1190 IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 10, NO. 5, SEPTEMBER 2013
Tensor Ensemble of Ground-Based Cloud Sequences:
Its Modeling, Classification, and Synthesis
Shuang Liu, Chunheng Wang, Baihua Xiao, Zhong Zhang, and Xiaozhong Cao
Abstract—Since clouds are one of the most important me-
teorological phenomena related to the hydrological cycle and
affect Earth radiation balance and climate changes, cloud analysis
is a crucial issue in meteorological research. Most researchers
only consider the classification task of cloud images while less
attention has been paid to the synthesis one. In addition, all the
existing research on cloud identification from sky images is based
on single cloud images. However, the cloud-measuring devices
on the ground actually take one image of the clouds every few
minutes and collect a series of cloud images. Thus, the existing
methods neglect the temporal information exhibited by contigu-
ous cloud images. To overcome this drawback, in this letter we
treat ground-based cloud sequences (GCSs) as dynamic texture.
We then propose the Tensor Ensemble of Ground-based Cloud
Sequences (eTGCS) model which represents the ensemble of GCSs
in a tensor manner. In the eTGCS model, all GCSs form a single
tensor, and each GCS is a subtensor of the single tensor. There
are two main characteristics of the eTGCS model: 1) All GCSs
share an identical mode subspace, which makes the classification
convenient, and 2) a new GCS can be synthesized as long as the
parameters of the eTGCS model are used. Therefore, less storage
space is required. Comprehensive experiments are conducted to
prove the superiority of our eTGCS model. The classification
accuracy achieves 92.31%, and the synthesized GCSs are similar
to the original ones in visual appearance.
Index Terms—Ground-based cloud sequences (GCSs), tensor
ensemble.
I. INTRODUCTION
C
LOUD analysis plays an essential role in meteorological
research because clouds affect the hydrological cycle,
Earth radiation balance, and climate changes. Cloud type classi-
fication is an important research field in the area of cloud anal-
ysis because successful classification of cloud types can help
to understand climatic conditions. Nowadays, cloud images
observed from the ground are classified by the observers who
have received professional training. However, this method takes
huge manual cost, and its quality depends on the experience
Manuscript received June 19, 2012; revised August 20, 2012, October 20,
2012, and November 25, 2012; accepted December 17, 2012. Date of pub-
lication February 5, 2013; date of current version June 13, 2013. This work
was supported by the Chinese National Natural Science Foundation through
Projects 61172103, 60933010, and 60835001.
S. Liu, C. Wang, B. Xiao, and Z. Zhang are with the State Key Laboratory
of Management and Intelligent Control of Complex Systems, Institute of
Automation, Chinese Academy of Sciences, Beijing 100190, China (e-mail:
shuang.liu@ia.ac.cn; chunheng.wang@ia.ac.cn; baihua.xiao@ia.ac.cn; zhong.
zhang@ia.ac.cn).
X. Cao is with the Meteorological Observation Centre, China Meteorological
Administration, Beijing 100081, China (e-mail: caoxzh@126.com).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/LGRS.2012.2236073
of the observers in a great degree. Thus, it is of great need to
automatically classify cloud images observed from the ground
in this area. It should be noted that cloud images observed from
the ground are referred to as ground-based cloud images for
convenience throughout this letter.
In the past few years, a lot of methods have been proposed
for ground-based cloud image classification. Bush et al. [1]
provided a classification method based on the Whole Sky
Imager (WSI) data, in which they applied binary decision trees
to classify the WSI images into five different sky conditions.
Singh and Glennen [2] utilized several features for cloud clas-
sification, such as co-occurrence and autocorrelation matrices.
Calbó and Sabburg [3] used Fourier transformation to classify
eight predefined sky conditions. Sun et al. [4] extracted local
binary pattern (LBP) as the representation features for cloud
classification. Heinle et al. [5] proposed an approach to extract
spectral features and some simple textural features, such as
energy and entropy for a fully automated classification algo-
rithm, in which seven different sky conditions are distinguished.
Liu et al. [6] extracted some cloud structure features from
infrared cloud images for classification. However, as a matter
of fact, the cloud-measuring devices on the ground take one
image of the clouds every few minutes and thus collect a series
of cloud images. All the aforementioned methods conduct
classification on only one single cloud image and neglect the
temporal information exhibited by contiguous cloud images.
It is reasonable to treat ground-based cloud sequences
(GCSs), a kind of natural texture, as dynamic texture (DT).
In the famous autoregressive moving average (ARMA) model
[7], DT is directly modeled at the image sequence level. Being
simple, straightforward, and effective, the ARMA model has
achieved great success. However, this model cannot capture the
temporal and spatial redundancy of appearance efficiently. It
is because the ARMA model is essentially linear and uses the
image-as-vector representation, where an image with the size
of I
1
∗ I
2
pixels is organized as a vector of K (K = I
1
∗ I
2
)
dimensions.
In this letter, we propose a novel model named Tensor
Ensemble of Ground-based Cloud Sequences (eTGCS) which
utilizes cloud sequences for classification. In the eTGCS model,
an ensemble of GCSs forms a single tensor, and each GCS
is a subtensor of the single tensor. To utilize the spatial re-
lationship, each frame of the cloud sequence uses the image-
as-matrix representation. In the framework of an ensemble, all
GCSs share an identical class mode subspace, which makes the
classification convenient. Furthermore, the eTGCS model goes
beyond mere classification, and we apply it to conduct other
tasks such as synthesis and compression in a very convenient
1545-598X/$31.00 © 2013 IEEE