Access to this full-text is provided by Frontiers.
Content available from Frontiers in Earth Science
This content is subject to copyright.
ORIGINAL RESEARCH
published: 24 February 2017
doi: 10.3389/feart.2017.00017
Frontiers in Earth Science | www.frontiersin.org 1February 2017 | Volume 5 | Article 17
Edited by:
Monique Petitdidier,
Institut Pierre-Simon Laplace, France
Reviewed by:
Karine Zeitouni,
University of Versailles Saint-Quentin,
France
Télesphore Yao Brou,
University of La Réunion, France
Fabrizio Niro,
European Space Research Institute,
Italy
*Correspondence:
Mykola Lavreniuk
nick_93@ukr.net
Specialty section:
This article was submitted to
Environmental Informatics,
a section of the journal
Frontiers in Earth Science
Received: 26 September 2016
Accepted: 10 February 2017
Published: 24 February 2017
Citation:
Shelestov A, Lavreniuk M, Kussul N,
Novikov A and Skakun S (2017)
Exploring Google Earth Engine
Platform for Big Data Processing:
Classification of Multi-Temporal
Satellite Imagery for Crop Mapping.
Front. Earth Sci. 5:17.
doi: 10.3389/feart.2017.00017
Exploring Google Earth Engine
Platform for Big Data Processing:
Classification of Multi-Temporal
Satellite Imagery for Crop Mapping
Andrii Shelestov 1, 2, Mykola Lavreniuk 1, 2*, Nataliia Kussul 1, 2 , Alexei Novikov 2and
Sergii Skakun 3, 4
1Department of Space Information Technologies and Systems, Space Research Institute (NASU-SSAU), Kyiv, Ukraine,
2Department of Information Security, National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute,” Kyiv,
Ukraine, 3Department of Geographical Sciences, University of Maryland, College Park, MD, USA, 4NASA Goddard Space
Flight Centre, Greenbelt, MD, USA
Many applied problems arising in agricultural monitoring and food security require reliable
crop maps at national or global scale. Large scale crop mapping requires processing
and management of large amount of heterogeneous satellite imagery acquired by various
sensors that consequently leads to a “Big Data” problem. The main objective of this study
is to explore efficiency of using the Google Earth Engine (GEE) platform when classifying
multi-temporal satellite imagery with potential to apply the platform for a larger scale
(e.g., country level) and multiple sensors (e.g., Landsat-8 and Sentinel-2). In particular,
multiple state-of-the-art classifiers available in the GEE platform are compared to produce
a high resolution (30 m) crop classification map for a large territory (∼28,100 km2and
1.0 M ha of cropland). Though this study does not involve large volumes of data, it
does address efficiency of the GEE platform to effectively execute complex workflows
of satellite data processing required with large scale applications such as crop mapping.
The study discusses strengths and weaknesses of classifiers, assesses accuracies that
can be achieved with different classifiers for the Ukrainian landscape, and compares
them to the benchmark classifier using a neural network approach that was developed
in our previous studies. The study is carried out for the Joint Experiment of Crop
Assessment and Monitoring (JECAM) test site in Ukraine covering the Kyiv region (North
of Ukraine) in 2013. We found that GEE provides very good performance in terms of
enabling access to the remote sensing products through the cloud platform and providing
pre-processing; however, in terms of classification accuracy, the neural network based
approach outperformed support vector machine (SVM), decision tree and random forest
classifiers available in GEE.
Keywords: Google Earth Engine, big data, classification, optical satellite imagery, land cover, land use, image
processing
Shelestov et al. GEE for Big Data Processing
INTRODUCTION
Information on land cover/land use (LCLU) geographical
distribution over large areas is extremely important for many
environmental and monitoring tasks, including climate change,
ecosystem dynamics analysis, food security, and others. Reliable
crop maps can be used for more accurate agriculture statistics
estimation (Gallego et al., 2010, 2012, 2014), stratification
purposes (Boryan and Yang, 2013), better crop yield prediction
(Kogan et al., 2013a,b; Kolotii et al., 2015), and drought risk
assessment (Kussul et al., 2010, 2011; Skakun et al., 2016b).
During the past decades, satellite imagery became the most
promising data source for solving such important tasks as
LCLU mapping. Yet, at present, there are no globally available
satellite-derived crop specific maps at high-spatial resolution.
Only coarse-resolution imagery (>250 m spatial resolution) has
been utilized to derive global cropland extent (e.g., GlobCover,
MODIS; Fritz et al., 2013). At present, a wide range of satellites
provide objective, open and free high spatial resolution data
on a regular basis. These new opportunities allow one to
build high-resolution LCLU maps on a regular basis and to
assess LCLU changes for large territories (Roy et al., 2014).
With launches of Sentinel-1, Sentinel-2, Proba-V and Landsat-
8 remote sensing satellites, there will be generated up to
petabyte of raw (unprocessed) images per year. The increasing
volume and variety of remote sensing data, dubbed as a “Big
Data” problem, creates new challenges in handling datasets
that require new approaches to extracting relevant information
from remote sensing (RS) data from data science perspective
(Kussul et al., 2015; Ma et al., 2015a,b). Generation of high
resolution crop maps for large areas (>10,000 sq. km) using Earth
observation data from space requires processing of large amount
of satellite images acquired by various sensors. Images acquired
at different dates during crop growth period are usually required
to discriminate certain crop types. The following issues should
be addressed while providing classification of multi-temporal
satellite images for large areas: (i) non-uniformity of coverage of
ground truth data and satellite scenes; (ii) seasonal differentiation
of crop groups (e.g., winter and summer crops) and the need for
incremental classification (to provide both in season and post
season maps); (iii) the need to store, manage and seamlessly
process large amount of data (big data issues).
The Google Earth Engine (GEE) provides a cloud platform
to access and seamlessly process large amount of freely available
satellite imagery, including those acquired by the Landsat-8
remote sensing satellite. The GEE also provides a set of the state-
of-the-art classifiers for pixel-based classification that can be used
for crop mapping. Though these methods are well-documented
in the literature, there were no previous studies to compare
all these methods for classification of multi-temporal satellite
imagery for large scale crop mapping. Since the GEE platform
does not include any neural network based models, we add a
neural network based classifier (Skakun et al., 2007; Kussul et al.,
2016; Lavreniuk et al., 2016; Skakun et al., 2016a) to the analysis
to provide a more complete comparison. Hence, the paper aims
to explore efficiency of using the GEE platform when classifying
multi-temporal satellite imagery for crop mapping with potential
to apply the platform for a larger scale (e.g., country level) and
multiple sensors (e.g., Landsat-8 and Sentinel-2). Results are
presented for a highly heterogeneous landscape with multiple
cropping systems on the Joint Experiment of Crop Assessment
and Monitoring (JECAM) test site in Ukraine with the area of
more than 28,000 km2.
STUDY AREA AND MATERIALS
DESCRIPTION
The proposed study is carried out for the Joint Experiment
for Crop Assessment and Monitoring (JECAM) test site in
Ukraine. Agriculture is a major part of Ukrainian’s economy
accounting for 12% of the Ukrainian Gross Domestic Product
(GDP). Globally, Ukraine was the largest sunflower producer
(11.6 MT) and exporter, and the ninth largest wheat producer
(22.2 MT) in the world in 2013, according to the U.S. Department
of Agriculture (USDA) Foreign Agricultural Service (FAS)
statistics.
The JECAM test site in Ukraine was established in 2011 as part
of the collaborative activities within the GEOGLAM Community
of Practice. The site covers the administrative region of Kyiv
region with the geographic area of 28,100 km2and almost 1.0
M ha of cropland (Figure 1). Major crop types in the region
include: winter wheat, maize, soybeans, vegetables, sunflower,
barley, winter rapeseed, and sugar beet. The crop calendar is
September-July for winter crops, and April-October for spring
and summer crops (Table 1). Fields in Ukraine are quite large
with size generally ranging up to 250 ha.
Ground surveys to collect data on crop types and other land
cover classes were conducted in 2013 in Kyiv region (Figure 2).
European Land Use and Cover Area frame Survey (LUCAS)
nomenclature was used in this study as a basis for land cover/land
use types. In total, 386 polygons were collected covering the
area of 22,700 ha (Table 2). Data were collected along the roads
following the JECAM adopted protocol (Waldner et al., 2016)
using mobile devices with built-in GPS. All surveyed fields were
randomly split into training set (50%) to train the classifiers and
testing set (50%) for testing purposes. Fields were selected in such
a way so there is no overlap between training and testing sets. All
classification results, in particular overall accuracy (OA), user’s
(UA), and producer’s (PA) accuracies are reported for testing
set (Congalton, 1991; Congalton and Green, 2008). The input
features were classified into one of the 13 classes.
Remote sensing images acquired by the Operational Land
Imager (OLI) sensor aboard Landsat-8 satellite were used for
crop mapping over the study region. Landsat-8/OLI acquires
images in eight spectral bands (bands 1–7, 9) at 30 m spatial
resolution and in panchromatic band 8 at 15 m resolution (Roy
et al., 2014). Only bands 2 through 7 acquired at different time
periods were used for crop classification maps. Bands 1 and
9 were not used due to strong atmospheric absorption. Three
scenes with path/row coordinates 181/24, 181/25, and 181/26
covered the test site region. Table 3 summarizes dates of image
acquisitions and the fraction of missing values in images due to
clouds and shadows.
Frontiers in Earth Science | www.frontiersin.org 2February 2017 | Volume 5 | Article 17
Shelestov et al. GEE for Big Data Processing
FIGURE 1 | Location of Ukraine and JECAM test site in Ukraine (Kyiv region, marked with bold boundaries).
TABLE 1 | Crop calendar with sowing and harvesting for Kyiv region.
NCrop Sowing Harvesting
1 Winter wheat End of September Mid of July
2 Winter rapeseed End of August Mid of July
3 Maize End of April Mid of August—begin of September
4 Soybean End of April—Begin of May Mid of September—begin of October
5 Spring crops Begin of April Mid of July—August
6 Sugar beet End of April—Mid of May End of September—Mid of November
7 Sunflower End of April—Begin of May Begin of September
FIGURE 2 | Location of fields surveyed during ground measurements in 2013. These data were used to train and assess performance of classifiers.
The main issues that need to be addressed while dealing with
satellite imagery for large areas (such as Kyiv region) are a non-
regular coverage over the region and the presence of missing
data due to the clouds and shadows. Therefore, before feeding
satellite data to the classifiers, a pre-processing step should be
performed to fill in missing values due to clouds and shadows.
Frontiers in Earth Science | www.frontiersin.org 3February 2017 | Volume 5 | Article 17
Shelestov et al. GEE for Big Data Processing
TABLE 2 | Number of polygons and total area of crops and land cover
types collected during the ground survey in 2013.
Polygons Area
NLUCAS class Class No. % ha %
1 Axx Artificial 6 1.6 23.0 0.1
2 B11 Winter wheat 51 13.2 3960.8 17.4
3 B32 Winter rapeseed 12 3.1 937.3 4.1
4 B12, B14 Spring crops 9 2.3 455.9 2.0
5 B16 Maize 87 22.5 7253.3 31.9
6 B22 Sugar beet 8 2.1 632.5 2.8
7 B31 Sunflower 30 7.8 2549.0 11.2
8 B33 Soybeans 60 15.5 3252.3 14.3
9 B19, B39, B40 Other cereals 32 8.3 1364.0 6.0
10 C10, B60 Forest 17 4.4 1014.3 4.5
11 E01, E02 Grassland 48 12.4 747.5 3.3
12 F00 Bare land 10 2.6 67.2 0.3
13 G01, G02 Water 16 4.1 448.3 2.0
Total 386 100 22705.3 100
TABLE 3 | List of Landsat-8/OLI image acquisitions and estimate of
missing values due to clouds and shadows.
Date (2013) Path/
row
Missing
values (%)
Date (2013) Path/
row
Missing
values (%)
April 16 181/24 4.77 June 19 181/24 62.58
181/25 0.60 181/25 33.37
181/26 0.02 181/26 26.43
May 02 181/24 0.01 July 05 181/24 35.30
181/25 0.77 181/25 21.06
181/26 4.38 181/26 15.69
May 18 181/24 9.06 August 06 181/24 24.35
181/25 14.93 181/25 12.69
181/26 14.32 181/26 40.86
At present, there is no a standard approach for dealing with these
issues. Compositing is a very popular approach, however missing
values can still happen in composite products (Yan and Roy,
2014). Another approach is to fill gaps using image processing
techniques or ancillary data such as MODIS (Gao et al., 2006;
Roy et al., 2008; Hilker et al., 2009). The main issue with this
approach is spatial discrepancy between the 250 and 500 m
MODIS products and the 30 m Landsat imagery. In this study,
two types of approaches were explored: (i) compositing products
available in GEE, so to benefit from products already available
in GEE; (ii) restored multi-temporal images without involving
ancillary data (Skakun and Basarab, 2014).
Composite Products Available at GEE
Different composites derived from Landsat-8 imagery and
available in GEE were analyzed in the study. Landsat 8 8-Day
Top-of-atmosphere (TOA) Reflectance Composites were used
from GEE (Figure 3). As to the time of composition, 8 day
composites were selected over 32 day composites to have a better
temporal resolution. The reason for that is that 32 composites are
composed based on the latest image, and this latest image can
be of not the best quality. These Landsat-8 composites are made
from Level L1T orthorectified scenes, using the computed TOA
reflectance. The composites include all the scenes in each 8-day
period beginning from the first day of the year and continuing
to the 360th day of the year. The last composite of the year,
beginning on day 361, will overlap the first composite of the
following year by 3 days. All the images from each 8-day period
are included in the composite, with the most recent pixel on top.
Landsat-8 Data Pre-Processing (Outside
GEE)
The following pre-processing steps were applied for all Landsat-8
images:
1. Conversion of digital numbers (DNs) values to the TOA
reflectance values using conversion coefficients in the
metadata file (Roy et al., 2014).
2. To decrease an impact of atmosphere to the image quality
conversion from the TOA reflectance to the surface
reflectance (SR) has been done using the Simplified
Model for Atmospheric Correction (SMAC; Rahman and
Dedieu, 1994). The source code for the model was acquired
from http://www.cesbio.ups-tlse.fr/multitemp/?p=2956.
Parameters of the atmosphere to run the model (in particular,
aerosol optical depth) were acquired from the Aeronet
network’s station in Kyiv (geographic coordinates +50.374N
and +30.497E). The differences between TOA and SR satellite
image is shown in Figure 4.
3. Detection of clouds and shadows were done using
Fmask algorithm (Zhu and Woodcock, 2012). For
this, a stand alone application was used from
https://code.google.com/p/cfmask/.
4. Landsat-8 satellite images were reconstructed from missing
pixel values (clouds and shadows) using self-organizing
Kohonen maps (SOMs; Skakun and Basarab, 2014).
These pre-processing steps were performed outside GEE
platform. After these products were generated they were
uploaded in the GEE cloud platform for the further classification
using classification algorithms available in GEE.
METHODOLOGY DESCRIPTION
In the study, classification of multi-temporal satellite imagery was
performed at per-pixel basis. Multiple classification techniques
were evaluated in the study. At first, all classification algorithms
available in GEE were analyzed and used for classifying
multi-temporal 8-days Landsat-8 TOA composites from GEE.
Then, the best classification algorithms in terms of overall
classification accuracy were compared with the neural network
classifier that used multi-temporal SR values generated outside
GEE. The presented approaches were compared in terms
of classification accuracy at pixel level. GEE offers several
Frontiers in Earth Science | www.frontiersin.org 4February 2017 | Volume 5 | Article 17
Shelestov et al. GEE for Big Data Processing
FIGURE 3 | An example of the Landsat-8 8-Day TOA Reflectance Composites products over the study area for 15.04.2013.
FIGURE 4 | An example of TOA (A) and SR (B) Landsat-8 image acquired
on the 8th of August 2013. A true color composition of Landsat-8 bands 4-3-2
is shown.
classification algorithms among which are decision trees, random
forests, support vector machine (SVM), and Naïve Bayes
classifier.
Support Vector Machine (SVM)
SVM became popular in a recent decade for solving problems
in classification, regression, and novelty detection. An important
property of SVMs is that the determination of the model
parameters corresponds to a convex optimization problem, and
so any local solution is also a global optimum (Bishop, 2006).
The SVM approaches classification problem through the concept
of the margin, which is defined to be the smallest distance
between the decision boundary and any of the samples. The
decision boundary is chosen to be the one for which the margin is
maximized. The margin is defined as the perpendicular distance
between the decision boundary and the closest of the data points.
Maximizing the margin leads to a particular choice of decision
boundary. The location of this boundary is determined by a
subset of the data points, known as support vectors.
Classification and Regression Tree (CART)
A decision tree (DT) classifier is built from a set of training data
using the concept of information entropy. At each node of the
tree, one attribute of the data that most effectively splits its set of
samples into subsets enriched in one class or the other is selected.
Its criterion is the normalized information gain that results from
choosing an attribute for splitting the data. The attribute with
the highest normalized information gain is chosen to make the
decision. The algorithm then recurs on the smaller sublists. One
disadvantage of the DT classifier is the considerable sensitivity to
the training dataset, so that a small change to the training data
can result in a very different set of subsets (Bishop, 2006).
Random Forests (RFs)
Since the major problem of the DT classifier is overfitting, RF
overcomes it by constructing an ensemble of DTs (Breiman,
2001). More specifically, RF operates by constructing a multitude
Frontiers in Earth Science | www.frontiersin.org 5February 2017 | Volume 5 | Article 17
Shelestov et al. GEE for Big Data Processing
of DT at training time and outputting the class that is the mode
of the classes (classification) of the individual trees. RFs correct
for DT habit of overfitting to their training set.
Other classifiers included in the GEE platform that are less
popular in the remote sensing community:
GMO Max Entropy
Multinomial logistic regression is generalization of linear
regression using the softmax transformation function and the
main task is to minimize an error function by taking the
negative logarithm of the likelihood, which means cross-entropy.
The main difference from other models and algorithms is the
outcome score that could be considered as a probability value
(Bishop, 2006; Haykin, 2008).
MultiClassPerceptron
This approach applied implementation of the linear perceptron
to multiclass problems. The main idea of the perceptron is
that the summing node of the neural model computes a linear
combination of the inputs applied to its synapses, as well as
incorporates an externally applied bias. The resulting sum, that is,
the induced local field, is applied to a hard limiter. Accordingly,
the neuron produces an output equal to 1 if the hard limiter input
is positive, and -1 if it is negative (Haykin, 2008). Unfortunately,
the perceptron is the simplest form of a neural network used for
the classification of patterns said to be linearly separable.
Naïve Bayes
Bayes classifier is a simple probabilistic approach which is based
on the Bayes theorem and assumption of independence between
input features. Within the learning procedure, it minimizes
the average risk of classification error. Main advantage of this
classifier is that it requires a small number of training data to
compute the decision surface. At the same time, its derivation is
contingent on the assumption that the underlying distributions
be Gaussian, which may limit its area of application (Haykin,
2008).
Intersection Kernel Passive Aggressive
Method for Information Retrieval (IKPamir)
It is the specialized version of the SVM which represents a
histogram intersection kernel SVMs (IKSVMs). The runtime
complexity of the classifier is logarithmic in the number of
support vectors as opposed to linear for the standard approach.
It allows to put IKSVM classifiers in the same order of
computational cost for evaluation as linear SVMs (Maji et al.,
2008).
Winnow
The algorithm can be expressed as a linear-threshold algorithm
that is similar to MultiClassPerceptron. However, while the
perceptron uses an additive weight-update scheme, the Winnow
classifier uses a multiplicative scheme. A primary advantage of
this algorithm is that the number of mistakes that it makes
is relatively little affected by the presence of large numbers of
irrelevant attributes in the examples. The number of mistakes
grows only logarithmically with the number of irrelevant
TABLE 4 | Overall classification accuracy achieved by GEE classifiers for
TOA 8-day composites as an input.
Classifier OA (%)
CART 75
GMO Max Entropy 72
Random Forest 68
MultiClassPerceptron 60
IKPamir 57
Winnow 49
FastNaiveBayes 32
Pegasos –
VotingSvm –
MarginSvm –
attributes. Classifier is computationally efficient (both in time and
space; Littlestone, 1988).
Primal Estimated sub-GrAdient SOlver for
Svm (Pegasos)
A variant of SVM with simple and effective sub-GrAdient SOlver
algorithm for approximately minimizing the objective function
that has a fast rate of convergence results. At each iteration, a
single training example is chosen at random and used to estimate
a sub-gradient of the objective, and a step with pre-determined
step-size is taken in the opposite direction. Solution is found in
probability solely due to the randomization steps employed by
the algorithm and not due to the data set. Therefore, the runtime
does not depend on the number of training examples and thus
Pegasos is especially suited for large datasets (Shalev-Shwartz
et al., 2011).
Ensemble of Neural Networks
It should be noted that neural network (NN) classifiers are
not available in GEE. Our proposed neural network approach
based on committee of NNs, in particular Multi-Layer Perceptron
(MLPs), is utilized to improve performance of individual
classifiers. The MLP classifier has a hyperbolic tangent activation
function for neurons in the hidden layer and logistic activation
function in the output layer. Within training cross-entropy (CE)
error function is minimized (Bishop, 2006)
E(w)= − ln p(T|w)= −
N
X
n=1
K
X
k=1
tnk ln ynk (1)
where E(w) is the CE error function that depends on the neurons’
weight coefficients w,Tis the set of vectors of target outputs
in the training set composed of Nsamples, Kis the number of
classes, tnk and ynk are the target and MLP outputs, respectively.
In the target output for class k, all components of vector tnare set
to 0, except for the k-th component which is set to 1. The CE error
E(w) is minimized by means of the scaled conjugate gradient
algorithm by varying weight coefficients w(Bishop, 2006).
A committee of MLPs was used to increase performance of
individual classifiers. The committee was formed using MLPs
Frontiers in Earth Science | www.frontiersin.org 6February 2017 | Volume 5 | Article 17
Shelestov et al. GEE for Big Data Processing
TABLE 5 | Classification results for CART, RF, and committee of NN for atmospherically corrected and restored Landsat-8 imagery.
CART RF Committee of MLPs
OA (%) 76.9 69.9 84.7
PA (%) UA (%) PA (%) UA (%) PA (%) UA (%)
1 Artificial 87.1 33.2 80.8 54.9 64.8 94.9
2 Winter wheat 87.8 91.3 82.7 90.0 91.3 95.0
3 Winter rapeseed 88.1 92.8 67.2 86.5 97.7 92.9
4 Spring crops 10.4 6.5 13.2 14.2 39.2 45.2
5 Maize 75.3 92.7 73.3 83.9 85.9 89.7
6 Sugar beet 56.1 48.4 44.8 27.3 91.1 95.3
7 Sunflower 75.0 75.9 67.0 61.7 87.5 83.9
8 Soybeans 74.0 50.8 56.8 51.3 77.0 68.0
9 Other cereals 74.1 53.6 64.8 34.9 82.4 69.5
10 Forest 93.8 89.8 89.4 94.8 82.7 97.3
11 Grassland 60.1 77.2 52.7 61.6 72.8 93.1
12 Bare land 90.1 83.9 84.6 80.2 98.5 85.2
13 Water 98.5 99.6 99.3 99.8 98.1 99.8
Classification metrics are presented per pixel basis.
with different parameters trained on the same training data.
Outputs from different MLPs were integrated using the technique
of average committee. Under this technique the average class
probability over classifiers is calculated, and the class with the
highest average posterior probability for the given input sample
is selected.
Ensemble based neural network model for crop classification
was recently validated in the JECAM experiment within study
areas in five different countries and agriculture conditions, and
provided the best result compare to SVM, maximum likelihood,
decision tree and logit regression (Waldner et al., 2016).
Therefore, this approach is used as a benchmark for assessing
classification techniques available in GEE.
RESULTS
Input (Product) Selection
The first set of experiments was carried out to select the
best input (TOA 8-day composites or restored values) and
evaluating different classifiers available in GEE. Table 4 shows
the derived OA on polygons from a testing set using TOA 8-
day composites as inputs. The best performance was achieved
for CART at 75%. Somewhat surprisingly, an ensemble of DTs,
i.e., RF, was outperformed by CART and yielded only 68%.
Logistic regression (GMO Max Entropy) gave 72% accuracy.
Linear classifiers, MultiClassPerceptron and Winnow, provided
up to 60% accuracy, while variants of SVM achieve moderate
accuracy of 57%. Unfortunately, it was unable to produce stable
classification results for SVM classifiers which usually resulted
into the Internal Server Error on invocation from Python.
Classifier Selection
One of the best GEE classifiers (CART and RF) on
atmospherically corrected Landsat-8 imagery were compared
to the committee of NN that was implemented outside GEE, in
the Matlab environment using a Netlab toolbox (http://www.
aston.ac.uk/eas/research/groups/ncrg/resources/netlab). Table 5
summarizes classification metrics, in particular OA, PA, and
UA for these three classification schemes. Committee of MLPs
considerably outperformed DT-based classifiers: by +14.8% RF
and by +7.8% DT.
DISCUSSION
Input Selection
The GEE platform offers powerful capabilities in handling large
volumes of remote sensing imagery that can be used, for
example, for classification purposes such as crop mapping for
large territories. In order to deal with irregular observation
and missing values due to clouds and shadows, a compositing
approach was applied. Several compositing products were
available. Also, using the rich JavaScript and/or Python APIs, it
is possible to filling missing values using MODIS images using,
for example, approaches developed by Roy et al. (2008),Hilker
et al. (2009), and Gao et al. (2006). But these are not trivial
procedures, are not available at hand and require a lot of efforts
from the end-user to be implemented. It substantially reduced
the quality of classification. In particular, the best OA achieved
on composites from the GEE was 75%, while on atmospherically
corrected and restored images the achieved accuracy was almost
77%. Therefore, while GEE offers several built-in composites,
substantial efforts (including programming) are required from
the end user to generate the required input data sets and to
remove clouds/shadow effects.
Classification Algorithms
The GEE platforms provides a set of classification algorithms.
The best results in the GEE were obtained for the DT-based
Frontiers in Earth Science | www.frontiersin.org 7February 2017 | Volume 5 | Article 17
Shelestov et al. GEE for Big Data Processing
FIGURE 5 | Final map obtained by classifying multi-temporal Landsat-8 imagery using a committee of MLP classifiers.
classifiers, namely CART and RF. The best accuracy achieved
on atmospherically corrected and restored Landsat-8 images was
76.9 and 69.9% for CART and RF, respectively. The classifier
GMO Max Entropy shows better result than RF but there is no
Python implementation (only Javascript). It is known that CART
and RF tend to overfit and require fine-tuning. These accuracies
were significantly lower than for the ensemble of MLPs that
obtained 84.7% overall accuracy (Figure 5). Also, most SVM-
based algorithms were not performing correctly within GEE,
and therefore, performance of SVM was not possible to evaluate
adequately.
Crop-Specific Classification Accuracies
Accuracy of 85% is usually considered as target accuracy for
agriculture applications (McNairn et al., 2009). Among classifiers,
considered in this study, the target accuracy was almost achieved
only by the MLPs ensemble. As to the specific crops, this accuracy
was achieved for the following crops:
•winter wheat (class 2, PA =91.3%, UA =95.0%): the major
confusion was with other cereals (class 9).
•winter rapeseed (class 3, PA =97.7%, UA =92.9%): the major
confusion was with other cereals (class 9).
•maize (class 5, PA =85.9%, UA =89.7%): the main confusion
was with the soybeans (class 8) and partly with sunflower (class
7).
•sugar beet (class 6, PA =91.1%, UA =95.3%): the main
confusion was with other cereals (class 9).
The crops that did not pass the 85% threshold for PA and
UA:
•spring crops (class 4, PA =39.2%, UA =45.2%): classification
using available set of satellite imagery failed to produce
reasonable performance for spring crops. The main
confusion of this class is with other cereals (class 9).
Confusion with other cereals can be explained by almost
identical vegetation cycle of spring barley and other
cereals produced in the region, namely with rye and
oats.
•sunflower (class 7, PA =87.5%, UA =83.9%): main confusion
was with soybeans, maize, and other cereals.
Frontiers in Earth Science | www.frontiersin.org 8February 2017 | Volume 5 | Article 17
Shelestov et al. GEE for Big Data Processing
•soybeans (class 8, PA =77.0%, UA =68.0%): this is the least
discriminated summer crop with main confusion with maize.
DISCUSSION AND CONCLUSIONS
The activities within this paper were targeted on the comparison
of pixel-based approaches to crop mapping in Ukraine, and
exploring efficiency of the GEE cloud platform for large scale
crop mapping with target to apply the platform for large areas
involving multiple sensors and large volumes of data (at the
order terabytes). Crop classification was performed by using
multi-temporal Landsat-8 imagery over the JECAM tests site
in Ukraine. Several inputs (products and composites) were
evaluated by means of different classifiers available at GEE
and our own approach. The classifiers included state-of-the-
art techniques such as SVM, decision tree, random forest, and
neural networks. Unlike other applications such as general land
cover mapping or forest mapping, crop discrimination requires
acquisition of multi-temporal profiles of crop growth dynamics.
Therefore, images at multiple dates (or composites at several
time intervals) need to be acquired or generated to discriminate
particular crops. At the same, it causes the problem of missing
data due to clouds and shadows when dealing with large territory,
for example Kyiv oblast in the study. Therefore, it is very difficult
to find an optimal solution in terms of temporal resolution and
cloud-free composites, and missing data still occur. One way to
handle this problem was to restore the missing values using self-
organizing Kohonen maps. Another approach would be to use
ancillary data such as MODIS to predict and fill in missing values
in the Landsat time-series. But these are not trivial procedures,
are not available at hand in GEE and require a lot of efforts
from the end-user to be implemented. Therefore, 8-day TOA
reflectance composites were used in GEE.
Better performance in terms of overall accuracy was shown
for atmospherically corrected and restored Landsat images over
the 8-day TOA reflectance composites from GEE. One of the
explanations would that quality of restored data was better than
TOA composites from GEE.
As to classification algorithms, ensemble of neural networks
outperformed SVM, decision tree, and random forest classifiers.
At some extent, it contradicts with recent studies where SMV
and RF show better performance than neural networks. In our
opinion, these are due: (i) in many cases, out-of-date neural
networks techniques are used and not full potential of neural
networks is explored in the remote sensing community; (ii) for
SVM, DT and RF built in GEE parameters were used.
For the following agriculture classes an 85% threshold of
producer’s and user’s accuracies was achieved: winter wheat,
winter rapeseed, maize, and sugar beet. Such crops as sunflower,
soybeans, and spring crops showed worse performance (below
85%).
In general, GEE provided very good performance in enabling
access to remote sensing products through the cloud platform,
and allowing seamless execution of complex workflow for
satellite data processing. The user at once gets access to many
satellite scenes or composites that are ready for processing.
Although our particular study did not directly deal with large
volumes of data, we still showed effectiveness of the GEE cloud
platform to build large scale applications when accessing and
processing satellite data with no reference to the volume. Indeed,
the approaches addressed in the study can be applied at country
level thanks to the capabilities offered in GEE. Another point
addressed in the study was validity—in other words, what kind
of accuracy can be expected for crop mapping with GEE when
utilizing existing products and processing chains.
However, in our opinion, several improvements should be
made to enable large scale crop mapping through GEE, especially
within operational context:
•Provide at hand tools to deal with missing data due to clouds
and shadows.
•To add neural networks classifiers (e.g., Tensorflow deep
learning library) and allow optimization of existing classifiers
especially SVM.
Future works should include:
•Large scale crop mapping (potentially at national scale) using
optical and SAR imagery taking into account the coverage of
Landsat-8, Proba-V, Sentinel-1, and Sentinel-2 imagery for the
whole country (such as Ukraine).
•In the future we will implement parcel-based classification
approach using GEE connected pixel method and some others,
to improve a pixel-based crop classification map.
AUTHOR CONTRIBUTIONS
AS was the project manager. ML provided crop classification
maps using ensemble of neural networks and conducted
experiments in Google Earth Engine. NK and AN were scientific
coordinators of the research and provided results interpretation.
SS has prepared training and test sets and preprocessed Landsat-8
data for crop mapping.
ACKNOWLEDGMENTS
This research was conducted in the framework of the “Large scale
crop mapping in Ukraine using SAR and optical data fusion”
Google Earth Engine Research Award funded by the Google Inc.
REFERENCES
Bishop, C. M. (2006). Pattern Recognition and Machine Learning. New York, NY:
Springer.
Boryan, C. G., and Yang, Z. (2013, July). “Deriving crop specific covariate data
sets from multi-year NASS geospatial cropland data layers,” in 2013 IEEE
International Geoscience and Remote Sensing Symposium-IGARSS (Melbourne,
VIC), 4225–4228.
Breiman, L. (2001). Random forests. Mach. Learn. 45, 5–32.
doi: 10.1023/A:1010933404324
Congalton, R. G. (1991). A review of assessing the accuracy of
classifications of remotely sensed data. Remote Sens. Environ. 37, 35–46.
doi: 10.1016/0034-4257(91)90048-B
Congalton, R. G., and Green, K. (2008). Assessing the Accuracy of
Remotely Sensed Data: Principles and Practices. Boca Raton, FL:
CRC Press.
Frontiers in Earth Science | www.frontiersin.org 9February 2017 | Volume 5 | Article 17
Shelestov et al. GEE for Big Data Processing
Fritz, S., See, L., You, L., Justice, C., Becker-Reshef, I., Bydekerke, L., and Gilliams,
S. (2013). The need for improved maps of global cropland. Eos Trans. Am.
Geophys. Union 94, 31–32. doi: 10.1002/2013EO030006
Gallego, F. J., Kussul, N., Skakun, S., Kravchenko, O., Shelestov, A., and
Kussul, O. (2014). Efficiency assessment of using satellite data for crop area
estimation in Ukraine. Int. J. Appl. Earth Observ. Geoinform. 29, 22–30.
doi: 10.1016/j.jag.2013.12.013
Gallego, J., Carfagna, E., and Baruth, B. (2010). “Accuracy, objectivity and
efficiency of remote sensing for agricultural statistics,” in Agricultural Survey
Methods, eds R. Benedetti, M. Bee, G. Espa, and F. Piersimoni (Chichester, UK:
John Wiley & Sons, Ltd.). doi: 10.1002/9780470665480.ch12
Gallego, J., Kravchenko, A. N., Kussul, N. N., Skakun, S. V., Shelestov, A. Y., and
Grypych, Y. A. (2012). Efficiency assessment of different approaches to crop
classification based on satellite and ground observations. J. Autom. Inf. Sci. 44,
67–80. doi: 10.1615/JAutomatInfScien.v44.i5.70
Gao, F., Masek, J., Schwaller, M., and Hall, F. (2006). On the blending
of the Landsat and MODIS surface reflectance: predicting daily Landsat
surface reflectance. IEEE Trans. Geosci. Remote Sens. 44, 2207–2218.
doi: 10.1109/TGRS.2006.872081
Haykin, S. (2008). Neural Networks and Learning Machines, 3rd Edn. Upper Saddle
River: NJ: Prentice Hall.
Hilker, T., Wulder, M. A., Coops, N. C., Seitz, N., White, J. C., Gao, F., et al. (2009).
Generation of dense time series synthetic Landsat data through data blending
with MODIS using a spatial and temporal adaptive reflectance fusion model.
Remote Sens. Environ. 113, 1988–1999. doi: 10.1016/j.rse.2009.05.011
Kogan, F., Kussul, N., Adamenko, T., Skakun, S., Kravchenko, O., Kryvobok,
O., et al. (2013a). Winter wheat yield forecasting in Ukraine based on Earth
observation, meteorological data and biophysical models. Int. J. Appl. Earth
Observ. Geoinform. 23, 192–203. doi: 10.1016/j.jag.2013.01.002
Kogan, F., Kussul, N. N., Adamenko, T. I., Skakun, S. V., Kravchenko, A. N.,
Krivobok, A. A., et al. (2013b). Winter wheat yield forecasting: A comparative
analysis of results of regression and biophysical models. J. Autom. Inf. Sci. 45,
68–81. doi: 10.1615/JAutomatInfScien.v45.i6.70
Kolotii, A., Kussul, N., Shelestov, A., Skakun, S., Yailymov, B., Basarab, R., and
Ostapenko, V. (2015). Comparison of biophysical and satellite predictors for
wheat yield forecasting in Ukraine. Int. Arch. Photogramm. Remote Sens. Spat.
Inf. Sci. 40, 39. doi: 10.5194/isprsarchives-XL-7-W3-39-2015
Kussul, N., Lemoine, G., Gallego, F. J., Skakun, S. V., Lavreniuk, M., and Shelestov,
A. Y. (2016). Parcel-based crop classification in ukraine using landsat-8 data
and sentinel-1A data. IEEE J. Select. Top. Appl. Earth Observ. Rem. Sens. 9,
2500–2508. doi: 10.1109/JSTARS.2016.2560141
Kussul, N. N., Sokolov, B. V., Zyelyk, Y. I., Zelentsov, V. A., Skakun,
S. V., and Shelestov, A. Y. (2010). Disaster risk assessment based on
heterogeneous geospatial information. J. Autom. Inf. Sci. 42, 32–45.
doi: 10.1615/JAutomatInfScien.v42.i12.40
Kussul, N., Shelestov, A., Basarab, R., Skakun, S., Kussul, O., and Lavrenyuk, M.
(2015). “Geospatial Intelligence and Data Fusion Techniques for Sustainable
Development Problems,” in ICTERI (Lviv), 196–203.
Kussul, N., Shelestov,A., and Skakun, S. (2011). “Grid te chnologiesfor s atellite data
processing and management within international disaster monitoring projects,”
in Grid and Cloud Database Management, eds S. Fiore and G. Aloisio (Berlin;
Heidelberg: Springer), 279–305.
Lavreniuk, M. S., Skakun, S. V., Shelestov, A. J., Yalimov, B. Y., Yanchevskii,
S. L., Yaschuk, D. J., et al. (2016). Large-scale classification of land
cover using retrospective satellite data. Cybern. Syst. Anal. 52, 127–138.
doi: 10.1007/s10559-016-9807-4
Littlestone, N. (1988). Learning quickly when irrelevant attributes abound: a new
linear-threshold algorithm. Mach. Learn. 2, 285–318. doi: 10.1007/BF00116827
Ma, Y., Wang, L., Liu, P., and Ranjan, R. (2015a). Towards building a data-
intensive index for big data computing–a case study of Remote Sensing data
processing. Inf. Sci. 319, 171–188. doi: 10.1016/j.ins.2014.10.006
Ma, Y., Wu, H., Wang, L., Huang, B., Ranjan, R., Zomaya, A., et al.
(2015b). Remote sensing big data computing: challenges and opportunities.
Future Generation Comput. Syst. 51, 47–60. doi: 10.1016/j.future.2014.
10.029
Maji, S., Berg, A. C., and Malik, J. (2008). “Classification using intersection kernel
support vector machines is efficient,” in IEEE Conference on Computer Vision
and Pattern Recognition (Anchorage, AK), 1–8.
McNairn, H., Champagne, C., Shang, J., Holmstrom, D. A., and Reichert, G.
(2009). Integration of optical and Synthetic Aperture Radar (SAR) imagery for
delivering operational annual crop inventories. ISPRS J. Photogramm. Remote
Sens. 64, 434–449. doi: 10.1016/j.isprsjprs.2008.07.006
Rahman, H., and Dedieu, G. (1994). SMAC: a simplified method for the
atmospheric correction of satellite measurements in the solar spectrum. Remote
Sens. 15, 123–143. doi: 10.1080/01431169408954055
Roy, D. P., Ju, J., Lewis, P., Schaaf, C., Gao, F., Hansen, M., et al. (2008). Multi-
temporal MODIS–Landsat data fusion for relative radiometric normalization,
gap filling, and prediction of Landsat data. Remote Sens. Environ. 112,
3112–3130. doi: 10.1016/j.rse.2008.03.009
Roy, D. P., Wulder, M. A., Loveland, T. R., Woodcock, C. E., Allen, R.
G., Anderson, M. C., et al. (2014). Landsat-8: science and product vision
for terrestrial global change research. Remote Sens. Environ. 145, 154–172.
doi: 10.1016/j.rse.2014.02.001
Shalev-Shwartz, S., Singer, Y., Srebro, N., and Cotter, A. (2011). Pegasos:
primal estimated sub-gradient solver for svm. Math. Programming 127, 3–30.
doi: 10.1007/s10107-010-0420-4
Skakun, S., Kussul, N., Shelestov, A., and Kussul, O. (2016b). The use of satellite
data for agriculture drought risk quantification in Ukraine. Geomatics Nat.
Hazards Risk, 7, 901–917. doi: 10.1080/19475705.2015.1016555
Skakun, S., Kussul, N., Shelestov, A. Y., Lavreniuk, M., and Kussul, O. (2016a).
Efficiency assessment of multitemporal C-Band Radarsat-2 intensity and
Landsat-8 surface reflectance satellite imagery for crop classification in
Ukraine. IEEE J. Select. Top. Appl. Earth Observ. Remote Sens. 9, 3712–3719.
doi: 10.1109/JSTARS.2015.2454297
Skakun, S. V., and Basarab, R. M. (2014). Reconstruction of missing data in time-
series of optical satellite images using self-organizing Kohonen maps. J. Autom.
Inf. Sci. 46, 19–26. doi: 10.1615/JAutomatInfScien.v46.i12.30
Skakun, S. V., Nasuro, E. V., Lavrenyuk, A. N., and Kussul, O. M. (2007). Analysis
of applicability of neural networks for classification of satellite data. J. Autom.
Inf. Sci. 39, 37–50. doi: 10.1615/JAutomatInfScien.v39.i3.40
Waldner, F., De Abelleyra, D., Verón, S. R., Zhang, M., Wu, B., Plotnikov, D.,
et al. (2016). Towards a set of agrosystem-specific cropland mapping methods
to address the global cropland diversity. Int. J. Remote Sens. 37, 3196–3231.
doi: 10.1080/01431161.2016.1194545
Yan, L., and Roy, D. P. (2014). Automated crop field extraction from multi-
temporal Web Enabled Landsat Data. Remote Sens. Environ. 144, 42–64.
doi: 10.1016/j.rse.2014.01.006
Zhu, Z., and Woodcock, C. E. (2012). Object-based cloud and cloud
shadow detection in Landsat imagery. Remote Sens. Environ. 118, 83–94.
doi: 10.1016/j.rse.2011.10.028
Conflict of Interest Statement: The authors declare that the research was
conducted in the absence of any commercial or financial relationships that could
be construed as a potential conflict of interest.
Copyright © 2017 Shelestov, Lavreniuk, Kussul, Novikov and Skakun. This is an
open-access article distributed under the terms of the Creative Commons Attribution
License (CC BY). The use, distribution or reproduction in other forums is permitted,
provided the original author(s) or licensor are credited and that the original
publication in this journal is cited, in accordance with accepted academic practice.
No use, distribution or reproduction is permitted which does not comply with these
terms.
Frontiers in Earth Science | www.frontiersin.org 10 February 2017 | Volume 5 | Article 17
Content uploaded by Andrey Yu. Shelestov
Author content
All content in this area was uploaded by Andrey Yu. Shelestov on Feb 24, 2017
Content may be subject to copyright.
