IEEE Geoscience and Remote Sensing Magazine - September 2015 - 62

Abstract-Because of sensor malfunction and poor
atmospheric conditions, there is usually a great deal of
missing information in optical remote sensing data,
which reduces the usage rate and hinders the follow-up
interpretation. In the past decades, missing information
reconstruction of remote sensing data has become an active research field, and a large number of algorithms have
been developed. However,
to the best of our knowledge, there has not, to date,
MIssINg INFORMaTION
been a study that has been
ReCONsTRUCTION OF
aimed at expatiating and
summarizing the current
ReMOTe seNsINg DaTa Is OF
situation. This is therefore
gReaT IMpORTaNCe FOR The
our motivation in this resUbseqUeNT applICaTIONs.
view. This paper provides
an introduction to the principles and theories of missing information reconstruction of remote sensing data.
We classify the established and emerging algorithms into
four main categories, followed by a comprehensive comparison of them from both experimental and theoretical
perspectives. This paper also predicts the promising future research directions.
I. INTRODUCTION
emote sensing instruments can capture information
about the atmosphere, ocean, and the Earth's surface.
They are one of the most frequently used and most powerful approaches to understanding and investigating our
planet. However, because of defective sensors and poor
atmospheric conditions (e.g., thick clouds), the acquired
remote sensing data are often so incomplete (we also say
that they have missing information) that the data usability
is greatly reduced.
For example, 15 of the 20 detectors in the Aqua Moderate Resolution Imaging Spectroradiometer (MODIS) band
6 are ineffective [1]; the scan line corrector (SLC) of the
Landsat enhanced thematic mapper plus (ETM+) sensor
has permanently failed [2]; and the ozone monitoring instrument (OMI) onboard the Aura satellite is subject to a

R

(a)

(b)

(c)

row anomaly problem. On average, at any one time, approximately 35% of the global land surface is obscured by
clouds [3], and in an individual country, the rate of cloud
cover can be much higher, e.g., in Canada, from 50% to
80% of the Earth's surface is covered by clouds in midmorning [4]. Since cloud cover is frequently inevitable in
the observation process, passive remote sensing data contain lots of missing information. Clouds in remote sensing data are useful information for liquid water research;
however, in this paper, they are considered as useless information that obstructs the observation. It is noteworthy
that only passive remote sensing imageries are affected by
atmospheric conditions; thus, when it comes to the atmosphere (or clouds) in the following descriptions, the data
are from passive sensor platforms.
Dead pixels resulting from sensor failure or random error are also a common phenomenon. Missing information
limits the regular and further applications of remote sensing data, e.g. classification [5], unmixing, target detection,
etc. As a result, reconstructing the missing information of
remote sensing data is of great significance for many practical applications (e.g., improving the subsequent interpretation accuracy and enhancing the data availability).
In order to intuitively perceive the information loss of
remote sensing data, we show some concrete examples in
Fig. 1. In short, Fig. 1(a)-(c) are the consequences of sensor failure, and Fig. 1(d)-(f) are the consequences of cloud
obscuration. The figures include digital number (DN) value
images and quantitative products of remote sensing, i.e., reflectance, land surface temperature (LST), the normalized
difference vegetation index (NDVI), and ozone.
In our opinion, according to the different sources of
the complementary information when reconstructing the
missing information, the present algorithms can be primarily classified into four categories: 1) spatial-based methods,
without any other auxiliary information source; 2) spectralbased methods, which extract the complementary information from other spectra; 3) temporal-based methods, which
extract the complementary information from other data
acquired at the same position and at different time periods
(hereafter, we say they are multitemporal); and 4) hybrid

(d)

(e)

(f)

FIgURe 1. Several examples of remote sensing data with missing information. In addition to the white clouds, the dark regions also represent
missing information. (a) Reflectance of Aqua MODIS band 6 with sensor failure. (b) DN value of Landsat ETM+ with the SLC-off problem. (c)
Ozone of Aura OMI with the row anomaly problem. (d) DN value of IKONOS-2 with cloud obscuration. (e) LST of MODIS with cloud obscuration. (f) NDVI of MODIS with cloud obscuration. Note that the dynamic range [-1, 1] is stretched linearly to [0, 255] for the visual effect.

62

ieee Geoscience and remote sensing magazine

september 2015
Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - September 2015
