Content uploaded by Chunxue Xu
Author content
All content in this area was uploaded by Chunxue Xu on Apr 29, 2021
Content may be subject to copyright.
Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=tcag20
Cartography and Geographic Information Science
ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/tcag20
Deep fake geography? When geospatial data
encounter Artificial Intelligence
Bo Zhao, Shaozeng Zhang, Chunxue Xu, Yifan Sun & Chengbin Deng
To cite this article: Bo Zhao, Shaozeng Zhang, Chunxue Xu, Yifan Sun & Chengbin Deng (2021):
Deep fake geography? When geospatial data encounter Artificial Intelligence, Cartography and
Geographic Information Science, DOI: 10.1080/15230406.2021.1910075
To link to this article: https://doi.org/10.1080/15230406.2021.1910075
Published online: 21 Apr 2021.
Submit your article to this journal
View related articles
View Crossmark data
Deep fake geography? When geospatial data encounter Articial Intelligence
Bo Zhao
a
, Shaozeng Zhang
b
, Chunxue Xu
c
, Yifan Sun
a
and Chengbin Deng
d
a
Department of Geography, University of Washington, Seattle, USA;
b
Department of Anthropology, Oregon State University, Corvallis, USA;
c
College of Earth, Ocean, and Atmospheric Sciences, Oregon State University, Corvallis, USA;
d
Department of Geography, Binghamton
University, State University of New York, USA
ABSTRACT
The developing convergence of Articial Intelligence and GIScience has raised a concern on the
emergence of deep fake geography and its potentials in transforming human perception of the
geographic world. Situating fake geography under the context of modern cartography and
GIScience, this paper presents an empirical study to dissect the algorithmic mechanism of falsifying
satellite images with non-existent landscape features. To demonstrate our pioneering attempt at
deep fake detection, a robust approach is then proposed and evaluated. Our proactive study warns
of the emergence and proliferation of deep fakes in geography just as “lies” in maps. We suggest
timely detections of deep fakes in geospatial data and proper coping strategies when necessary.
More importantly, it is encouraged to cultivate a critical geospatial data literacy and thus to
understand the multi-faceted impacts of deep fake geography on individuals and human society.
ARTICLE HISTORY
Received 19 September 2020
Accepted 25 March 2021
KEYWORDS
Artificial Intelligence;
geospatial data; deep fake;
fake geography; GeoAI; fake
satellite image; Generative
Adversarial Networks
1. Introduction
Geospatial Artificial Intelligence (GeoAI), for its potential
to provide groundbreaking capabilities to leverage
GIScience with a series of Artificial Intelligence (AI)
advances, such as natural language process, unstructured
data classification, computer vision, or map style transfer
(Hu et al., 2019; Kamel Boulos et al., 2019; Sirosh, 2018),
has been hailed by both industry pundits (e.g. the seam-
less integration of deep learning functions in ArcGIS Pro,
GeoAI solution launched on Microsoft Azure) and scho-
lars alike (e.g. a series of GeoAI sessions in AAG annual
conferences 2018, 2019, and 2020, Critical GeoAI session
in AAG annual conference 2021, GeoAI workshops at
ACM SIGSPATIAL 2017, 2018, and 2019, special issues
on GeoAI sponsored by International Journal of
Geographical Information Science and International
Journal of Geo-Information). Such a wide applause of
GeoAI is not the first time when GIS practitioners paid
close attention to the use of AI in improving our capacity
of understanding various geographical phenomena
though; similar efforts can be traced back to mid-1980s
(Couclelis, 1986; Estes et al., 1986; Nystuen, 1984; Smith,
1984). AI was a major driving force to form the
subfields like automated geography (Dobson, 1983) and
GeoComputation (Openshaw & Abrahart, 1996), which
later became significant components of today’s prosper-
ous GIScience. This early wave of AI in GIScience
was well documented in Openshaw’s book “Artificial
Intelligence in Geography” (Openshaw & Openshaw,
1997).
Besides the above-mentioned technical merits brought
by AI, scholars have also witnessed problematic and
unexpected implications of the convergence of AI and
GIScience, such as fabricated GPS signals (Tippenhauer
et al., 2011), fake locational information on social media
(Zhao & Sui, 2017), simulated trajectories of online game
bots (Pao et al., 2010), and fake photos of geographical
environments (Isola et al., 2017). Even so, deep fake, as
a problematic use of AI, has not widely proliferated in
GIScience yet. Deep fake is often referred to as the decep-
tive and/or misleading synthetic media (e.g. image, audio,
or video) that are created by AI. The deep fakes of
politician speech and celebrity pornography spreading
on social media that have received wide public attention
in recent years. It has been regarded as a serious threat to
individual privacy and national security (Chesney &
Citron, 2019; Sayler & Harris, 2019), and has thus
incurred responses from both industry and government
to restrain its use. Big tech companies, including
Amazon, Facebook and Microsoft, have jointly launched
a deep fake detection challenge (DFDC, 2019), and
Microsoft published a tool to identify artificially manipu-
lated media (Burt, 2020). The proliferating misuse of AI
also brought up serious concerns with the appearance of
deep fakes in geography. For example, the automation
lead at the National Geospatial-Intelligence Agency
CONTACT Bo Zhao zhaobo@uw.edu
CARTOGRAPHY AND GEOGRAPHIC INFORMATION SCIENCE
https://doi.org/10.1080/15230406.2021.1910075
© 2021 Cartography and Geographic Information Society
(NGA, known as the National Imagery and Mapping
Agency from 1996–2003), a combat support agency
under the United States Department of Defense and
a member of the United States Intelligence, openly
unveiled that AI was used to manipulate scenes and pixels
to create artifacts on satellite images for malicious pur-
poses (Tucker, 2019). Due to the often-sensitive nature of
deep fake satellite imagery in similar settings as such, we
could not get convenient and safe access to existing deep
fake satellite images for this study and publication. Even
though, we cannot ignore the appearance, or underesti-
mate the development, of deep fake in satellite images or
other types of geospatial data.
While many GIS practitioners have been celebrating
the technical merits of deep learning and other types of
AI for geographical problem solving, few have publicly
recognized or criticized the potential threats of deep
fake to the field of geography or beyond. Therefore, we
would like to take the lead to explore the potential
influences of deep fake on geospatial data and
GIScience. Indeed, the emergence of deep fakes in
GIScience is inevitable just as “lies” are essential in
maps. As Monmonier (1991, p. 1) argued, “Not only is
it easy to lie with maps, it’s essential . . . To present
a useful and truthful picture, an accurate map must
tell white lies.” Therefore, expecting the proliferating
deep fakes of geospatial data, it is necessary to develop
proper coping strategies and critically analyze their
complicated social implications. Thus, in the remaining
sections of this paper, we review the development of
fake geography since before AI and introduce the basic
technical details of deep fake in relevance to geography
today. Then, we detail a case study of fake satellite
images of Tacoma, Washington in order to closely
examine the algorithmic mechanism of deep fake tech-
niques in simulating fake satellite images – a primary
type of geospatial data. Next, we introduce a feasible
detecting approach to assessing the authenticity of
a satellite image. We conclude this paper with
a summary of our research findings and a critical dis-
cussion of “fake” from a broader humanistic geography
perspective.
2. Related works
2.1. Fake geography
Although the term “Fake Geography” first appeared to
describe an AI-generated fake digital geographical
environment and warn us of its detrimental effects
(Maclenan, 2018), its theoretical connotation and
potential significance for geography is far more pro-
found and broader. We could trace the origin of fake
geography all the way back to the false or mythological
interpretation of the world that could be illustrated
from some ancient maps such as the Babylonian cunei-
form map in the 5th century B.C. However, in this
paper we situated fake geography in the context of
modern science and technology. In doing so, we rea-
lized the importance of the correspondence theory of
truth as epistemological guidance on the determina-
tion of what is true and its opposites (David, 2016).
This theory is premised on a clear binary relationship
between an object and its measurement: if the mea-
surement is in correspondence with the object, a truth
is thus established and the object can be represented by
the measurement. For those measurements that are not
deemed as “truth,” various terms have been used, such
as “error,” “false,” “outlier” or “anomaly” that are often
used to indicate inconsistent measurements in scienti-
fic research, and also “lie,” “fake,” “misinformation” or
“disinformation” that are commonly used in public
media and political debates to describe a deliberately
generated inconsistent representation.
Monmonier is one of the first geographers whose work
can be enlightening for today’s debates on fake geography.
In his famous book “How to lie with maps,” a variety of
ways in which maps (or geospatial data) distortedly repre-
sent the real world have been systematically explored
(Monmonier, 1991). Early fake geographies would also
include, for example, propaganda maps in wartime that
distortedly illustrated the real battle situations in order to
shake the enemy’s morale (Herb, 2002); fictitious geogra-
phical entries, also called paper towns, phantom settle-
ments, or trap streets, that are labeled on the map to help
unveil copyright infringements (S. Zhang, 2015). It is
worth noting that the term “lie” in the book title cannot
be simply taken as some negative intentions in map mak-
ing. Indeed, cartographic generalization is a type of “white
lie” – any map needs to simplify and thus reduce the
complexity of the real-world phenomenon in order to
enable an efficient and legible visual communication.
Monmonier’s book, republished several times by
now, has influenced generations of cartographers and
GIScientists. It did not foresee, but inspired us to under-
stand more critically and holistically, the emerging “lies” or
fake geographies in today’s data-intensive and networked
environments. For example, GPS signals were spoofed to
mislead superyachts off the course (Shepard et al., 2012),
and selfies in fake scenery spots (e.g. beach, national parks)
were shared on social media to show off “fakations” (a.k.a.
fake vacations) (M. Zhang, 2015). Starting from 2017,
Zhao and his collaborators have conducted a series of
studies on location spoofing and its existence on multiple
digital platforms, such as Twitter (Zhao & Sui, 2017),
Facebook (S. Zhang et al., 2020) and the online mobile
2B. ZHAO ET AL.
game Pokémon GO (Zhao & Zhang, 2019). Location
spoofing, a relatively new geographical phenomenon of
fake geography, refers to a deliberate inconsistency
between the reported geospatial information and the
ground truth. Zhao and Sui (2017) also proposed
a detection approach through combining time geography
principles and the Bayesian statistics, and further explored
Twitter users’ intentions in generating fake geo-tags. Zhao
and Zhang (2019) further explored the spoofing issues in
Pokémon GO and discussed its underlying social implica-
tions. As indicated by this study, although location spoof-
ing was considered as cheating by the game company as
well as some game players, it can be used to overcome the
spatial disparity of game resources (e.g. between black and
white neighborhoods in New York City) and promote
fairness in accessing game resources. Moreover, S. Zhang
et al. (2020) examined a cyber protest on Facebook. During
this protest, the AI-powered recommendation algorithm
referred the posts about the protest to Facebook users who
may be interested in this topic. As a result, a great number
of Facebook users remotely spoofed their location check-
ins to show their support to the local protesters. AI plays an
increasingly significant role in building fake geographies
that are essential to the recent debates on misinformation
and post-truth (Maclenan, 2018; Oscar, 2018; S. Zhang
et al., 2020).
The fast penetration of AI in various areas of today’s
society is driving fake geography to another level, deep
fake geography, which has triggered heated debates on
its controversial capacity and unforeseeable impact on
society. The NGA, as mentioned earlier, has seriously
reminded us of the risk of deep fake satellite images
being used as a terrifying AI-powered weapon
(Tucker, 2019). Considering the increasing number of
fake satellite images emerged during the past two years,
such as satellite images of night light in India during
“Diwali” – a Hindu festival of lights (Kundu, 2019) or of
fake fire in the central park of New York City (Markuse,
2019), it is highly likely in the near future if not yet that
deep fake techniques could be implemented to create
fake satellite images containing uncannily real landscape
features. If so, deep fake can potentially develop into
a new mode of unpredictable and even terrifying fake
geography (Kwok & Koh, 2020; Maclenan, 2018;
Tucker, 2019).
2.2. Deep fake and its detection
To understand such a new mode of fake geography, it is
necessary to comprehend the basic algorithm of deep
fake techniques in making fake geospatial data and thus
to inspire us to explore possible detection approaches.
From an algorithmic perspective, deep fake techniques
primarily rely on Generative Adversarial Networks
(GANs), which is a class of unsupervised deep learning
algorithms that can simulate synthetic media (e.g.
image, video, audio) that appear authentic (Charleer,
2018; Oscar, 2018). The GANs generate two networks –
a “generator” and a “discriminator”; and enable them to
contest with one another through a multiple-epoch
training process. In the training process, the generator
creates a latent space of candidate datasets, and then the
discriminator evaluates whether the candidate datasets
are qualified by satisfying an evolving statistical charac-
teristics criterion. The candidate data from the genera-
tor, after several training epochs of tuning, can reach an
acceptable similarity to the required statistical charac-
teristics (Goodfellow et al., 2014; Salimans et al., 2016).
Similarly, if we use a GAN to simulate geospatial data,
the GAN’s generator will create candidates of geospatial
data and ask the discriminator whether the candidates
meet the characteristics of a typical geospatial data.
Here, the geospatial data can be as simple as a point,
polyline or polygon, or relative complex data like satel-
lite images, or even 3D point clouds. After several
epochs’ training, the candidates could eventually meet
the criteria of qualified geospatial data. At this stage, the
candidates, recognized as seemingly authentic geospa-
tial data, embody a new mode of fake geography.
With a thorough review of the existing deep fake
detection methods, we categorized these methods into
two groups based on the detecting feature selection
process – manually defined or automatically extracted
(Afchar et al., 2018; Galbally & Marcel, 2014; Hsu et al.,
2020; Matern et al., 2019; Zhu et al., 2017). The detec-
tion methods using manually defined features were
developed prior to those using automatically extracted
features. Galbally and Marcel (2014) proposed 14 gen-
eral image quality metrics to distinguish between legit-
imate face images and impostor samples generated by
deep learning algorithms and achieved competitive
results. When it comes to videos, by summarizing visual
artifacts arising from global consistency, illumination
estimation and geometry estimation, Matern et al.
(2019) were able to recognize face manipulations in
videos with acceptable accuracy only using shallow clas-
sifiers such as logistic regression models. For detection
methods using automatically extracted features, Hsu
et al. (2020) proposed a two-step approach for general
fake image detection. The first step extracts the discri-
minative features using the common fake feature
network (CFFN) learning process, and the following
step feeds these salient features into a small convolu-
tional neural network (CNN) concatenated to the last
CARTOGRAPHY AND GEOGRAPHIC INFORMATION SCIENCE 3
convolutional layer of CFFN. This method achieved
a precision at least 0.92 on a variety of image datasets
generated by state-of-the-art GANs, significantly super-
ior to other existing fake image detectors.
3. Data and method
Despite the significant progress in deep fake detection,
specific methods for detecting deep fake satellite images
have not been explored yet. We thus designed an
empirical study to closely examine deep fake techniques
and explore feasible means to detect deep fakes in satel-
lite images. Since no existing GANs-generated satellite
image has been publicized or easily accessible, this
empirical study began with our own experiment of
simulating a baseline dataset of satellite imagery of
Tacoma, Washington. The simulated satellite images
were developed on the basic urban structure on the
CartoDB basemap, but with the landscape features
extracted from two other cities, Seattle, Washington
and Beijing, China. Such GANs-generated satellite
images could be viewed as fake since the displaying
landscapes did not exist in the real world and hence
were used experimentally in this study for testing our
deep fake detection approach. It has never been our
objective to show how to fake satellite images; in fact,
we acknowledged that the satellite image simulation
process not only provided a baseline of simulated or
fake satellite images, but also offered a demonstrative
example of the essential deep fake mechanism. The
baseline dataset enabled us to analyze the characteristics
of fake satellite images, thereby facilitating the process
of proposing an approach to detecting the deep fakes in
satellite images.
3.1. Deep faking satellite image using GANs
Cycle-Consistent Adversarial Networks (CycleGAN), as
a popular model of GANs, is frequently adopted for
generating deep fakes (Zhu et al., 2017). In our study,
we used CycleGAN to translate the basemap of a city to
satellite images into landscape features of other cities. If
the newly simulated images embodied any fake geogra-
phical environment but appeared to be real, we would
consider them as fake images.
Specifically, CycleGAN translates between two differ-
ent domains (i.e. X and Y), where X and Y should share
some underlying relationship. CycleGAN aims to learn
the relationship by developing two mapping functions,
G: X → Y and F: Y → X. Two associated adversarial
discriminators D
Y
and D
X
are developed to facilitate the
mapping between the two domains by encouraging the
G and F functions to generate the output indistinguish-
able from the corresponding domain (i.e. Y and X,
respectively). CycleGAN aims to solve Equation (1)
(Zhu et al., 2017):
G;F¼L G;F;Dx;Dyð Þ ¼ LGAN G;Dy;X;Yð Þ
þLGAN F;Dx;Y;Xð Þ
þβLcyc G;Fð Þ
(1)
where β is a parameter that controls the relative impor-
tance of the two losses: adversarial losses LGAN and cycle
consistency losses Lcyc G;Fð Þ;which are defined as
follows:
LGAN G;Dy;X;Yð Þ ¼ Ey,pdata yð ÞðlogDYyð Þ½
þEx,pdata xð Þ log 1DYG xð Þð Þð Þ½ (2)
Lcyc G;Fð Þ ¼ Ex,pdata xð Þ kF G xð Þð Þ xk1
½
þEy,pdata yð Þ kG F yð Þð Þ yk1
½ (3)
In Equation (2), the first item is the expectation (E) of
discriminator (D) output given y where y is sampled
from a data distribution p_data(y). For the second item,
the input to the discriminator is the output from the
generator G, after being feeded in data from X domain.
In Equation (3), the loss of CycleGAN is to minimize
the forward cycle consistency (the first item) and back-
ward cycle consistency (the second item). The cycle
consistency is used to depict the process that the original
image (x or y) should be reconstructed (y or x) after
image transformation (mapping function G, F) using
CycleGAN.
To demonstrate the deep faking process, we con-
ducted an experiment to simulate satellite images of
City A to embody the landscape features of City
B using CycleGAN (Figure 1). Two types of web map
tile datasets from Google Earth’s satellite imagery and
CartoDB positron basemap were collected as the model
input. The high-resolution imagery in Google Earth
provided fine details of the spatial pattern at different
zoom levels and has been widely used as the ground
truth reference for image interpretation. The mono-
chrome CartoDB positron basemap presented basic
urban structural information of the geospatial context
without any geoname label. Collected through the QTile
plugin in QGIS, both datasets are in 256*256-pixel tiles
at the zoom level 16, which is equivalent to the scale of
1:8,000.
In our empirical study, we collected the landscape
features from two big cities – Seattle and Beijing (See
Figure 2). Seattle is located between the saltwater Puget
Sound to the west and Lake Washington to the east, on
4B. ZHAO ET AL.
the northwest coast of the United States. It is featured as
one of the largest employment bases in the country but
being ranked as low in terms of the population density
compared to other big cities in the United States. Beijing
is located in North China, and is the world’s most
populous capital city, combining both modern and tra-
ditional architectures. Beijing presents rapid urban
growth in the past few decades, and it witnessed the
most intense conversion of the land through the urban
sprawl and renewal process. These two cities present
different spatial arrangements and configurations,
which can be identified by the spectral and spatial fea-
tures at the street scale on the high-resolution satellite
imagery. Next, the dataset of the basic urban structural
information was collected from Tacoma, Washington.
Tacoma (see Figure 2) is a mid-sized port city located in
Pierce County, Washington, and is adjacent to the
southwest of Seattle. Geographically, the spatial features
of Tacoma are similar to Seattle. At the zoom level 16,
1196 and 1122 pairs of satellite imagery and map tiles
were collected for Seattle and Beijing, respectively, and
758 image tiles were covering for Tacoma.
Figure 3 shows the training process with five different
training epochs from the beginning to end. At the early
epoch (i.e. epoch = 1 in the figure), the CycleGAN
generally caught the structure of the road network, but
it falsely generated a large area of shadow for the road
network. The land parcels were made in gray color
which rarely existed in the real world. Besides,
a noticeable haloing feature could be clearly found on
the top edge of the image. This haloing feature did not
exist in the real world. It is yet unknown why CycleGAN
created such a non-geographical object. One possible
reason is that, compared to the base map, the satellite
image had relatively more geospatial texture; thus, it was
more challenging for CycleGAN to capture the under-
lying relationship when mapping from the base map to
the satellite image. As the epoch increased from 50 to
100, more geographical features appeared, such as green
space, open land, water bodies, and buildings. Due to
the cyan color, it was difficult to differentiate the green
spaces from water bodies. The area of shadows for the
road network disappeared. However, some urban areas
such as local communities seemed clustered and thereby
hard to recognize its land use type and internal struc-
ture. For the last two epochs (150 and 200), an authen-
tically visual feeling of the simulated satellite images
increased to a great extent: more geographical details
were found; and the colors for open land and water
bodies became differentiated and natural. Overall, with
the 200-epoch training, CycleGAN successfully gener-
ated the mapping from the base map to the satellite
images. This pilot experiment demonstrates the capabil-
ity of GANs in generating satellite images with non-
existent geographical features. The simulated satellite
imagery is not perfect yet, but it looks authentic and
uncovers the potential of the algorithms in falsifying
satellite images that cannot be easily identified by
humans.
3.2. Detecting fake satellite images
With our own eyes, it was nearly impossible to tell
whether the simulated satellite image (see the epoch
200 in Figure 3) was authentic or fake. To investigate
whether and to what extent these fake images can be
Figure 1. Dataflow of CycleGAN model in this study. The CycleGAN model is trained using the Basemap and satellite imagery dataset
of City B to build the mapping relationships between two data domains. In this model, G function maps satellite imagery to basemap,
while F function maps basemap to satellite imagery. Then the CycleGAN model is applied to the basemap of City A to generate the
simulated satellite imagery. This CycleGAN illustration is modified from Zhu et al. (2017).
CARTOGRAPHY AND GEOGRAPHIC INFORMATION SCIENCE 5
identified, we have learned some clues from previous
studies. Scholars found that GAN-generated fake
images were different from authentic ones in multiple
visual features such as color, texture and details, and in
frequency domain features such as a certain type of
periodic replications (Galbally & Marcel, 2014; Wang
et al., 2020; X. Zhang et al., 2019). Thus, to incorporate
these early findings into our detection approach, we
examined the GAN-generated satellite images using
a suite of graphical features of color histogram, spatial
domains, and frequency domains as shown in Table 1.
To explore the spatial domain, we chose Image
Colorfulness Index (CFI) to describe the difference in
the perception of colorfulness between the authentic
image and the fake ones (Haralick et al., 1973). Besides,
an image with higher definition appears with sharper
edges and has larger gradient value, so the image clarity
can be characterized by summing up the image gradient
values. Three non-reference image quality evaluation fea-
tures using Brenne, Tenengrad and Laplacian gradients
were employed in this study to measure the clarity of
satellite images. The clearer the image, the greater the
value of these features (Mittal et al., 2012).
Among a variety of image texture description meth-
ods, gray level concurrence matrix (GLCM) is simple
and effective. Haralick et al. (1973) defined 14 GLCM
quadratic statistics in image classification research.
Ulaby et al. (1986) further pointed out that among the
aforementioned quadratic statistics, four independent
features – angular second moment (ASM), contrast
(CON), entropy (ENT), and inverse different moment
(IDM) – are effective in recognizing land use variations.
These four features were used in this study to describe
the texture of satellite images. In addition to the texture
of the spatial domain, replications in the frequency
domain can be considered as a texture characteristic as
Figure 2. The study areas – Tacoma, Seattle and Beijing and their local landscape characteristics.
6B. ZHAO ET AL.
well. Therefore, the aforementioned four quadratic sta-
tistics were recalculated based on GLCM of frequency
spectra to describe the frequency characteristics.
Furthermore, a number of features were calculated to
quantify the difference between the histogram of the
authentic images and that of the fake ones. First, we used
the mean (MEAN), standard deviation (STD), skewness
(SKEW), kurtosis (KURT), and entropy (GET) to describe
the grayscale histogram. Then, color moments (CM) were
employed to indicate the characteristics of single channel
histograms. Since the color distribution information is
mainly concentrated in low-order moments, only the first
(CM1), second (CM2), and third moments of color (CM3)
are sufficient to express the color distribution of an image
(Stricker & Orengo, 1995).
The above-mentioned 26 features can assist us to
develop specific strategies of fake satellite image detection.
In practice, an independent t-test is conducted on these 26
features in order to identify salient features that have sig-
nificant mean value difference between all authentic and
fake satellite images and thus to use these identified fea-
tures for fake satellite image detection (Galbally & Marcel,
2014; Matern et al., 2019). Moreover, considering prior
studies in satellite data classification (Maghsoudi et al.,
Figure 3. Training process illustrated by five image pairs of
different epochs.
Table 1. Features of authentic and fake satellite images.
Code Feature description
Spatial
CFI Image Colorfulness Index: A larger value indicates a more
colorful image
BIQ Brenne Image Quality Index: A larger value indicates a clearer
image
TIQ Tenengrad Image Quality Index: A larger value indicates
a clearer image
LIQ Laplacian Image Quality Index: A larger value indicates
a clearer image
ASM Angular Second Moment of GLCM: A larger value indicates
a more uniform and regularly changing texture pattern
CON Contrast of GLCM: The greater the CON, the deeper the
grooves of the texture, and the clearer the visual effect
ENT Entropy of GLCM: The more complex and uneven the texture
in the image, the greater the ENT value
IDM Inverse Different Moment of GLCM: The larger the IDM, the
smaller the change between areas of the image texture, or
the local pattern is more uniform
Frequency
FASM ASM at Frequency Domain: Similar to ASM
FCON CON at Frequency Domain: Similar to CON
FENT ENT at Frequency Domain: Similar to ENT
FIDM IDM at Frequency Domain: Similar to IDM
Histogram
MEAN Mean of GLH, The larger the value the brighter the image
STD Standard Deviation of GLH, the larger the value less
concentrated the GLH
SKEW Skewness of GLH, the larger the value more skewed the GLH
KURT Kurtosis of GLH, the larger the value the steeper the GLH
GET Entropy of GLH, the larger the value, the less even the GLH
CM1_R/
G/B
First Order Color Moment of Red (Green/Blue): mean of color
histogram
CM2_R/
G/B
Second Order Color Moment of Red (Green/Blue): variance of
color histogram
CM3_R/
G/B
Third Order Color Moment of Red (Green/Blue): skewness of
color histogram
GLCM means gray level concurrence matrix; GLH refers to gray level
histogram.
CARTOGRAPHY AND GEOGRAPHIC INFORMATION SCIENCE 7
2013) and deep fake detection (Matern et al., 2019), we
proposed a fake satellite image detection approach by
feeding these salient features to a Support Vector
Machine (SVM). By constructing a hyperplane that has
the largest distance to the nearest sample of any class in
a high-dimensional space, a SVM can efficiently perform
data classification functions and then differentiate fake
satellite images from the authentic ones. We further
employed indicators like precision, recall and F1 score to
evaluate the performance of our approach. In the context
of fake satellite image detection, precision is the ratio of
successfully detected fake satellite images to the total num-
ber of satellite images that our approach considers fake.
This indicator reflects the credibility of our approach in
judging a satellite image as fake or not; Recall is the ratio of
successfully detected fake satellite images to all fake satellite
images. This indicator measures the capability of our
approach to detect fake satellite images. F1 score is the
weighted average of precision and recall. This indicator
evaluates the overall performance of our approach in
detecting fake satellite images.
4. Results
4.1. Satellite images with non-existent landscape
features
After applying the deep fake approach proposed in sec-
tion 3.1, we simulated/faked satellite images of Tacoma
that embodied landscape features similar to those of
Beijing and Seattle (see Figure 4). Since these landscape
features do not exist in Tacoma, we considered the simu-
lated satellite images of Tacoma as fake. In general, the
road network, green space, and buildings are captured by
fake satellite images in the visual pattern of either Seattle
or Beijing, whereas the geospatial details differ from one
another. The fake satellite image in the visual pattern of
Seattle presents similar landscape features such as low-
rise buildings, whereas the one in the visual pattern of
Beijing presents different landscapes such as high-rise
compact buildings with large shadow areas.
Moreover, the CycleGAN models show stability in
generating fake satellite images in a large geographical
region. Figure 5(b,c) shows two fake satellite images
covering a neighborhood in Tacoma made up of four
fake mosaic tiles in the visual patterns of Beijing and
Seattle, respectively. The fake satellite image in the
visual pattern of Beijing contains more landscape details
compared to those in the visual pattern of Seattle, espe-
cially in open areas where there is lack of geospatial
information on the CartoDB basemap. For example,
buildings were generated at the left-bottom corner on
Figure 5(b) but did not appear on Figure 5(c).
According to our results, CycleGAN performs excel-
lently in recognizing and simulating green space, which
may be due to its simple color feature. Unsurprisingly,
the number of details differ between the two visual
patterns. Overall, the above experiment demonstrates
that satellite imagery can be faked by CycleGAN.
4.2. Deep fake detection
With the successfully simulated fake satellite imagery,
this section presents our pioneering attempt in detecting
the deep fakes in satellite imagery. Specifically, we con-
structed a deep fake detection dataset containing 8064
satellite images in the size of 256*256 pixels. The dataset
includes authentic satellite images of Tacoma (2016
pieces), Seattle (1008 pieces), and Beijing (1008 pieces),
as well as fake satellite images of Tacoma in the visual
pattern of Seattle (2016 pieces) and in the visual pattern
of Beijing (2016 pieces).
According to the result of the independent two-
sample t-test (Table 2), 21 out of 25 features have
a significantly different mean value, and they are further
considered salient features. These salient features indi-
cate a significant difference between authentic and fake
satellite images. To be more specific, from the view of
the spatial domain, a significantly smaller image color-
fulness index (CFI) of fake satellite images shows
a relatively less colorful visual perception in comparison
to authentic ones. When it comes to the clarity of
satellite images, a significantly larger Brenne Image
Quality Index (BIQ) of fake satellite images shows
more sharp edges on them. Besides, fake satellite images
have a significantly smaller angular second moment
(ASM) and inverse different moment (IDM) of gray
level concurrence matrix (GLCM), and a significantly
larger entropy of GLCM (ENT). It indicates that fake
satellite images have more complex and uneven texture
in comparison to authentic satellite images.
Regarding the histogram features, fake satellite
images have a significantly smaller mean value of gray
level histogram (MEAN) and first-order color moment
of both red, green and blue (CM1_R/G/B). It means fake
satellite images have a dimmer visual appearance, and
we can observe a gray level or color histogram closer to
the left. Besides, the second-order color moment of the
fake satellite image is significantly smaller on the red
channel (CM2_R) and larger on the green channel
(CM2_G). Since the skewness, kurtosis, and entropy of
gray level histogram (SKEW, KURT, GET) of fake satel-
lite images is significantly larger than authentic ones, the
gray level histogram of fake satellite images appears
more skewed, steeper and less even. A significant larger
third color moment in all channels (CM3_R/G/B) of
8B. ZHAO ET AL.
fake satellite images indicates that color histograms of
fake ones tend to be more skewed.
From a frequency domain perspective, fake satellite
images have a significantly smaller FASM and FCON,
and a significantly larger FNET and FIDM in compar-
ison to authentic ones. It indicates the texture on fre-
quency domain of authentic satellite images is more
uniform and has deeper grooves, while the frequency
domain of the fake satellite images a more complex but
locally consistent texture.
When comparing specific authentic and fake satellite
images, we can visually observe differences in the salient
features. For example, the three authentic satellite images
in Figure 6 are brighter and more colorful than the two fake
ones, and these fake ones have more sharp edges such as the
edges of the road. Moreover, the authentic satellite image of
Figure 4. Fake satellite images of a neighborhood in Tacoma with landscape features of other cities. (a) The original CartoDB basemap
tile; (b) the corresponding satellite image tile. The fake satellite image in the visual patterns of (c) Seattle and (d) Beijing.
CARTOGRAPHY AND GEOGRAPHIC INFORMATION SCIENCE 9
Seattle contains clear light-colored roofs, whereas those
light-colored roofs on the fake image in the visual pattern
of Seattle contain some mottled textures. This textural dif-
ference can be used to identify the fakes. Besides, similar to
the empirical results found by X. Zhang et al. (2019) and
Wang et al. (2020), periodic replications of spectra can be
observed in fake satellite images in either the visual pattern of
Seattle or that of Beijing.
Taking one or more types (spatial, histogram, and fre-
quency) of salient features as input, SVMs are trained and
adjusted. The performance of different models in the test
dataset is shown in Table 3. All models achieved a recall
more than 0.94, which indicates more than 94% fake
satellite images can be detected. When only taking fre-
quency salient features into consideration, the precision
had a relative lower value 0.7283 in comparison to other
models. In other words, when the model labels a satellite
image as fake, it is only 73% reliable. Taking its highest
recall 0.9697 into consideration, its relative lower precision
may be due to that periodic replications of spectra may also
be observed in some authentic satellite images. The model
fed with only salient features of spatial domain obtains an
F1 score of 0.9399, it indicates that we can distinguish the
fake satellite images by taking a closer look at their color,
edge clarity, and texture characteristics. When all spatial,
histogram, and frequency salient features are taken into
Figure 5. (a) A CartoDB basemap covering a neighborhood area
in Tacoma; a fake satellite image with the transferred visual
pattern of (b) Beijing; and (c) Seattle.
Table 2. The independent t-test between features of authentic
and fake satellite images.
Feature Mean Fake Mean Authentic Diff.
Spatial
CFI 12.1343 20.6602 −8.5259***
BIQ 496.2755 462.3757 33.8998**
TIQ 19284.5431 19472.7016 −188.1585
LIQ 4897.4112 4937.9481 −40.5369
ASM 0.0089 0.0258 −0.0169***
CON 373.3451 364.6969 8.6482
ENT 7.7601 7.5506 0.2096**
IDM 0.1767 0.2764 −0.0997***
Histogram
MEAN 79.0960 91.1394 −12.0434***
STD 28.8618 28.5261 0.3357
SKEW 0.9246 0.7465 0.1781***
KURT 5.9737 3.5810 2.3927***
GET 6.2897 6.1589 0.1308**
CM1_R 80.3252 89.2334 −8.9084***
CM2_R 30.4465 31.710 −1.2675***
CM3_R 25.1773 13.3325 11.8447***
CM1_G 80.7005 92.9016 −12.2011***
CM2_G 28.9633 28.5668 0.3965**
CM3_G 26.4078 15.7421 10.6658***
CM1_B 75.4713 88.3756 −12.9042***
CM2_B 29.1654 29.6676 −0.5022
CM3_B 28.6201 23.1306 5.4895***
Frequency
FASM 0.0007 0.0047 −0.0041***
FCON 245.7877 251.9979 −6.2102***
FENT 8.2935 8.2288 0.0648***
FIDM 0.0956 0.0903 0.0054***
Mean Fake (Authentic) refers to the mean value of different features of all
fake (authentic) samples, Diff. indicates difference between the mean
value of features for all authentic samples and all fake samples; *,
p < 0.05; **, p < 0.01; ***, p < 0.001.
10 B. ZHAO ET AL.
Figure 6. The comparison between authentic and fake satellite image records with respect to their spatial domains, color histograms,
and frequency domains. Three authentic satellite image records showing (a) an area in Tacoma, (b) an area in Seattle, and (c) another
area in Beijing, respectively. Two fake satellite image records of Tacoma in (d) a transferred visual pattern of Seattle and in (e) another
transferred visual pattern of Beijing, respectively.
CARTOGRAPHY AND GEOGRAPHIC INFORMATION SCIENCE 11
consideration, the F1 score will rise to a competitive level
of 0.9530.
Therefore, the results indicate that the proposed
approach can effectively detect CycleGAN-generated
fake satellite images. To enhance the current approach,
it is necessary to include a few other cities rather than
just Tacoma to our empirical study as a means to repre-
sent a variety of landscape structures. Moreover, if this
approach is applied to fake satellite images that are
generated by other GAN models (e.g. pix2pix,
styleGAN), its performance may decline. Further, the
current approach can only provide a binary result – an
image is either authentic or totally fake, it is still
a challenge to detect whether an image contains both
authentic and fake landscape features, or even to deline-
ate which landscape in an image is fake or not. To
address the above-listed issues, the proposed approach
can be further improved by establishing a more com-
prehensive database of satellite images that represents
different types of existent and non-existent landscape
features, and thereby training a baseline dataset that
incorporates a variety of authentic and fake satellite
image scenarios. In addition to an enlarged dataset, we
should also optimize the algorithmic mechanism to
detect fake images generated by other deep fake techni-
ques, such as FaceForensics++ (Rössler et al., 2019).
Although further enhancement can be effectively made
through the above-listed directions by us or other scho-
lars who are also interested in this timely topic, our
ultimate goal of this paper is not to develop a universal
approach that can effectively detect all kinds of fake
satellite images. A universal approach is perhaps
a mission impossible anyways due to the complexity of
the landscape features in the real world and the diversity
of deep fake approaches.
5. Concluding remarks
In this paper, we took a proactive and critical approach to the
well-recognized capabilities of GeoAI and attempted to raise
public awareness of how this technology may transform our
perceptions about the geographic world. We used GANs –
a deep learning technique to transfer satellite images of
Tacoma into new or fake ones in order to demonstrate
whether and how GeoAI can falsify satellite images.
Further, based on a close examination of a series of satellite
image characteristics, we proposed and applied a feasible
approach to detecting fake satellite images. The approach
achieved a F1 score more than 0.95 when taking all salient
features into consideration in an independent test set. The
methods and results in our study could be very useful as they
enable us to better understand the deep fakes in geospatial
data expecting further development and potential impacts of
fake geography. However, few existent fake satellite images
do not mean the topic is unnecessary to investigate. Instead,
it is one of the key missions for scholars to envision pro-
spective developments and suggest proper coping strategies.
This study is meant to encourage proper precautions
toward the upcoming development of deep fake in geogra-
phy. If we continue being unaware of and unprepared for
deep fake, we run the risk of entering a “fake geography”
dystopia (Maclenan, 2018). With this warning, the detection
approach proposed in this paper would be crucial to discover
fake satellite images, and to inspire our fellow cartographers
and GIS practitioners to develop approaches to coping with
other types of fake geospatial data. That said, we also need to
remind ourselves of another extreme situation – when the
existence of a few deep fake cases may force us to verify the
trustworthiness of every piece of geospatial data. This could
be so far beyond inconvenient that we would feel helpless
and anxious, especially in this data-intensive era– when
geospatial data has become such a fundamental resource
for everyday life such as in real-time traffic provided by
Google Maps, Autopilot offered by Tesla vehicles, and loca-
tion-based restaurant recommendations provided by Yelp.
Therefore, we humans may easily fall into a dilemma, in
which we are uncertain of implementing all-weather detec-
tion toward the rare cases of deep fakes in geography or
simply taking no interventions. The former makes us
anxious while the latter put us in danger. Instead of provid-
ing a simple solution to this dilemma, we suggest recalling
the humanistic geography perspective toward fake geogra-
phy introduced earlier with Monmonier. The emergence of
deep fakes in GIScience and human society at large is
inevitable just as “lies” are essential in maps. We ought to
admit that fake, for good or ill, is an inevitable component of
human civilization. In some cases, lyings, deceptions, or
spoofings can smooth our social life. An apt example is the
story “The Emperor’s New Clothes,” in which two dress-
makers, chancellors, the public, and even the king himself
would rather believe, for various reasons, that the non-
existent clothes are the “most beautiful” in the world, and
eventually, only the “uncivilized” child is willing to expose
that lie. This may imply that the cost of civilization is the
integration of lying into human societies no matter whether
we humans like or dislike fakes. In this sense, the existence of
Table 3. Performance of different fake satellite images detection
models.
Model F1 score Precision Recall
Spatial 0.9399 0.9316 0.9483
Histogram 0.8795 0.8196 0.9484
Frequency 0.8324 0.7283 0.9697
Spatial + Histogram 0.9484 0.9472 0.9516
Spatial + Frequency 0.9387 0.9308 0.9468
Histogram + Frequency 0.8879 0.8347 0.9481
Spatial + Histogram + Frequency 0.9530 0.9482 0.9579
12 B. ZHAO ET AL.
fakes seems inevitable or even natural; Deep fakes in geo-
graphy as well as other deceptive phenomena in this world
may also facilitate our social life and become an integrated
part of human civilization. Indeed, deep fakes of satellite
imagery can be misleading or even threatening to national
security, but can also be very useful, such as in predicting
land-use change scenarios, reconstructing and preserving
historical scenes, and automated making of reference and
topographic maps (Kang et al., 2019). From this broader
human geography perspective, what is urgent for human
society is to properly utilize the underlying GeoAI technique
(e.g. GANs), and to understand more comprehensively and
critically its fast emergence and powerful impacts. We
indeed need timely detection of deep fakes in geospatial
data and proper coping strategies when necessary, for exam-
ple, when satellite images of fake fires in the city center to
trigger panic (Markuse, 2019), or 3d point cloud of fake road
obstacles being transmitted to an autonomous driving vehi-
cle to mislead its navigation system (Cao et al., 2019).
Overall, in today’s increasingly data-intensive envir-
onment, neither a simple optimistic nor a pessimistic
attitude would be the most helpful. While recognizing
the tremendous opportunities expected in the latest AI
advances, GIS practitioners should also be aware of
possible falsification of geospatial data and get prepared
for that by developing detection approaches for identi-
fying fake geospatial data and utilizing such an approach
when necessary. Further, it is encouraged to cultivate
a critical geospatial data literacy instead of anxiously
implementing the detection all the time. This critical
data literacy would not only enable us to interpret fake
geospatial data, but also further raise the public aware-
ness toward the complicated and multi-faceted social
implications of geospatial technologies that are being
transformed dramatically by AI advances.
Acknowledgments
We express our gratitude to the editor and anonymous
reviewers for their efforts during the COVID-19 pandemic.
Disclosure statement
No potential conflict of interest was reported by the authors.
ORCID
Bo Zhao http://orcid.org/0000-0002-7834-6672
Shaozeng Zhang http://orcid.org/0000-0002-7871-8384
Chunxue Xu http://orcid.org/0000-0003-4561-9491
Yifan Sun http://orcid.org/0000-0001-7969-1147
Chengbin Deng http://orcid.org/0000-0002-5459-5586
Data availability statement
The data and codes that support the findings of this study are
available in “figshare.com” with the identifier at https://fig
share.com/s/eeedcd150e759ef4353c.
References
Afchar, D., Nozick, V., Yamagishi, J., & Echizen, I. (2018).
MesoNet: A compact facial video forgery detection
network. In 2018 IEEE International Workshop on
Information Forensics and Security (WIFS) (pp. 1–7).
IEEE. https://doi.org/10.1109/WIFS.2018.8630761
Burt, T. (2020). New steps to combat disinformation. Microsoft on
the Issues. https://blogs.microsoft.com/on-the-issues/2020/09/
01/disinformation-deepfakes-newsguard-video-authenticator/
Cao, Y., Zhou, Y., Chen, Q. A., Xiao, C., Park, W., Fu, K.,
Cyr, B., Rampazzi, S., & Morley Mao, Z. (2019). Adversarial
sensor attack on LiDAR-based perception in autonomous
driving. In L. Cavallaro & J. Kinder (Eds.), Proceedings of
the 2019 ACM SIGSAC conference on computer and com-
munications security (pp. 2267–2281). ACM. https://doi.
org/10.1145/3319535.3339815
Charleer, S. (2018). Family fun with deepfakes: Or how I got my
wife onto the Tonight Show [Sven Charleer, PhD]. http://
svencharleer.com/2018/02/02/family-fun-with-deepfakes-
or-how-i-got-my-wife-onto-the-tonight-show/
Chesney, B., & Citron, D. (2019). Deep fakes: A looming
challenge for privacy, democracy, and national security.
California Law Review, 107(6), 1753–1819. https://doi.org/
10.2139/ssrn.3213954
Couclelis, H. (1986). Artificial Intelligence in geography:
Conjectures on the shape of things to come. Professional
Geographer, 38(1), 1–11. https://doi.org/10.1111/j.0033-
0124.1986.00001.x
David, M. (2016). The correspondence theory of truth. In
E. N. Zalta (Ed.), The Stanford encyclopedia of philosophy
(Fall 2016) (pp. 1–65). Metaphysics Research Lab, Stanford
University. https://plato.stanford.edu/archives/fall2016/
entries/truth-correspondence/
DFDC. (2019). Deepfake detection challenge: Identify videos
with facial or voice manipulations. Kaggle. https://www.
kaggle.com/c/deepfake-detection-challenge
Dobson, J. E. (1983). Automated geography. The Professional
Geographer, 35(2), 135–143. https://doi.org/10.1111/j.0033-
0124.1983.00135.x
Estes, J. E., Sailer, C., & Tinney, L. R. (1986). Applications of
artificial intelligence techniques to remote sensing.
Professional Geographer, 38(2), 133–141. https://doi.org/
10.1111/j.0033-0124.1986.00133.x
Galbally, J., & Marcel, S. (2014). Face anti-spoofing based on
general image quality assessment. In 2014 22nd interna-
tional conference on pattern recognition (pp. 1173–1178).
https://doi.org/10.1109/ICPR.2014.211
Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley,
D., Ozair, S., Courville, A., & Bengio, Y. (2014). Generative
adversarial nets. In Z. Ghahramani, M. Welling, C. Cortes,
N. Lawrence, & K. Weinberger (Eds.), Proceedings of the 27th
international conference on neural information processing sys-
tems - Volume 2 (pp. 2672–2680). MIT Press. https://doi.org/10.
5555/2969033.2969125
CARTOGRAPHY AND GEOGRAPHIC INFORMATION SCIENCE 13
Haralick, R. M., Shanmugam, K., & Dinstein, I. H. (1973).
Textural features for image classification. IEEE
Transactions on Systems, Man, and Cybernetics, SMC-3(6),
610–621. https://doi.org/10.1109/TSMC.1973.4309314
Herb, G. H. (2002). Under the map of Germany: Nationalism and
propaganda 1918–1945 (1st ed.). Routledge. https://doi.org/10.
4324/9780203433720
Hsu, C. C., Zhuang, Y. X., & Lee, C. Y. (2020). Deep fake
image detection based on pairwise learning. Applied
Sciences (Switzerland), 10(1), 370. https://doi.org/10.3390/
app10010370
Hu, Y., Gao, S., Newsam, S. D., & Lunga, D. D. (2019). GeoAI
2018 workshop report the 2nd ACM SIGSPATIAL inter-
national workshop on GeoAI: AI for geographic knowledge
discovery Seattle, WA, USA - November 6, 2018, In
A. Züfle (Ed.), SIGSPATIAL special (Vol. 10(3), pp. 2–29).
ACM. https://doi.org/10.1145/3307599.3307609
Isola, P., Zhu, J.-Y., Zhou, T., & Efros, A. A. (2017). Image-to-
image translation with conditional adversarial networks. In
R. Chellappa, Z. Zhang, & A. Hoogs (Eds.), 2017 IEEE
Conference on Computer Vision and Pattern Recognition
(CVPR) (pp. 5967–5976). IEEE. https://doi.org/10.1109/
CVPR.2017.632
Kamel Boulos, M. N., Peng, G., & Vopham, T. (2019). An
overview of GeoAI applications in health and healthcare.
International Journal of Health Geographics, 18(1), 1–9.
https://doi.org/10.1186/s12942-019-0171-2
Kang, Y., Gao, S., & Roth, R. E. (2019). Transferring
multiscale map styles using Generative Adversarial
Networks. International Journal of Cartography, 5
(2–3), 115–141. https://doi.org/10.1080/23729333.2019.
1615729
Kundu, C. (2019). Fact check: All that glitters is not Diwali. Indian
Today. https://www.indiatoday.in/fact-check/story/fact-check
-all-that-glitters-is-not-diwali-1613464-2019-10-28
Kwok, A. O. J., & Koh, S. G. M. (2020). Deepfake: A social
construction of technology perspective. Current Issues in
Tourism, 1–5. Advance online publication. https://doi.org/
10.1080/13683500.2020.1738357 .
Maclenan, A. (2018). Fake geography. GeoConnexion
International. https://flickread.com/edition/html/
5a4b7cdbe3415#18
Maghsoudi, Y., Collins, M. J., & Leckie, D. G. (2013).
Radarsat-2 polarimetric SAR data for boreal forest classifi-
cation using SVM and a wrapper feature selector. IEEE
Journal of Selected Topics in Applied Earth Observations
and Remote Sensing, 6(3), 1531–1538. https://doi.org/10.
1109/JSTARS.2013.2259219
Markuse, P. (2019). It’s a faaaake . . . — Or not? Medium.
https://medium.com/sentinel-hub/its-a-faaaake-or-not-
bace4f0c01ec
Matern, F., Riess, C., & Stamminger, M. (2019). Exploiting
visual artifacts to expose deepfakes and face manipulations.
In 2019 IEEE Winter Applications of Computer Vision
Workshops (WACVW) (pp. 83–92). IEEE. https://doi.org/
10.1109/WACVW.2019.00020
Mittal, A., Moorthy, A. K., & Bovik, A. C. (2012). No-
reference image quality assessment in the spatial domain.
IEEE Transactions on Image Processing, 21(12), 4695–4708.
https://doi.org/10.1109/TIP.2012.2214050
Monmonier, M. (1991). How to lie with maps. University of
Chicago Press. https://doi.org/10.7202/022356ar
Nystuen, J. D. (1984). Comment on “artificial intelligence and
its applicability to geographical problem solving”.
Professional Geographer, 36(3), 358–359. https://doi.org/
10.1111/j.0033-0124.1984.00358.x
Openshaw, S., & Abrahart, R. J. (1996). Geocomputation. In
R. J. Abrahart (Ed.), Proceedings of the 1st international confer-
ence on geocomputation (pp. 665–666). University of Leeds.
Openshaw, S., & Openshaw, C. (1997). Artificial intelligence in
geography. Wiley. https://doi.org/book/10.5555/549314
Oscar, S. (2018). You thought fake news was bad? Deep fakes
are where truth goes to die. The Guardian. https://www.
theguardian.com/technology/2018/nov/12/deep-fakes-fake
-news-truth
Pao, H. K., Chen, K. T., & Chang, H. C. (2010). Game bot
detection via avatar trajectory analysis. IEEE Transactions
on Computational Intelligence and AI in Games, 2(3),
162–175. https://doi.org/10.1109/TCIAIG.2010.2072506
Rössler, A., Cozzolino, D., Verdoliva, L., Riess, C., Thies, J., &
Niessner, M. (2019). FaceForensics++: Learning to detect
manipulated facial images. In K. Lee, D. Forsyth,
M. Pollefeys, & X. Tang (Eds.), 2019 IEEE/CVF
International Conference on Computer Vision (ICCV) (pp.
1–11). IEEE. https://doi.org/10.1109/ICCV.2019.00009
Salimans, T., Goodfellow, I., Zaremba, W., Cheung, V.,
Radford, A., & Chen, X. (2016). Improved techniques for
training GANs. In D. Lee, U. Luxburg, R. Garnett,
M. Sugiyama, & I. Guyon (Eds.), Proceedings of the 30th
international conference on neural information processing
systems (pp. 2234–2242). ACM. https://doi.org/10.5555/
3157096.3157346
Sayler, K. M., & Harris, L. A. (2019). Deep fakes and national
security (CRS Report No. IF11333). Congressional Research
Service. https://crsreports.congress.gov/product/pdf/IF/
IF11333
Shepard, D. P., Humphreys, T. E., & Fansler, A. A. (2012).
Evaluation of the vulnerability of phasor measurement
units to GPS spoofing attacks. International Journal of
Critical Infrastructure Protection, 5(3), 146–153. https://
doi.org/10.1016/j.ijcip.2012.09.003
Sirosh, J. (2018). Microsoft and Esri launch geospatial AI on Azure.
Microsoft Azure Blog. https://azure.microsoft.com/en-us
/blog/microsoft-and-esri-launch-geospatial-ai-on-azure/
Smith, T. R. (1984). Artificial intelligence and its applicability
to geographical problem solving. The Professional
Geographer, 36(2), 147–158. https://doi.org/10.1111/j.
0033-0124.1984.00147.x
Stricker, M. A., & Orengo, M. (1995). Similarity of color images.
Storage and Retrieval for Image and Video Databases III, 2420,
381–392. https://doi.org/10.1117/12.205308
Tippenhauer, N. O., Pöpper, C., Rasmussen, K. B., &
Capkun, S. (2011). On the requirements for successful
GPS spoofing attacks. In Proceedings of the 18th ACM
conference on computer and communications security
(pp.75–86). https://doi.org/10.1145/2046707.2046719
Tucker, P. (2019). The newest AI-enabled weapon: ‘deep-
faking’ photos of the earth. Defense One. https://www.defen
seone.com/technology/2019/03/next-phase-ai-deep-fak
ing-whole-world-and-china-ahead/155944/
Ulaby, F. T., Kouyate, F., Brisco, B., & Williams, T. H. L.
(1986). Textural Information in SAR Images. IEEE
Transactions on Geoscience and Remote Sensing, GE-24(2),
235–245. https://doi.org/10.1109/TGRS.1986.289643
14 B. ZHAO ET AL.
Wang, S.-Y., Wang, O., Zhang, R., Owens, A., & Efros, A. A.
(2020). CNN-generated images are surprisingly easy to
spot . . . for now. In T. Boult, G. Medioni, & R. Zabih
(Eds.), 2020 IEEE/CVF conference on computer vision and
pattern recognition (pp. 8692–8701). IEEE. https://doi.org/
10.1109/cvpr42600.2020.00872
Zhang, M. (2015). KFC’s new tray liners let you take “fakation”
photos. PetaPixel. https://petapixel.com/2015/08/05/kfcs-
new-tray-liners-let-you-take-fakation-photos/
Zhang, S. (2015). The fake places that only exist to catch
copycat mapmakers. Gizmodo. https://gizmodo.com/the-
fake-places-that-only-exist-to-catch-copycat-cartog-
1695414770
Zhang, S., Zhao, B., Tian, Y., & Chen, S. (2020). Stand with
#StandingRock: Envisioning an epistemological shift in
understanding geospatial big data in the “post-truth” era.
Annals of the American Association of Geographers, 1–21.
Advance online publication. https://doi.org/10.1080/
24694452.2020.1782166 .
Zhang, X., Karaman, S., & Chang, S. F. (2019). Detecting and
simulating artifacts in GAN fake images. In 2019 IEEE
international Workshop on Information Forensics and
Security (WIFS) (pp. 1–6). IEEE. https://doi.org/10.1109/
WIFS47025.2019.9035107
Zhao, B., & Sui, D. Z. (2017). True lies in geospatial big data:
Detecting location spoofing in social media. Annals of GIS,
23(1), 1–14. https://doi.org/10.1080/19475683.2017.
1280536
Zhao, B., & Zhang, S. (2019). Rethinking spatial data quality:
Pokémon Go as a case study of location spoofing.
Professional Geographer, 71(1), 96–108. https://doi.org/10.
1080/00330124.2018.1479973
Zhu, J. Y., Park, T., Isola, P., & Efros, A. A. (2017). Unpaired
image-to-image translation using cycle-consistent adver-
sarial networks. In K. Ikeuchi, G. Medioni, & M. Pelillo
(Eds.), 2017 IEEE International Conference on Computer
Vision (ICCV) (pp. 2242–2251). IEEE. https://doi.org/10.
1109/ICCV.2017.244
CARTOGRAPHY AND GEOGRAPHIC INFORMATION SCIENCE 15
