Zhu 2017
Zhu 2017
a r t i c l e i n f o a b s t r a c t
Article history: The free and open access to all archived Landsat images in 2008 has completely changed the way of using
Received 8 February 2017 Landsat data. Many novel change detection algorithms based on Landsat time series have been developed
Received in revised form 30 June 2017 We present a comprehensive review of four important aspects of change detection studies based on
Accepted 30 June 2017
Landsat time series, including frequencies, preprocessing, algorithms, and applications. We observed
the trend that the more recent the study, the higher the frequency of Landsat time series used. We
reviewed a series of image preprocessing steps, including atmospheric correction, cloud and cloud
Keywords:
shadow detection, and composite/fusion/metrics techniques. We divided all change detection algorithms
Review
Landsat
into six categories, including thresholding, differencing, segmentation, trajectory classification, statistical
Change detection boundary, and regression. Within each category, six major characteristics of different algorithms,
Time series such as frequency, change index, univariate/multivariate, online/offline, abrupt/gradual change, and
sub-pixel/pixel/spatial were analyzed. Moreover, some of the widely-used change detection algorithms
were also discussed. Finally, we reviewed different change detection applications by dividing these
applications into two categories, change target and change agent detection.
Ó 2017 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Published by Elsevier
B.V. All rights reserved.
Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
2. Frequencies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
3. Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
3.1. Atmospheric correction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
3.2. Cloud and cloud shadow detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
3.3. Composite, fusion, and metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
4. Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
4.1. Thresholding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
4.2. Differencing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
4.3. Segmentation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
4.4. Trajectory classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
4.5. Statistical boundary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
4.6. Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
4.7. Some of the widely-used algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
5. Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
5.1. Change target . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
5.2. Change agent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
⇑ Address: Department of Geosciences, Texas Tech University, Lubbock, TX 79409, United States.
E-mail address: [email protected]
http://dx.doi.org/10.1016/j.isprsjprs.2017.06.013
0924-2716/Ó 2017 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Published by Elsevier B.V. All rights reserved.
Z. Zhu / ISPRS Journal of Photogrammetry and Remote Sensing 130 (2017) 370–384 371
6. Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
Fig. 1. Yearly publications from 1994 to 2017 indexed by Scopus. The Search was conduction on June 27, 2017 by search ‘‘change detection”, ‘‘Landsat”, and ‘‘time series” in
article tile, abstract, and keywords. A total of 254 articles were found (document type only included research article).
372 Z. Zhu / ISPRS Journal of Photogrammetry and Remote Sensing 130 (2017) 370–384
Table 1
Top 15 most relevant journals, most prolific authors, and major research institutions from 1994 to 2017 indexed by Scopus. The numbers in the parentheses represent the total
number of publications found in each category. The Search was conduction on June 27, 2017 by searching ‘‘change detection”, ‘‘Landsat”, and ‘‘time series” in article tile, abstract,
and keywords. A total of 254 articles were found (document type only included research article).
applications for a total of 102 articles that used Landsat time series frequencies of Landsat time series, we used the highest frequency
for change detection published between 2000 and 2016. to represent the studies. If the study only provided a total number
of Landsat images for several path/rows, we used the average fre-
quency (across several path/rows) to represent the frequency of
2. Frequencies the study. Based on the statistics in Fig. 2, it is rare to see articles
that used more than one Landsat image per year before 2008; in
The Frequencies of Landsat time series used for change detec- fact, many of them were using one Landsat images every two
tion have increased substantially in the recent years. A single Land- (0.5 images per year) or four years (0.25 images per year). After
sat satellite visits the same location in every 16 days, which means 2008, the story changed completely. We not only observed more
it can collect 22–23 images per year for a given location (without articles (Fig. 1), but also observed a substantial increase in the
considering overlap areas). Two Landsat satellites can provide a number of Landsat images used per year in each article (Fig. 2).
maximum of 45–46 images per year for the same location. Among Studies that used one Landsat image per year or even 20–30
the 102 articles that use Landsat time series for change detection, images per year appeared (Fig. 2A). As most of the time series stud-
89 of them contain the information to calculate the number of ies were only interested in producing annual or biannual change
images used per year (Fig. 2). For those studies with multiple maps, they tended to select multiple images (partly cloudy images)
30
Number of images per year
Original
25 Composite
Metric
20 Fusion
Average # of images/year
A 15
10
0
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016
Year of publication
Number of images per year
1.5
B 1
0.5
0
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016
Year of publication
Fig. 2. Frequencies of Landsat time series used for change detection. The upper image (Figure A) shows the number of Landsat images used per year for all 89 articles. The
lower image (Figure B) shows the number of Landsat images used per year for all 89 articles with the number of Landsat images used per year less or equal to 2. The ‘‘crosses”
are the articles that used the original Landsat images as their inputs. The ‘‘circles” are the articles that used composite Landsat images as their inputs. The ‘‘diamonds” are the
articles that used statistical metrics as their inputs. The ‘‘triangles” are the articles that used image fusion of Landsat and MODIS images as their inputs. The ‘‘dashed line” is
the average values of number of Landsat images used per year from all the articles published in the same year.
Z. Zhu / ISPRS Journal of Photogrammetry and Remote Sensing 130 (2017) 370–384 373
acquired in the same season and the same year to produce cloud- 3.1. Atmospheric correction
free composite images as their inputs (‘‘circles” in Fig. 2). In this
way, seasonal differences caused by solar angle differences and To reduce influences from atmosphere, atmospheric correction
vegetation phenological changes were minimized, and the data is a common preprocessing step before detecting change. Gener-
volume was reduced substantially. There are also studies that ally, there are two categories of atmospheric correction
fused Landsat data with MODIS data and were capable of predict- approaches: (1) relative normalization (Schroeder et al., 2006)
ing Landsat image in every 16 days (22 images per year) for detect- and absolute correction (Chávez, 1996; Song and Woodcock,
ing forest change (Hilker et al., 2009) (‘‘triangles” in Fig. 2A). After 2003). Relative normalization involves adjusting the radiometric
2012, the average number of Landsat images used per year is of Landsat time series to a reference image based on the relation-
approximately 10 (‘‘dashed line” in Fig. 2) – another big increase ship between pseudo-invariant features from multi-date images
compared to 2008. In 2012, new approaches that used all available (Song et al., 2001). Absolute correction can be further divided into
Landsat images appeared in the literature (‘‘crosses” in Fig. 2). two categories: empirical and physical-based approaches. The
These approaches can model the intra-annual seasonal changes Dark-Object Subtraction (DOS) method is a widely used empirical
in the time series and detect change as quick as every 8 days method for estimating the path radiance based on the darkest
(Zhu et al., 2012). Note that there are also approaches that used value in the image (Song and Woodcock, 2003). DOS is relatively
statistical metrics, such as percentiles and slopes, as inputs for simple, but it does not consider the pixel-to-pixel variation in
change detection (‘‘diamonds” in Fig. 2). It is obvious that the more atmospheric effects. The physical-based approaches, such as Atmo-
the recent the study, the more frequent the Landsat time series spheric/Topographic CORrection (ATCOR; Richter, 1997), MODer-
used. The use of more frequent Landsat time series provides the ate resolution atmospheric TRANsmission (MODTRAN; Berk et al.,
possibility of detecting change at a much faster pace. The dense 1998), and the Satellite Signal in the Solar Spectrum (6S) code
Landsat time series also provides the possibility of capturing the (Vermote et al., 1997), are able to consider the heterogeneity of
intra-annual seasonal changes, and this information can be very the atmosphere, but need many complicated steps and manual
helpful for improving change detection accuracy. operations, which make them difficult to process large amount of
Landsat time series. However, the Landsat Ecosystem Disturbance
3. Preprocessing Adaptive Processing System (LEDAPS) software (Masek et al.,
2006), which has adopted the 6S code, has made atmospheric cor-
Before applying different change detection algorithms for Land- rection for Landsats 4–7 fully automated. Recently, Vermote et al.
sat time series, we need to perform a series of image preprocessing (2016) developed an improved atmospheric correction algorithm
steps, including atmospheric correction, cloud and cloud shadow for Landsat 8 (L8SR), which has shown an improvement over the
detection, and composite/fusion/metrics. To make sure the time ad-hoc Landsats 5–7 LEDAPS product. In 2012, Landsats 4–7 sur-
series are well aligned, most of the change detection algorithms face reflectance product generated from LEDAPS were provided
only select Level 1 terrain-corrected (L1T) Landsat images as their by the USGS EROS Center. Later in 2014, provisional Landsat 8 sur-
inputs. The L1T Landsat images are reported to have high geomet- face reflectance products generated from L8SR were also provided
ric accuracies (RMSE less than 30 meters in more than 99 percent by the USGS EROS Center. The free distribution of Landsat surface
of the data; http://landsat.usgs.gov/geometry.php), and geometric reflectance has greatly increased the use of LEDAPS- and L8SR-
correction is generally considered unnecessary if only L1T Landsat based surface reflectance products in change detection with Land-
images are included in change detection (Zhu and Woodcock, sat time series (‘‘dashed line” in Fig. 3).
2014a). Moreover, in 2016, the USGS EROS Center started reorga-
nizing the Landsat archive into a formal tiered data Collection 3.2. Cloud and cloud shadow detection
structure, which ensures that Landsat Level-1 products provides
a consistent archive of known data quality (Tier 1, Tier 2, and The presence of clouds and their shadows complicate the use of
Real-Time) to support time series analyses and data ‘‘stacking”. Landsat data, making the detection of both items an evitable step
Highest available data quality Landsat images (image-to-image tol- prior to change detection. For a long time, cloud and cloud shadow
erances of <12 m RMSE; https://landsat.usgs.gov/landsat-collec- detection at pixel level only existed for coarse spatial resolution
tions) are placed into Tier 1 and are considered most suitable for sensors, such as Advanced Very High Resolution Radiometer
time series analysis. (AVHRR) and Moderate Resolution Imaging Spectroradiometer
15
Number of articles published per year
Fmask/CFmask
LEDAPS/L8SR
10
0
2009 2010 2011 2012 2013 2014 2015 2016
Year of publication
Fig. 3. Number of articles that used Fmask/CFmask (‘‘dashed line”) or LEDAPS/L8SR (‘‘solid line”) per year between 2009 and 2016. The statistics is derived from the 102
articles that used Landsat time series to detect change.
374 Z. Zhu / ISPRS Journal of Photogrammetry and Remote Sensing 130 (2017) 370–384
(MODIS) (Ackerman et al., 1998; Derrien et al., 1993; Saunders and images is quite challenging, as MSS does not have thermal and
Kriebel, 1988). Though Landsat has been collecting data for dec- SWIR bands, both of which are critical in cloud detection. However,
ades, there is hardly any operational algorithm that is capable of Braaten et al. (2015) used a rule-based approach to detect both
providing cloud and cloud shadow masks at pixel level. The Auto- clouds and cloud shadows in Landsat MSS images, and achieved
mated Cloud Cover Assessment (ACCA) system (Irish, 2000; Irish comparable accuracies to the Fmask algorithm.
et al., 2006) works well for estimating the percentage of cloud The machine learning based algorithms identify cloud and
cover for each individual Landsat image, but it does not provide cloud shadow using a supervised classifier trained by previously
clouds and cloud shadows masks for every Landsat pixel. collected training data (Hughes and Hayes, 2014; Potapov et al.,
Since the free access to the entire Landsat archive in 2008, many 2011; Roy et al., 2010; Scaramuzza et al., 2012). Roy et al. (2010)
cloud and cloud shadow detection algorithms have been developed and Potapov et al. (2011) used decision tree classifier to classify
based on a single-date Landsat image (Braaten et al., 2015; Huang clouds and the classifiers were trained based on many manually
et al., 2010b; Oreopoulos et al., 2011; Potapov et al., 2011; Roy interpreted cloud masks. Scaramuzza et al. (2012) developed two
et al., 2010; Scaramuzza et al., 2012; Qiu et al., 2017; Vermote algorithms for detecting clouds based on Landsat 8 OLI (without
et al., 2016; Vermote and Saleous, 2007; Zhu and Woodcock, using TIRS bands), with one from an off-the-shelf machine learning
2014b, 2012). Most of the algorithms are designed for Thematic package (See5) and the other based on ACCA but enhanced by a
Mapper (TM) and Enhanced Thematic Mapper Plus (ETM+), and neural network (AT-ACCA). Both algorithms have shown good
only a few algorithms are designed for Operational Land Imager accuracies and neither of them needs the thermal band as input.
(OLI)/Thermal Infrared Sensor (TIRS) (Scaramuzza et al., 2012; Hughes and Hayes (2014) also explored the use of a neural net-
Vermote et al., 2016; Zhu et al., 2015a), and Multispectral Scanner work classifier and spatial post-processing to identify clouds and
(MSS) (Braaten et al., 2015). Though all the algorithms are devel- cloud shadows in Landsat images. Compared to the Fmask algo-
oped for the same purpose, the mechanism for identifying clouds rithm, this approach achieved lower omission errors in cloud sha-
and their shadows are quite different. Generally, we can divide dow detection and slightly higher omission errors in cloud
the algorithms into two categories: physical rules based and detection. Though all these studies pointed out the usefulness of
machine learning based algorithms. machine learning based algorithms for cloud and cloud shadow
The physical rule based algorithms detect clouds and their detection, these algorithms require a certain level of knowledge
shadows based on their physical characteristics (Braaten et al., of clouds, cloud shadows, and land surface condition within the
2015; Huang et al., 2010b; Oreopoulos et al., 2011; Qiu et al., image (as the training data) and these algorithms can easily fail
2017; Vermote et al., 2016; Vermote and Saleous, 2007; Zhu and to detect cloud and cloud shadow for certain unique conditions
Woodcock, 2014b, 2012). Vermote and Saleous (2007) proposed (Huang et al., 2010b).
a cloud and cloud shadow detection algorithm for Landsat data The recent progress of the fully automated cloud and cloud sha-
as one of the internal products in the LEDAPS software. This algo- dow detection algorithms based on a single-date Landsat image
rithm needs other auxiliary data like surface temperature from the has facilitated many remote sensing applications that need to pro-
National Centers for Environmental Prediction (NCEP) to generate cess large number of Landsat images. However, for change detec-
a surface temperature reference layer for detecting clouds. Later, tion using Landsat time series, the accuracy of single-date
Vermote et al. (2016) proposed a cloud and cloud shadow detec- algorithms remains insufficient. To further improve cloud and
tion algorithm for Landsat 8 data within the L8SR framework, cloud shadow detection accuracies, algorithms based on multitem-
which takes the advantages of some of the new spectral bands poral Landsat images have been developed. Jin et al. (2013) and
(e.g., the cirrus band and the new blue band). Huang et al. Wang et al. (1999) proposed to detect clouds and their shadows
(2010b) proposed an algorithm that detects clouds based on tem- by comparing a cloud-free reference image to the observed images.
perature that are normalized by DEM and pre-classified confident This approach is relatively simple and can provide accurate results,
forest pixels, and the cloud shadows are detected based on the but it is highly dependent on the availability and quality of the ref-
solar-sensor geometry. Oreopoulos et al. (2011) modified a cloud erence image, which makes it difficult to work operationally.
detection algorithm that was originally developed for MODIS Recently, cloud and cloud shadow detection algorithms based on
(Luo et al., 2008), and applied the algorithm to Landsat data. This Landsat time series have been developed, which are able to provide
algorithm performs on par with the ACCA algorithm, without using accurate detection results, and at the same time fully automated
the thermal band. Zhu and Woodcock (2012) and Zhu et al. (2015b) (Goodwin and Collett, 2014; Hagolle et al., 2010; Zhu and
proposed a method called Fmask (Function of mask) to detect Woodcock, 2014b). One disadvantage of these time series algo-
cloud, cloud shadow, and snow in Landsats 4–8 images based on rithms is that they may also identify some ephemeral changes
the spectral and spatial information in the Landsat data. Clouds (e.g., soil wetness change) as clouds or cloud shadows (Zhu and
are identified based on a cloud probability layer and a scene- Woodcock, 2014b). Considering the importance of accurate cloud
based dynamic threshold, and cloud shadows are matched with and cloud shadow detection in analyzing Landsat time series, more
clouds based on similarity and the corresponding solar-sensor time series based cloud and cloud shadow detection algorithms are
geometry. Recently, Qiu et al. (2017) integrated DEM with the anticipated in the near future.
Fmask algorithm and developed a MFmask (Mountainous Fmask)
algorithm that showed better cloud and cloud shadow detection 3.3. Composite, fusion, and metrics
results in mountainous regions. In 2013, the Fmask algorithm has
been reprogramed in C (CFMask) by the USGS EROS Center, and Before detecting change, some algorithms need to create cloud-
the CFMask results are distributed by the USGS EROS Center along free composite Landsat images, fuse Landsat images with other low
with the surface reflectance product. This free distribution of spatial resolution images, or calculate statistical metrics from the
CFMask results has greatly increased the use of the Fmask algo- Landsat time series.
rithm, and we can see a dramatic increase in using Fmask/CFMask Image compositing is a good tool for reducing data volume and
in change detection using Landsat time series since 2013 (‘‘solid minimizing atmosphere influences, but most of the image com-
line” in Fig. 3). Moreover, within the formal tiered data Collection positing algorithms are only designed for coarse spatial resolution
structure provided by USGS in 2016, the CFmask algorithm is used images, such as AVHRR (Holben, 1986) and MODIS (Luo et al.,
to provide the Quality Assessment (QA) band for all Collection 1 2008), with only a few studies that are designed for Landsat
products. Cloud and cloud shadow detection in Landsat MSS (Griffiths et al., 2013; Potapov et al., 2011; Roy et al., 2010;
Z. Zhu / ISPRS Journal of Photogrammetry and Remote Sensing 130 (2017) 370–384 375
White et al., 2014; Zhu et al., 2015b). Roy et al. (2010) first pro- cation change detection method (Hansen et al., 2014, 2013;
posed an image composite method for Landsat ETM+ images Potapov et al., 2011, 2015, 2012). There are many kinds of metrics
mainly based on a combination of maximum Normalized Differ- that can be derived from Landsat time series, such as individual
ence Vegetation Index (NDVI) and highest brightness temperature ranks, means, and regression slopes of spectral bands and vegeta-
criteria. This method has been applied to the entire conterminous tion indices, and they are generally calculated either based on
U.S. and the final composites are provided for free download. time-sequential reflectance or reflectance ranking. One advantage
Potapov et al. (2011) used the median values of the Near Infrared of this method is that the statistics derived from Landsat time ser-
(NIR) band as the criteria for selecting the ‘‘best” observations ies are relatively robust to noise and contains temporal informa-
and this approach performed better than the conventional maxi- tion that are important inputs for multi-date classification
mum NDVI compositing method. Griffiths et al. (2013) developed change detection method.
a method that calculates scores for every Landsat observation, with
image compositing rules determined by the weighted scores that
are calculated based on acquisition year, acquisition day of year, 4. Algorithms
distance of a given pixel to cloud (from Fmask). Similarly, White
et al. (2014) proposed a pixel-based image compositing method Based on the mathematical approach used for detecting change,
that calculates pixel scores based on sensor type, day of year, dis- the Landsat time series change detection algorithms can be divided
tance to cloud or cloud shadow (from Fmask), and opacity (from into six major categories, including thresholding, differencing, seg-
LEDAPS). Zhu et al. (2015b) proposed to use all available clear mentation, trajectory classification, statistical boundary, and
Landsat data to estimate time series models for each pixel and each regression (Fig. 4 and Table 2). Within each category, six character-
spectral band, and used the estimated time series models to pre- istics, including frequency, change index, univariate/multivariate,
dict daily clear-sky synthetic Landsat data. Among all these meth- online/offline, abrupt/gradual change, and sub-pixel/pixel/spatial
ods, it is hard to quantify which method works the best, as there is were discussed. The detailed definitions of these characteristics
not a standard set of reference images to compare with. Most of the are as follows.
evaluations are made by visual check of the results based on natu- Frequency: Number of Landsat images used for the same loca-
ral color composite. tion. We used three levels to indicate the frequency of Landsat data
As the frequency of Landsat time series is relatively low, it may used: low (a few years per image), medium (approximately 1
take a few weeks or months to generate cloud-free composite images per year), and high (more than 1 images per year). Algo-
Landsat images, and the ‘‘best” observations selected from image rithms that use multiple cloudy Landsat images per year to derive
compositing may show large seasonal differences. To overcome annual cloud-free composite images are categorized into medium
this limitation, fusing Landsat data with coarse resolution images, frequency.
such as MODIS data provides a solution (Gao et al., 2006; Hilker Change index: The index used for detecting change, such as spec-
et al., 2009; Roy et al., 2008; Zhu et al., 2010). Gao et al. (2006) pro- tral bands, vegetation indices, tasseled cap transformations, land
posed the Spatial and Temporal Adaptive Reflectance model cover class, or land cover fractions.
(STARFM), which is capable of predicting Landsat-scale observa- Univariate/multivariate: Most of the change detection algo-
tions on MODIS observation dates. STARFM has been later modified rithms use a single variable at each time point (univariate), but
to form the Spatial Temporal Adaptive Algorithm for mapping there are some algorithms use multivariate vector at each time
Reflectance Change (STAARCH) for detecting forest disturbance point (multivariate).
(Hilker et al., 2009), and the Enhanced STARFM (ESTARFM) for bet- Online/offline: Online change detection assumes the time series
ter handling of heterogeneous areas (Zhu et al., 2010). Gao et al. data are coming in at a certain rate, and the main goal is to detect
(2015) reviewed STARFM, STAARCH, and ESTARFM, and suggested changes in near real-time (or continuously), with minimum delay.
that though these approaches can produce images with high spa- Offline change detection assumes the time series data already exist
tial temporal resolution, they are still dependent on the availability and the main goal is to detect when the characteristics of the time
of actual satellite images and the quality of the remote sensing series changed.
products, therefore, cannot replace the actual satellite missions. Abrupt/gradual change: Abrupt change refers to large magnitude
Roy et al. (2008) proposed a semi-physical fusion approach that changes which are usually occurred in a short time. Abrupt change
uses MODIS Bidirectional Reflectance Distribution Function can be caused by disturbances such as deforestation, floods, fires,
(BRDF)/Albedo land surface characterization products to predict or urbanization. Gradual change refers to small magnitude changes
Landsat images, and achieved good results. The synthetic Landsat which are usually occurred in a long time. Gradual change can be
images generated by fusing Landsat and MODIS can substantially triggered by factors, such as interannual climate variability or
densify the Landsat time series, which can be critical for time series gradual change in land management or land degradation.
analysis in places that are frequently covered by clouds. Subpixel/pixel/spatial: Most of algorithms use pixel as their
Metrics from Landsat time series can provide extra generic fea- smallest unit (pixel), but there are some algorithms use sub-pixel
ture space, which is particularly useful for the multi-date classifi- information to detect change (subpixel). Most of the algorithms
Fig. 4. Categories and subcategories of Landsat time series change detection algorithms.
Table 2
376
Characteristics of different Landsat time series change detection algorithms.
Category Subcategory Temporal Change Index Univariate Online Abrupt Subpixel References
Frequency Multivariate Offline Gradual Pixel
Both Spatial
Thresholding Low Principal Components (PCs) Multivariate Offline Abrupt Pixel Hayes and Sader (2001)
Low NDVI Univariate Offline Abrupt Pixel Lee (2008)
Medium NDVI Univariate Offline Abrupt Pixel J. Li et al. (2015a)
Medium Disturbance index Univariate Offline Abrupt Pixel Pickell et al. (2014) and Sieber et al. (2013)
Medium Z-scores based on Tasseled Cap Transformation (TCT) Univariate Offline Abrupt Pixel Kayastha et al. (2012)
Medium Integrated Forest Z-score (IFZ) Univariate Offline Abrupt Pixel Chen et al. (2013) and Huang et al. (2010a, 2009)
High Multivariate Alteration Detection (MAD) variates Multivariate Offline Abrupt Pixel Alaibakhsh et al. (2015)
High Disturbance Index (DI), Tasseled Cap Brightness Multivariate Offline Abrupt Pixel Hilker et al. (2009)
(TCB), Tasseled Cap Wetness (TCW), & normalized
NDVI
Z. Zhu / ISPRS Journal of Photogrammetry and Remote Sensing 130 (2017) 370–384
Differencing Classification Low Land cover Univariate Offline Abrupt Pixel Dekker et al. (2005), Giri et al. (2008), Giri and Muhlhausen
(2008), Mihai et al. (2015), Muttitanon and Tripathi (2005),
Palandro et al. (2003), Pillay et al. (2014), Ramadan et al.
(2004), Su (2000), Unger et al. (2015), Yang et al. (2015),
Zhang et al. (2013) and Zhao et al. (2014)
Low Land cover Univariate Offline Abrupt Spatial Liu and Cai (2012)
Medium Land cover Univariate Offline Abrupt Pixel Palandro et al. (2008)
High Land cover Univariate Offline Abrupt Pixel Knudby et al. (2010), Kontgis et al. (2015), Li and Narayanan
(2003), X. Li et al. (2015), Nutini et al. (2013), Pardo-Pascual
et al. (2014), Potapov et al. (2015) and Tulbure and Broich
(2013)
Spectral Low Impervious fraction Univariate Offline Abrupt Subpixel Powell et al. (2008)
Mixture Low Vegetation and non-vegetation fractions Multivariate Offline Abrupt Subpixel Cunningham et al. (2015)
Analysis Medium Wetland fraction Univariate Offline Abrupt Subpixel Dingle Robertson et al. (2015)
Spectral/ Low NDVI Univariate Offline Abrupt Pixel Hayes and Sader (2001) and Marzen et al. (2011)
Index Low Normalized Difference Wetness Index (NDWI) Univariate Offline Abrupt Pixel Jin and Sader (2005)
Low Change vector Multivariate Offline Abrupt Pixel Fraser et al. (2009) and Vorovencii (2014)
Medium Normalized Burn Ratio (NBR) Univariate Offline Abrupt Pixel Bolton et al. (2015) and Parker et al. (2015)
Medium Chang vector Multivariate Offline Abrupt Pixel Zanotta et al. (2015)
Medium DI and NBR Multivariate Offline Abrupt Pixel Neigh et al. (2014)
Medium Enhanced wetness Univariate Online Abrupt Pixel Linke et al. (2009)
High Red and blue bands Multivariate Offline Abrupt Pixel Hagolle et al. (2010)
Segmentation Medium NBR Univariate Offline Both Pixel Chance et al. (2016), Franklin et al. (2015), Kennedy et al.
(2010), Liang et al. (2014), Meigs et al. (2015, 2011) and Senf
et al. (2015)
Medium NBR Univariate Offline Both Spatial Hermosilla et al. (2015a,b) and Kennedy et al. (2015)
Medium TCW Univariate Offline Both Pixel Frazier et al. (2015), Griffiths et al. (2012) and Grogan et al.
(2015)
Trajectory Hypothesized Low NDVI & Normalized Difference Water Index Multivariate Offline Abrupt Pixel Xue et al. (2014)
Classification Trajectory (MNDWI)
Low Tasseled Cap Angle (TCA) Univariate Offline Abrupt Pixel Ahmed et al. (2014)
Low Short-wave Infrared (SWIR) band Univariate Offline Abrupt Pixel Gillanders et al. (2008)
Medium SWIR band Univariate Offline Both Pixel Kennedy et al. (2007)
Multi-date Low NDVI Univariate Offline Abrupt Pixel Hayes and Sader (2001) and Sader et al. (2003)
Classification Low NDVI or NDWI Univariate Offline Abrupt Pixel Wilson and Sader (2002)
Low Spectral bands Multivariate Offline Abrupt Pixel Coops et al. (2010) and Gavier-Pizarro et al. (2012)
Low Spectral bands Multivariate Offline Abrupt Spatial Boucher et al. (2006)
Low Normalized Difference Moisture Index (NDMI) Univariate Offline Abrupt Pixel Sader and Legaard (2008)
Medium Spectral bands Multivariate Offline Abrupt Pixel Kaufmann and Seto (2001), Margono et al. (2012) and Sieber
et al. (2013)
Medium NDVI Univariate Offline Abrupt Pixel Maxwell and Sylvester (2012)
High Spectral bands metrics Multivariate Offline Abrupt Pixel Hansen et al. (2014, 2013) and Potapov et al. (2011, 2012)
Z. Zhu / ISPRS Journal of Photogrammetry and Remote Sensing 130 (2017) 370–384 377
4.1. Thresholding
Collett (2014) and Zhu and Woodcock (2014a,b)
(Hilker et al., 2009; Pickell et al., 2014; Sieber et al., 2013), or Inte-
grated Forest Z-score (IFZ) (Chen et al., 2013; Huang et al., 2010a,
References
Subpixel
(Alaibakhsh et al., 2015) have been used to extract the data dimen-
Pixel, &
Spatial
Spatial
Pixel
Pixel
Pixel
Pixel
Pixel
Pixel
Pixel
Pixel
Pixel
Pixel
Pixel
Pixel
Pixel
Pixel
Gradual
Gradual
Gradual
Gradual
Gradual
Abrupt
Abrupt
Abrupt
Abrupt
Abrupt
Abrupt
Abrupt
Abrupt
Both
Both
Both
Both
Offline
Offline
Offline
Offline
Offline
Offline
Online
Online
Online
Online
Online
Online
Online
Online
Online
Online
Online
4.2. Differencing
Multivariate
Multivariate
Multivariate
Multivariate
Univariate
Univariate
Univariate
Univariate
Univariate
Univariate
Univariate
Univariate
Univariate
Univariate
Univariate
Univariate
Univariate
Moreover, this method can also provide ‘‘from-to” land cover infor-
mation. However, the accuracy of this method is highly dependent
Spatially normalized NDVI
NDVI or NDWI
Spectral bands
Spectral bands
Change Index
NDVI
TCA
derived from SMA to detect change for different land cover types
Frequency
Temporal
et al., 2008). One benefit of this method is that we can get sub-
High
High
High
High
High
High
High
High
High
High
High
High
Low
Low
Regression
Statistical
(Bolton et al., 2015; Neigh et al., 2014; Parker et al., 2015), change
Category
the studies used a single index (Bolton et al., 2015; Hayes and line, except Hansen et al. (2016) developed an alerting system that
Sader, 2001; Jin and Sader, 2005; Linke et al., 2009; Marzen is able to monitor humid tropical forest in near-real time based on
et al., 2011; Parker et al., 2015), while others used multiple indices newly collected Landsat images. The trajectory classification
to detect change (Fraser et al., 2009; Hagolle et al., 2010; Neigh method is mainly designed to detect abrupt changes, but
et al., 2014; Vorovencii, 2014; Zanotta et al., 2015). The spectral/ Kennedy et al. (2007) detected both gradual and abrupt changes
index method requires accurate radiometric calibration to make based on the hypothesized trajectory method. Moreover, the
the spectral/index comparable between the dates, but as long as hypothesized trajectory method favors univariate time series
the data are well calibrated, the accuracy from this subcategory (Ahmed et al., 2014; Gillanders et al., 2008; Kennedy et al.,
will be a lot higher than the other two subcategories. All the differ- 2007), while the multi-date classification method favors multivari-
encing methods discussed here are offline and only detect abrupt ate time series (Boucher et al., 2006; Coops et al., 2010; Gavier-
changes. The differencing method is relatively simple to use, but Pizarro et al., 2012; Hansen et al., 2016, 2014, 2013; Kaufmann
is highly dependent on the consistency of the images to be com- and Seto, 2001; Margono et al., 2012; Potapov et al., 2012, 2011,
pared with. Methods of classification and SMA differencing are 2015; Sieber et al., 2013). The trajectory classification method is
generally not recommended for change detection due to the likeli- computational expensive, and this method requires fully under-
hood of including compounded errors from classification or SMA. standing of the characteristics of different kinds of change.
The segmentation method is an offline approach that needs to The statistical boundary method expects the time series to fol-
have all the historical time series data available at the same time. low a statistical boundary and any significant departure from the
By segmenting the time series into a series of straight line seg- boundary is detected as a change. All the studies in this category
ments based on the residual-error and angle criterions, both abrupt used a high frequency of Landsat time series. Most of the time
and gradual changes can be detected based on the derived straight- Landsat time series are decomposed into trend, seasonal change,
line segments at pixel level. Most of the studies in this category are and noise components before estimating the statistical boundary.
focused on forest change and only use a medium frequency of Algorithms based on statistical quality control charts (Brooks
Landsat time series. NBR (Chance et al., 2016; Franklin et al., et al., 2014), econometrics structural change monitoring (DeVries
2015; Hermosilla et al., 2015a, 2015b; Kennedy et al., 2015, et al., 2015; Dutrieux et al., 2015; Hamunyela et al., 2016; Reiche
2010; Liang et al., 2014; Meigs et al., 2015, 2011; Senf et al., et al., 2015), and model prediction (Fu and Weng, 2016; Goodwin
2015) and Tasseled Cap Wetness (TCW) (Frazier et al., 2015; et al., 2013; Goodwin and Collett, 2014; Vogelmann et al., 2015;
Griffiths et al., 2012; Grogan et al., 2015) are the only change Zhu et al., 2016a, 2012; Zhu and Woodcock, 2014a, 2014b) are used
indices used in this category. Two studies incorporated spatial for estimating the statistical boundaries. Most of the studies are
information to help detect change, in which Kennedy et al. able to detect change continuously (online), but some of them need
(2015) used spatial adjacency and temporal coherence to improve to have all the time series available before detecting change (off-
the final maps, and Hermosilla et al. (2015a,b) relabeled low- line) (Goodwin et al., 2013; Goodwin and Collett, 2014; Zhu and
reliability changes based on spatially-adjacent high reliability Woodcock, 2014b). Univariate time series are widely used, and
change. The segmentation method is computational expensive, only a few of them use multiple spectral bands of Landsat time ser-
but has shown great potentials in detecting forest change at large ies (Fu and Weng, 2016; Goodwin et al., 2013; Goodwin and
scale. Collett, 2014; Zhu et al., 2016a; Zhu and Woodcock, 2014a,
2014b). Most of the studies did not use spatial information to help
4.4. Trajectory classification with change detection, except Hamunyela et al. (2016), which used
NDVI values that are spatially normalized (sNDVI) based on differ-
The trajectory classification method first extracts information ent window sizes to reduce phenological differences. They sug-
from Landsat time series for places that have undergone certain gested that by integrating this spatial information, change can be
kinds of change (for training purpose), and later uses this informa- identified with less delay. Most of the algorithms in this category
tion to further classify every Landsat time series in the image. have the potential of detecting both gradual and abrupt changes,
Based on how the trajectory of Landsat time series is classified, but only Brooks et al. (2014) and Zhu et al. (2016a) demonstrated
we can further divide this category into two subcategories: this capability. The statistical boundary method is very computa-
hypothesized trajectory and multi-date classification. For the tional expensive and requires large storage. However, this method
hypothesized trajectory method, the rules for classifying different can detect change much faster and is usually less influenced by the
kinds of changes are based on the hypothesized trajectories repre- seasonal differences.
senting signatures specific to different kinds of changes. For the
multi-date classification method, there is no hypothesis on the 4.6. Regression
time series trajectories, but rather supervised classifiers are used
to classify different change types. In the hypothesized trajectory The regression method assumes there is a linear relationship
approach, both low (Ahmed et al., 2014; Gillanders et al., 2008; between predictor and response variables and uses regression to
Xue et al., 2014) and medium (Kennedy et al., 2007) frequencies get the answer. The predictor variables are usually the time of
of Landsat time series have been used. In the multi-date classifica- the observations, and the response variables are usually the
tion method, most of the studies are based on low (Boucher et al., observed values (spectral bands or indices). The main purpose of
2006; Coops et al., 2010; Gavier-Pizarro et al., 2012; Hayes and the regression method is to estimate the long-term movements
Sader, 2001; Sader et al., 2003; Sader and Legaard, 2008; Wilson or trends in time series, which are mostly gradual changes. All
and Sader, 2002) and high (Hansen et al., 2016, 2014, 2013; regression studies are univariate and changes are detected in an
Hansen and Loveland, 2012; Potapov et al., 2011, 2015, 2012) fre- offline manner. Variables such as NDVI, NDWI, Soil-adjusted Vege-
quencies of Landsat time series, while only a few studies are based tation Index (SAVI), TCT components, green vegetation fraction,
on a medium frequency of Landsat time series (Kaufmann and Seto, six-band discriminant index, Short-wave Infrared (SWIR)/NIR ratio,
2001; Margono et al., 2012; Maxwell and Sylvester, 2012; Sieber are used as change indices. Most of the studies use pixel as the
et al., 2013). Most of the trajectory classification methods are off- smallest unit, except Sonnenschein et al. (2011) used green vegeta-
Z. Zhu / ISPRS Journal of Photogrammetry and Remote Sensing 130 (2017) 370–384 379
tion fraction as the predictor variable (sub-pixel scale). The regres- algorithms use a medium frequency of Landsat time series, and
sion method is heavily reliant on the accurate calibration of differ- are capable of providing change information on an annual time-
ent Landsat sensors. Landsat sensors including MSS, TM, and ETM+ scale. The Breaks for Additive Season and Trend (BFAST) algorithm
have been placed onto a consistent radiometric scale (Markham belongs to the statistical boundary method, and it was rooted in the
and Helder, 2012). However, the OLI data are reported to have econometrics discipline for monitoring structural change (Chu
some differences when compared with previous Landsat sensors et al., 1995; Leisch et al., 2000; Zeileis et al., 2005). The BFAST algo-
(Holden and Woodcock, 2016; Roy et al., 2016; Zhu et al., rithm decomposes time series into trend, season, and noise compo-
2016a). Moreover, Zhang and Roy (2016) found there are some nents, which makes it possible to detect abrupt change as well as
inconsistencies in the Landsat 5 NDVI time series caused by satel- gradual change. This algorithm was originally developed for detect-
lite orbit change, while Sulla-Menashe et al. (2016) suggested that ing vegetation change in 16-day MODIS composite time series in an
the red reflectance from Landsat 7 is lower than that from Landsat offline mode (Verbesselt et al., 2010), and was later modified and
5, which introduces artificial long-term trends in NDVI. All fre- renamed (BFAST Monitor) to detect drought-related vegetation dis-
quencies (low, medium, and high) have been used in the regression turbance in near real-time using MODIS time series (online)
method. For low and medium frequencies, seasonal changes are (Verbesselt et al., 2012). Recently, the BFAST Monitor algorithm
the main sources of noise in trend analysis, such that images from has been applied to Landsat time series for detecting forest change
near-anniversary date and within growing season are preferred. (DeVries et al., 2015; Dutrieux et al., 2015; Hamunyela et al., 2016;
Normally, the regression method assumes there is no abrupt Reiche et al., 2015). The BFAST Monitor algorithm uses a high fre-
change in the time series, except Vogelmann et al. (2016) consid- quency of Landsat time series, and is a univariate (e.g., NDVI most
ered the possibility of both gradual and abrupt changes in Landsat of the time) approach. The Continuous Change Detection and Clas-
time series. The regression method is relatively simple to use, but it sification (CCDC; Zhu and Woodcock, 2014a) algorithm is a statisti-
is mostly designed to quantify long-term trends in the time series, cal boundary method that uses all available Landsat data (high
and is less ideal for detecting changes that are abrupt in the Land- frequency). It is evolved from a Continuous Monitoring of Forest
sat time series. Moreover, as the regression method is highly Disturbance Algorithm (CMFDA), which is designed to detect forest
dependent on the consistency of different sensors, it is important disturbance (abrupt changes) based on all available Landsat data
to calibrate all sensors to make sure the significant trends are sta- (Zhu et al., 2012). The CCDC algorithm has expanded the change tar-
tistically meaningful. get from forest to many land cover types and added a slope compo-
nent to detect gradual changes. To detect many kinds of surface
change, all spectral bands (multivariate) were used to define a
4.7. Some of the widely-used algorithms change; the ‘‘continuous” character makes CCDC capable of detect-
ing change as soon as new Landsat images are collected (online).
Table 3 lists some of the widely-used change detection algo- The CCDC algorithm has been used to study abrupt changes such
rithms that use Landsat time series as their inputs. The Landsat- as land cover change (Fu and Weng, 2016; Zhu and Woodcock,
based detection of Trends in Disturbance and Recovery (Land- 2014a), as well as gradual vegetation changes (Vogelmann et al.,
Trendr; Kennedy et al., 2010) algorithm uses the segmentation 2015; Zhu et al., 2016a). Recently, Zhu et al. (2015b) made several
method to detect abrupt change (forest disturbance) and, between major updates to improve its change detection results, and Zhu
the abrupt changes, a slope is fitted for each segment to capture the et al. (2016b) optimized the strategy for selecting training and aux-
gradual changes (forest recovery in this study). The LandTrendr iliary data for improving classification results. Currently, the CCDC
algorithm is an offline and univariate approach that uses NBR as algorithm has been chosen as the change detection algorithm for
its main change index. This algorithm has been used for many the USGS Land Change Monitoring, Assessment, and Projection
applications, such as attribution of disturbance change agents (LCMAP) program for generating land change products for the Uni-
(Kennedy et al., 2015), insect infestation detection (Liang et al., ted States (Pengra et al., 2016; Zhu et al., 2016b). Hansen et al.
2014; Meigs et al., 2015, 2011; Senf et al., 2015), forest change (2013) produced the first global map of annual forest change based
detection (Griffiths et al., 2012; Grogan et al., 2015; Kennedy on a high frequency of Landsat time series using the multi-date
et al., 2012, 2010), and land cover change detection (Franklin classification method. This approach uses multiple Landsat spectral
et al., 2015). The Vegetation Change Tracker (VCT; Huang et al., bands (multivariate) and is an offline algorithm. Similar approaches
2010a) algorithm normalizes each Landsat image into a forest prob- have been applied to other parts of the world, mainly for detecting
ability index called integrated forest z-score and uses the thresh- forest change (Hansen et al., 2014; Margono et al., 2012; Potapov
olding method to detect forest disturbance. The VCT algorithm is et al., 2015, 2012), and recently it has been modified to an online
an offline and univariate method. It is mainly focused on detecting approach for disturbance alerts in humid tropical forest (Hansen
abrupt changes in vegetation, such as forest disturbance (Chen et al., 2016). One common character of these widely-used algo-
et al., 2013; Huang et al., 2010a, 2009; Pickell et al., 2014) and wet- rithms is that all of these algorithms are fully automated and are
land change (Kayastha et al., 2012). Both the LandTrendr and VCT
Table 3
Characteristics of some of the widely-used algorithms.
Algorithm Temporal Change Univariate Online Abrupt Sub-pixel Pixel Applications References
Name Frequency Index Multivariate Offline gradual spatial
LandTrendr Medium NBR Univariate Offline Both Pixel Forest disturbance and Kennedy et al.
recovery (2010)
VCT Medium IFZ Univariate Offline Abrupt Pixel Forest disturbance Huang et al. (2010a)
BFAST High NDVI Univariate Online Both Pixel Drought-related vegetation Verbesselt et al.
monitor disturbance (2012)
CCDC High Spectral Multivariate Online Both Pixel Land cover change Zhu and Woodcock
bands (2014a)
NA High Spectral Multivariate Offline Both Pixel Forest gain and loss Hansen et al. (2013)
bands
380 Z. Zhu / ISPRS Journal of Photogrammetry and Remote Sensing 130 (2017) 370–384
capable of detecting change at large scale (from regional to global) 5.2. Change agent
and within a short time period (from annual to weekly). The other
common character is that many of these algorithms (i.e., Understanding the cause of change – the ‘‘change agent” – is
LandTrendr, BFAST, and CCDC) are publicly available, which further also valuable in change detection. For natural changes, the agent
popularized the use of these algorithms. refers to different kinds of natural events, such as fire, insect infes-
tation, storm, flooding, and drought. For anthropogenic changes,
the agent refers to the human induced change (mostly land use
5. Applications
related), such as urban development, farming, logging, and mining.
Most of the change agent studies have only focused on a single
Change detection based on Landsat time series has many
change agent, such as urban development, farming, insect infesta-
applications. Generally, we can divide these applications into two
tion, fire, logging, mining, storm, and petroleum exploration/pro-
categories: change target and change agent detection.
duction. For urban development, classification differencing is the
main method used to identify urban related changes (X. Li et al.,
5.1. Change target 2015; Mihai et al., 2015; Ramadan et al., 2004). SMA differencing
can also be used to quantify impervious surface change
Knowing what is changing – the ‘‘change target” – is important (Cunningham et al., 2015; Powell et al., 2008). Hypothesized tra-
in change detection. By limiting changes detected within the tar- jectory is another approach for detecting urban growth (Xue
geted land cover and use types, change can be better detected et al., 2014). For farming related changes, Kontgis et al. (2015) used
based on thresholds and change indices that are determined and classification differencing to map rice paddy extent and intensifica-
selected specifically for the change target. tion, and Maxwell and Sylvester (2012) used multi-date classifica-
The change target is usually quite broad, which involves changes tion to map ‘‘ever-cropped” land. For insect infestation related
in a variety of land cover and land use classes. Classification differ- change, segmentation methods, such as LandTrendr have been
encing is the most widely used method (Liu and Cai, 2012; widely used (Liang et al., 2014; Meigs et al., 2015, 2011; Senf
Muttitanon and Tripathi, 2005; Nutini et al., 2013; Pillay et al., et al., 2015). Coops et al. (2010) also tested using multi-date clas-
2014; Zhang et al., 2013; Zhao et al., 2014). There are also methods sification method to detect forest changes following insect infesta-
based on spectral/index differencing (Fraser et al., 2009; Linke et al., tion. For fire related changes, Parker et al. (2015) used spectral/
2009; Vorovencii, 2014), trajectory classification (Boucher et al., index differencing and mapped fire severity based on differences
2006; Gillanders et al., 2008; Kaufmann and Seto, 2001; Sieber in NBR, and Goodwin and Collett (2014) proposed a statistical
et al., 2013), segmentation (Franklin et al., 2015), and statistical boundary method for mapping fire history. For forest logging/har-
boundary (Fu and Weng, 2016; Zhu and Woodcock, 2014a). vest related change, the multi-date classification approach is the
Some change detection studies have focused on only one land most common method used (Schroeder et al., 2012; Wilson and
cover type. The most studies target has been forest, in which the Sader, 2002). For mining related changes, J. Li et al. (2015a) used
methods of statistical boundary (Brooks et al., 2014; Dutrieux the thresholding method to reconstruct mining activity and
et al., 2015; Hamunyela et al., 2016; Reiche et al., 2015; Zhu Gillanders et al. (2008) explored mining impacts on land cover
et al., 2012), thresholding (Chen et al., 2013; Hilker et al., 2009; change based on the hypothesized trajectory method. For storm
Huang et al., 2010a, 2009; Lee, 2008), multi-date classification related changes, Pardo-Pascual et al. (2014) used the classification
(Hansen et al., 2016, 2014; Margono et al., 2012; Potapov et al., differencing method to evaluate the impact of storm on sandy bea-
2015, 2012; Sader and Legaard, 2008; Wilson and Sader, 2002), dif- ches. Unger et al. (2015) studied the impacts of petroleum explo-
ferencing (Bolton et al., 2015; Hayes and Sader, 2001; Jin and ration and production on land cover based on the classification
Sader, 2005; Zanotta et al., 2015), segmentation (Chance et al., differencing method.
2016; DeVries et al., 2015; Frazier et al., 2015; Griffiths et al., Some of the studies can also detect and classify multiple change
2012; Grogan et al., 2015; Hermosilla et al., 2015a; Kennedy agents. Neigh et al. (2014) detected all forest changes based on
et al., 2012, 2010), and trajectory classification (Ahmed et al., spectral/index differencing and then used supervised classification
2014; Sader et al., 2003) have been used. There are also change to separate different kinds of change agents, including logging, fire,
detection studies that are mainly focusing on studying the long- and insect. Kennedy et al. (2015) used segmentation method
term trends of vegetation using the regression method (Latifovic (Landtrendr) to detect changes in NBR time series and then used
and Pouliot, 2014; Sonnenschein et al., 2011; Vogelmann et al., topographic data, change related information, and patch shape
2012). Cloud can be detected based on the statistical boundary index to classify many change agents, including agriculture, forest
change detection method (Goodwin et al., 2013; Zhu and management, natural change, and riparian. Alaibakhsh et al. (2015)
Woodcock, 2014b). Wetland change can be detected based on used the thresholding method (MAD); change agents, such as cli-
regression (Fickas et al., 2016), thresholding (Kayastha et al., mate variability, fire events, and mining activity were detected
2012), and SMA differencing (Dingle Robertson et al., 2015). There based on each of the six MAD variates. Note that by detecting
are also studies focused on changes in coral reefs (Palandro et al., change based on its agent, we will be able to detect change even
2003, 2008), water (Li and Narayanan, 2003; Tulbure and Broich, if the land cover/use type is still the same before and after the
2013), seagrass (Dekker et al., 2005), and mangrove (Giri et al., change event. However, for this application, we need to have good
2008; Giri and Muhlhausen, 2008) based on the classification dif- understanding of the different change agents within the study
ferencing method. Note that by focusing on one or a few change area. Moreover, for a long time, many algorithms were only
targets, the identified changes are usually related to the categorical designed to detect a single change agent, but algorithms with the
changes for these change targets. If a change event occurred to the capability of detecting multiple change agents are attracting more
change target but did not change the type of the cover (for exam- and more attentions.
ple, forest impacted by beetle infestation is still a forest pixel), this
change will not be identified in this kind of application. Moreover,
there is a growing trend that many more studies are focusing on a 6. Conclusions
single change target instead of many land cover/use type, due to
the benefit of choosing more precise thresholds and better change The free and open Landsat data distribution policy started in
indices for the interested change target. 2008 has completely revolutionized the way of utilizing Landsat
Z. Zhu / ISPRS Journal of Photogrammetry and Remote Sensing 130 (2017) 370–384 381
data and has stimulated many novel change detection algorithms charts and landsat data. IEEE Trans. Geosci. Remote Sens. 52, 3316–3332. http://
dx.doi.org/10.1109/TGRS.2013.2272545.
based on Landsat time series. We have endeavored to keep
Chance, C.M., Hermosilla, T., Coops, N.C., Wulder, M.A., White, J.C., 2016. Effect of
updated with all the newly published studies involving the use topographic correction on forest change detection using spectral trend analysis
of Landsat time series for change detection, but there are always of Landsat pixel-based composites. Int. J. Appl. Earth Obs. Geoinf. 44, 186–194.
new algorithms and applications coming out, even during the pro- http://dx.doi.org/10.1016/j.jag.2015.09.003.
Chávez, P.S.J., 1996. Image-based atmospheric corrections - revisited and improved.
cess of writing this review. Among all the studies, a general trend is Photogramm. Eng. Remote Sensing 62, 1025–1036. doi: 0099-1112/96/6209-
observed: the more recent the study, the higher the frequency of 1025.
the Landsat time series used. At the same time, automated image Chen, X., Vogelmann, J.E., Chander, G., Ji, L., Tolk, B., Huang, C., Rollins, M., 2013.
Cross-sensor comparisons between Landsat 5 TM and IRS-P6 AWiFS and
preprocessing algorithms like LEDAPS and Fmask are being widely disturbance detection using integrated Landsat and AWiFS time-series images.
used, which further facilitate the use of Landsat time series. Based Int. J. Remote Sens. 34, 2432–2453. http://dx.doi.org/10.1080/
on the mathematical methods used for detecting change, we divide 01431161.2012.743690.
Chu, C., shang, J., Hornik, K., Kaun, C.M., 1995. Mosum tests for parameter
all change detection algorithms into six categories, including constancy. Biometrika 82, 603–617. http://dx.doi.org/10.1093/biomet/82.3.603.
thresholding, differencing, segmentation, trajectory classification, Coops, N.C., Gillanders, S.N., Wulder, M.A., Gergel, S.E., Nelson, T., Goodwin, N.R.,
statistical boundary, and regression. Each category has its own lim- 2010. Assessing changes in forest fragmentation following infestation using
time series Landsat imagery. For. Ecol. Manage. 259, 2355–2365. http://dx.doi.
itations and strengths, and users should understand each method org/10.1016/j.foreco.2010.03.008.
well before conducing change detection with Landsat time series. Coppin, P., Jonckheere, I., Nackaerts, K., Muys, B., Lambin, E., 2004. Review
Currently, most of the time series algorithms are heavily ArticleDigital change detection methods in ecosystem monitoring: a review.
Int. J. Remote Sens. 25, 1565–1596. http://dx.doi.org/10.1080/
focused on the temporal domain of the Landsat data, and the spa-
0143116031000101675.
tial domain of the data is almost entirely neglected. More algo- Cunningham, S., Rogan, J., Martin, D., DeLauer, V., McCauley, S., Shatz, A., 2015.
rithms that use the spatial–temporal techniques are anticipated. Mapping land development through periods of economic bubble and bust in
Though many online change detection algorithms have been devel- Massachusetts using Landsat time series data. GIScience Remote Sens. 1603.
http://dx.doi.org/10.1080/15481603.2015.1045277.
oped, most of the change detection products derived from Landsat Dekker, A.G., Brando, V.E., Anstee, J.M., 2005. Retrospective seagrass change
time series are at least one years old, which are not management detection in a shallow coastal tidal Australian lake. Remote Sens. Environ. 97,
relevant. More near real-time change detection products derived 415–433. http://dx.doi.org/10.1016/j.rse.2005.02.017.
Derrien, M., Farki, B., Harang, L., LeGléau, H., Noyalet, A., Pochic, D., Sairouni, A.,
from Landsat time series are expected. The change detection appli- 1993. Automatic cloud detection applied to NOAA-11 /AVHRR imagery. Remote
cations of the Landsat time series cover a wide range of topics, Sens. Environ. 46, 246–267. http://dx.doi.org/10.1016/0034-4257(93)90046-Z.
however, at present, most of the applications are limited to small DeVries, B., Decuyper, M., Verbesselt, J., Zeileis, A., Herold, M., Joseph, S., 2015.
Tracking disturbance-regrowth dynamics in tropical forests using structural
areas due to constraints of storage and computing resources. With change detection and Landsat time series. Remote Sens. Environ. 169, 320–334.
the advances in high-performance computing and cheaper storage, http://dx.doi.org/10.1016/j.rse.2015.08.020.
applications based on Landsat time series at continental or even Dingle Robertson, L., King, D.J., Davies, C., 2015. Assessing land cover change and
anthropogenic disturbance in wetlands using vegetation fractions derived from
global scale will be the mainstream in the next a few years. landsat 5 TM imagery (1984–2010). Wetlands 35, 1077–1091. http://dx.doi.org/
10.1007/s13157-015-0696-5.
Dutrieux, L.P., Verbesselt, J., Kooistra, L., Herold, M., 2015. Monitoring forest cover
Acknowledgements loss using multiple data streams, a case study of a tropical dry forest in Bolivia.
ISPRS J. Photogramm. Remote Sens. 107, 112–125. http://dx.doi.org/10.1016/j.
isprsjprs.2015.03.015.
We gratefully acknowledge the USGS leadership in the data pol- Fickas, K.C., Cohen, W.B., Yang, Z., 2016. Landsat-based monitoring of annual
icy that facilitated the advances in Landsat time series analysis. wetland change in the Willamette Valley of Oregon, USA from 1972 to 2012.
This research was supported by USGS Great Plains Cooperative Wetl. Ecol. Manage. 24, 73–92. http://dx.doi.org/10.1007/s11273-015-9452-0.
Franklin, S.E., Ahmed, O.S., Wulder, M.A., White, J.C., Hermosilla, T., Coops, N.C.,
Ecosystem Studies Unit (CESU) Program (grant number
2015. Large area mapping of annual land cover dynamics using multitemporal
G17AC00057). change detection and classification of landsat time series data. Can. J. Remote
Sens. 41, 293–314. http://dx.doi.org/10.1080/07038992.2015.1089401.
Fraser, R.H., Olthof, I., Pouliot, D., 2009. Monitoring land cover change and ecological
References integrity in Canada’s national parks. Remote Sens. Environ. 113, 1397–1409.
http://dx.doi.org/10.1016/j.rse.2008.06.019.
Frazier, R.J., Coops, N.C., Wulder, M.A., 2015. Boreal Shield forest disturbance and
Ackerman, S.A., Strabala, K.I., Menzel, W.P., Frey, R.a., Moeller, C.C., Gumley, L.E.,
recovery trends using Landsat time series. Remote Sens. Environ. 170, 317–327.
1998. Discriminating clear sky from clouds with MODIS. J. Geophys. Res. 103,
http://dx.doi.org/10.1016/j.rse.2015.09.015.
32141–32157. http://dx.doi.org/10.1029/1998JD200032.
Fu, P., Weng, Q., 2016. A time series analysis of urbanization induced land use and
Ahmed, O.S., Franklin, S.E., Wulder, M.A., 2014. Interpretation of forest disturbance
land cover change and its impact on land surface temperature with Landsat
using a time series of Landsat imagery and canopy structure from airborne lidar.
imagery. Remote Sens. Environ. 175, 205–214. http://dx.doi.org/10.1016/j.
Can. J. Remote Sens. 39, 521–542. http://dx.doi.org/10.5589/M14-004.
rse.2015.12.040.
Alaibakhsh, M., Emelyanova, I., Barron, O., Khiadani, M., 2015. Multivariate
Gao, F., Hilker, T., Zhu, X., Anderson, M., Masek, J., Wang, P., Yang, Y., 2015. Fusing
detection and attribution of land-cover changes in the Central Pilbara,
landsat and MODIS data for vegetation monitoring. IEEE Geosci. Remote Sens.
Western Australia. Int. J. Remote. http://dx.doi.org/10.1080/
Mag. 3, 47–60. http://dx.doi.org/10.1109/MGRS.2015.2434351.
01431161.2015.1042595.
Gao, F., Masek, J., Schwaller, M., Hall, F., 2006. On the blending of the landsat and
Banskota, A., Kayastha, N., Falkowski, M.J., Wulder, M.A., Froese, R.E., White, J.C.,
MODIS surface reflectance: Predicting daily landsat surface reflectance. IEEE
2014. Forest monitoring using landsat time series data: a review. Can. J. Remote
Trans. Geosci. Remote Sens. 44, 2207–2218. http://dx.doi.org/10.1109/
Sens. 40, 362–384. http://dx.doi.org/10.1080/07038992.2014.987376.
TGRS.2006.872081.
Berk, A., Bernstein, L.S., Anderson, G.P., Acharya, P.K., Robertson, D.C., Chetwynd, J.
Gavier-Pizarro, G.I., Kuemmerle, T., Hoyos, L.E., Stewart, S.I., Huebner, C.D., Keuler,
H., 1998. MODTRAN Cloud and Multiple Scattering Upgrades with Application
N.S., Radeloff, V.C., 2012. Monitoring the invasion of an exotic tree (Ligustrum
to AVIRIS 4257.
lucidum) from 1983 to 2006 with Landsat TM/ETM + satellite data and Support
Bolton, D.K., Coops, N.C., Wulder, M.A., 2015. Characterizing residual structure and
Vector Machines in Córdoba, Argentina. Remote Sens. Environ. 122, 134–145.
forest recovery following high-severity fire in the western boreal of Canada
http://dx.doi.org/10.1016/j.rse.2011.09.023.
using Landsat time-series and airborne lidar data. Remote Sens. Environ. 163,
Gillanders, S.N., Coops, N.C., Wulder, M.A., Goodwin, N.R., 2008. Application of
48–60. http://dx.doi.org/10.1016/j.rse.2015.03.004.
Landsat satellite imagery to monitor land-cover changes at the Athabasca Oil
Boriah, S., 2010. Time series change detection: Algorithms for land cover change
Sands, Alberta, Canada. Can. Geogr. 52, 466–485. http://dx.doi.org/10.1111/
160.
j.1541-0064.2008.00225.x.
Boucher, a., Seto, K.C., Journel, a.G., 2006. A novel method for mapping land cover
Giri, C., Muhlhausen, J., 2008. Mangrove forest distributions and dynamics in
changes: incorporating time and space with geostatistics. IEEE Trans. Geosci.
Madagascar (1975–2005). Sensors 8, 2104–2117. http://dx.doi.org/10.3390/
Remote Sens. 44, 3427–3435. http://dx.doi.org/10.1109/TGRS.2006.879113.
s8042104.
Braaten, J.D., Cohen, W.B., Yang, Z., 2015. Automated cloud and cloud shadow
Giri, C., Zhu, Z., Tieszen, L.L., Singh, A., Gillette, S., Kelmelis, J.A., 2008. Mangrove
identification in Landsat MSS imagery for temperate ecosystems. Remote Sens.
forest distributions and dynamics (19752005) of the tsunami-affected region of
Environ. 169, 128–138. http://dx.doi.org/10.1016/j.rse.2015.08.006.
Asia. J. Biogeogr. 35, 519–528. http://dx.doi.org/10.1111/j.1365-
Brooks, E.B., Wynne, R.H., Thomas, V.A., Blinn, C.E., Coulston, J.W., 2014. On-the-fly
2699.2007.01806.x.
massively multitemporal change detection using statistical quality control
382 Z. Zhu / ISPRS Journal of Photogrammetry and Remote Sensing 130 (2017) 370–384
Goodwin, N.R., Collett, L.J., 2014. Development of an automated method for Irish, R.R., Barker, J.L., Goward, S.N., Arvidson, T., 2006. Characterization of the
mapping fire history captured in Landsat TM and ETM + time series across Landsat-7 ETM + Automated Cloud-Cover Assessment (ACCA) Algorithm.
Queensland, Australia. Remote Sens. Environ. 148, 206–221. http://dx.doi.org/ Photogramm. Eng. Remote Sens. 72, 1179–1188. http://dx.doi.org/10.14358/
10.1016/j.rse.2014.03.021. PERS.72.10.1179.
Goodwin, N.R., Collett, L.J., Denham, R.J., Flood, N., Tindall, D., 2013. Cloud and cloud Jin, S., Homer, C., Yang, L., Xian, G., Fry, J., Danielson, P., Townsend, P.a., 2013.
shadow screening across Queensland, Australia: An automated method for Automated cloud and shadow detection and filling using two-date Landsat
Landsat TM/ETM + time series. Remote Sens. Environ. 134, 50–65. http://dx.doi. imagery in the USA. Int. J. Remote Sens. 34, 1540–1560. http://dx.doi.org/
org/10.1016/j.rse.2013.02.019. 10.1080/01431161.2012.720045.
Griffiths, P., Kuemmerle, T., Kennedy, R.E., Abrudan, I.V., Knorn, J., Hostert, P., 2012. Jin, S., Sader, S.A., 2005. Comparison of time series tasseled cap wetness and the
Using annual time-series of Landsat images to assess the effects of forest normalized difference moisture index in detecting forest disturbances. Remote
restitution in post-socialist Romania. Remote Sens. Environ. 118, 199–214. Sens. Environ. 94, 364–372. http://dx.doi.org/10.1016/j.rse.2004.10.012.
http://dx.doi.org/10.1016/j.rse.2011.11.006. Kaufmann, R.K., Seto, K.C., 2001. Change detection, accuracy, and bias in a
Griffiths, P., van der Linden, S., Kuemmerle, T., Hostert, P., 2013. A pixel-based sequential analysis of Landsat imagery in the Pearl River Delta, China:
landsat compositing algorithm for large area land cover mapping. Sel. Top. Appl. Econometric techniques. Agric. Ecosyst. Environ. 85, 95–105. http://dx.doi.org/
Earth Obs. Remote Sensing, IEEE J. 6, 2088–2101. http://dx.doi.org/10.1109/ 10.1016/S0167-8809(01)00190-6.
JSTARS.2012.2228167. Kayastha, N., Thomas, V., Galbraith, J., Banskota, A., 2012. Monitoring wetland
Grogan, K., Pflugmacher, D., Hostert, P., Kennedy, R., Fensholt, R., 2015. Cross-border change using inter-annual landsat time-series data. Wetlands 32, 1149–1162.
forest disturbance and the role of natural rubber in mainland Southeast Asia http://dx.doi.org/10.1007/s13157-012-0345-1.
using annual Landsat time series. Remote Sens. Environ. 169, 438–453. http:// Kennedy, R.E., Andréfouët, S., Cohen, W.B., Gómez, C., Griffiths, P., Hais, M., Healey,
dx.doi.org/10.1016/j.rse.2015.03.001. S.P., Helmer, E.H., Hostert, P., Lyons, M.B., Meigs, G.W., Pflugmacher, D., Phinn, S.
Hagolle, O., Huc, M., Pascual, D.V., Dedieu, G., 2010. A multi-temporal method for R., Powell, S.L., Scarth, P., Sen, S., Schroeder, T.A., Schneider, A., Sonnenschein, R.,
cloud detection, applied to FORMOSAT-2, VENmS, LANDSAT and SENTINEL-2 Vogelmann, J.E., Wulder, M.A., Zhu, Z., 2014. Bringing an ecological view of
images. Remote Sens. Environ. 114, 1747–1755. http://dx.doi.org/10.1016/j. change to landsat-based remote sensing. Front. Ecol. Environ. 12, 339–346.
rse.2010.03.002. http://dx.doi.org/10.1890/130066.
Hamunyela, E., Verbesselt, J., Herold, M., 2016. Using spatial context to improve Kennedy, R.E., Cohen, W.B., Schroeder, T.A., 2007. Trajectory-based change
early detection of deforestation from Landsat time series. Remote Sens. Environ. detection for automated characterization of forest disturbance dynamics.
172, 126–138. http://dx.doi.org/10.1016/j.rse.2015.11.006. Remote Sens. Environ. 110, 370–386. http://dx.doi.org/10.1016/j.
Hansen, M.C., Egorov, A., Potapov, P.V., Stehman, S.V., Tyukavina, A., Turubanova, S. rse.2007.03.010.
A., Roy, D.P., Goetz, S.J., Loveland, T.R., Ju, J., Kommareddy, A., Kovalskyy, V., Kennedy, R.E., Yang, Z., Braaten, J., Copass, C., Antonova, N., Jordan, C., Nelson, P.,
Forsyth, C., Bents, T., 2014. Monitoring conterminous United States (CONUS) 2015. Attribution of disturbance change agent from Landsat time-series in
land cover change with Web-Enabled Landsat Data (WELD). Remote Sens. support of habitat monitoring in the Puget Sound region, USA. Remote Sens.
Environ. 140, 466–484. http://dx.doi.org/10.1016/j.rse.2013.08.014. Environ. 166, 271–285. http://dx.doi.org/10.1016/j.rse.2015.05.005.
Hansen, M.C., Krylov, A., Tyukavina, A., Potapov, P.V., Turubanova, S., Zutta, B., Ifo, S., Kennedy, R.E., Yang, Z., Cohen, W.B., 2010. Detecting trends in forest disturbance
Margono, B., Stolle, F., Moore, R., 2016. Humid tropical forest disturbance alerts and recovery using yearly Landsat time series: 1. LandTrendr - Temporal
using Landsat data. Environ. Res. Lett. 11, 34008. http://dx.doi.org/10.1088/ segmentation algorithms. Remote Sens. Environ. 114, 2897–2910. http://dx.doi.
1748-9326/11/3/034008. org/10.1016/j.rse.2010.07.008.
Hansen, M.C., Loveland, T.R., 2012. A review of large area monitoring of land cover Kennedy, R.E., Yang, Z., Cohen, W.B., Pfaff, E., Braaten, J., Nelson, P., 2012. Spatial and
change using Landsat data. Remote Sens. Environ. 122, 66–74. http://dx.doi.org/ temporal patterns of forest disturbance and regrowth within the area of the
10.1016/j.rse.2011.08.024. Northwest Forest Plan. Remote Sens. Environ. 122, 117–133. http://dx.doi.org/
Hansen, M.C.C., Potapov, P.V., Moore, R., Hancher, M., Turubanova, S.A.a., Tyukavina, 10.1016/j.rse.2011.09.024.
A., Thau, D., Stehman, S.V.V., Goetz, S.J.J., Loveland, T.R.R., Kommareddy, a., Knudby, A., Newman, C., Shaghude, Y., Muhando, C., 2010. Simple and effective
Egorov, A., Chini, L., Justice, C.O.O., Townshend, J.R.G.R.G., Patapov, P.V., Moore, monitoring of historic changes in nearshore environments using the free
R., Hancher, M., Turubanova, S.A.a., Tyukavina, A., Thau, D., Stehman, S.V.V., archive of Landsat imagery. Int. J. Appl. Earth Obs. Geoinf. 12, 116–122. http://
Goetz, S.J.J., Loveland, T.R.R., Kommaredy, A., Egorov, A., Chini, L., Justice, C.O.O., dx.doi.org/10.1016/j.jag.2009.09.002.
Townshend, J.R.G.R.G., 2013. High-Resolution Global Maps of. Science (80-) 342, Kontgis, C., Schneider, A., Ozdogan, M., 2015. Mapping rice paddy extent and
850–854. http://dx.doi.org/10.1126/science.1244693. intensification in the Vietnamese Mekong River Delta with dense time stacks of
Hayes, D.J., Sader, S.a., 2001. Comparison of change detection techniques for Landsat data. Remote Sens. Environ. 169, 255–269. http://dx.doi.org/10.1016/j.
monitoring tropical forest clearing and vegetation regrowth in a time series. rse.2015.08.004.
Photogramm. Eng. Remote Sens. 67 (9), 1067–1075. doi:citeulike-article- Latifovic, R., Pouliot, D., 2014. Monitoring cumulative long-term vegetation changes
id:7954186. over the Athabasca Oil Sands region. IEEE J. Sel. Top. Appl. Earth Obs. Remote
Hermosilla, T., Wulder, M.A., White, J.C., Coops, N.C., Hobart, G.W., 2015a. Regional Sens. 7, 3380–3392. http://dx.doi.org/10.1109/JSTARS.2014.2321058.
detection, characterization, and attribution of annual forest change from 1984 Lawes, R.A., Wallace, J.F., 2008. Monitoring an invasive perennial at the landscape
to 2012 using Landsat-derived time-series metrics. Remote Sens. Environ. 170, scale with remote sensing. Ecol. Manage. Restor. 9, 53–59. http://dx.doi.org/
121–132. http://dx.doi.org/10.1016/j.rse.2015.09.004. 10.1111/j.1442-8903.2008.00387.x.
Hermosilla, T., Wulder, M.A., White, J.C., Coops, N.C., Hobart, G.W., 2015b. An Lee, H., 2008. Mapping deforestation and age of evergreen trees by applying a
integrated Landsat time series protocol for change detection and generation of binary coding method to time-series landsat november images. IEEE Trans.
annual gap-free surface reflectance composites. Remote Sens. Environ. 158, Geosci. Remote Sens. 46, 3926–3936. http://dx.doi.org/10.1109/
220–234. http://dx.doi.org/10.1016/j.rse.2014.11.005. TGRS.2008.2001158.
Hilker, T., Wulder, M.A., Coops, N.C., Linke, J., McDermid, G., Masek, J.G., Gao, F., Leisch, F., Hornik, K., Kuan, C.-M., 2000. Monitoring structural changes with the
White, J.C., 2009. A new data fusion model for high spatial- and temporal- generalized fluctuation test. Econom. Theory 16, 835–854. doi:null.
resolution mapping of forest disturbance based on Landsat and MODIS. Remote Li, J., Narayanan, R.M., 2003. A shape-based approach to change detection of lakes
Sens. Environ. 113, 1613–1627. http://dx.doi.org/10.1016/j.rse.2009.03.007. using time series remote sensing images. Geosci. Remote Sensing, IEEE Trans.
Holben, B.N., 1986. Characteristics of maximum-value composite images from 41, 2466–2477.
temporal AVHRR data. Int. J. Remote Sens. 7, 1417–1434. http://dx.doi.org/ Li, J., Zipper, C.E., Donovan, P.F., Wynne, R.H., Oliphant, A.J., 2015a. Reconstructing
10.1080/01431168608948945. disturbance history for an intensively mined region by time-series analysis of
Holden, C.E., Woodcock, C.E., 2016. An analysis of Landsat 7 and Landsat 8 Landsat imagery. Environ. Monit. Assess. 187. http://dx.doi.org/10.1007/
underflight data and the implications for time series investigations. Remote s10661-015-4766-1.
Sens. Environ. 185, 16–36. http://dx.doi.org/10.1016/j.rse.2016.02.052. Li, X., Gong, P., Liang, L., 2015. A 30-year (1984–2013) record of annual urban
Huang, C., Goward, S.N., Masek, J.G., Thomas, N., Zhu, Z., Vogelmann, J.E., 2010a. An dynamics of Beijing City derived from Landsat data. Remote Sens. Environ. 166,
automated approach for reconstructing recent forest disturbance history using 78–90. http://dx.doi.org/10.1016/j.rse.2015.06.007.
dense Landsat time series stacks. Remote Sens. Environ. 114, 183–198. http:// Liang, L., Hawbaker, T.J., Chen, Y., Zhu, Z., Gong, P., 2014. Characterizing recent and
dx.doi.org/10.1016/j.rse.2009.08.017. projecting future potential patterns of mountain pine beetle outbreaks in the
Huang, C., Goward, S.N., Schleeweis, K., Thomas, N., Masek, J.G., Zhu, Z., 2009. Southern Rocky Mountains. Appl. Geogr. 55, 165–175. http://dx.doi.org/
Dynamics of national forests assessed using the Landsat record: Case studies in 10.1016/j.apgeog.2014.09.012.
eastern United States. Remote Sens. Environ. 113, 1430–1442. http://dx.doi.org/ Linke, J., McDermid, G.J., Laskin, D.N., McLane, a.J., Pape, A.D., Cranston, J., Hall-
10.1016/j.rse.2008.06.016. Beyer, M., Franklin, S.E., 2009. A disturbance-inventory framework for flexible
Huang, C., Thomas, N., Goward, S.N., Masek, J.G., Zhu, Z., Townshend, J.R.G., and reliable landscape monitoring. Photogramm. Eng. Remote Sensing 75, 981–
Vogelmann, J.E., 2010b. Automated masking of cloud and cloud shadow for 995. http://dx.doi.org/10.14358/PERS.75.8.981.
forest change analysis using Landsat images. Int. J. Remote Sens. 31, 5449–5464. Liu, D., Cai, S., 2012. A spatial-temporal modeling approach to reconstructing land-
http://dx.doi.org/10.1080/01431160903369642. cover change trajectories from multi-temporal satellite imagery. Ann. Assoc.
Hughes, M.J., Hayes, D.J., 2014. Automated detection of cloud and cloud shadow in Am. Geogr. 102, 1329–1347. http://dx.doi.org/10.1080/00045608.2011.596357.
single-date Landsat imagery using neural networks and spatial post-processing. Loveland, T.R., Dwyer, J.L., 2012. Landsat: building a strong future. Remote Sens.
Remote Sens. 6, 4907–4926. http://dx.doi.org/10.3390/rs6064907. Environ. 122, 22–29. http://dx.doi.org/10.1016/j.rse.2011.09.022.
Irish, R.R., 2000. Landsat 7 automatic cloud cover assessment. AeroSense 2000 Lu, D., Mausel, P., Brondizio, E., Moran, E., 2004. Change detection techniques. Int. J.
(4049), 348–355. http://dx.doi.org/10.1117/12.410358. Remote Sens. 25 (12), 2365–2401.
Z. Zhu / ISPRS Journal of Photogrammetry and Remote Sensing 130 (2017) 370–384 383
Luo, Y., Trishchenko, A.P., Khlopenkov, K.V., 2008. Developing clear-sky, cloud and 2012 quantified from the full Landsat archive. Remote Sens. Environ. 159, 28–
cloud shadow mask for producing clear-sky composites at 250-meter spatial 43. http://dx.doi.org/10.1016/j.rse.2014.11.027.
resolution for the seven MODIS land bands over Canada and North America. Powell, S.L., Cohen, W.B., Yang, Z., Pierce, J.D., Alberti, M., 2008. Quantification of
Remote Sens. Environ. 112, 4167–4185. http://dx.doi.org/10.1016/j. impervious surface in the Snohomish Water Resources Inventory Area of
rse.2008.06.010. Western Washington from 1972–2006. Remote Sens. Environ. 112, 1895–1908.
Margono, B.A., Turubanova, S., Zhuravleva, I., Potapov, P., Tyukavina, A., Baccini, A., http://dx.doi.org/10.1016/j.rse.2007.09.010.
Goetz, S., Hansen, M.C., 2012. Mapping and monitoring deforestation and forest Qiu, S., He, B., Zhu, Z., Liao, Z., Quan, X., 2017. Improving Fmask cloud and cloud
degradation in Sumatra (Indonesia) using Landsat time series data sets from shadow detection in mouantianous area for Landsats 4–8. Remote Sens.
1990 to 2010. Environ. Res. Lett. 7, 34010. http://dx.doi.org/10.1088/1748- Environ. Accepted.
9326/7/3/034010. Ramadan, E., Feng, X., Cheng, Z., 2004. Satellite remote sensing for urban growth
Markham, B.L., Helder, D.L., 2012. Forty-year calibrated record of earth-reflected assessment in Shaoxing City, Zhejiang Province. J. Zhejiang Univ. Sci. 5, 1095–
radiance from Landsat: A review. Remote Sens. Environ. http://dx.doi.org/ 1101. http://dx.doi.org/10.1631/jzus.2004.1095.
10.1016/j.rse.2011.06.026. Reiche, J., Verbesselt, J., Hoekman, D., Herold, M., 2015. Fusing Landsat and SAR time
Marzen, L.J., Szantoi, Z., Harrington, L.M.B., Harrington, J.a., 2011. Implications of series to detect deforestation in the tropics. Remote Sens. Environ. 156, 276–
management strategies and vegetation change in the Mount St. Helens blast 293. http://dx.doi.org/10.1016/j.rse.2014.10.001.
zone. Geocarto Int. 26, 359–376. http://dx.doi.org/10.1080/ Richter, R., 1997. Correction of atmospheric and topographic effects for high spatial
10106049.2011.584977. resolution satellite imagery. Int. J. Remote Sens. 18, 1099–1111. http://dx.doi.
Masek, J.G., Vermote, E.F., Saleous, N.E., Wolfe, R., Hall, F.G., Huemmrich, K.F., Lim, T. org/10.1080/014311697218593.
K., 2006. A Landsat surface reflectance dataset for North America, 1990–2000. Roy, D.P., Ju, J., Kline, K., Scaramuzza, P.L., Kovalskyy, V., Hansen, M., Loveland, T.R.,
IEEE Geosci. Remote Sens. Lett. 3 (1), 68–72. Vermote, E., Zhang, C., 2010. Web-enabled Landsat Data (WELD): Landsat ETM
Maxwell, S.K., Sylvester, K.M., 2012. Identification of ‘‘ever-cropped” land (1984– + composited mosaics of the conterminous United States. Remote Sens. Environ.
2010) using Landsat annual maximum NDVI image composites: Southwestern 114, 35–49. http://dx.doi.org/10.1016/j.rse.2009.08.011.
Kansas case study. Remote Sens. Environ. 121, 186–195. http://dx.doi.org/ Roy, D.P., Ju, J., Lewis, P., Schaaf, C., Gao, F., Hansen, M., Lindquist, E., 2008.
10.1016/j.rse.2012.01.022. Multi-temporal MODIS-Landsat data fusion for relative radiometric
Meigs, G.W., Kennedy, R.E., Cohen, W.B., 2011. A Landsat time series approach to normalization, gap filling, and prediction of Landsat data. Remote Sens.
characterize bark beetle and defoliator impacts on tree mortality and surface Environ. 112, 3112–3130. http://dx.doi.org/10.1016/j.rse.2008.03.009.
fuels in conifer forests. Remote Sens. Environ. 115, 3707–3718. http://dx.doi. Roy, D.P., Kovalskyy, V., Zhang, H.K., Vermote, E.F., Yan, L., Kumar, S.S., Egorov, A.,
org/10.1016/j.rse.2011.09.009. 2016. Characterization of Landsat-7 to Landsat-8 reflective wavelength and
Meigs, G.W., Kennedy, R.E., Gray, A.N., Gregory, M.J., 2015. Spatiotemporal dynamics normalized difference vegetation index continuity. Remote Sens. Environ. 185,
of recent mountain pine beetle and western spruce budworm outbreaks across 57–70. http://dx.doi.org/10.1016/j.rse.2015.12.024.
the Pacific Northwest Region, USA. For. Ecol. Manage. 339, 71–86. http://dx.doi. Roy, D.P., Wulder, M.A., Loveland, T.R., C.E., W., Allen, R.G., Anderson, M.C., Helder,
org/10.1016/j.foreco.2014.11.030. D., Irons, J.R., Johnson, D.M., Kennedy, R., Scambos, T.A., Schaaf, C.B., Schott, J.R.,
Mihai, B., Nistor, C., Simion, G., 2015. Post-socialist urban growth of Bucharest, Sheng, Y., Vermote, E.F., Belward, A.S., Bindschadler, R., Cohen, W.B., Gao, F.,
Romania – a change detection analysis on Landsat imagery (1984–2010). Acta Hipple, J.D., Hostert, P., Huntington, J., Justice, C.O., Kilic, A., Kovalskyy, V., Lee, Z.
Geogr. Slov. 55, 223–234. http://dx.doi.org/10.3986/AGS.709. P., Lymburner, L., Masek, J.G., McCorkel, J., Shuai, Y., Trezza, R., Vogelmann, J.,
Muttitanon, W., Tripathi, N.K., 2005. Land use/land cover changes in the coastal Wynne, R.H., Zhu, Z., 2014. Landsat-8: Science and product vision for terrestrial
zone of Ban Don Bay, Thailand using Landsat 5 TM data. Int. J. Remote Sens. 26, global change research. Remote Sens. Environ. 145, 154–172. http://dx.doi.org/
2311–2323. http://dx.doi.org/10.1080/0143116051233132666. 10.1016/j.rse.2014.02.001.
Neigh, C.S.R., Bolton, D.K., Williams, J.J., Diabate, M., 2014. Evaluating an automated Sader, S.A., Bertrand, M., Wilson, E.H., 2003. Satellite change detection of forest
approach for monitoring forest disturbances in the Pacific Northwest from harvest patterns on an industrial forest landscape. For. Sci. 49, 341–353.
logging, fire and insect outbreaks with landsat time series data. Forests 5, 3169– Sader, S.A., Legaard, K.R., 2008. Inclusion of forest harvest legacies, forest type, and
3198. http://dx.doi.org/10.3390/f5123169. regeneration spatial patterns in updated forest maps: A comparison of mapping
Nutini, F., Boschetti, M., Brivio, P.A., Bocchi, S., Antoninetti, M., 2013. Land-use and results. For. Ecol. Manage. 255, 3846–3856. http://dx.doi.org/10.1016/
land-cover change detection in a semi-arid area of Niger using multi-temporal j.foreco.2008.03.047.
analysis of Landsat images. Int. J. Remote Sens. 34, 4769–4790. http://dx.doi. Saunders, R.W., Kriebel, K.T., 1988. An improved method for detecting clear sky and
org/10.1080/01431161.2013.781702. cloudy radiances from AVHRR data. Int. J. Remote Sens. 9, 123–150. http://dx.
Oreopoulos, L., Wilson, M.J., Várnai, T., 2011. Implementation on landsat data of a doi.org/10.1080/01431168808954841.
simple cloud-mask algorithm developed for MODIS land bands. IEEE Geosci. Scaramuzza, P.L., Bouchard, M.A., Dwyer, J.L., 2012. Development of the
Remote Sens. Lett. 8, 597–601. http://dx.doi.org/10.1109/LGRS.2010.2095409. landsat data continuity mission cloud-cover assessment algorithms. IEEE
Palandro, D.A., Andréfouët, S., Hu, C., Hallock, P., Müller-Karger, F.E., Dustan, P., Trans. Geosci. Remote Sens. 50, 1140–1154. http://dx.doi.org/10.1109/
Callahan, M.K., Kranenburg, C., Beaver, C.R., 2008. Quantification of two decades TGRS.2011.2164087.
of shallow-water coral reef habitat decline in the Florida Keys National Marine Schott, J.R., Gerace, A., Woodcock, C.E., Wang, S., Zhu, Z., Wynne, R.H., Blinn, C.E.,
Sanctuary using Landsat data (1984–2002). Remote Sens. Environ. 112, 3388– 2016. The impact of improved signal-to-noise ratios on algorithm performance:
3399. http://dx.doi.org/10.1016/j.rse.2008.02.015. Case studies for Landsat class instruments. Remote Sens. Environ. 185, 37–45.
Palandro, D., Andréfouët, S., Muller-Karger, F.E., Dustan, P., Hu, C., Hallock, P., 2003. http://dx.doi.org/10.1016/j.rse.2016.04.015.
Detection of changes in coral reef communities using Landsat-5 TM and Schroeder, T.A., Wulder, M.A., Healey, S.P., Moisen, G.G., 2012. Detecting post-fire
Landsat-7 ETM + data. Can. J. Remote Sens. 29, 201–209. http://dx.doi.org/ salvage logging from Landsat change maps and national fire survey data.
10.5589/m02-095. Remote Sens. Environ. 122, 166–174. http://dx.doi.org/10.1016/j.rse.2011.10.
Pardo-Pascual, J.E., Almonacid-Caballer, J., Ruiz, L.A., Palomar-Vázquez, J., Rodrigo- 031.
Alemany, R., 2014. Evaluation of storm impact on sandy beaches of the Gulf of Schroeder, T.a., Cohen, W.B., Song, C., Canty, M.J.M.J., Yang, Z., Song, S., Canty, M.J.M.
Valencia using Landsat imagery series. Geomorphology 214, 388–401. http:// J., Zhiqiang, Y., 2006. Radiometric correction of multi-temporal Landsat data for
dx.doi.org/10.1016/j.geomorph.2014.02.020. charecterization of early successional forest patterns in western Oregon.
Parker, B.M., Lewis, T., Srivastava, S.K., 2015. Estimation and evaluation of multi- Remote Sens. Environ. 103, 16–26. http://dx.doi.org/10.1016/j.rse.2006.03.008.
decadal fire severity patterns using Landsat sensors. Remote Sens. Environ. 170, Senf, C., Pflugmacher, D., Wulder, M.A., Hostert, P., 2015. Characterizing spectral-
340–349. http://dx.doi.org/10.1016/j.rse.2015.09.014. temporal patterns of defoliator and bark beetle disturbances using Landsat time
Pengra, B., Gallant, A., Zhu, Z., Dahal, D., 2016. Evaluation of the initial thematic series. Remote Sens. Environ. 170, 166–177. http://dx.doi.org/10.1016/j.
output from a continuous change-detection algorithm for use in automated rse.2015.09.019.
operational land-change mapping by the U.S. Geological Survey. Remote Sens. 8, Sieber, A., Kuemmerle, T., Prishchepov, A.V., Wendland, K.J., Baumann, M., Radeloff,
811. http://dx.doi.org/10.3390/rs8100811. V.C., Baskin, L.M., Hostert, P., 2013. Landsat-based mapping of post-Soviet land-
Pickell, P.D., Hermosilla, T., Coops, N.C., Masek, J.G., Franks, S., Huang, C., 2014. use change to assess the effectiveness of the Oksky and Mordovsky protected
Monitoring anthropogenic disturbance trends in an industrialized boreal forest areas in European Russia. Remote Sens. Environ. 133, 38–51. http://dx.doi.org/
with Landsat time series. Remote Sens. Lett. 5, 783–792. http://dx.doi.org/ 10.1016/j.rse.2013.01.021.
10.1080/2150704X.2014.967881. Singh, A., 1989. Digital change detection techniques using remotely-sensed
Pillay, K., Agjee, N.H., Pillay, S., 2014. Modelling changes in land cover patterns in data. Int. J. Remote Sens. 10, 989–1003. http://dx.doi.org/10.1080/
Mtunzini, South Africa using satellite imagery. J. Indian Soc. Remote Sens. 42, 01431168908903939.
51–60. http://dx.doi.org/10.1007/s12524-013-0312-1. Song, C., Woodcock, C.E., 2003. Monitoring forest succession with multitemporal
Potapov, P., Turubanova, S., Hansen, M.C., 2011. Regional-scale boreal forest cover landsat images: factors of uncertainty. IEEE Trans. Geosci. Remote Sens. 41,
and change mapping using Landsat data composites for European Russia. 2557–2567. http://dx.doi.org/10.1109/TGRS.2003.818367.
Remote Sens. Environ. 115, 548–561. http://dx.doi.org/10.1016/j. Song, C., Woodcock, C.E., Seto, K.C., Lenney, M.P., Macomber, S.A., 2001.
rse.2010.10.001. Classification and change detection using Landsat TM data: When and how to
Potapov, P.V., Turubanova, S.A., Hansen, M.C., Adusei, B., Broich, M., Altstatt, A., correct atmospheric effects? Remote Sens. Environ. 75, 230–244. http://dx.doi.
Mane, L., Justice, C.O., 2012. Quantifying forest cover loss in Democratic org/10.1016/S0034-4257(00)00169-3.
Republic of the Congo, 2000–2010, with Landsat ETM + data. Remote Sens. Sonnenschein, R., Kuemmerle, T., Udelhoven, T., Stellmes, M., Hostert, P., 2011.
Environ. 122, 106–116. http://dx.doi.org/10.1016/j.rse.2011.08.027. Differences in Landsat-based trend analyses in drylands due to the choice of
Potapov, P.V., Turubanova, S.A., Tyukavina, A., Krylov, A.M., McCarty, J.L., Radeloff, V. vegetation estimate. Remote Sens. Environ. 115, 1408–1420. http://dx.doi.org/
C., Hansen, M.C., 2015. Eastern Europe’s forest cover dynamics from 1985 to 10.1016/j.rse.2011.01.021.
384 Z. Zhu / ISPRS Journal of Photogrammetry and Remote Sensing 130 (2017) 370–384
Su, Z., 2000. Remote sensing of land use and vegetation for mesoscale hydrological Status, consolidation, and direction. Remote Sens. Environ. 185, 271–283.
studies. Int. J. Remote Sens. 21, 213–233. http://dx.doi.org/10.1080/ http://dx.doi.org/10.1016/j.rse.2015.11.032.
014311600210803. Xue, X., Liu, H., Mu, X., Liu, J., 2014. Trajectory-based detection of urban expansion
Sulla-Menashe, D., Friedl, M.A., Woodcock, C.E., 2016. Sources of bias and variability using Landsat time series. Int. J. Remote Sens. 35, 1450–1465. http://dx.doi.org/
in long-term Landsat time series over Canadian boreal forests. Remote Sens. 10.1080/01431161.2013.878058.
Environ. 177, 206–219. http://dx.doi.org/10.1016/j.rse.2016.02.041. Yang, Y., Zhang, S., Wang, D., Yang, J., Xing, X., 2015. Spatiotemporal changes of
Thonfeld, F., Hecheltjen, A., Menz, G., 2015. Bi-temporal change detection, change farming-pastoral ecotone in Northern China, 1954–2005: A case study in
trajectories and time series analysis for forest monitoring. Photogramm. - Zhenlai County, Jilin Province. Sustain. 7, 1–22. http://dx.doi.org/
Fernerkundung - Geoinf. 2015, 129–141. http://dx.doi.org/10.1127/pfg/2015/ 10.3390/su7010001.
0259. Zanotta, D.C., Bruzzone, L., Bovolo, F., Shimabukuro, Y.E., 2015. An adaptive
Tulbure, M.G., Broich, M., 2013. Spatiotemporal dynamic of surface water bodies semisupervised approach to the detection of user-defined recurrent changes
using Landsat time-series data from 1999 to 2011. ISPRS J. Photogramm. in image time series. IEEE Trans. Geosci. Remote Sens. 53, 3707–3719. http://dx.
Remote Sens. 79, 44–52. http://dx.doi.org/10.1016/j.isprsjprs.2013.01.010. doi.org/10.1109/TGRS.2014.2381645.
Unger, D., Hung, I.-K., Farrish, K., Dans, D., 2015. Quantifying land cover change due Zeileis, A., Leisch, F., Kleiber, C., Hornik, K., 2005. Monitoring structural change in
to petroleum exploration and production in the Haynesville shale region using dynamic econometric models. J. Appl. Econom. 20, 99–121. http://dx.doi.org/
remote sensing. Int. J. Appl. Geospatial Res. 6, 1–17. http://dx.doi.org/10.4018/ 10.1002/jae.776.
ijagr.2015040101. Zhang, H., Qi, Z.fang, Ye, X.yue, Cai, Y.bin, Ma, W.chun, Chen, M.nan, 2013. Analysis
Verbesselt, J., Hyndman, R., Zeileis, A., Culvenor, D., 2010. Phenological change of land use/land cover change, population shift, and their effects on
detection while accounting for abrupt and gradual trends in satellite image spatiotemporal patterns of urban heat islands in metropolitan Shanghai.
time series. Remote Sens. Environ. 114, 2970–2980. http://dx.doi.org/10.1016/j. China. Appl. Geogr. 44, 121–133. http://dx.doi.org/10.1016/j.
rse.2010.08.003. apgeog.2013.07.021.
Verbesselt, J., Zeileis, A., Herold, M., 2012. Near real-time disturbance detection Zhang, H.K., Roy, D.P., 2016. Landsat 5 Thematic Mapper reflectance and NDVI 27-
using satellite image time series. Remote Sens. Environ. 123, 98–108. http://dx. year time series inconsistencies due to satellite orbit change. Remote Sens.
doi.org/10.1016/j.rse.2012.02.022. Environ. 186, 217–233. http://dx.doi.org/10.1016/j.rse.2016.08.022.
Vermote, E., Justice, C., Claverie, M., Franch, B., 2016. Preliminary analysis of the Zhao, J., Guo, W., Huang, W., Huang, L., Zhang, D., Yang, H., Yuan, L., 2014.
performance of the Landsat 8/OLI land surface reflectance product. Remote Characterizing spatiotemporal dynamics of land cover with multi-temporal
Sens. Environ. 185, 46–56. http://dx.doi.org/10.1016/j.rse.2016.04.008. remotely sensed imagery in Beijing during 1978–2010. Arab. J. Geosci. 7, 3945–
Vermote, E., Saleous, N., 2007. LEDAPS surface reflectance product description 1–21. 3959. http://dx.doi.org/10.1007/s12517-013-1072-5.
Vermote, E.F., Tanré, D., Deuzé, J.L., Herman, M., Morcrette, J.J., 1997. Second Zhu, X., Chen, J., Gao, F., Chen, X., Masek, J.G., 2010. An enhanced spatial and
simulation of the satellite signal in the solar spectrum, 6 s: an overview. IEEE temporal adaptive reflectance fusion model for complex heterogeneous regions.
Trans. Geosci. Remote Sens. 35, 675–686. http://dx.doi.org/10.1109/36.581987. Remote Sens. Environ. 114, 2610–2623. http://dx.doi.org/10.1016/j.
Vogelmann, J.E., Gallant, A.L., Shi, H., Zhu, Z., 2016. Perspectives on monitoring rse.2010.05.032.
gradual change across the continuity of Landsat sensors using time-series data. Zhu, Z., Fu, Y., Woodcock, C.E., Olofsson, P., Vogelmann, J.E., Holden, C., Wang, M.,
Remote Sens. Environ. 185, 258–270. http://dx.doi.org/10.1016/j. Dai, S., Yu, Y., 2016a. Including land cover change in analysis of greenness
rse.2016.02.060. trends using all available Landsat 5, 7, and 8 images: A case study from
Vogelmann, J.E., Gallant, A.L., Shi, H., Zhu, Z., 2015. Perspectives on monitoring Guangzhou, China (2000–2014). Remote Sens. Environ. 185, 243–257. http://dx.
gradual change across the continuity of Landsat sensors using time-series data. doi.org/10.1016/j.rse.2016.03.036.
Remote Sens. Environ. http://dx.doi.org/10.1016/j.rse.2016.02.060. Zhu, Z., Gallant, A.L., Woodcock, C.E., Pengra, B., Olofsson, P., Loveland, T.R., Jin, S.,
Vogelmann, J.E., Xian, G., Homer, C., Tolk, B., 2012. Monitoring gradual ecosystem Dahal, D., Yang, L., Auch, R.F., 2016b. Optimizing selection of training and
change using Landsat time series analyses: Case studies in selected forest and auxiliary data for operational land cover classification for the LCMAP initiative.
rangeland ecosystems. Remote Sens. Environ. 122, 92–105. http://dx.doi.org/ ISPRS J. Photogramm. Remote Sens. 122, 206–221. http://dx.doi.org/10.1016/j.
10.1016/j.rse.2011.06.027. isprsjprs.2016.11.004.
Vorovencii, I., 2014. A change vector analysis technique for monitoring land cover Zhu, Z., Wang, S., Woodcock, C.E., 2015a. Improvement and expansion of the Fmask
changes in Copsa Mica, Romania, in the period 1985–2011. Environ. Monit. algorithm: Cloud, cloud shadow, and snow detection for Landsats 4–7, 8, and
Assess. 186, 5951–5968. http://dx.doi.org/10.1007/s10661-014-3831-5. Sentinel 2 images. Remote Sens. Environ. 159, 269–277. http://dx.doi.org/
Wang, B., Ono, A., Muramatsu, K., Fujiwara, N., 1999. Automated detection and 10.1016/j.rse.2014.12.014.
removal of clouds and their shadows from landsat TM images. Ieice Trans. Inf. Zhu, Z., Woodcock, C.E., 2014a. Continuous change detection and classification of
Syst. E82D, 453–460. land cover using all available Landsat data. Remote Sens. Environ. 144, 152–
White, J.C., Wulder, M.A., Hobart, G.W., Luther, J.E., Hermosilla, T., Griffiths, P., 171. http://dx.doi.org/10.1016/j.rse.2014.01.011.
Coops, N.C., Hall, R.J., Hostert, P., Dyk, A., Guindon, L., 2014. Pixel-based image Zhu, Z., Woodcock, C.E., 2014b. Automated cloud, cloud shadow, and snow
compositing for large-area dense time series applications and science. Can. J. detection in multitemporal Landsat data: An algorithm designed specifically
Remote Sens. 40, 192–212. http://dx.doi.org/10.1080/07038992.2014.945827. for monitoring land cover change. Remote Sens. Environ. 152, 217–234. http://
Wilson, E.H., Sader, S.A., 2002. Detection of forest harvest type using multiple dates dx.doi.org/10.1016/j.rse.2014.06.012.
of Landsat TM imagery. Remote Sens. Environ. 80, 385–396. http://dx.doi.org/ Zhu, Z., Woodcock, C.E., 2012. Object-based cloud and cloud shadow detection in
10.1016/S0034-4257(01)00318-2. Landsat imagery. Remote Sens. Environ. 118, 83–94. http://dx.doi.org/10.1016/
Woodcock, C., E.A., 2008. 1018 Free Access to Landsat Imagery Teach by the Book j.rse.2011.10.028.
Science Education : Science (80-.). 320, 1011–1012. http://dx.doi.org/10. Zhu, Z., Woodcock, C.E., Holden, C., Yang, Z., 2015b. Generating synthetic Landsat
1126/science.320.5879.1011a. images based on all available Landsat data: Predicting Landsat surface
Wulder, M.A., Masek, J.G., Cohen, W.B., Loveland, T.R., Woodcock, C.E., 2012. reflectance at any given time. Remote Sens. Environ. 162, 67–83. http://dx.
Opening the archive: How free data has enabled the science and monitoring doi.org/10.1016/j.rse.2015.02.009.
promise of Landsat. Remote Sens. Environ. 122, 2–10. http://dx.doi.org/10.1016/ Zhu, Z., Woodcock, C.E., Olofsson, P., 2012. Continuous monitoring of forest
j.rse.2012.01.010. disturbance using all available Landsat imagery. Remote Sens. Environ. 122,
Wulder, M.A., White, J.C., Loveland, T.R., Woodcock, C.E., Belward, A.S., Cohen, W.B., 75–91. http://dx.doi.org/10.1016/j.rse.2011.10.030.
Fosnight, E.A., Shaw, J., Masek, J.G., Roy, D.P., 2016. The global Landsat archive: