. Huete. 1991. Interpreting vegetation of Applied Ecology 51:839–848.
indices. Preventive Veterinary Medicine 11:185–200. Pinty, B., and M. M. Verstraete. 1992. GEMI: a non-­linear
Jones, H. G., and R. A. Vaughan. 2010. Remote sensing of veg- ­index to monitor global vegetation from satellites. Vegetatio
etation: principles, techniques, and applications. Oxford 101:15–20.
University Press, Oxford, England, UK. Riaño, D., E. Chuvieco, J. Salas, and I. Aguado. 2003. Assessment
Kennedy, R. E., P. A. Townsend, J. E. Gross, W. B. Cohen, of different topographic corrections in Landsat-­TM data for
P. Bolstad, Y. Q. Wang, and P. Adams. 2009. Remote sensing mapping vegetation types (2003). IEEE Transactions on
change detection tools for natural resource managers: under- Geoscience and Remote Sensing 41:1056–1061.
standing concepts and tradeoffs in the design of landscape Richter, R., T. Kellenberger, and H. Kaufmann. 2009.
monitoring projects. Remote Sensing of Environment 113: Comparison of topographic correction methods. Remote
1382–1396. Sensing 1:184–196.
Kennedy, R. E., et al. 2014. Bringing an ecological view of Roy, D. P., J. Ju, K. Kline, P. L. Scaramuzza, V. Kovalskyy,
change to Landsat-­ based remote sensing. Frontiers in M. Hansen, T. R. Loveland, E. Vermote, and C. Zhang. 2010.
Ecology and the Environment 12:339–346. Web-­enabled Landsat Data (WELD): Landsat ETM+ com-
Laraby, K. 2016. Landsat surface temperature product: global posited mosaics of the conterminous United States. Remote

Concepts & Synthesis

validation and cloud analysis. Landsat Science Team 2016 Sensing of Environment 114:35–49.
Winter Meeting, Blacksburg, Virginia, USA. Presentation. Roy, D. P., V. Kovalskyy, H. K. Zhang, E. F. Vermote, L. Yan,
Lawrence, R. L., and C. J. Moran. 2015. The AmericaView S. S. Kumar, and A. Egorov. 2016. Characterization of
­classification methods accuracy comparison project: a rigor- Landsat-­7 to Landsat-­8 reflective wavelength and normalized
ous approach for model selection. Remote Sensing of Envir­ difference vegetation index continuity. Remote Sensing of
onment 170:115–120. Environment 185:57–70. https://doi.org/10.1016/j.rse.2015.
Leprieur, C. E., J. M. Durand, and J. L. Peyron. 1988. Influence 12.024
of topography on forest reflectance using Landsat Thematic Roy, D. P., et al. 2014. Landsat-­8: science and product vision
Mapper and digital terrain data. Photogrammetric Engi­ for terrestrial global change research. Remote Sensing of
neering and Remote Sensing 54:491–496. Environment 145:154–172.
Li, J., M. Menenti, A. Mousivand, and S. M. Luthi. 2014. Non-­ Savage, S. L., R. L. Lawrence, and J. R. Squires. 2015.
vegetated playa morphodynamics using multi-­ temporal Predicting relative species composition within mixed conifer
Landsat imagery in a semi-­arid endorheic basin: Salar de forest pixels using zero-­inflated models and Landsat imagery.
Uyuni, Bolivia. Remote Sensing 6:10131–10151. Remote Sensing of Environment 171:326–336.
Loveland, T. R., and J. L. Dwyer. 2012. Landsat: building a Schaepman-Strub, G., M. E. Schaepman, T. H. Painter,
strong future. Remote Sensing of Environment 122:22–29. S. Dangel, and J. V. Martonchik. 2006. Reflectance quantities
Lozano, F. J., S. Suárez-Seoane, and E. de Luis. 2007. in optical remote sensing—definitions and case studies.
Assessment of several spectral indices derived from multi-­ Remote Sensing of Environment 103:27–42.
temporal Landsat data for fire occurrence probability model- Schott, J. R., C. Salvaggio, and W. J. Volchok. 1988. Radio­
ling. Remote Sensing of Environment 107:533–544. metric scene normalization using pseudoinvariant features.
Lu, D., P. Mausel, E. Brondizio, and E. Moran. 2002. Remote Sensing of Environment 26:1–16.
Assessment of atmospheric correction methods for Landsat Schowengerdt, R. A. 2007. Remote sensing: models and meth-
TM data applicable to Amazon basin LBA research. ods for image processing. Third edition. Elsevier Academic
International Journal of Remote Sensing 23:2651–2671. Press, Cambridge, Massachusetts, USA.
Markham, B. L., and D. L. Helder. 2012. Forty-­year calibrated Schroeder, T. A., W. B. Cohen, C. Song, M. J. Canty, and
record of earth-­reflected radiance from Landsat: a review. Z. Yang. 2006. Radiometric correction of multi-­ temporal
Remote Sensing of Environment 122:30–40. Landsat data for characterization of early successional forest
Masek, J. G., E. F. Vermote, N. E. Saleous, R. Wolfe, F. G. patterns in western Oregon. Remote Sensing of Environment
Hall, K. F. Huemmrich, F. Gao, J. Kutler, and T.-K. Lim. 103:16–26.
2006. A Landsat surface reflectance dataset for North Shepherd, J. D., and J. R. Dymond. 2003. Correcting satellite
America, 1990–2000. IEEE Geoscience and Remote Sensing imagery for the variance of reflectance and illumination with
Letters 3:68–72. topography. International Journal of Remote Sensing
Meyer, P., K. I. Itten, T. Kellenberger, S. Sandmeier, and 24:3503–3514.
R. Sandmeier. 1993. Radiometric corrections of topographi- Sola, I., M. González-Audícana, and J. Álvarez-Mozos. 2016.
cally induced effects on Landsat TM data in an alpine envi- Multi-­criteria evaluation of topographic correction methods.
ronment. ISPRS Journal of Photogrammetry and Remote Remote Sensing of Environment 184:247–262.
Sensing 48:17–28. Song, C., C. E. Woodcock, K. C. Seto, M. P. Lenney, and S. A.
Miller, H., L. Richardson, N. Sexton, L. Koontz, J. Loomis, Macomber. 2001. Classification and change detection using
S. Koontz, and C. Hermans. 2011. The users, uses, and value Landsat TM data: When and how to correct atmospheric
of Landsat and other Moderate-Resolution Satellite Imagery ­effects? Remote Sensing of Environment 75:230–244.
 932 NICHOLAS E. YOUNG ET AL. Ecology, Vol. 98, No. 4

Storey, J., D. P. Roy, J. Masek, F. Gascon, J. Dwyer, and U.S. Geological Survey. 2016b. Product guide: Landsat surface
M. Choate. 2016. A note on the temporary misregistration of reflectance-derived spectral indices. Version 2.7. U.S. Geo­
Landsat-­8 Operational Land Imager (OLI) and Sentinel-­2 logical Survey, EROS Data Center. https://landsat.usgs.gov/
Multi Spectral Instrument (MSI) imagery. Remote Sensing of sites/default/files/documents/si_product_guide.pdf
Environment 186:121–122. U.S. Geological Survey. 2016c. Landsat collections. https://
Strahler, A. H., C. E. Woodcock, and J. A. Smith. 1986. On the landsat.usgs.gov/landsat-collections
nature of models in remote sensing. Remote Sensing of U.S. Geological Survey. 2016d. ESPA ordering interface. http://
Environment 20:121–139. espa.cr.usgs.gov
Sun, L., and K. Schulz. 2015. The improvement of land cover Vanonckelen, S., S. Lhermitte, and A. Van Rompaey. 2013.
classification by thermal remote sensing. Remote Sensing The effect of atmospheric and topographic correction meth-
7:8368–8390. ods on land cover classification accuracy. International
Sundaresan, A., P. K. Varshney, and M. K. Arora. 2007. Journal of Applied Earth Observation and Geoinformation
Robustness of change detection algorithms in the presence of 24:9–21.
registration errors. Photogrammetric Engineering & Remote Vogelmann, J. E., A. L. Gallant, H. Shi, and Z. Zhu. 2016.
Sensing 73:375–383. Perspectives on monitoring gradual change across the continu-
Teillet, P. M., B. Guindon, and D. G. Goodenough. 1982. On ity of Landsat sensors using time-­series data. Remote Sensing
the slope-­ aspect correction of multispectral scanner data. of Environment 185:258–270. https://doi.org/10.1016/j.rse.
Canadian Journal of Remote Sensing 8:84–106. 2016.02.060
Turner, W., et al. 2015. Free and open-­access satellite data are Vuolo, F., M. Mattiuzzi, and C. Atzberger. 2015. Compari­son
key to biodiversity conservation. Biological Conservation of the Landsat Surface Reflectance Climate Data Record
182:173–176. (CDR) and manually atmospherically corrected data in a
U.S. Geological Survey. 2014. Landsat 8 data users handbook. semi-­arid European study area. International Journal
USGS Earth Resources Observation and Science, Sioux of Applied Earth Observation and Geoinformation 42:
Falls, South Dakota, USA. 1–10.
U.S. Geological Survey. 2015a. What are the band designations Wang, K., S. E. Franklin, X. Guo, and M. Cattet. 2010. Remote
for the Landsat satellites? https://landsat.usgs.gov/what-are- sensing of ecology, biodiversity and conservation: a review
Concepts & Synthesis

band-designations-landsat-satellites from the perspective of remote sensing specialists. Sensors

U.S. Geological Survey. 2015b. USGS EarthExplorer. http:// 10:9647–9667.
earthexplorer.usgs.gov Willis, K. S. 2015. Remote sensing change detection for ecolog-
U.S. Geological Survey. 2015c. Landsat processing details. ical monitoring in United States protected areas. Biological
https://landsat.usgs.gov/Landsat-Processing-Details Conservation 182:233–242.
U.S. Geological Survey. 2015d. Landsat surface reflectance high Wulder, M. A., J. G. Masek, W. B. Cohen, T. R. Loveland, and
level data products. http://landsat.usgs.gov/CDR_LSR.php C. E. Woodcock. 2012. Opening the archive: How free data
U.S. Geological Survey. 2016a. User guide: Earth resources has enabled the science and monitoring promise of Landsat.
­observation and science (EROS) center science processing Remote Sensing of Environment 122:2–10.
­architecture (ESPA) on demand interface, version 3.3. https:// Zitová, B., and J. Flusser. 2003. Image registration methods:
pubs.er.usgs.gov/publication/70057873 a survey. Image and Vision Computing 21:977–1000.

Supporting Information
Additional supporting information may be found in the online version of this article at http://onlinelibrary.wiley.com/
doi/10.1002/ecy.1730/suppinfo

View publication stats