IEEE Geoscience and Remote Sensing Magazine - September 2015 - 58

the original program that requires to run 30 times individually. Since we do prediction pixel by pixel, it is also
straightforward to enable parallel computing for a multiprocessor system. These improvements allow applications
of the data fusion approach over large areas. The recently
improved STARFM software package can be downloaded
from the USDA ARS software website.
Many of the improvements discussed here for the operational STARFM data fusion framework can also benefit
other data fusion approaches as well.
V. DISCUSSION
Data fusion approaches have been applied to synthesize
remote sensing imagery from multiple sources with different spatial, temporal and spectral characteristics, such
that the fused result can convey more information than
can an individual sensor alone. Traditional data fusion
approaches such as the IHS-like methods aim to sharpen
multi-spectral band imagery using panchromatic band
imagery. That approach focuses on fusing spectral information with spatial features. The data fusion approaches
presented in this paper aim to fuse images acquired from
different sensors and dates at very different spatial resolutions. The main objective of this kind of data fusion is
to generate dense time-series of remote sensing data at
a finer spatial resolution and thus the focus is on fusing
spatial and temporal features using same multi-spectral
band imagery.
The currently existing body of literature has demonstrated the capability of STARFM, STAARCH and ESTARFM
to monitor changes of terrestrial ecosystems with added
benefit over using single sensor approaches. Most data fusion techniques, including the approaches and examples
in this paper, are based on Landsat and MODIS data since
they have similar spectral bands and more importantly they
are freely and can be easily accessed to the public. However,
the techniques described in this paper could equally be extended to other, similar sensors. For example, the Visible
Infrared Imaging Radiometer Suite (VIIRS) on board the
Suomi-National Polar-orbiting Partnership (NPP) mission
(launched on October 28, 2011) and planned Joint Polar
Satellite System (JPSS) missions provide visible to shortwave
spectral bands at 375 and 750 m resolution, as well as TIR
data. The VIIRS instrument inherits attributes from MODIS
and AVHRR. Its configuration makes it an ideal high frequency data source for data fusion approaches. The recently
launched Sentinel-2 mission from European Space Agency
(ESA) provides additional multiple spectral band images
at 10 to 20 m resolution. The Sentinel-2 data will be freely
available and should be an ideal high spatial resolution data
source for data fusion.
Even though the data fusion approach can be used
to make synthetic images from multiple sources, these
fused images cannot replace actual images from satellite missions especially when surface changes are rapid
and subtle. Changes that are missed (e.g. due to clouds)
58

or not sensible (too small) in the coarse resolution images cannot be reconstructed. The data fusion models
have limitations on predicting dates that are distant from
the available image pairs. For instance, the predictability of changes in the fine resolution synthetic images
depends on the capacity of at least one sensor to detect
these changes in space and time, particularly when they
occur in vegetation structure or stand composition or at
sub-pixel ranges [7]. More actual medium resolution images can therefore help to capture small changes and to
shorten prediction period and thus reduce uncertainties
of data fusion results. By combining existing satellite observations and fused results we will be able to perform
time-series analyses at medium spatial resolution and to
investigate landscape changes at spatiotemporal resolutions that have been heretofore inaccessible.
AUthOr INfOrmAtION
Feng Gao received the B.A. degree in geology and the M.E. degree in remote sensing from Zhejiang University, Hangzhou,
China, in 1989 and 1992, respectively,
the Ph.D. degree in geography from Beijing Normal University, Beijing, China, in
1998, and the M.S. degree in computer science from Boston University, Boston, MA, USA, in 2003. From
1992 to 1998, he was a Research Assistant at the Nanjing Institute of Geography and Limnology, Chinese Academy of Science,
Nanjing, China. From 1998 to 2004, he was a Research Associate Professor at the Department of Geography and a Researcher
in the Center for Remote Sensing, Boston University. From 2004
to 2011, he was a Research Scientist with Earth Resources Technology, Inc. and the NASA Goddard Space Flight Center. Since
2011, he has been a Research Scientist with the Hydrology and
Remote Sensing Laboratory, USDA Agricultural Research Service.
His recent research interests include remote sensing modeling,
multi-sensor data fusion, and vegetation biophysical parameter
retrieving for crop and ecosystem condition monitoring. He is currently a member of the Landsat Science Team and the MODIS
Science Team.
Thomas Hilker obtained a PhD from the University of British Columbia (UBC), Canada,
in 2008 and have worked at UBC (2008-
2011) and NASA's Goddard Space Flight Center (2011-2012). Currently, he is holding a
position as Assistant Professor at Oregon State
University. His research interest is in remote
sensing of terrestrial ecosystems, particularly the carbon, water
and energy balance. Earth system science is currently at the crossroads as it is faced with the daunting task of assessing the risks of
anthropogenic changes to Earth and atmosphere, their feedbacks
on global ecosystems and consequences for live on earth. To meet
these challenges, development and application of models and algorithms is needed for retrieving ecosystem states and fates from remote sensing data. His research philosophy and guiding principles
ieee Geoscience and remote sensing magazine

september 2015

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