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

net/publication/344056372

Google Earth Engine Cloud Computing Platform for Remote Sensing Big Data
Applications: A Comprehensive Review

Article  in  IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · September 2020
DOI: 10.1109/JSTARS.2020.3021052

CITATIONS READS
4 2,517

12 authors, including:

Meisam Amani Arsalan Ghorbanian

Wood Environment & Infrastructure Solutions Khaje Nasir Toosi University of Technology
60 PUBLICATIONS   497 CITATIONS    8 PUBLICATIONS   29 CITATIONS   

SEE PROFILE SEE PROFILE

S.Ali Ahmadi Mohammad Kakooei

Khaje Nasir Toosi University of Technology Babol Noshirvani University of Technology
6 PUBLICATIONS   15 CITATIONS    19 PUBLICATIONS   68 CITATIONS   

SEE PROFILE SEE PROFILE

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

Applications in remote sensing View project

Soil Moisture Retrieval Using Synthetic Aperture RADAR (SAR) Images View project

All content following this page was uploaded by Meisam Amani on 18 September 2020.

The user has requested enhancement of the downloaded file.

5326 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020

Google Earth Engine Cloud Computing Platform for

Remote Sensing Big Data Applications: A
Comprehensive Review
Meisam Amani , Senior Member, IEEE, Arsalan Ghorbanian , Seyed Ali Ahmadi , Mohammad Kakooei ,
Armin Moghimi , S. Mohammad Mirmazloumi, Student Member, IEEE, Sayyed Hamed Alizadeh Moghaddam ,
Sahel Mahdavi, Masoud Ghahremanloo, Saeid Parsian,
Qiusheng Wu , and Brian Brisco

Abstract—Remote sensing (RS) systems have been collecting GEE publications have significantly increased during the past few
massive volumes of datasets for decades, managing and analyzing years, and it is expected that GEE will be utilized by more users
of which are not practical using common software packages and from different fields to resolve their big data processing challenges.
desktop computing resources. In this regard, Google has developed
a cloud computing platform, called Google Earth Engine (GEE), to Index Terms—Big data, cloud computing, Google Earth Engine
effectively address the challenges of big data analysis. In particular, (GEE), remote sensing (RS).
this platform facilitates processing big geo data over large areas and
monitoring the environment for long periods of time. Although this
platform was launched in 2010 and has proved its high potential for
different applications, it has not been fully investigated and utilized I. INTRODUCTION
for RS applications until recent years. Therefore, this study aims
N RECENT years, there has been a significant increase in the
to comprehensively explore different aspects of the GEE platform,
including its datasets, functions, advantages/limitations, and vari-
ous applications. For this purpose, 450 journal articles published in
I number of remote sensing (RS) datasets acquired by various
spaceborne and airborne sensors with different characteristics
150 journals between January 2010 and May 2020 were studied. It (e.g., spectral, spatial, temporal, and radiometric resolutions)
was observed that Landsat and Sentinel datasets were extensively [1]. This trend is expected to continue due to the availability of
utilized by GEE users. Moreover, supervised machine learning
algorithms, such as Random Forest, were more widely applied to
more open-access RS datasets and daily advancement in sensor,
image classification tasks. GEE has also been employed in a broad image processing, and computer vision technologies [2].
range of applications, such as Land Cover/land Use classification, Working with petabytes of RS datasets is a challenging task
hydrology, urban planning, natural disaster, climate analyses, and and has its own special requirements. The challenges of big data
image processing. It was generally observed that the number of processing and analyzing can be divided into two categories:
common and individual facets [3]. The common challenges
Manuscript received June 7, 2020; revised July 18, 2020 and August 11, 2020; are more related to handling big data and include big data
accepted August 26, 2020. Date of publication September 1, 2020; date of current
version September 17, 2020. (Corresponding author: Meisam Amani.) computing, big data collaboration, and big data methodologies.
Meisam Amani and Sahel Mahdavi are with the Wood Environ- The individual challenges are related to big data life cycle in
ment & Infrastructure Solutions, Ottawa, ON K2E 7L5, Canada (e-mail: different applications, such as the appropriate data identifica-
meisam.amani@woodplc.com; sahel.mahdavi@woodplc.com).
Arsalan Ghorbanian, Seyed Ali Ahmadi, Armin Moghimi, and tion, data deployment, data representation, data fusion, as well
Sayyed Hamed Alizadeh Moghaddam are with the Faculty of Geodesy as data visualization and interpretation. In order to provide a
and Geomatics Engineering, Department of Remote Sensing and Pho- comprehensive solution that can meet a wide range of current
togrammetry, K. N. Toosi University of Technology, Tehran 1996715433,
Iran (e-mail: a.ghorbanian@email.kntu.ac.ir; cpt.ahmadisnipiol@yahoo.com; and future challenges and requirements in RS applications, one
moghimi.armin@gmail.com; h.alizadeh@email.kntu.ac.ir). of the most important steps is to develop a safe, efficient, and
Mohammad Kakooei is with the Department of Electronic Engineering, advanced cloud computing platform [3], [4].
Babol Noshirvani University of Technology, Babol 4714871167, Iran (e-mail:
kakooei.mohammad@stu.nit.ac.ir). Cloud computing platforms are efficient ways of storing,
S. Mohammad Mirmazloumi is with the Centre Tecnològic de Telecomu- accessing, and analyzing datasets on very powerful servers,
nicacions de Catalunya (CTTC/CERCA), 08860 Castelldefels, Spain (e-mail: which virtualize supercomputers for the user. These systems
sm.mirmazloumi@cttc.es).
Masoud Ghahremanloo is with the Department of Earth and Atmo- provide infrastructure, platform, storage services, and software
spheric Sciences, University of Houston, Houston, TX 77004 USA (e-mail: packages in a variety of ways for the customers [3], [4]. Several
mghahremanloo@uh.edu). cloud computing platforms have so far been developed. For
Saeid Parsian is with the Department of Surveying Engineering, Tafresh
University, Tafresh 395187611, Iran (e-mail: saeid90parsian@gmail.com). example, Amazon Web Services (AWS) is a pay-as-you-go
Qiusheng Wu is with the Department of Geography, University of Tennessee, platform, where users pay based on the hours that they use the
Knoxville, TN 37996 USA (e-mail: qwu18@utk.edu). services [2]. AWS has a dedicated cloud Earth Observation (EO)
Brian Brisco is with the Canada Center for Mapping and Earth Observation,
Ottawa, ON K1S 5K2, Canada (e-mail: brian.brisco@canada.ca). offering called “Earth on AWS” as part of its Public Dataset
Digital Object Identifier 10.1109/JSTARS.2020.3021052 Program, which includes open data from several satellites such
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
AMANI et al.: GOOGLE EARTH ENGINE CLOUD COMPUTING PLATFORM FOR REMOTE SENSING BIG DATA APPLICATIONS 5327

as Landsat-8, Sentinel-1, Sentinel-2, China–Brazil Earth Re- and Mutanga [9] also briefly discussed the publication and au-
sources Satellite program, National Oceanographic, and Atmo- thorship trends, datasets, study areas, and applications of GEE by
spheric Administration Advanced (NOAA) image datasets, as reviewing 300 journal papers. Furthermore, Mutanga and Kumar
well as global model outputs. AWS also hosts open data supplied [10] briefly discussed four main applications of GEE. More
by DigitalGlobe with its SpaceNet challenges. Moreover, AWS recently, Tamiminia et al. [2] also discussed various aspects of
hosts the largest suite of machine learning services [4]. Azure GEE by reviewing 349 journal papers. The authors provided
is another cloud computing platform hosted by Microsoft. This comprehensive information about the GEE publications based
platform has established the Artificial Intelligence (AI) for earth on study areas, number of publications, datasets and products,
initiative to facilitate the use of its AI tools for addressing functions, sensor type and resolutions, classification accuracies,
environmental challenges in four main areas of climate, agri- and various applications.
culture, biodiversity, and water. Azure only contains Landsat There is still need for a more comprehensive review to discuss
and Sentinel-2 products for North America, since 2013, as well various aspects of the GEE platform. Therefore, in this study,
as moderate resolution imaging spectroradiometer (MODIS) 450 journal articles along with peer-reviewed conference papers
imagery. Azure is also a pay-as-you-go platform which provides were investigated through eight main sections: Section I pro-
virtual systems for the users [5]. vides an introduction to GEE; Section II provides an overview
Google Earth Engine (GEE) is another cloud computing of the GEE platform; Section III presents different datasets
platform which was launched by Google, in 2010. GEE included in this platform; Section IV discusses various GEE
uses Google’s computational infrastructure and available open- functions and algorithms; Section V provides comprehensive
access RS datasets [6]. GEE is the most popular big geo data information about the advantages and limitations of GEE; Sec-
processing platform, facilitating the scientific discovery process tion VI analyzes the pattern of GEE publications over one
by providing users with free access to numerous remotely sensed decade; Section VII discusses different applications of GEE;
datasets [1], [2]. Users can access GEE via an internet-based and finally Section VIII provides several case studies, in which
Application Programming Interface (API) and a web-based In- GEE was applied to process and analyze big data over large areas
teractive Development Environment [2], [6]. Additionally, users and within a long period of time.
do not need to have expertise in web programming or HyperText
Markup Language to use GEE for different applications [6].
GEE has the features of an automatic parallel processing and fast II. GEE PLATFORM OVERVIEW
computational platform to effectively deal with the challenges GEE is mainly composed of the following three platforms:
of big data processing [6], [7]. For instance, according to Hansen 1) Earth Engine (EE) Explorer;
et al. [8], it only took 100 h to process 654 178 Landsat-7 images 2) EE Code Editor;
(about 707 terabytes) within GEE and produce a global map of 3) EE Timelapse.
forests. This was reported as a great achievement because if The details of each platform are discussed in the following
they did not use GEE, this process would have taken a million sections.
hours to complete. Furthermore, users do not need to download
the available dataset within GEE in order to use them or install
any software to perform the processing tasks existing in GEE. A. EE Explorer
However, GEE users can utilize complementary software pack- EE Explorer (see Fig. 1) is a data viewer platform which
ages or process their own private datasets within this platform. allows users to access the massive datasets available in the EE
This platform also contains various built-in algorithms, such as Data Catalog. The Data Catalog houses millions of publicly
classification algorithms, to analyze data at a planetary scale and available datasets, including a complete series of Landsat-4, -
also helps scientists to develop their own algorithms with less 5, -7, and -8, MODIS, Sentinel-1, -2, -3, and -5P imagery, as
effort than before [1], [2], [9]. well as several atmospheric, meteorological, and vector datasets,
As discussed, the remarkable capabilities of GEE provide which will be further discussed in Section III. The Data Catalog
unprecedented opportunities to employ this platform for big receives approximately 4000 new datasets every day [11].
data processing and interpretation and, therefore, it is effectively As illustrated in Fig. 1, the EE Explorer is composed of the
employed in a broad variety of disciplines in all branches of Earth Workspace [see Fig. 1(a)] and the Data Catalog [see Fig. 1 (b)]. In
science studies. It is also expected that users will more frequently the Data Catalog, users can search among massive datasets and
use this cloud computing service considering the trends of GEE import them to the Workspace. In the Workspace, users can man-
studies in recent years. There are currently four GEE literature age and visualize datasets. The Workspace also enables users for
review studies conducted by Gorelick et al. [6], Kumar and a quick view, zoom, and pan. Additionally, it allows users to set
Mutanga [9], Mutanga and Kumar [10], and Tamiminia et al. [2], parameters related to the visualization setting, such as contrast,
published between 2017 and 2020, respectively. Gorelick et al. brightness, and opacity levels. To better inspect any changes over
[6] was the first comprehensive GEE review paper conducted time, users can add multiple layers to the Workspace. Users
by the main GEE developers. The authors comprehensively can display the layers in a three-band RGB or a single-band
discussed different aspects of GEE, including data catalog, grayscale/pseudocolor representation [6]. For example, Fig. 1(a)
system architecture, functions, data distribution models, effi- demonstrates a true color composite of a MODIS bidirectional
ciency, along with several applications and challenges. Kumar reflectance distribution function (BRDF)-adjusted image.
5328 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020

Fig. 1. Earth Engine Explorer platform. (a) Workspace. (b) Data Catalog.

B. EE Code Editor Asset tab. It is also possible to interactively query the map using
While the EE Explorer platform is designed to visualize the Inspector tab. Finally, the Geometry tools allow users to draw
datasets, the EE Code Editor (see Fig. 2) is designated to process geometric features, such as points, lines, and polygons, which
can be used in further analyses [6].
big data using a JavaScript programming language and to de-
velop EE applications. According to Fig. 2, the EE Code Editor is
composed of the following elements: Code editor, Map, Layer C. EE Time-Lapse
manager, Geometry tools, and several tabs, including Script, GEE combines petabytes of RS datasets over four decades
Doc, Assets, Inspector, Console, and Tasks. and produces a global, zoomable, and cloud-free video over
The central panel allows users to write their JavaScript code. space and time in its EE Time-laps platform [6]. The Timelapse
GEE processes the written codes and illustrates the results as platform is an example of the great computational power of the
images in the Map panel or as messages in the Console Tab. GEE platform. This platform provides the most comprehensive
Similar to the EE Explorer, users can set the visualization picture of the Earth revealing how its residents are treating it.
parameters via the Layer manager in the Code Editor (see Fig. 2). For instance, through EE Time-lapse, one can easily observe
In the Script tab, numerous examples of scripts facilitate devel- the fast retreat of Mendenhall Glacier in Alaska, decapitation of
oping applications. There are more than 800 prebuilt functions West Virginia Mountains by the mining industry, forest loss in
(discussed in detail in Section IV) in the EE library, users can the Amazon, and drying Urmia lake in Iran over time.
become familiar with them using the Doc tab, providing API
reference documentation [6].
As previously mentioned, GEE includes big open-access III. GEE DATASETS
datasets. Users, however, are not restricted to use only these As discussed, GEE contains an immense number of datasets,
datasets. They can upload and manage their own data using the including raw datasets, preprocessed data, elevation models, and
AMANI et al.: GOOGLE EARTH ENGINE CLOUD COMPUTING PLATFORM FOR REMOTE SENSING BIG DATA APPLICATIONS 5329

Fig. 2. Overview of the Earth Engine Code Editor.

products at global, national, and regional extents. Table IV in the estimated satellite-derived bathymetry (SDB) of three regions
Appendix provides all available datasets within GEE along with in the Aegean Sea using Sentinel-2 time-series analysis.
a brief description of each. Some of these datasets, which are GEE includes MODIS images. MODIS has a great potential
frequently utilized by users are discussed in more detail in the in near-real-time (NRT) mapping of the ground surface in na-
following. tional and global scales. MODIS acquires images in 36 spectral
Landsat datasets are valuable resources to perform tem- bands, the spatial resolutions of which vary from 250 m to 1
poral analysis. Landsat collection includes seven multispec- km. MODIS time series are available in GEE Data Catalog
tral satellites: Landsat 1–3 (1972–1983), Landsat-4 (1982– from 2000 to present, facilitating temporal analysis over globe.
1993), Landsat-5 (1984–2012), Landsat-7 (1999–present), and Campos-Taberner et al. [21] developed a temporal investigation
Landsat-8 (2013–present). Landsat satellites have optical sen- on MODIS-based indices, including the global Leaf Area Index,
sors, the images of which may be obscured by clouds. Therefore, Canopy water content, Fraction Vegetation Cover, and Fraction
temporal cloud detecting, masking, and removing are essential of Absorbed Photosynthetically Active Radiation.
preprocessing steps in different applications, such as image clas-
sifications using multitemporal imagery [12]. Additionally, the
availability of the multitemporal Landsat datasets has facilitated IV. GEE FUNCTIONS
national and global scale analysis [13]. Landsat-based datasets GEE provides various functions to perform spectral and spa-
within GEE have been employed in various applications. For in- tial operations on either a single image or a batch of images.
stance, Landsat data available in GEE have been widely utilized Different operations within the GEE platform, ranging from
in generating Land Cover/Land Use (LCLU) maps (e.g., [14]– simple mathematical operations to advanced image processing
[16]). Moreover, urban detection and extraction is an impor- and machine learning algorithms are illustrated in Fig. 3. Various
tant task in the economic investigation due to rapid population pixel-based spectral operations, which have high potential to be
growth. Therefore, several studies have utilized Landsat data in implemented in parallel on cloud architecture, are included in
urban monitoring [17], [18]. GEE. However, GEE supports fewer spatial functions, such as
GEE includes datasets acquired by Sentinel satellites, devel- Gaussian and Laplacian filters, edge detection methods (e.g.,
oped by the European Space Agency (ESA). Sentinel collection Sobel, Roberts, and Canny), line detection via the Hough Trans-
includes Sentinel-1 Synthetic Aperture RADAR (SAR) (2014– form, and morphological operators (e.g., dilation and erosion)
present), Sentinel-2 multispectral (2015–present), Sentinel-3 due to parallel implementation issues. Moreover, GEE currently
Ocean and Land Color (2016–present), and Sentinel-5P Tro- does not support several functions, including frequency do-
pospheric Monitoring (2018–present) datasets. Sentinel-1 and main algorithms (e.g., FFT and Wavelet), hierarchical algo-
Sentinel-2 have been extensively utilized by GEE users for rithms (e.g., hierarchical clustering), graph-based methods (e.g.,
different applications. Their 10 m spatial resolution makes it graphcut), geometric descriptors (e.g., Haar, SIFT, SURF), and
possible to analyze objects in a better resolution compared physical-based models (e.g., radiative transfer models).
to Landsat images. They can also simplify the procedure of Both supervised and unsupervised machine learning algo-
training and validation steps in image classification tasks. Man- rithms are accessible through the GEE library. For example,
dal et al. [19] applied Sentinel-1 SAR data to map rice and the classification and regression tree (CART), support vector
monitor its temporal changes. Additionally, Traganos et al. [20] machine (SVM), and random forest (RF) classifiers are among
5330 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020

Fig. 3. Overview of different supporting functions within GEE.

the supervised classification algorithms within GEE. Labeled detection of Trends (LandTrendr) [32] are available. CCDC
samples are required in supervised classification methods to train fits harmonic functions to temporal data to detect points with
the classifiers, for which both sampling and training functions significant variations. EWMACD calculates a model according
are available in GEE. There are also many clustering algorithms to the training data. Then, the difference between the model and
in GEE, such as K-means. K-means is a popular clustering real data points are found according to the Shewhart X-bar charts
method in the data mining area. The algorithm requires users and an exponentially weighted moving average. LandTrendr is
to define the number of clusters (K) and the stopping criteria specially designed for Landsat data and finds the pixel-based
[22]–[24]. Besides the original K-means, two modified versions spectral change in temporal analysis. Vegetation analysis is also
of K-means (i.e., Cascade K-means [25] and X-means [26]), a popular subject in temporal analysis. Therefore, GEE has
in which the number of clusters is estimated automatically, are several algorithms, such as vegetation change tracker (VCT) [33]
available in GEE. Cobweb is another clustering algorithm which and vegetation regeneration and disturbance estimates through
hierarchically handles data instances data instances. It constructs time (VERDET) [34], which are specifically developed for this
a classification tree and manages it through merging and splitting purpose. VCT can automatically analyze Landsat time-series
steps [27]. Simple noniterative clustering (SNIC) is another images to generate forest disturbance history. VERDET catego-
clustering-based segmentation method, which is initiated with rizes forest change into three types, including disturbed, stable,
randomly/manually determined seeds and generates segments and regenerating. The analysis is based on the total variation
[28]. SNIC is widely utilized by users to perform object-based regularization in the spatial and temporal domain [34].
image classifications (e.g., [29]).
As mentioned before, GEE contains over 40 years of datasets,
facilitating temporal and change analyses. For temporal analysis V. GEE ADVANTAGES AND LIMITATIONS
purposes, several functions, such as continuous change detection GEE is a valuable tool in analyzing geospatial data that
and classification (CCDC) [30], exponentially weighted moving provides many capabilities for researchers, especially for the RS
average change detection (EWMACD) [31], and Landsat-based community. However, there are also several limitations that users
AMANI et al.: GOOGLE EARTH ENGINE CLOUD COMPUTING PLATFORM FOR REMOTE SENSING BIG DATA APPLICATIONS 5331

TABLE I
MAIN ADVANTAGES OF GEE BIG GEO DATA PROCESSING PLATFORM

should be aware of. The key advantages and limitations of GEE GEE stores and analyzes RS imagery based on a pyramiding
are summarized in Table I and discussed in more detail in the and tiling concept [39]. Every image ingested into GEE has its
following section. As illustrated in Table I, the advantages and pyramid at different pixel resolutions [6]. Furthermore, every
disadvantages of GEE are investigated within the four categories tool used in GEE processes images on 256×256 tiles. Thus,
of cloud infrastructure, API, data, and functions. different scales of the pyramid are used at various zoom levels.
This enables GEE to visualize large areas of processed imagery
quickly and efficiently.
A. Advantages Fast filtering and sorting capabilities are provided within
1) Cloud Infrastructure: GEE is mainly a free cloud-based GEE, inherited from Google. This enables users to select their
service without having to download and manage data locally desired data out of millions of images based on various spatial
[35]. It is built upon the Google cloud computing infrastructure and temporal specifications [40].
and computations are automatically handled by Google itself. 2) API: GEE is combined with a powerful web-based pro-
All operations are automatically performed in bulk and parallel gramming interface. Users can easily access archived RS data
on the Google CPUs and GPUs [6]. The complexities of parallel through the JavaScript and Python API. The straightforward
computing are hidden due to this automation in processes [17]. concept of using both APIs allows users to focus on the logic
Since GEE was mainly created and optimized for geospatial of data selection and programmable workflow. Only a log-in is
data analysis, it can process petabyte of RS data both in large required to access all GEE power. An online code editor is also
geographical scales and in long temporal coverages [17]. Thus, available to write scripts, debug them, and see the results just
it is a great tool for analyzing regional, national, continental, and after compilation.
global-scale applications. Both the JavaScript and Python APIs provide access to the
Besides various datasets, which are already available within same set of EE objects and methods, except for a few methods
GEE, researchers can easily upload and share their own datasets which are capitalized differently (e.g., .and() versus .And()) [41],
as well as their scripts and models through URLs [9]. Other [42].
maps and products are generated on-the-fly [28], [29], once any Most of the libraries in GEE are similar to existing open source
user wants to run the code [36], [37]. Additionally, there is no components, such as OpenCV, and GDAL. Therefore, there is a
need to install third-party software packages, such as ENVI and minimum requirement to learn new concepts.
ERDAS, because almost all of the required tools are already The Python API provides a programmatic and flexible
available on GEE [38]. interface to EE [41], [43]. It allows for automating batch
5332 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020

processing tasks, piping EE processed data to Python packages Machine learning, image processing, vector processing, geo-
for postprocessing and leveraging the power of the command metrical analysis, different visualizations, and multiple special-
line. Additionally, the Jupyter notebook interface of the Google ized algorithms are gathered into the GEE platform and enable
Colaboratory platform delivers a highly interactive and collab- users to implement their idea. The GEE functions usually satisfy
orative experience and is without the burden of local system the needs of a typical scientific project. Additionally, users can
setup and management as a hosted service. In summary, the always implement their own algorithms outside GEE and return
EE code editor has a high ease of setup and use, while the the result for postprocessing. For instance, TensorFlow is a better
Python API is more flexible. Combining GEE and Python option in the deep learning section, for which more complex
APIs inside a Jupyter notebook provides the advantage of both models, larger training datasets, more input properties, or longer
to users. training times are required [47], [48]. TensorFlow models are
In order to compare JavaScript with Python based on [43], developed, trained, and deployed outside EE [49]. For easier
it can be argued that JavaScript is easy to get started and share interoperability, the EE API provides methods to import/export
scripts, while it cannot share code between scripts. However, data in TFRecord format [47]. This facilitates generating training
Python is easy to share code between scripts and is easier and evaluation datasets in EE and exporting them to a format
to be transformed into a web application. Moreover, Python where they can be ingested to a TensorFlow model.
has many plotting options, which requires several assembly A complete API reference and tutorial with runnable code
and maintenance. Finally, the code editor enables the user to examples are available for beginner to advanced users (e.g.,
store, share, and control their codes in a behind-the-scene git [47]). The tutorials are detailed and cross referenced to each
environment. other to guide users through different applications and important
3) Datasets: As discussed in Section III, GEE contains a notes. Outputs of these algorithms can be directly embedded in
large catalog of RS, geophysical, and meteorological datasets. different applications.
It contains most of the important and temporal datasets in
RS, including Landsat, MODIS, and Sentinel. Furthermore,
the combination of different sources of imagery improves the B. Limitations
temporal density of datasets and can help fusion algorithms to GEE limitations are relatively minor, but it is essential to be
have more power. Moreover, several NRT datasets are uploaded familiar with the constraints. Several main limitations of GEE
to GEE in a daily manner. If a dataset is not in the GEE Data are discussed in the following.
Catalog, it can also be uploaded to the servers. Datasets are also 1) Although data is kept as private in the user’s account, it is
downloadable to continue from a desktop workstation at any still stored in the servers of a private company, which is not
point of the workflow. acceptable for many governmental agencies and private
GEE stores datasets in their original projection with all orig- companies [50].
inal data and metadata. Resolutions are managed directly by 2) GEE-based image analysis is restricted to existing tools
the platform. Data are stored in its original resolution, but a within the GEE API. For example, several standard image
pyramid of images is also constructed and stored beside every preprocessing methods (e.g., atmospheric correction tech-
image which is used in different zoom levels for the sake of niques) are currently not implemented in GEE. Moreover,
efficiency. As mentioned, users can also easily search for their developing new tools is not trivial and requires knowledge
desired data using the tags provided within data categorization, about all GEE algorithms and their functionality along
which is very well handled in GEE. with performance considerations about cloud-based com-
Several preprocessing steps have been already applied to the puting on Google servers.
datasets and, thus, users can use corrected data besides raw data. 3) GEE is limited to selected data mining models for classi-
For instance, the orthorectified, atmospherically corrected, and fication and regression. There are only a few classification
Calibrated Top of Atmosphere Landsat data are easily accessible and regression algorithms, such as CART, RF, and SVM.
apart from the raw data [44], [45]. Analysis-ready SAR datasets 4) Image classification as one of the important applications of
on GEE represent a significant step forward because SAR pre- RS can be considerably improved by object-based image
processing is relatively complex (especially for regular users). analysis. However, currently, there is not an efficient and
For example, GEE hosts Sentinel-1 GRD data preprocessed with accurate segmentation algorithm within GEE [1].
ESA’s SNAP software [46]. 5) One of the main approaches to improve classification
GEE makes many derivative products available. Multiple accuracy is increasing the number of training samples or
popular spectral indices (e.g., NDVI) are already calculated. input features. However, users are limited to employ only a
Since storage is more expensive than computation, most of these certain amount of samples or a limited number of features
derivative products are computed on-the-fly upon users’ request. within classification methods [1], [16].
4) Functions: As discussed in Section IV, a large set of 6) Complex machine/deep learning algorithms which require
functions and algorithms are available within GEE library large training datasets or longer training times are not
for analyzing various datasets. All algorithms are parallel in performed in GEE due to computational restrictions. Thus,
nature and can automatically handle data management over users need to implement these algorithms outside of this
servers. environment [48].
AMANI et al.: GOOGLE EARTH ENGINE CLOUD COMPUTING PLATFORM FOR REMOTE SENSING BIG DATA APPLICATIONS 5333

TABLE II
TOP 10 JOURNALS PUBLISHING GEE-RELATED ARTICLES ALONG WITH THE NUMBER OF PUBLICATIONS PER JOURNAL

TABLE III
GEE STUDIES PRESENTED AT THE TOP EIGHT CONFERENCES ALONG WITH THE NUMBER OF PRESENTED ARTICLES

7) When trying to download processed data in the middle of because most of them had a relatively lower academic level
the workflow for further analysis in a third-party software or had later been converted to journal papers. Only the top
environment, users face a time-consuming process due to conferences, where GEE studies were presented, were provided
huge map size and internet speed limitations. in Table III. The Google Earth Engine and GEE search queries
8) Complex SAR phase data are not stored in GEE because were performed in the journal articles’ titles, abstracts, and
they are not compatible with the tiling concept of the keywords from January 2010 to mid-May 2020. The EndNote
infrastructure [51]. This limits the Polarimetric SAR and software was then used to remove the duplicate articles, which
Interferometric SAR applications, which relies on the resulted in 462 peer-reviewed journal articles. Subsequently, 12
phase information. papers, which discussed unrelated topics (e.g., using GEE for
gaming development and analyzing the computational perfor-
VI. GEE PATTERN OF PUBLICATIONS mance of GEE) were discarded. Finally, 450 journal articles
were selected for further analyses.
In this study, 450 journal papers, published between January
2010 and May 2020, were assessed to depict the pattern of GEE
publications. Several investigations, including keyword analy- B. Keyword Analysis
sis, annual publication numbers, and geographical distribution
Fig. 4 illustrates a word cloud visualization based on the
are provided in the following sections. Additionally, the top
keywords in these GEE studies. The more frequent the term
journals and conferences, which have published GEE papers
appears within the keyword analysis, the larger the word de-
are discussed in Section VI-E.
picts in the figure. As clear, Google Earth Engine, Landsat,
Remote Sensing, Sentinel-2, Random Forest, Cloud Computing,
A. Analysis Method NDVI, Machine Learning, and Land Cover were the mostly
A meta-analysis was performed in the Elsevier’s Scopus (the used keywords, respectively. For example, Google Earth Engine
largest abstract and citation database of peer-reviewed literature keyword was utilized in 278 papers. The name of different
covering over 5000 publishers) and Web of Science (formerly satellites and machine learning algorithms are also widely used
known as ISI Thomson) to provide a comprehensive literature in GEE publications. Landsat, Sentinel-2, MODIS, Landsat-8,
trend conducted using GEE. It is worth noting that conference and Sentinel-1 are the satellites and Random Forest is the clas-
articles and presentations were also reviewed during the course sification method, which are frequently utilized in the keyword
of this study; however, they were not considered in this study lists of GEE journal papers. It was also observed that NDVI, land
5334 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020

presentation of historical 4-D city models. As clear from Fig. 5,

a substantial increase in the number of GEE publications was
observed from 2017, when Gorelick et al. [6] (GEE develop-
ers) discussed a comprehensive utility of this cloud computing
platform, particularly for RS applications. Additionally, it was
observed that the increasing trend in the number of GEE pub-
lications is getting more substantial. For instance, 35 journal
articles were published within the last 1.5 months (April 1–May
15, 2020).

D. Geographical Distribution
The study areas of the peer-reviewed GEE journal papers were
investigated to provide a picture of the geographic distribution of
GEE studies. Fig. 6 illustrates the geographical distribution of
GEE studies after removing ten papers, which did not belong
to any study areas (e.g., literature review papers and papers
Fig. 4. Word cloud of the keywords from the GEE journal articles.
related to the theoretical aspects and development of the GEE
platform). Additionally, 35 studies which were conducted over
the continental scales (7 and 1 papers covered the entire Africa
and Europe, respectively) and global scales (27 papers cov-
ered the entire world) were not considered in a country-based
enumeration. Furthermore, if a study was conducted over sev-
eral countries, all of them were counted separately. Moreover,
studies with sub-country scales (e.g., small study site, city, or
province) were considered in the number of publications for
the corresponding countries. Finally, it was observed that GEE
studies were conducted over 138 countries. The highest number
of GEE publications have been conducted over the United States
(97 articles), China (96 articles), Brazil (29 articles), Canada
(25 articles), and India (25 articles), followed by Australia
(19 articles), and Indonesia (15 articles), respectively. On the
continental scale, 37.5%, 24%, 18.5%, 9%,7.5%, 3.5%, and 0%
of studies were conducted over Asia, North America, Africa,
Europe, South America, Australia, and Antarctica, respectively.
Fig. 5. Number of journal articles, which utilized GEE. 38% of studies conducted over Asia were related to China.

cover, classification, and Urbanization were among the most E. Journals and Conferences
used keywords, indicating the popularity of LCLU classification
Table II provides the top journals, in which GEE studies have
applications. Additionally, the Time Series, Change Detection,
been published. 450 journal papers have been published in 150
Climate Change, Land Cover Change, and Time Series Analysis
journals, 95 of which have published only one GEE paper. Based
keywords were frequently utilized in the GEE publications,
on the results, the Remote Sensing, Remote Sensing of Environ-
indicating the importance of the archived open-access remote
ment, ISPRS Journal of Photogrammetry and Remote Sensing,
sensing datasets in change detection studies. Furthermore, multi-
and International Journal of Applied Earth Observation and
ple journal publications used China, United States, and Africa in
Geoinformation were the top four journals, which published 126,
the keywords, demonstrating the leadership of the corresponding
61, 14, and 12 papers, respectively.
countries in utilizing GEE in their studies.
Table III provides the name of several conferences, in which
GEE studies have been most frequently presented. GEE studies
C. Annual Publication Numbers
are among the most presented research topics in the prominent
The statistical analysis of the number of publications related international RS conferences, such as the International Archives
to GEE is provided in Fig. 5. The first peer-reviewed journal of the Photogrammetry Remote Sensing and Spatial Information
paper was published in 2011 by Keller et al. [52] in PFG - Jour- Sciences ISPRS Archive, the International Geoscience and Re-
nal of Photogrammetry, Remote Sensing and Geoinformation mote Sensing Symposium IGARSS, and the Proceedings of SPIE
Science. This study investigated the automated generation and The International Society for Optical Engineering.
AMANI et al.: GOOGLE EARTH ENGINE CLOUD COMPUTING PLATFORM FOR REMOTE SENSING BIG DATA APPLICATIONS 5335

Fig. 6. Number of GEE studies conducted over each country.

VII. GEE APPLICATIONS interacting with water, soil, and air [54]. Such cycles are im-
The 450 selected GEE journal articles were studied to decide portant for global vegetation pattern and climate studies and,
about the main disciplines. It is worth noting that seven papers thus, vegetation is also important for biodiversity conservation
which were review articles were initially removed from the and climate change mitigation [55]. Moreover, vegetations are
analysis. Consequently, all the 443 journal papers were divided the primary source of converting dioxide carbon to oxygen,
into 10 categories as illustrated in Fig. 7 along with several enabling aerobic metabolism on the globe [56]. Considering the
keywords describing each category. The articles which include important services of vegetation, it is highly required to monitor
more than one application were considered in the most relevant the current state and dynamics of various vegetation types. GEE
category by an in-depth review of the paper. It is worth noting that leverages cloud computing services for long-term monitoring
although only the journal articles were investigated to adopt the of vegetation covers. Furthermore, the publicly available RS
main disciplines, it was observed that other sorts of publications data within GEE enable researchers to employ this platform
(e.g., conference papers) correspond well with the application for vegetation monitoring at various spatial scales. In particu-
types considered in this study. lar, the existence of several vegetation indices in GEE allows
Fig. 8 illustrates the number of journal articles related to conducting vegetation studies in efficient and quick manners.
each application provided in Fig. 7. The highest number of GEE has been widely used for vegetation mapping [57], [58],
contributions were in the Vegetation category with 90 papers vegetation dynamics monitoring [59], [60], deforestation [61],
followed by 77 papers in Agriculture, 68 papers in Hydrology, 53 [62], vegetation and forest expansion [63], [64], forest health
papers in Land cover, 40 papers in Urban, 40 papers in Natural monitoring [65], [66], forest mapping [67], [68], pasture mon-
disaster, 31 papers in Atmosphere and climate, 17 papers in itoring [49], [69], and rangeland assessment [70], [71]. For
Image processing, and 14 papers in Pedosphere. Moreover, 13 instance, the full archive of the Landsat imagery was processed
papers, which were not related to any of the 10 application types within GEE to map the vegetation dynamics from 1988 to 2017
or their numbers and not enough to be assigned to a new category in Queensland, Australia [59]. Field observations were utilized
were considered in the Others category. to evaluate the performance of the proposed algorithm and an
In the following sections, more information about each overall accuracy of 82.6% was reported. Finally, the suitability
of the GEE applications along with several case studies are of GEE for large-scale and long-term vegetation monitoring
discussed. was reported along with an approximately 20% decrease in the
vegetation cover in this study area. The authors emphasized
the high computational efficiency of GEE compared to when
A. Vegetation they did the same analysis using traditional methods. In another
Vegetation (e.g. forest, grassland, rangeland, and shrub) can study, an algorithm was developed within GEE by employing
be considered as one of the most vital components of the spectral mixture analysis to detect degradation and deforestation
Earth’s biosphere because it serves critical functions to both in the Brazilian state of Rondônia [62]. To this end, Landsat
humans and the environment [53]. Vegetation is also impor- archived images from 1990 to 2013 were used. All the required
tant in many biochemical cycles that are directly or indirectly processing steps were performed within GEE to produce annual
5336 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020

Fig. 7. GEE applications (LC: Land Cover).

forest disturbances maps. Landsat data were transformed into and Agriculture Organization (FAO) has set its goal to achieve
spectral endmember fraction and were applied to calculate the food security around the globe [72], [73]. Agricultural products
Normalized Degradation Fraction Index. The presented method not only play a vital role in human life, but also are critical
obtained producer accuracies of 68.1% and 85.3% for degrada- from economic aspects. Therefore, agriculture can be considered
tion and deforestation maps, respectively. as a source of livelihood and a contributor to national revenue
[74]. Moreover, monitoring agricultural products is required for
policy-makers and governments to ensure the path to economic
B. Agriculture growth and self-sufficiency of the country [72]. RS datasets
Mapping and monitoring croplands and plantations are es- allow frequent and cost-effective monitoring of croplands
sential for food security. Food security could be stated as one of and plantations. GEE hosted extensive publicly available RS
the most significant issues in the current era and, thus, the Food datasets that can be effectively utilized for productivity, quality,
AMANI et al.: GOOGLE EARTH ENGINE CLOUD COMPUTING PLATFORM FOR REMOTE SENSING BIG DATA APPLICATIONS 5337

Fig. 8. Number and percentage of journal papers related to GEE applications, published in each discipline provided in Fig. 7.

profitability, and sustainability studies of agriculture produc- is a vital need. Publicly available datasets within GEE along with
tion. Researchers have applied GEE to plantation mapping and its high computing performance allow for accurate monitoring of
monitoring [75], [76], phenology-based classification [77], [78], water resources with adequate temporal and spatial resolutions.
cropland mapping [79], [80], crop condition monitoring [81], Consequently, GEE was efficiently employed for surface water
[82], crop yield estimation [83], [84], irrigation mapping [85], dynamics monitoring [93], [94], bathymetry [20], [95], shoreline
[86], and other agricultural studies [87], [88]. For example, and coastal studies [96], [97], lake and reservoir mapping and
seasonal median composites of Sentinel-1 and Sentinel-2 were monitoring [98], [99], glacier studies [90], [100], snow ablation
calculated in GEE to predict the Maize yield in Kenya and and snow mapping [92], [101], suspended sediments and river
Tanzania [83]. The use of RF resulted in the production of studies [102], [103], and water health assessment [104], [105].
Maize/none Maize maps in Kenya and Tanzania with 63% and For instance, Nguyen et al. [93] introduced a fully automatic
79% overall accuracies, respectively. Finally, satellite obser- method for water extraction in New Zealand. The GEE and
vations along with gridded soil datasets were ingested into a Landsat-8 images between 2014 and 2018 were employed to
scalable harmonic regression to estimate Maize yield. Moreover, map lakes and reservoirs using an Automatic Water Extraction
multitemporal Landsat-8, Landsat-7, and Sentinel-2 imagery Index with an overall accuracy of 85%. In a different study,
were employed to calculate composite NDVI images for winter GEE was used to combine MODIS fractional snow cover with
cropland mapping in an area of over 200 000 km2 [77]. Then, the Sentinel-1 wet snow mask data to develop an algorithm to
multitemporal NDVI curve was inserted into a CART algorithm produce a monthly wet-dry snow map [92]. In this study, 2.5
to produce a phenology-based map of winter cropland with an years were studied in the Indian Himalayan region covering
overall accuracy of 96.22%. The authors reported that lacking around 55 000 km2 . It is worth noting that the underestimation
remote sensing images with high temporal frequency in GEE of the wet snow area was corrected by DEM. In another study,
was one of the limitations of their work and, thus, suggested to blue and green bands of Sentinel-2 were processed to develop
use Chinese GaoFen satellite data with four days revisit time for an empirical model for satellite-derived bathymetry maps [20].
the future cropland classifications. In this regard, cloud masking, sun glint correction, radiometric
calibration, and normalization were performed within GEE in
three sites of the Aegean Sea in the Eastern Mediterranean.
C. Hydrology Finally, based on 9818 reference points, the proposed approach
Water is an essential element for life whether in liquid form achieved R2 and RMSE of 0.9 and 1.67 m, respectively. The au-
(e.g., lake, reservoir, and river) or solid forms (e.g., snow, ice, thors argued that GEE time-out error was the main limitation in
and glaciers) in the cryosphere and, thus, obtaining reliable in- their work, because their empirical method required estimation
formation about water resources is a high necessity. In addition, of the regression between the image composite values and water
monitoring inland, coastal, and arctic water resources are bene- depth over large region and a long period of time.
ficial in climate change studies [89]. Moreover, investigating
the size and behavior of glaciers along with the amount of
snow ablation could render supporting information about the D. Urban
Cryosphere–Atmosphere interactions and climate change [90], Urban areas are regions with concentrated people and human
[91]. Furthermore, drought and flood disasters are relatively infrastructure and usually expand through the time for better
associated with the dynamics of water resources [92]. Therefore, livelihood. These regions have become the central point of
persistent and precise monitoring of all types of water resources economic, social, cultural, and recreational activities, as well as
5338 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020

resource consumption [106], [107]. Therefore, urban areas could an automatic land cover mapping was developed within GEE
be considered as the primary source of human interaction with through the integration of Landsat imagery and RF algorithm
the surrounding environment. The environment and the urban over the north of China [14]. The reference samples were col-
areas affect each other mutually since the environmental changes lected by rules of pixel and spectral filtering from MODIS land
could influence human life. On the other hand, unrestricted urban cover products with the International Geosphere-Biosphere Pro-
growth causes severe damage to natural resources and can nega- gram theme in ten classes. Two types of monthly and percentile
tively alter the atmosphere and climate [108], [109]. Conducting features were utilized separately, and the best result was obtained
urban studies are essential to support sustainable development. through the usage of monthly features by achieving over 80%
In this regard, RS datasets enable the quantification and profound accuracy. In another study, GEE was used to produce a sharpened
analysis of urban dynamics that are fundamental for devising land cover map over Mato Grosso, Brazil [15]. Their proposed
suitable approaches for urban development and urban planning algorithm (BULC-U) fused the 300 m the GlobeCover product
[110]. GEE promotes long-term monitoring of urban conditions with Landsat imagery to produce a 30 m land cover map. In this
to effectively study the urban environment from different as- regard, Landsat images were segmented and then the ISODATA
pects. Urban expansion and extent mapping [18], [111], urban algorithm was applied to generate an unsupervised map in 20
morphology and local climate zone monitoring [112], [113], clusters. Finally, the unsupervised classification result was fused
urban 4-D modeling [52], urban green space classification [114], to the GlobeCover product. More recently, Ghorbanian et al.
[115], urban temperature and urban heat island identification [126] produced an improved version of the land cover map of
[17], [110] are some of the main urban studies conducted within Iran using Sentinel-1/2 imagery within GEE. They also proposed
GEE. For instance, Ravanelli et al. [17] studied the long-term an automatic workflow to update this map every year without the
monitoring of Surface Urban Heat Island (SUHI) and its relation need to collect additional in situ data using migrated samples.
to urban land cover changes over six metropolitan areas of the
United States. More than 6000 Landsat images were interpreted
between 1992 and 2011 by Detrended Rate Matrix analysis to F. Natural Disaster
illustrate the land cover change versus SUHI. It was reported Extreme and unexpected phenomena caused by the natural
that GEE was the best solution for their applications in terms of process of the Earth are called natural disasters. These events
efficiency in time, cost, and computation. The results revealed bring destruction to the surrounding environment and human
a definite increase of SUHI due to urban growth. Moreover, life [127]. Profound research should be carried out to investi-
Gong et al. [18] investigated the urban expansion dynamics gate the characteristics and behavior of these phenomena and,
by producing annual global artificial impervious surfaces that consequently, to reduce the amount of damage. The importance
are predominate indicators of human settlement. To this end, of geospatial data for monitoring and damage assessment of
the full archives of Landsat satellite data between 1985 and natural disasters is undeniable [128]. Long-term and NRT pub-
2018 were processed within GEE. Sentinel-1 SAR data and licly available RS datasets within GEE along with its high-
nighttime images were also used to improve the final results performance computing promote this cloud-based platform for
in arid areas. The implementation of the Exclusion–Inclusion monitoring, forecasting, prevention, vulnerability, and resilience
algorithm combined with the temporal consistency check within studies of natural disasters. In particular, GEE was utilized
GEE yielded the overall accuracy of over 90% in mapping annual for drought monitoring [129], [130], flood mapping and flood
global impervious surfaces. risk assessment [131], [132], wildfire severity mapping [133],
[134], landslides analyses [135], hurricane studies [136], and
tsunami studies [137]. For instance, MODIS and meteorological
E. Land Cover
datasets were employed within GEE to study the temporal and
The dominant land cover types of a region determine the spatial variations of drought events in Potohar Plateau of Punjab,
terrestrial surface characteristics of the corresponding area. Veg- Pakistan between 2000 and 2015 [129]. In this regard, multiple
etation, water, and soil are the main land cover types spread features of standard precipitation index, standard precipitation-
across the globe. These land cover types form environmental evapotranspiration index, vegetation condition index, precipita-
conditions for the habitat of various flora and fauna [116], [117]. tion condition index, soil moisture condition index, and temper-
Furthermore, the distribution of land covers defines the physical ature condition index were utilized for drought monitoring. In
interaction between Earth’s surface and the surrounding environ- addition, 44 Sentinel-1 GRD dual-polarized data were employed
ment. Recognizing the significant impacts of land covers on the within GEE to develop an operational methodology for rapid
environment and investigating the current condition along with flood inundation mapping in Bangladesh [131]. Moreover, a
monitoring long-term dynamics of land covers are essential for potential flood damage map was generated to support efficient
sustainable development, climate change modeling, biodiversity decision making. The proposed method obtained 96.44% overall
studies, and natural resource monitoring [118], [119]. GEE accuracy by incorporating 4500 reference samples. Finally, a
hosted enormous publicly RS datasets in various spectral and preflood Landsat-8 image was used to generate a land cover map
spatial resolutions to conduct land cover mapping [14], [120], for further estimation of flood damages to cropland and rural
land cover dynamics monitoring [121], [122], coastal mapping settlements. It was reported that the developed algorithm within
[123], [124], and wetland classification [125]. For instance, GEE could be effectively used for monitoring land covers in
AMANI et al.: GOOGLE EARTH ENGINE CLOUD COMPUTING PLATFORM FOR REMOTE SENSING BIG DATA APPLICATIONS 5339

a cost-efficient approach because open-access Landsat datasets every input data directly affects the final accuracy of studies,
are regularly inserted into GEE. In a different study, very high- image processing must be considered a necessity. Precision,
resolution oblique images were processed within GEE to detect level of automation, reliability, computational complexity, and
irregularity in façade and rooftop areas caused by hurricane time-consumption are the most critical criteria in developing
events [136]. First, a vertical building map was produced from image processing algorithms [149], [150]. Therefore, to ensure
a temporal analysis of predisaster images through an edge- high-quality results, it is inevitable to develop and enhance the
based/knowledge-based approach. Then, pre- and postdisaster existing image processing algorithms within GEE protocols.
images were fused in the data level followed by spectral-only In this regard, researchers have employed GEE to develop
and geospectral classifiers through the RF algorithm. The results various efficient and useful image processing algorithms, such
obtained a significant reduction in false-positive error. as cloud masking [12], [149], data selection and enhancement
[13], [150], image-based sensor calibration [151], [152], and
G. Atmosphere and Climate training sample migration [153]. For instance, Kong et al.
[150] introduced weighted Whittaker with a dynamic param-
As a principal component of the natural process of the Earth
eter (wWHd) denoising method within GEE to reconstruct the
system, land interacts with the atmosphere through biophysical
vegetation phenology based on 500 m MODIS EVI products.
and biochemical processes mutually [138]. Constant population
A large number of reference samples were used to compare
growth and human activities result in significant changes in the
the proposed method with four well-known denoising methods.
atmospheric constituents [139]. Climate change and air pollu-
The results, in terms of RMSE, roughness, and computational
tion are two decisive consequences of these disturbances that
efficiency revealed the superiority of the proposed method. Fur-
directly impact the surrounding environment and human health
thermore, Li et al. [13] developed an algorithm to improve GEE’s
[140], [141]. Therefore, it is essential to monitor and control
processing to efficiently acquire large-scale cloud-free Landsat
air quality and climate conditions to avoid severe outcomes.
images to support further applications. This method comprises
The availability of climate products accompanied by surface
cloud and shadow masking, snow/ice masking, and low-quality
products within GEE, make this platform a great tool for climate
pixels removal by incorporating the quality band. Therefore, this
studies and air quality monitoring. These advantages create a
method can efficiently prepare high-quality data for each region
rising interest in the research community to use GEE for air
of interest. It was discussed that their algorithm was developed
pollution analyses [142], [143], climate change and monitoring
within GEE, and the open-access codes within this platform
[144], [145], biophysical variable studies [21], [146], evapotran-
provided a simple framework with a flexible user-friendly inter-
spiration estimation [147], and precipitation mapping [148]. For
face. Finally, Kakooei et al. [154] proposed a global Sentinel-1
instance, GEE was employed to map exposed mine waste areas
foreshortening mask to improve the reliability of SAR-based
to estimate the corresponding emission of particulate matter to
analysis.
the atmosphere [142]. Four benchmark years of 1990, 2000,
2010, and 2018 as a part of Canada’s Air Pollutant Emission
Inventory were studied. Landsat-5, Landsat-8, Sentinel-1, and I. Pedosphere
Sentinel-2 satellite data were used to map exposed mine through
The Pedosphere is the outermost layer of Earth which dynam-
an RF algorithm. Finally, the authors reported that GEE was
ically interacts with the Biosphere and atmosphere [155]. Moni-
an invaluable platform for monitoring long-term emission from
toring and studying the Pedosphere and the corresponding cate-
exposed mine waste. Furthermore, GEE was used along with
gories (e.g., soil, geology, and geomorphology) are prerequisites
version 1 Tropical Rainfall Measuring Mission (TRMM) pre-
for sustainable development, especially in the climate modeling
cipitation products to study the spatial and temporal patterns
context [156]. Soil is the most significant component of the Pedo-
of precipitation in the Zambezi River basin [148]. To this end,
sphere that has straight impacts on the surrounding environment
TRMM data from 1998 to 2017 were processed in GEE to
and, thus, essential for biodiversity conservation and climate
investigate the precipitation trends and magnitudes by Kendall’s
regulations [157]–[160]. The availability of RS datasets in GEE
correlation and Sen’s slope reducer respectively. A “dry gets
makes it an appealing platform for the Pedosphere studies at
dryer, wet gets wetter” pattern was observed and reported in the
diverse scales. GEE was utilized for digital soil mapping [50],
study region.
[161], geology and mining [162], [163], geomorphology studies
[164], soil topography mapping [165], soil moisture derivation
H. Image Processing
[166], and soil carbon and salinity estimation [167], [168]. For
In the current era, almost all EO platforms are equipped example, Ivushkin et al. [168] applied Landsat-5 and Landsat-8
by digital sensors and, thus, terabytes of data are generated datasets within GEE to produce a global soil salinity map based
and stored in digital formats every day. As discussed, GEE on the thermal anomaly. They incorporated 15 188 reference
hosts an immense number of digital images. The RS images points from ISRIC-world soil information. Seven soil salinity
are extensively utilized in various applications and for different indicators of sand content, silt content, clay content, PH, bulk
purposes. Therefore, it is highly required to develop and en- density, organic carbon content, and cation exchange capacity
hance digital image processing algorithms to efficiently exploit with thermal anomaly were fed to the RF algorithm. The final
the potential of digital images. Moreover, since the quality of soil salinity map obtained overall accuracies of 67%–70% for
5340 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020

six different times. Moreover, GEE capabilities and Landsat A. Globe

imagery were combined to automatically delineate the annual
Long et al. [133] proposed an automatic method for producing
extent of surface coal mining in Central Appalachia between a global annual burned area maps using all available Landsat
1985 and 2015 [162]. To this end, the urban areas were masked
images acquired between 2014 and 2015 within the GEE cloud
using publicly available datasets and the mining zones were
computing platform. The map of the burn degree was first gen-
identified by low values in NDVI images. The proposed algo- erated using the RF classifier. Then, several logical filters (e.g.,
rithm achieved Kappa coefficients varying from 0.62 to 0.93 for
NDVI, Normalized Burned Ratio (NBR), and temporal filters)
different years.
were implemented to select candidate seeds of the burned area.
Finally, the global annual burned area map of 2015 (GABAM
J. Others 2015) was produced by employing an iterative seed-growing
process. A strong correlation (R2 = 0.74) was observed between
Other than previously mentioned applications, there are mul- the spatial distribution of the burned surfaces from the GABAM
tiple articles related to other applications of GEE, which were 2015 and the annual 250 m MODIS Vegetation Continuous
conducted with lower frequency. Therefore, their number was Fields (VCF) Collection 5.1 (MOD44B) product.
not enough to have a separate category and, thus, were as- Hansen et al. [8] analyzed forest cover changes at the global
signed to the Others applications category. These studies are scale between 2000 and 2012 using Landsat time-series images
mainly related to archaeology [169]–[171], 3-D printing [172], within GEE. Based on the results, the authors reported the
wildlife [173], [174], oil platform detection [175], and crashed following:
airplane detection [176]. For example, 300 Landsat-8 images 1) the tropical domain had the highest forest cover change
between 2013 and 2018 were processed in GEE to detect pos- (loss and gain) with annual deforestation rate of approxi-
sible crashed airplane in the Cambodian jungle [176]. NDVI, mately 2101 km2 /year;
albedo, thermal bands, spectral information, and panchromatic 2) most forests in the subtropical climate domain were con-
features were utilized in this study. Moreover, Sentinel-1 SAR sidered as croplands, because the existence of long-lived
data were used to automatically identify and delineate offshore natural forests in this domain was relatively rare;
oil platforms [175]. The proposed method was evaluated by 3) the trend of change in temperate forests was almost con-
1577 reference samples and obtained an overall accuracy of stant and had a low ratio of loss compared to gain;
96.09% over the Gulf of Mexico. Furthermore, GEE was re- 4) fire was the most important cause of deforestation in the
ported as a suitable platform to process high-resolution drone boreal domain;
imagery for pottery shreds identification [169]. In this regard, 5) the speed of deforestation in Brazil was more than other
texture and gradient features from RGB drone imagery were countries.
calculated within GEE and were ingested into the RF classifier. In [177], a grid-based Mountain Green Cover Index (MGCI)
The developed algorithm was able to identify pottery shreds was implemented to monitor mountain ecosystems at large
with 32.9% and 76.8% accuracies for two separate regions. scales. A novel frequency- and phenology-based technique was
Moreover, GEE was employed to process drone imagery to applied to generate the global green vegetation cover using all
estimate the wildlife aggregation population [173]. To this end, available Landsat-8 images within the GEE platform. Then,
the RF algorithm was applied to map targets of interest (bird nest) the real surface area generated from ASTER GDEM Version
pursuit using a predictive model to estimate the population. The 2 was applied to calculate the MGCI model instead of the
proposed approach obtained overall accuracies ranging from planimetric surface. The results showed that the generated data
86% to 96% over four different water bird colonies. Finally, had a high correlation (R2 = 0.9548) with FAO MGCI baseline
a web-application called TouchTerrian was developed to sim- data.
plify the 3-D terrain model printing [172]. After determining In [178], global surface water and its long-term changes were
the region of interest, the corresponding DEM was obtained mapped over three decades of Landsat satellite images (three
through GEE to be used for final 3-D printing. It was re- million images) within the GEE platform. The result of this
ported that users with any level of expertise could easily utilize global assessment demonstrated the following:
their model within GEE with minimum computing resources 1) permanent water bodies disappeared by approximately 90
requirements. 000 km2 and new water bodies covering 184 000 km2
formed between 1984 and 2015;
2) the permanent net water of all continental regions in-
VIII. GEE LARGE-SCALE CASE STUDIES creased except for Oceania;
As discussed, the enormous capabilities of GEE resolve the 3) over 70% of global net permanent water loss occurred
existing challenges of processing big data over large-scale areas. in the Middle East and Central Asia due to drought and
Therefore, GEE has been recognized as an efficient platform for human actions (e.g., river damming).
regional to global LC mapping and monitoring over long periods It is finally argued that the proposed strategy within GEE can
of time. In this section, ten studies conducted over the globe, be effectively used for water resources management.
continents, and big countries (e.g., the United States, Canada, Scherler et al. [90] proposed a novel automatic method to map
and China) are discussed in detail. supraglacial debris cover over the globe using multitemporal
AMANI et al.: GOOGLE EARTH ENGINE CLOUD COMPUTING PLATFORM FOR REMOTE SENSING BIG DATA APPLICATIONS 5341

optical satellite images within GEE. In this study, debris-covered Goldblatt et al. [183] used GEE for temporal analysis of large
ice surfaces were generated by thresholding of three indices, urban areas in India using multitemporal Landsat-7 and Landsat-
including red to Shortwave Infrared band ratio, the Normalized 8 images. In order to generate high-quality maps of built-up
Difference Snow Index, and linear spectral unmixing-derived areas, the country was classified into the built-up and non-built-
Fractional Debris Cover. These indices were generated based on up regions using 21 030 training datasets and three types of
Landsat-8 and Sentinel-2 optical satellite images in 19 glacier supervised classification algorithms (i.e., SVM, CART, and RF).
areas at the world-scale from 2013 to 2015. The results showed It was reported that the proposed GEE approach generated a
that 4.4% (about 26 000 km²) of all glacier areas is affected with high-quality map of built-up areas in India and can be potentially
debris. Furthermore, an inverse relationship between glacier employed in other countries.
size and percentage of debris was also reported, indicating
continuous shrinking glaciers due to the debris effects.
IX. CONCLUSION
The proliferation of big geo data and the recent advance in
B. Continent and Big Countries cloud computing and big data processing services are changing
Amani et al. [1] produced the first Canadian wetland inventory the future of RS. In this regard, GEE is effectively paving the road
(CWI) map using Landsat-8 imagery and several advanced for researchers, scientists, and developers to be able to easily
algorithms available within GEE. In this study, 30 000 scenes extract valuable information from big RS datasets without the
of Landsat-8 images were used along with machine learning burdens of traditional data analysis methods. The massive troves
algorithms in GEE. The RF algorithm was applied to classify of RS datasets available with GEE (e.g., archived Landsat and
wetlands over the entire Canada. The CWI map was based on Sentinel images) helps researchers to address global challenges
five wetland classes, defined by the Canadian Wetland Classifi- and environmental issues, such as global warming, climate
cation System: bog, fen, marsh, swamp, and shallow water. The change, LCLU classification over large areas, and monitoring
quantity and quality of the results showed that the generated landscape over several decades. GEE also contains hundreds of
CWI map had reasonable accuracy considering the challenges prebuilt functions which can be easily understood and utilized by
existing over this immense country (9.985 million km2 ). different users. Through a basic knowledge of JavaScript, users
Li et al. [179] generated African LCLU map at a 10 m can also implement their own algorithms. These advantages
resolution within GEE using multisource RS datasets, including make any user employ this cloud computing platform for various
Sentinel-2, Landsat-8, Global Human Settlement Layer, Night applications related to LCLU, agriculture, hydrology, natural
Time Light data, Shuttle Radar Topography Mission (SRTM), disaster, etc. Besides all the advantages, it also has several
and MODIS Land Surface Temperature images. The RF al- limitations, such as limited storage of 250 GBs for each user
gorithm was applied to classify the area into five categories and limited memory to train machine learning algorithms, which
of urban, trees, low plants, bare soil, and water. The results may push a new user backward. However, it is undeniable that
showed that the LCLU map generated by this method had a better GEE presents a novel way of processing geospatial data and
performance than that of the FROM-GLC10 [180] in detecting resolves several big data challenges existed for RS researchers.
urban class and distinguishing trees from low plants in rural Based on the GEE publication trends, it is also clear that this
areas. platform is becoming more popular not only among the RS
Beresford et al. [181] developed an NRT monitoring frame- researchers but also within any community interested in using
work for conservation of the Key Biodiversity Areas (KBAs) in EO datasets.
Africa using the GEE platform. In this study, simple repeatable
techniques were proposed to detect changes in fire rate, tree
AUTHORS’ CONTRIBUTION
loss, and nighttime lights between 1992 and 2013. The results
showed that fire rate, nighttime lights, and rate of forest loss Meisam Amani designed and supervised the study, profes-
considerably increased in KBAs and ecoregions. Moreover, the sionally optimized all sections, acquired funding, wrote the
authors argued that the method implemented within GEE has Abstract, Introduction, and Conclusion; Arsalan Ghorbanian
a high potential for monitoring changes over any geographic wrote the “GEE Applications” section; Seyed Ali Ahmadi
area and using different RS data types and could be effectively wrote the “GEE Advantages and Limitations” section; Moham-
utilized by conservation end-users. mad Kakooei wrote the “GEE Datasets” and “GEE Functions”
Teluguntla et al. [182] developed a precise Landsat-based sections; Armin Moghimi wrote the “GEE Large-scale Case
cropland extent product over Australia and China using machine Studies” section; Seyed Mohammad Mirmazloumi gathered the
learning tools in GEE. In this study, cropland maps were pro- required articles and wrote the “GEE Pattern of Publication”
duced by applying RF to Landsat-8 images. The RF classifier section; Sayyed Hamed Alizadeh Moghaddam wrote the “GEE
was trained and validated using ground truth data obtained from Platform Overview” section; Sahel Mahdavi professionally op-
different resources, such as field surveys, very high spatial res- timized all sections; Masoud Ghahremanloo helped in initial
olution (5 m) imagery, and several other auxiliary information. literature review; Saeid Parsian helped in gathering the required
Based on their results, the total cropland areas of Australia articles; Qiusheng Wu and Brian Brisco professionally opti-
and China were estimated as 35.1 and 165.2 million hectares, mized all sections. Finally, all authors read and approved the
respectively. final manuscript.
5342 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020

APPENDIX

TABLE IV
LIST OF AVAILABLE DATASETS WITHIN GEE
AMANI et al.: GOOGLE EARTH ENGINE CLOUD COMPUTING PLATFORM FOR REMOTE SENSING BIG DATA APPLICATIONS 5343

TABLE IV
CONTINUED
5344 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020

TABLE IV
CONTINUED
AMANI et al.: GOOGLE EARTH ENGINE CLOUD COMPUTING PLATFORM FOR REMOTE SENSING BIG DATA APPLICATIONS 5345

REFERENCES [24] A. Moghimi, S. Khazai, and A. Mohammadzadeh, “An improved fast

level set method initialized with a combination of k-means clustering and
[1] M. Amani et al., “Canadian wetland inventory using Google Earth Otsu thresholding for unsupervised change detection from SAR images,”
Engine: The first map and preliminary results,” Remote Sens., vol. 11, Arab. J. Geosci., vol. 10, no. 13, p. 293, Jul. 2017.
no. 7, Apr. 2019, Art. no. 842. [25] T. Calinski and J. Harabasz, “A dendrite method for cluster analysis,”
[2] H. Tamiminia, B. Salehi, M. Mahdianpari, L. Quackenbush, S. Adeli, Commun. Statist., Theory Methods, vol. 3, no. 1, pp. 1–27, 1974.
and B. Brisco, “Google Earth Engine for geo-big data applications: A [26] A. W. M. Dan Pelleg, “X-means: Extending k-means with efficient
meta-analysis and systematic review,” ISPRS J. Photogrammetry Remote estimation of the number of clusters,” in Proc. 17th Int. Conf. Mach.
Sens., vol. 164, pp. 152–170, Jun. 2020. Learning, 2000, pp. 727–734.
[3] M. Chi, A. Plaza, J. A. Benediktsson, Z. Sun, J. Shen, and [27] J. H. Gennari, P. Langley, and D. Fisher, “Models of incremental concept
Y. Zhu, “Big data for remote sensing: Challenges and opportunities,” formation,” Artif. Intell., vol. 40, nos. 1–3, pp. 11–61, Sep. 1989.
Proc. IEEE, vol. 104, no. 11, pp. 2207–2219, Nov. 2016. [28] R. Achanta and S. Susstrunk, “Superpixels and polygons using simple
[4] Y. Ma et al., “Remote sensing big data computing: Challenges and op- non-iterative clustering,” in Proc. IEEE Conf. Comput. Vis. Pattern
portunities,” Future Gener. Comput. Syst., vol. 51, pp. 47–60, Oct. 2015. Recognit., Jul. 2017, pp. 4895–4904.
[5] B. Wilder, Cloud Architecture Patterns: Using Microsoft Azure. Newton, [29] M. Mahdianpari, B. Salehi, F. Mohammadimanesh, S. Homayouni, and
MA, USA: O’Reilly Media, 2012. E. Gill, “The first wetland inventory map of Newfoundland at a spatial
[6] N. Gorelick, M. Hancher, M. Dixon, S. Ilyushchenko, D. Thau, and R. resolution of 10 m using Sentinel-1 and Sentinel-2 data on the Google
Moore, “Google earth engine: Planetary-scale geospatial analysis for Earth Engine cloud computing platform,” Remote Sens., vol. 11, no. 1,
everyone,” Remote Sens. Environ., vol. 202, pp. 18–27, Dec. 2017. Dec. 2018, Art. no. 43.
[7] J. Hird, E. DeLancey, G. McDermid, and J. Kariyeva, “Google Earth [30] Z. Zhu and C. E. Woodcock, “Continuous change detection and clas-
Engine, open-access satellite data, and machine learning in support of sification of land cover using all available Landsat data,” Remote Sens.
large-area probabilistic wetland mapping,” Remote Sens., vol. 9, no. 12, Environ., vol. 144, pp. 152–171, Mar. 2014.
Dec. 2017, Art. no. 1315. [31] E. B. Brooks, R. H. Wynne, V. A. Thomas, C. E. Blinn, and J. W. Coulston,
[8] M. C. Hansen et al., “High-resolution global maps of 21st-century forest “On-the-Fly massively multitemporal change detection using statistical
cover change,” Science, vol. 342, no. 6160, pp. 850–853, Nov. 2013. quality control charts and Landsat data,” IEEE Trans. Geosci. Remote
[9] L. Kumar and O. Mutanga, “Google Earth Engine applications since Sens., vol. 52, no. 6, pp. 3316–3332, Jun. 2014.
inception: Usage, trends, and potential,” Remote Sens., vol. 10, no. 10, [32] R. E. Kennedy, Z. Yang, and W. B. Cohen, “Detecting trends in forest
Sep. 2018, Art. no. 1509. disturbance and recovery using yearly Landsat time series: 1. LandTrendr
[10] O. Mutanga and L. Kumar, “Google Earth Engine applications,” Remote — Temporal segmentation algorithms,” Remote Sens. Environ., vol. 114,
Sens., vol. 11, no. 5, pp. 11–14, 2019. no. 12, pp. 2897–2910, Dec. 2010.
[11] L. Bi, B. L. Fu, P. Q. Lou, and T. Y. Tang, “Delineation water of pearl [33] C. Huang, S. N. Goward, J. G. Masek, N. Thomas, Z. Zhu, and J. E.
river basin using Landsat images from Google Earth Engine,” ISPRS, Int. Vogelmann, “An automated approach for reconstructing recent forest
Arch. Photogrammetry Remote Sens. Spatial Inf. Sci., vol. XLII-3/W10, disturbance history using dense Landsat time series stacks,” Remote Sens.
pp. 5–10, Feb. 2020, pp. 5–10. Environ., vol. 114, no. 1, pp. 183–198, Jan. 2010.
[12] G. Mateo-García, L. Gómez-Chova, J. Amorós-López, J. Muñoz-Marí, [34] M. Hughes, S. Kaylor, and D. Hayes, “Patch-based forest change detec-
and G. Camps-Valls, “Multitemporal cloud masking in the Google Earth tion from Landsat time series,” Forests, vol. 8, no. 5, May 2017, Art.
Engine,” Remote Sens., vol. 10, no. 7, Jul. 2018, Art. no. 1079. no. 166.
[13] H. Li et al., “A Google Earth Engine-enabled software for efficiently [35] A. M. Shew and A. Ghosh, “Using multi-temporal remote sensing data
generating high-quality user-ready Landsat mosaic images,” Environ. to analyze the spatio-temporal patterns of dry season rice production in
Model. Softw., vol. 112, pp. 16–22, Feb. 2019. Bangladesh,” ISPRS Ann. Photogrammetry Remote Sens. Spatial Inf. Sci.,
[14] S. Xie, L. Liu, X. Zhang, J. Yang, X. Chen, and Y. Gao, “Automatic land- vol. IV-4/W2, pp. 61–68, Oct. 2017.
cover mapping using Landsat time-series data based on Google Earth [36] D. Parastatidis, Z. Mitraka, N. Chrysoulakis, and M. Abrams, “Online
Engine,” Remote Sens., vol. 11, no. 24, Dec. 2019, Art. no. 3023. global land surface temperature estimation from Landsat,” Remote Sens.,
[15] J. Lee, J. Cardille, and M. Coe, “BULC-U: Sharpening resolution and vol. 9, no. 12, Nov. 2017, Art. no. 1208.
improving accuracy of land-use/land-cover classifications in Google [37] Z. Yang, W. Li, Q. Chen, S. Wu, S. Liu, and J. Gong, “A scalable
Earth Engine,” Remote Sens., vol. 10, no. 9, Sep. 2018, Art. no. 1455. cyberinfrastructure and cloud computing platform for forest aboveground
[16] M. Amani et al., “A generalized supervised classification scheme to biomass estimation based on the Google Earth Engine,” Int. J. Digit.
produce provincial wetland inventory maps: An application of Google Earth, vol. 12, no. 9, pp. 995–1012, Sep. 2019.
Earth Engine for big geo data processing,” Big Earth Data, vol. 3, no. 4, [38] H. Xu and Y. Bai, “Evaluation of urban lake evolution using Google
pp. 378–394, Oct. 2019. Earth Engine—A case study of Wuhan, China,” in Proc. 4th Int. Conf.
[17] R. Ravanelli et al., “Monitoring the impact of land cover change on Agro-Geoinformat., Jul. 2015, pp. 322–325.
surface urban heat island through Google Earth Engine: Proposal of [39] I. Fuentes, J. Padarian, F. van Ogtrop, and R. W. Vervoort, “Comparison
a global methodology, first applications and problems,” Remote Sens., of surface water volume estimation methodologies that couple surface re-
vol. 10, no. 9, Sep. 2018, Art. no. 1488. flectance data and digital terrain models,” Water, vol. 11, no. 4, Apr. 2019,
[18] P. Gong et al., “Annual maps of global artificial impervious area (GAIA) Art. no. 780.
between 1985 and 2018,” Remote Sens. Environ., vol. 236, Jan. 2020, [40] S. S. Ray, “Exploring machine learning classification algorithms for
Art. no. 111510. crop classification using Sentinel 2 data,” ISPRS, Int. Arch. Photogram-
[19] D. Mandal, V. Kumar, A. Bhattacharya, Y. S. Rao, P. Siqueira, and S. Bera, metry Remote Sens. Spatial Inf. Sci., vol. XLII-3/W6, pp. 573–578,
“Sen4Rice: A processing chain for differentiating early and late trans- Jul. 2019.
planted rice using time-series Sentinel-1 SAR data with Google Earth [41] “Python vs. Javascript APIs.” [Online]. Available: https://developers.
Engine,” IEEE Geosci. Remote Sens. Lett., vol. 15, no. 12, pp. 1947–1951, google.com/earth-engine/python_install. Accesed: Jan. 2020.
Dec. 2018. [42] “GEE developers forum.” [Online]. Available: https://developers.google.
[20] D. Traganos, D. Poursanidis, B. Aggarwal, N. Chrysoulakis, and P. com/earth-engine/
Reinartz, “Estimating satellite-derived bathymetry (SDB) with the google [43] “GEE Python API Docs.” [Online]. Available: ttps://colab.research.
earth engine and Sentinel-2,” Remote Sens., vol. 10, no. 6, Jun. 2018, Art. google.com/github/csaybar/EEwPython/blob/dev/1_Introduction.
no. 859. ipynb#scrollTo=9CRoDhQNiHC8, Accesed: Jan. 2020.
[21] M. Campos-Taberner et al., “Global estimation of biophysical variables [44] A. F. Sunar, N. Yagmur, and A. Dervisoglu, “Flood analysis with re-
from Google Earth Engine platform,” Remote Sens., vol. 10, no. 8, mote sensing data – A case study: Maritsa river, Edirne,” ISPRS, Int.
Jul. 2018, Art. no. 1167. Arch. Photogrammetry Remote Sens. Spatial Inf. Sci., vol. XLII-3/W8,
[22] R. Xu and D. WunschII, “Survey of clustering algorithms,” IEEE Trans. pp. 497–502, Aug. 2019.
Neural Netw., vol. 16, no. 3, pp. 645–678, May 2005. [45] E. Çolak, M. Chandra, and F. Sunar, “The use of multi-temporal Sentinel
[23] A. Ghorbanian and A. Mohammadzadeh, “An unsupervised feature satellites in the analysis of land cover/land use changes caused by
extraction method based on band correlation clustering for hyperspectral the nuclear power plant construction,” ISPRS, Int. Arch. Photogram-
image classification using limited training samples,” Remote Sens. Lett., metry Remote Sens. Spatial Inf. Sci., vol. XLII-3/W8, pp. 491–495,
vol. 9, no. 10, pp. 982–991, Oct. 2018. Aug. 2019.
5346 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020

[46] M. Di Tullio, F. Nocchi, A. Camplani, N. Emanuelli, A. Nascetti, and M. [68] K. Shimizu, T. Ota, and N. Mizoue, “Detecting forest changes using dense
Crespi, “Copernicus big data and Google Earth Engine for glacier surface Landsat 8 and Sentinel-1 time series data in tropical seasonal forests,”
velocity field monitoring: Feasibility demonstration on San Rafael and Remote Sens., vol. 11, no. 16, Aug. 2019, Art. no. 1899.
San Quintin glaciers,” ISPRS, Int. Arch. Photogrammetry Remote Sens. [69] L. Parente, V. Mesquita, F. Miziara, L. Baumann, and L. Ferreira, “As-
Spatial Inf. Sci., vol. XLII–3, pp. 289–294, Apr. 2018. sessing the pasturelands and livestock dynamics in Brazil, from 1985
[47] TFrecord and Earth Engine. [Online]. Available: https://developers. to 2017: A novel approach based on high spatial resolution imagery and
google.com/earth-engine/tfrecord, Accesed: Jan. 2020. Google Earth Engine cloud computing,” Remote Sens. Environ., vol. 232,
[48] E. R. DeLancey, J. F. Simms, M. Mahdianpari, B. Brisco, C. Mahoney, Oct. 2019, Art. no. 111301.
and J. Kariyeva, “Comparing deep learning and shallow learning for [70] M. O. Jones et al., “Innovation in rangeland monitoring: annual, 30 m,
large-scale wetland classification in Alberta, Canada,” Remote Sens., plant functional type percent cover maps for U.S. rangelands, 1984-
vol. 12, no. 1, Dec. 2019, Art. no. 2. 2017,” Ecosphere, vol. 9, no. 9, Sep. 2018, Art. no. e02430.
[49] L. Parente, E. Taquary, A. Silva, C. Souza, and L. Ferreira, “Next [71] V. Jansen, C. Kolden, and H. Schmalz, “The development of near real-
generation mapping: Combining deep learning, cloud computing, and time biomass and cover estimates for adaptive rangeland management
big remote sensing data,” Remote Sens., vol. 11, no. 23, Dec. 2019, Art. using Landsat 7 and Landsat 8 surface reflectance products,” Remote
no. 2881. Sens., vol. 10, no. 7, Jul. 2018, Art. no. 1057.
[50] J. Padarian, B. Minasny, and A. B. McBratney, “Using Google’s cloud- [72] S. Xu, G. Li, and Z. Li, “China agricultural outlook for 2015–2024 based
based platform for digital soil mapping,” Comput. Geosci., vol. 83, on China Agricultural Monitoring and Early-warning System (CAMES),”
pp. 80–88, Oct. 2015. J. Integrative Agriculture, vol. 14, no. 9, pp. 1889–1902, Sep. 2015.
[51] “Sentinel-1 algorithms.” [Online]. Available: https://developers.google. [73] J. Schindler, F. Graef, H. J. König, and D. Mchau, “Develop-
com/earth-engine/sentinel1 ing community-based food security criteria in rural Tanzania,” Food
[52] F. Keller, J. Sänger, T. Kersten, and J. Schiewe, “Historical 4D city model Secur., vol. 9, no. 6, pp. 1285–1298, Dec. 2017.
of the free and Hanseatic city of Hamburg – automated generation and [74] J. W. McArthur and J. D. Sachs, “Agriculture, aid, and economic growth
presentation within the Google Earth Engine,” Photogrammtrie Fern- in Africa,” World Bank Econ. Rev., vol. 33, no. 1, pp. 1–20, Feb. 2019.
erkundung, Geoinf., vol. 2011, no. 3, pp. 155–169, Jun. 2011. [75] P. Han, J. Chen, Y. Han, L. Yi, Y. Zhang, and X. Jiang, “Monitoring rubber
[53] M. Amani and M. R. Mobasheri, “A parametric method for estimation plantation distribution on Hainan Island using Landsat OLI imagery,” Int.
of leaf area index using Landsat ETM+ data,” GIScience Remote Sens., J. Remote Sens., vol. 39, no. 8, pp. 2189–2206, Apr. 2018.
vol. 52, no. 4, pp. 478–497, Jul. 2015. [76] A. Poortinga et al., “Mapping plantations in Myanmar by fusing Landsat-
[54] B. Smith et al., “Implications of incorporating N cycling and N limitations 8, Sentinel-2 and Sentinel-1 data along with systematic error quantifica-
on primary production in an individual-based dynamic vegetation model,” tion,” Remote Sens., vol. 11, no. 7, Apr. 2019, Art. no. 831.
Biogeosciences, vol. 11, no. 7, pp. 2027–2054, Apr. 2014. [77] H. Tian, N. Huang, Z. Niu, Y. Qin, J. Pei, and J. Wang, “Mapping winter
[55] T. G. Workie and H. J. Debella, “Climate change and its effects on vegeta- crops in China with multi-source satellite imagery and phenology-based
tion phenology across ecoregions of Ethiopia,” Glob. Ecol. Conservation, algorithm,” Remote Sens., vol. 11, no. 7, Apr. 2019, Art. no. 820.
vol. 13, Jan. 2018, Art. no. e00366. [78] G. Ghazaryan et al., “A rule-based approach for crop identification using
[56] S. Besnard et al., “Memory effects of climate and vegetation affecting multi-temporal and multi-sensor phenological metrics,” Eur. J. Remote
net ecosystem CO2 fluxes in global forests,” PLoS One, vol. 14, no. 2, Sens., vol. 51, no. 1, pp. 511–524, Jan. 2018.
Feb. 2019, Art. no. e0211510. [79] L. Liang, B. R. K. Runkle, B. B. Sapkota, and M. L. Reba, “Automated
[57] B. G. Peter and J. P. Messina, “Errors in time-series remote sensing and an mapping of rice fields using multi-year training sample normalization,”
open access application for detecting and visualizing spatial data outliers Int. J. Remote Sens., vol. 40, no. 18, pp. 7252–7271, Sep. 2019.
using Google Earth Engine,” IEEE J. Sel. Topics Appl. Earth Obs. Remote [80] S. Wang, G. Azzari, and D. B. Lobell, “Crop type mapping without
Sens., vol. 12, no. 4, pp. 1165–1174, Apr. 2019. field-level labels: Random forest transfer and unsupervised clustering
[58] M. Zhang, P. Gong, S. Qi, C. Liu, and T. Xiong, “Mapping bamboo with techniques,” Remote Sens. Environ., vol. 222, pp. 303–317, Mar. 2019.
regional phenological characteristics derived from dense Landsat time [81] F. Rembold et al., “ASAP: A new global early warning system to detect
series using Google Earth Engine,” Int. J. Remote Sens., vol. 40, no. 24, anomaly hot spots of agricultural production for food security analysis,”
pp. 9541–9555, Dec. 2019. Agricultural Syst., vol. 168, pp. 247–257, Jan. 2019.
[59] Z. Xie et al., “Using Landsat observations (1988–2017) and Google [82] M. Rudiyanto, S. Minasny, R. M. Shah, N. C. Soh, C. Arif, and B. I.
Earth Engine to detect vegetation cover changes in rangelands—A first Setiawan, “Automated near-real-time mapping and monitoring of rice
step towards identifying degraded lands for conservation,” Remote Sens. extent, cropping patterns, and growth stages in Southeast Asia using
Environ., vol. 232, Oct. 2019, Art. no. 111317. Sentinel-1 time series on a Google Earth Engine platform,” Remote Sens.,
[60] J. Pei et al., “Time series of Landsat imagery shows vegetation recovery vol. 11, no. 14, Jul. 2019, Art. no. 1666.
in two fragile karst watersheds in Southwest China from 1988 to 2016,” [83] Z. Jin et al., “Smallholder maize area and yield mapping at national
Remote Sens., vol. 11, no. 17, Aug. 2019, Art. no. 2044. scales with Google Earth Engine,” Remote Sens. Environ., vol. 228,
[61] C. D. Mendenhall and A. M. Wrona, “Improving tree cover estimates pp. 115–128, Jul. 2019.
for fine-scale landscape ecology,” Landscape Ecol., vol. 33, no. 10, [84] D. B. Lobell, D. Thau, C. Seifert, E. Engle, and B. Little, “A scalable
pp. 1691–1696, Oct. 2018. satellite-based crop yield mapper,” Remote Sens. Environ., vol. 164,
[62] E. L. Bullock, C. E. Woodcock, and P. Olofsson, “Monitoring tropical pp. 324–333, Jul. 2015.
forest degradation using spectral unmixing and Landsat time series [85] J. M. Deines, A. D. Kendall, and D. W. Hyndman, “Annual irrigation
analysis,” Remote Sens. Environ., vol. 238, Mar. 2020, Art. no. 110968. dynamics in the U.S. Northern High Plains derived from Landsat satellite
[63] F. Ramdani, “Recent expansion of oil palm plantation in the most eastern data,” Geophys. Res. Lett., vol. 44, no. 18, pp. 9350–9360, Sep. 2017.
part of Indonesia: feature extraction with polarimetric SAR,” Int. J. [86] S. Ragettli, T. Herberz, and T. Siegfried, “An unsupervised classification
Remote Sens., vol. 40, no. 19, pp. 7371–7388, Oct. 2019. algorithm for multi-temporal irrigated area mapping in central Asia,”
[64] K. Anderson, D. Fawcett, A. Cugulliere, S. Benford, D. Jones, and R. Remote Sens., vol. 10, no. 11, Nov. 2018, Art. no. 1823.
Leng, “Vegetation expansion in the subnival Hindu Kush Himalaya,” [87] S. G. Yalew, A. van Griensven, and P. van der Zaag, “AgriSuit:
Glob. Change Biol., vol. 26, no. 3, pp. 1608–1625, Mar. 2020. A web-based GIS-MCDA framework for agricultural land suitabil-
[65] B. Cao, G. M. Domke, M. B. Russell, and B. F. Walters, “Spatial modeling ity assessment,” Comput. Electron. Agriculture, vol. 128, pp. 1–8,
of litter and soil carbon stocks on forest land in the conterminous United Oct. 2016.
States,” Sci. Total Environ., vol. 654, pp. 94–106, Mar. 2019. [88] N. You and J. Dong, “Examining earliest identifiable timing of crops using
[66] J. Pérez-Romero, R. M. Navarro-Cerrillo, G. Palacios-Rodriguez, C. all available Sentinel 1/2 imagery and Google Earth Engine,” ISPRS J.
Acosta, and F. J. Mesas-Carrascosa, “Improvement of remote sensing- Photogrammetry Remote Sens., vol. 161, pp. 109–123, Mar. 2020.
based assessment of defoliation of Pinus spp. caused by Thaume- [89] F. Yao, J. Wang, C. Wang, and J.-F. Crétaux, “Constructing long-
topoea Pityocampa Denis and Schiffermüller and related environmental term high-frequency time series of global lake and reservoir areas
drivers in Southeastern Spain,” Remote Sens., vol. 11, no. 14, Jul. 2019, using Landsat imagery,” Remote Sens. Environ., vol. 232, Oct. 2019,
Art. no. 1736. Art. no. 111210.
[67] C. Wei, D. N. Karger, and A. M. Wilson, “Spatial detection of alpine [90] D. Scherler, H. Wulf, and N. Gorelick, “Global assessment of
treeline ecotones in the Western United States,” Remote Sens. Environ., supraglacial debris-cover extents,” Geophys. Res. Lett., vol. 45, no. 21,
vol. 240, Apr. 2020, Art. no. 111672. pp. 11798–11805, Nov. 2018.
AMANI et al.: GOOGLE EARTH ENGINE CLOUD COMPUTING PLATFORM FOR REMOTE SENSING BIG DATA APPLICATIONS 5347

[91] T. Tedesche, E. D. Trochin, S. R. Fassnacht, and G. J. Wolken, “Extent [113] A. Middel, J. Lukasczyk, R. Maciejewski, M. Demuzere, and M. Roth,
changes in the perennial snowfields of gates of the Arctic national park “Sky view factor footprints for urban climate modeling,” Urban Climate,
and preserve, Alaska,” Hydrology, vol. 6, no. 2, Jun. 2019, Art. no. 53. vol. 25, pp. 120–134, Sep. 2018.
[92] B. Snapir, A. Momblanch, S. K. Jain, T. W. Waine, and I. P. Holman, [114] Q. Duan, M. Tan, Y. Guo, X. Wang, and L. Xin, “Understanding the spatial
“A method for monthly mapping of wet and dry snow using Sentinel-1 distribution of urban forests in china using Sentinel-2 images with google
and MODIS: Application to a Himalayan river basin,” Int. J. Appl. Earth earth engine,” Forests, vol. 10, no. 9, Aug. 2019, Art. no. 729.
Observ. Geoinf., vol. 74, pp. 222–230, Feb. 2019. [115] C. Huang, J. Yang, and P. Jiang, “Assessing impacts of urban form on
[93] U. N. T. Nguyen, L. T. H. Pham, and T. D. Dang, “An automatic water landscape structure of urban green spaces in China using Landsat images
detection approach using Landsat 8 OLI and Google Earth Engine cloud based on Google Earth Engine,” Remote Sens., vol. 10, no. 10, Oct. 2018,
computing to map lakes and reservoirs in New Zealand,” Environmental Art. no. 1569.
Monitoring Assessment, vol. 191, no. 4, Apr. 2019, Art. no. 235. [116] A. G. Auffret, A. Kimberley, J. Plue, and E. Waldén, “Super-regional
[94] Y. Zhou et al., “Continuous monitoring of lake dynamics on the Mon- land-use change and effects on the grassland specialist flora,” Nature
golian Plateau using all available Landsat imagery and Google Earth Commun., vol. 9, no. 1, Dec. 2018, Art. no. 3464.
Engine,” Sci. Total Environ., vol. 689, pp. 366–380, Nov. 2019. [117] S. Mahdavi, B. Salehi, J. Granger, M. Amani, B. Brisco, and W. Huang,
[95] T. Sagawa, Y. Yamashita, T. Okumura, and T. Yamanokuchi, “Satellite “Remote sensing for wetland classification: A comprehensive review,”
Derived Bathymetry using machine learning and multi-temporal satellite GIScience Remote Sens., vol. 55, no. 5, pp. 623–658, Sep. 2018.
images,” Remote Sens., vol. 11, no. 10, May 2019, Art. no. 1155. [118] A. Midekisa et al., “Mapping land cover change over continental Africa
[96] K. Vos, M. D. Harley, K. D. Splinter, J. A. Simmons, and I. L. using Landsat and Google Earth Engine cloud computing,” PLoS One,
Turner, “Sub-annual to multi-decadal shoreline variability from pub- vol. 12, no. 9, Sep. 2017, Art. no. e0184926.
licly available satellite imagery,” Coastal Eng., vol. 150, pp. 160–174, [119] N. J. Murray, D. A. Keith, D. Simpson, J. H. Wilshire, and R. M. Lucas,
Aug. 2019. “Remap: An online remote sensing application for land cover classifica-
[97] W. Cao, Y. Zhou, R. Li, and X. Li, “Mapping changes in coastlines and tion and monitoring,” Methods Ecol. Evol., vol. 9, no. 9, pp. 2019–2027,
tidal flats in developing islands using the full time series of Landsat Sep. 2018.
images,” Remote Sens. Environ., vol. 239, Mar. 2020, Art. no. 111665. [120] J. Miettinen, C. Shi, and S. C. Liew, “Towards automated 10–30 m
[98] F. Chen, M. Zhang, B. Tian, and Z. Li, “Extraction of glacial lake outlines resolution land cover mapping in insular South-East Asia,” Geocarto
in Tibet plateau using Landsat 8 imagery and Google Earth Engine,” Int., vol. 34, no. 4, pp. 443–457, Mar. 2019.
IEEE J. Sel. Topics Appl. Earth Obs. Remote Sens., vol. 10, no. 9, [121] Y. Ge et al., “Mapping annual land use changes in China’s poverty-
pp. 4002–4009, Sep. 2017. stricken areas from 2013 to 2018,” Remote Sens. Environ., vol. 232,
[99] S. Murphy, R. Wright, and D. Rouwet, “Color and temperature of the Oct. 2019, Art. no. 111285.
crater lakes at Kelimutu volcano through time,” Bull. Volcanol., vol. 80, [122] Y. Zhou et al., “Are there sufficient Landsat observations for retrospective
no. 1, Jan. 2018, Art. no. 2. and continuous monitoring of land cover changes in China?,” Remote
[100] M. Zhang, F. Chen, and B. Tian, “An automated method for glacial Sens., vol. 11, no. 15, Aug. 2019, Art. no. 1808.
lake mapping in High Mountain Asia using Landsat 8 imagery,” J. [123] X. Wang et al., “Tracking annual changes of coastal tidal flats in China
Mountain Sci., vol. 15, no. 1, pp. 13–24, Jan. 2018. during 1986–2016 through analyses of Landsat images with Google Earth
[101] N. E. Wayand, C. B. Marsh, J. M. Shea, and J. W. Pomeroy, “Glob- Engine,” Remote Sens. Environ., vol. 238, Mar. 2020, Art. no. 110987.
ally scalable alpine snow metrics,” Remote Sens. Environ., vol. 213, [124] G. Muthusankar et al., “When socio-economic plans exacerbate vulner-
pp. 61–72, Aug. 2018. ability to physical coastal processes on the South East coast of India,” J.
[102] C. G. Griffin, J. W. McClelland, K. E. Frey, G. Fiske, and R. M. Holmes, Coastal Res., vol. 85, pp. 1446–1450, May 2018.
“Quantifying CDOM and DOC in major Arctic rivers during ice-free [125] Q. Wu et al., “Integrating LiDAR data and multi-temporal aerial im-
conditions using Landsat TM and ETM+ data,” Remote Sens. Environ., agery to map wetland inundation dynamics using Google Earth Engine,”
vol. 209, pp. 395–409, May 2018. Remote Sens. Environ., vol. 228, pp. 1–13, Jul. 2019.
[103] Z. Cao, R. Ma, H. Duan, K. Xue, and M. Shen, “Effect of satellite temporal [126] A. Ghorbanian, M. Kakooei, M. Amani, S. Mahdavi, A. Moham-
resolution on long-term suspended particulate matter in inland lakes,” madzadeh, and M. Hasanlou, “Improved land cover map of Iran using
Remote Sens., vol. 11, no. 23, Nov. 2019, Art. no. 2785. Sentinel imagery within Google Earth Engine and a novel automatic
[104] E. Medina-Lopez and L. Ureña-Fuentes, “High-resolution sea surface workflow for land cover classification using migrated training sam-
temperature and salinity in the global ocean from raw satellite data,” ples,” ISPRS J. Photogrammetry Remote Sens., vol. 167, pp. 276–288,
Remote Sens., vol. 11, no. 19, Sep. 2019, Art. no. 2191. Sep. 2020.
[105] S. Lin, L. N. Novitski, J. Qi, and R. J. Stevenson, “Landsat TM/ETM+ [127] Y. Bayissa, T. Tadesse, G. Demisse, and A. Shiferaw, “Evaluation of
and machine-learning algorithms for limnological studies and algal satellite-based rainfall estimates and application to monitor meteorolog-
bloom management of inland lakes,” J. Appl. Remote Sens., vol. 12, ical drought for the Upper Blue Nile Basin, Ethiopia,” Remote Sens.,
no. 02, Apr. 2018, Art. no. 1. vol. 9, no. 7, Jun. 2017, Art. no. 669.
[106] S. Meerow, J. P. Newell, and M. Stults, “Defining urban resilience: A [128] M. Amani, B. Salehi, S. Mahdavi, A. Masjedi, and S. Dehnavi,
review,” Landscape Urban Planning, vol. 147, pp. 38–49, Mar. 2016. “Temperature-Vegetation-soil Moisture Dryness Index (TVMDI),” Re-
[107] S. A. Ahmadi, A. Ghorbanian, and A. Mohammadzadeh, “Moving ve- mote Sens. Environ., vol. 197, pp. 1–14, Aug. 2017.
hicle detection, tracking and traffic parameter estimation from a satellite [129] R. Khan, H. Gilani, N. Iqbal, and I. Shahid, “Satellite-based (2000–2015)
video: A perspective on a smarter city,” Int. J. Remote Sens., vol. 40, drought hazard assessment with indices, mapping, and monitoring of
no. 22, pp. 8379–8394, Nov. 2019. Potohar plateau, Punjab, Pakistan,” Environ. Earth Sci., vol. 79, no. 1,
[108] Q. Zhong, J. Ma, B. Zhao, X. Wang, J. Zong, and X. Xiao, “Assessing Jan. 2020, Art. no. 23.
spatial-temporal dynamics of urban expansion, vegetation greenness and [130] L. J. Roberts, R. Burnett, J. Tietz, and S. Veloz, “Recent drought and tree
photosynthesis in megacity Shanghai, China during 2000–2016,” Remote mortality effects on the avian community in southern Sierra Nevada:
Sens. Environ., vol. 233, Nov. 2019, Art no. 111374. A glimpse of the future?,” Ecol. Appl., vol. 29, no. 2, Mar. 2019,
[109] S. Parsian and M. Amani, “Building extraction from fused LiDAR Art. no. e01848.
and hyperspectral data using Random Forest Algorithm,” GEOMATICA, [131] K. Uddin, M. A. Matin, and F. J. Meyer, “Operational flood mapping using
vol. 71, no. 4, pp. 185–193, Dec. 2017. multi-temporal Sentinel-1 SAR images: A case study from Bangladesh,”
[110] M. Ranagalage and S. Dissanayake, “The Impacts of landscape changes Remote Sens., vol. 11, no. 13, Jul. 2019, Art. no. 1581.
on annual mean land surface temperature in the tropical mountain city of [132] D. Amitrano, R. Guida, and G. Ruello, “Multitemporal SAR RGB pro-
Sri Lanka: A case study of Nuwara Eliya (1996–2017),” Sustainability, cessing for Sentinel-1 GRD products: Methodology and applications,”
vol. 11, no. 19, Oct. 2019, Art. no. 5517. IEEE J. Sel. Topics Appl. Earth Obs. Remote Sens., vol. 12, no. 5,
[111] D. Liu, N. Chen, X. Zhang, C. Wang, and W. Du, “Annual large-scale pp. 1497–1507, May 2019.
urban land mapping based on Landsat time series in Google Earth Engine [133] T. Long et al., “30 m resolution global annual burned area mapping based
and OpenStreetMap data: A case study in the middle Yangtze River on Landsat Images and Google Earth Engine,” Remote Sens., vol. 11,
basin,” ISPRS J. Photogrammetry Remote Sens., vol. 159, pp. 337–351, no. 5, Feb. 2019, Art. no. 489.
Jan. 2020. [134] P. Sarricolea et al., “Recent wildfires in central Chile: Detecting links
[112] M. Demuzere, B. Bechtel, and G. Mills, “Global transferability of local between burned areas and population exposure in the wildland urban
climate zone models,” Urban Climate, vol. 27, pp. 46–63, Mar. 2019. interface,” Sci. Total Environ., vol. 706, Mar. 2020, Art. no. 135894.
5348 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020

[135] B. Yu, F. Chen, and S. Muhammad, “Analysis of satellite-derived land- [157] H.-J. Vogel et al., “A systemic approach for modeling soil functions,”
slide at Central Nepal from 2011 to 2016,” Environ. Earth Sci., vol. 77, SOIL, vol. 4, no. 1, pp. 83–92, Mar. 2018.
no. 9, May 2018, Art. no. 331. [158] A. Ghorbanian, M. R. Sahebi, and A. Mohammadzadeh, “Optimization
[136] M. Kakooei and Y. Baleghi, “A two-level fusion for building irregularity approach to retrieve soil surface parameters from single-acquisition
detection in post-disaster VHR oblique images,” Earth Sci. Informat., single-configuration SAR data,” Comptes Rendus Geosci., vol. 351, no. 4,
vol. 13, pp. 459–477, Feb. 2020. pp. 332–339, Apr. 2019.
[137] E. Meilianda et al., “Assessment of post-tsunami disaster land use/land [159] S. M. MirMazloumi and M. R. Sahebi, “Assessment of different backscat-
cover change and potential impact of future sea-level rise to low-lying tering models for bare soil surface parameters estimation from SAR data
coastal areas: A case study of Banda Aceh coast of Indonesia,” Int. J. in band C, L and P,” Eur. J. Remote Sens., vol. 49, no. 1, pp. 261–278,
Disaster Risk Reduction, vol. 41, Dec. 2019, Art. no. 101292. Jan. 2016.
[138] B. Hao et al., “Land use change and climate variation in the three [160] H. R. Mirsoleimani, M. R. Sahebi, N. Baghdadi, and M. El Hajj, “Bare
gorges reservoir catchment from 2000 to 2015 based on the Google Earth soil surface moisture retrieval from Sentinel-1 SAR data based on the
Engine,” Sensors, vol. 19, no. 9, May 2019, Art. no. 2118. calibrated IEM and Dubois models using neural networks,” Sensors,
[139] J.-C. Yeh and C.-H. Liao, “Impact of population and economic growth vol. 19, no. 14, Jul. 2019, Art. no. 3209.
on carbon emissions in Taiwan using an analytic tool STIRPAT,” Sustain. [161] R. R. Poppiel et al., “Mapping at 30 m resolution of soil attributes
Environ. Res., vol. 27, no. 1, pp. 41–48, Jan. 2017. at multiple depths in midwest Brazil,” Remote Sens., vol. 11, no. 24,
[140] I. R. Orimoloye, S. P. Mazinyo, A. M. Kalumba, O. Y. Ekundayo, and W. Dec. 2019, Art. no. 2905.
Nel, “Implications of climate variability and change on urban and human [162] A. A. Pericak et al., “Mapping the yearly extent of surface coal mining
health: A review,” Cities, vol. 91, pp. 213–223, Aug. 2019. in Central Appalachia using Landsat and Google Earth Engine,” PLoS
[141] H. Gu, Y. Cao, E. Elahi, and S. K. Jha, “Human health damages related One, vol. 13, no. 7, Jul. 2018, Art. no. e0197758.
to air pollution in China,” Environ. Sci. Pollut. Res., vol. 26, no. 13, [163] E. Sun, A. Nieto, and Z. Li, “GPS and Google Earth based 3D assisted
pp. 13115–13125, May 2019. driving system for trucks in surface mines,” Mining Sci. Technol., vol. 20,
[142] M. Fuentes, K. Millard, and E. Laurin, “Big geospatial data analysis no. 1, pp. 138–142, Jan. 2010.
for Canada’s Air Pollutant Emissions Inventory (APEI): using google [164] S. A. S. Brooke, M. D’Arcy, P. J. Mason, and A. C. Whittaker,
earth engine to estimate particulate matter from exposed mine disturbance “Rapid multispectral data sampling using Google Earth Engine,” Comput.
areas,” GIScience Remote Sens., vol. 57, no. 2, pp. 245–257, Feb. 2020. Geosci., vol. 135, Feb. 2020, Art. no. 104366.
[143] W. Chen, H. Wang, H. Zhao, and K. Qin, “Google Earth Engine–assisted [165] H. A. Orengo and C. A. Petrie, “Multi-scale relief model (MSRM): A new
black carbon radiative forcing calculation over a heavy industrial city algorithm for the visualization of subtle topographic change of variable
in China,” Air Quality Atmosphere Health, vol. 13, no. 3, pp. 329–338, size in digital elevation models,” Earth Surface Processes Landforms,
Mar. 2020. vol. 43, no. 6, pp. 1361–1369, May 2018.
[144] J. L. Huntington et al., “Climate Engine: Cloud computing and visual- [166] G. Castelli, L. A. A. Oliveira, F. Abdelli, H. Dhaou, E. Bresci, and M.
ization of climate and remote sensing data for advanced natural resource Ouessar, “Effect of traditional check dams (jessour) on soil and olive
monitoring and process understanding,” Bull. Amer. Meteorological Soc., trees water status in Tunisia,” Sci. Total Environ., vol. 690, pp. 226–236,
vol. 98, no. 11, pp. 2397–2410, Nov. 2017. Nov. 2019.
[145] T. Klein, M. Nilsson, A. Persson, and B. Håkansson, “From open data [167] C. H. Wilson, T. T. Caughlin, S. W. Rifai, E. H. Boughton, M. C. Mack,
to open analyses—New opportunities for environmental applications?” and S. L. Flory, “Multi-decadal time series of remotely sensed vegetation
Environments, vol. 4, no. 2, Apr. 2017, Art. no. 32. improves prediction of soil carbon in a subtropical grassland,” Ecological
[146] S. J. Weber, D. R. Mishra, S. B. Wilde, and E. Kramer, “Risks for Appl., vol. 27, no. 5, pp. 1646–1656, Jul. 2017.
cyanobacterial harmful algal blooms due to land management and climate [168] K. Ivushkin, H. Bartholomeus, A. K. Bregt, A. Pulatov, B. Kempen,
interactions,” Sci. Total Environ., vol. 703, Feb. 2020, Art. no. 134608. and L. de Sousa, “Global mapping of soil salinity change,” Remote Sens.
[147] E. Walker and V. Venturini, “Land surface evapotranspiration estimation Environ., vol. 231, Sep. 2019, Art. no. 111260.
combining soil texture information and global reanalysis datasets in [169] H. A. Orengo and A. Garcia-Molsosa, “A brave new world for archae-
Google Earth Engine,” Remote Sens. Lett., vol. 10, no. 10, pp. 929–938, ological survey: Automated machine learning-based potsherd detection
Oct. 2019. using high-resolution drone imagery,” J. Archaeological Sci., vol. 112,
[148] H. Zeng et al., “Spatiotemporal analysis of precipitation in the sparsely Dec. 2019, Art. no. 105013.
gauged Zambezi river basin using remote sensing and Google Earth [170] H. Orengo and C. Petrie, “Large-scale, multi-temporal remote sensing
Engine,” Remote Sens., vol. 11, no. 24, Dec. 2019, Art. no. 2977. of Palaeo-River networks: A case study from Northwest India and its
[149] L. Gómez-Chova, J. Amorós-López, G. Mateo-García, J. Muñoz-Marí, implications for the Indus Civilisation,” Remote Sens., vol. 9, no. 7,
and G. Camps-Valls, “Cloud masking and removal in remote sensing Jul. 2017, Art. no. 735.
image time series,” J. Appl. Remote Sens., vol. 11, no. 1, Jan. 2017, [171] B. Liss, M. D. Howland, and T. E. Levy, “Testing Google Earth Engine for
Art. no. 015005. the automatic identification and vectorization of archaeological features:
[150] D. Kong, Y. Zhang, X. Gu, and D. Wang, “A robust method for recon- A case study from Faynan, Jordan,” J. Archaeological Sci. Rep., vol. 15,
structing global MODIS EVI time series on the Google Earth Engine,” pp. 299–304, Oct. 2017.
ISPRS J. Photogrammetry Remote Sens., vol. 155, pp. 13–24, Sep. 2019. [172] F. J. Hasiuk, C. Harding, A. R. Renner, and E. Winer, “TouchTerrain: A
[151] R. Chastain, I. Housman, J. Goldstein, M. Finco, and K. Tenneson, “Em- simple web-tool for creating 3D-printable topographic models,” Comput.
pirical cross sensor comparison of Sentinel-2A and 2B MSI, Landsat-8 Geosci., vol. 109, pp. 25–31, Dec. 2017.
OLI, and Landsat-7 ETM+ top of atmosphere spectral characteristics [173] M. B. Lyons et al., “Monitoring large and complex wildlife aggrega-
over the conterminous United States,” Remote Sens. Environ., vol. 221, tions with drones,” Methods Ecol. Evol., vol. 10, no. 7, pp. 1024–1035,
pp. 274–285, Feb. 2019. Jul. 2019.
[152] M. N. Hasan, M. Shrestha, L. Leigh, and D. Helder, “Evaluation of an [174] F. Ascensão, D. Yogui, M. Alves, E. P. Medici, and A. Desbiez,
Extended PICS (EPICS) for calibration and stability monitoring of optical “Predicting spatiotemporal patterns of road mortality for medium-
satellite sensors,” Remote Sens., vol. 11, no. 15, Jul. 2019, Art. no. 1755. large mammals,” J. Environ. Manage., vol. 248, Oct. 2019, Art. no.
[153] H. Huang, J. Wang, C. Liu, L. Liang, C. Li, and P. Gong, “The migration 109320.
of training samples towards dynamic global land cover mapping,” ISPRS [175] B. A. Wong, C. Thomas, and P. Halpin, “Automating offshore infrastruc-
J. Photogrammetry Remote Sens., vol. 161, pp. 27–36, Mar. 2020. ture extractions using synthetic aperture radar & Google Earth Engine,”
[154] M. Kakooei, A. Nascetti, and Y. Ban, “Sentinel-1 global coverage fore- Remote Sens. Environ., vol. 233, Nov. 2019, Art. no. 111412.
shortening mask extraction: An open source implementation based on [176] A. Rasul, “An investigation into the location of the crashed aircraft
Google Earth Engine,” in Proc. IEEE Int. Geosci. Remote Sens. Symp., through the use of free satellite images,” J. Photogrammetry Remote
Jul. 2018, pp. 6836–6839. Sens. Geoinf. Sci., vol. 87, no. 3, pp. 119–122, Sep. 2019.
[155] R. Lal, J. Kimble, and R. F. Follett, “Pedospheric processes and the carbon [177] J. Bian, A. Li, G. Lei, Z. Zhang, and X. Nan, “Global high-resolution
cycle,” in Soil Processes and the Carbon Cycle. Boca Raton, FL, USA: mountain green cover index mapping based on Landsat images and
CRC Press, 2018, pp. 1–8. Google Earth Engine,” ISPRS J. Photogrammetry Remote Sens., vol. 162,
[156] G. Tóth, T. Hermann, M. R. da Silva, and L. Montanarella, “Moni- pp. 63–76, Apr. 2020.
toring soil for sustainable development and land degradation neutral- [178] J.-F. Pekel, A. Cottam, N. Gorelick, and A. S. Belward, “High-resolution
ity,” Environmental Monitoring Assessment, vol. 190, no. 2, Feb. 2018, mapping of global surface water and its long-term changes,” Nature,
Art. no. 57. vol. 540, no. 7633, pp. 418–422, Dec. 2016.
AMANI et al.: GOOGLE EARTH ENGINE CLOUD COMPUTING PLATFORM FOR REMOTE SENSING BIG DATA APPLICATIONS 5349

[179] Q. Li, C. Qiu, L. Ma, M. Schmitt, and X. X. Zhu, “Mapping the land cover Seyed Ali Ahmadi received the B.Sc. and M.Sc.
of Africa at 10 m resolution from multi-source remote sensing data with degrees from the Faculty of Geodesy and Geomat-
Google Earth Engine,” Remote Sens., vol. 12, no. 4, p. 602, Feb. 2020. ics, K. N. Toosi University of Technology, Tehran,
[180] P. Gong et al., “Stable classification with limited sample: transferring a Iran, where he is currently working toward the Ph.D.
30-m resolution sample set collected in 2015 to mapping 10-m resolution degree.
global land cover in 2017,” Sci. Bull., vol. 64, no. 6, pp. 370–373, He worked on image classification and segmen-
Mar. 2019. tation techniques, machine learning algorithms, and
[181] A. E. Beresford, P. F. Donald, and G. M. Buchanan, “Repeatable and LiDAR data processing. His thesis was focused on
standardised monitoring of threats to key biodiversity areas in Africa classifying hyperspectral and LiDAR data sets by
using Google Earth Engine,” Ecol. Indic., vol. 109, Feb. 2020, Art. no. combining spectral and spatial features in order to
105763. increase the classification accuracy. His research in-
[182] P. Teluguntla et al., “A 30-m landsat-derived cropland extent prod- terests include machine learning, deep learning, LiDAR data processing, hy-
uct of Australia and China using random forest machine learning al- perspectral image analysis, data fusion, image processing, and computer vision
gorithm on Google Earth Engine cloud computing platform,” ISPRS techniques.
J. Photogrammetry Remote Sens., vol. 144, pp. 325–340, Oct. 2018,
doi: 10.1016/j.isprsjprs.2018.07.017.
[183] R. Goldblatt, W. You, G. Hanson, and A. Khandelwal, “Detecting the
boundaries of urban areas in India: A dataset for pixel-based image
classification in Google Earth Engine,” Remote Sens., vol. 8, no. 8,
Aug. 2016, Art. no. 634.

Mohammad Kakooei received the B.S. degree from

Shahid Beheshti University, Tehran, Iran, in 2011,
the M.Sc. degree from Iran University of Science and
Technology, Tehran, in 2014, and the Ph.D. degree in
electronic engineering from Babol Noshirvani Uni-
versity of Technology (NIT), Babol, Iran, in 2020.
He is currently a Researcher with the Depart-
ment of Computer & Electrical Engineering, NIT.
Meisam Amani (Senior Member, IEEE) received His research interests include image processing, ma-
the B.Eng. degree in geomatics engineering from chine learning, remote sensing, parallel processing,
the University of Tabriz, Tabriz, Iran, in 2012, the GPGPU, CUDA, and data mining applications.
M.Eng. degree in remote sensing engineering from
K.N. Toosi University of Technology, Tehran, Iran,
in 2014, and the Ph.D. degree in electrical engineer-
ing from Memorial University of Newfoundland, St.
John’s, NL, USA, in 2018.
He is currently a Senior Remote Sensing Scientist
and the Key Specialty Leader of Data Analytics at
a global consulting and engineering company, called
Wood plc, where he manages and leads various academic, governmental, and Armin Moghimi received the B.Sc. and M.Sc. de-
industrial remote sensing projects worldwide. He has worked on different grees in photogrammetry and geomatics engineer-
applications of remote sensing, including but not limited to land cover/land ing from the K. N. Toosi University of Technology,
use classification, soil moisture estimation, drought monitoring, water quality Tehran, Iran, in 2013 and 2015, respectively. He is
assessment, watershed management, power/transmission line monitoring, fog currently working toward the Ph.D. degree in pho-
detection and nowcasting, and ocean wind estimation. To do these, he has utilized togrammetry and remote sensing at the K. N. Toosi
various remote sensing datasets and different machine learning and big data University.
processing algorithms. His research interests include change detection
Dr. Amani is an Associate Editor in IEEE JSTARS and the lead guest editor techniques, image preprocessing, image registration,
for a special issue in the Remote Sensing journal. He also serves as a regular and Lidar.
Reviewer in about 15 international remote sensing journals. A list of his research
works can be found at https://www.researchgate.net/profile/Meisam_Amani3.

S. Mohammad Mirmazloumi (Student Member,

IEEE) received the degree B.Sc. in geomatics from
the University of Tabriz, Tabriz, Iran, in 2012, and the
M.Sc. degree in remote sensing from K. N. Toosi Uni-
Arsalan Ghorbanian received the B.Sc. degree in versity of Technology, Tehran, Iran, in 2015, where
geodesy and geomatics and the M.Sc. degree in re- his research focused on retrieval of soil surface param-
mote sensing, in 2016 and 2018, respectively, from eters using AIRSAR. She is currently working toward
K. N. Toosi University of Technology, Tehran, Iran, the Ph.D. degree in aerospace science and technology
where he is currently working toward the Ph.D. degree in the Polytechnic University of Catalonia, Barcelona,
in remote sensing. Spain.
His research interests include land cover mapping, He is currently a Research Assistant with the
big data processing, image and video processing, soil Remote Sensing Department, Centre Tecnològic de Telecomunicacions de
moisture estimation from SAR data, and hyperspec- Catalunya (CTTC). His research interests include SAR data applications, DIn-
tral dimension reduction. SAR, land cover classification, and soil surface parameters’ retrievals.
5350 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020

Sayyed Hamed Alizadeh Moghaddam was born in Qiusheng Wu received the Ph.D. degree in geog-
1994. He received the B.S. degree in survey engineer- raphy from the University of Cincinnati, Cincinnati,
ing from the University of Isfahan, Isfahan, Iran, in OH, USA, in 2015.
2016, and the M.Sc. degree in remote sensing from He is currently an Assistant Professor with the
the Khajeh Nasir Toosi University of Technology, Department of Geography, University of Tennessee,
Tehran, Iran, in 2018. Knoxville, TN, USA. His research interests include
His research interests include geometric processing remote sensing, open-source GIS, wetland hydrology,
of satellite images, hyperspectral remote sensing, di- and Google Earth Engine.
mensionality reduction, computational intelligence,
and oil spill detection using SAR data.

Brian Brisco received the B.Sc. degree in ecology

Sahel Mahdavi received the Ph.D. degree in electri-
and the M.Sc. degree in soil science both from the
cal engineering from Memorial University of New-
University of Guelph, Guelph, ON, Canada, in 1977
foundland, St. John’s, NL, USA, in 2018.
and 1981, respectively. He also received the Ph.D. de-
Having almost ten years of academic background
gree in remote sensing/physical geography from the
in Remote Sensing, she is familiar with a wide array
University of Kansas, Lawrence, KS, USA, in 1985.
of topics relevant to RS/GIS and their applications in
He is an internationally recognized authority on
various environmental aspects. These topics include
Synthetic Aperture Radar (SAR) and its application
object-based wetland classification using a combina-
to a wide range of environmental monitoring applica-
tion of optical and full-polarimetric SAR data, feature
tions. He has been involved in remote sensing since
selection, soil moisture retrieval using SAR images,
1975 and participated in the SURSAT project from
image segmentation, speckle reduction in SAR im-
1977 to 1980 before spending 4 years at the Remote Sensing Laboratory at the
ages, target detection in multispectral optical images, and the relationship
University of Kansas under the supervision of Dr. F.T. Ulaby, widely recognized
between environmental conditions and SAR images. She also coauthored a
as one of the world’s leading authorities on radar. He worked for Intera from 1989
book entitled Principles of SAR Remote Sensing. She has authored more than
until 1997 as a Research Associate after completion of an NSERC Postdoctoral
25 journals. She was a member of a provincial project on wetland classification
fellowship served at the Canada Centre for Remote Sensing. From 1997 to 2004
during her Ph.D. when she identified the problem with wetland classification
he worked for Noetix Research Inc., where he was the Director of Research and
using Remote Sensing and, subsequently, proposed a novel scheme for wetland
Applications Development. In 2004, he joined CCRS as a Research Scientist. His
mapping. She is currently affiliated with the Data Analytics team at Wood PLC.
research activities focus on using remote sensing, particularly synthetic aperture
Dr. Mahdavi was the recipient of the Newfoundland and Labrador Branch of
radar (SAR), for mapping and managing renewable resources. His extensive
Canadian Institute of Geomatics Scholarship (2015) and the Emera Graduate
publications include studies on vegetation characterization, crop identification
Scholarship for Distinctive Women in Engineering for three consecutive years
and monitoring, conservation farming/ soil erosion mapping, soil moisture esti-
(2016–2018).
mation, land cover mapping, wetland mapping, rangeland management, forestry,
and developing tools, and techniques for ground truth data acquisition. His work
has included experience with interferometry, polarimetry, and radar backscatter
modeling including software development and operational implementation. He
has authored or coauthored more than 200 publications including more than
50 peer-reviewed journal publications and is the author of two chapters in
the Manual of Remote Sensing volume on radar applications published by
Masoud Ghahremanloo received the B.S. degree ASPRS. He provides peer review services to all the major remote sensing
in geodesy and geomatics from Zanjan University, journals and participates as an external examiner for many graduate students at
Zanjan, Iran, in 2012, and the M.S. degree in remote various universities in Canada and abroad. He has been consulted or contracted
sensing from K.N. Toosi University of Technology, by government and non-government organizations on a wide range of SAR
Tehran, Iran, in 2018. He is currently working toward applications and system development including NRCan, CSA, DND, AAFC,
the Ph.D. degree in Earth and atmospheric sciences EC, NASA, ESA, NASDA, NOAA, USDA, CAS, etc. He has extensive contacts
from the University of Houston, Houston, TX, USA. in the SAR community worldwide and has worked in China, Vietnam, Malaysia,
His research interests include satellite remote sens- Thailand, Indonesia, South Africa, Argentina, Uruguay, Chile, Brazil, Columbia,
ing and application of artificial intelligence, machine and Costa Rica.
learning (ML), and deep learning (DL) in study of Dr. Brisco is a Past-President of the Canadian Remote Sensing Society (CRSS)
Earth and atmosphere. He is actively involved with and the Canadian Aeronautics and Space Institute.
ML/DL algorithm development and scientific data utilization of different satel-
lite remote sensing missions.

Saeid Parsian received the B.Eng. degree in geomat-

ics engineering from the University of Tabriz, Tabriz,
Iran, in 2012, and the M.Eng. degree in photogram-
metry from Tafresh University, Tehran, Iran, in 2015.
He has extensive work experience as a surveying
operator, and has contributed to several civil and
surveying projects, such as road and tunnel design,
land registration, and satellite image processing and
analysis. His current research interests include uti-
lizing multisource remote sensing datasets for flood
mapping and risk assessment.

View publication stats