Hierarchical Bayesian Model Based on
Robust Fixed Rank Filter for
Fusing M
MSR-E SST
Yuxin Zhu, Emily Lei Kang, Yanchen Bo, Jinzong Zhang, Yuexiang Wang, Qingxin Tang
Abstract
Spatiotemporal complete sea surface temperature (
SST
)
dataset with higher accuracy and resolution is desirable for
many studies in atmospheric science and climate change. The
purpose of this study is to establish the spatiotemporal data
fusion model, the Hierarchical Bayesian Model (
HBM
) based
on Robust Fixed Rank Filter (
R-FRF
), that merge Moderate
Resolution Imaging Spectroradiometer (
MODIS
)
SST
with 4-km
resolution and Advanced Microwave Scanning Radiometer-
Earth Observing System (
AMSR-E
)
SST
with 25-km resolution
through their spatiotemporal complementarity to obtain
fusion
SST
with complete coverage, high spatial resolution,
and fine spatial pattern. First, a bias correction model was
applied to correct satellite
SST
. Second, a spatiotemporal
model called
R-FRF
was established to model potential
spatiotemporal process of
SST
. Third, the
R-FRF
model was
embedded in the hierarchical Bayesian framework, and
the corrected
MODIS
and
AMSR-E
SST
are merged. Finally,
the accuracy, spatial pattern and spatial completeness of
the fusion
SST
were assessed. The results of this study are
the following: (a) It is necessary to carry out bias correction
before data fusion. (b) The
R-FRF
model could simulate
SST
spatiotemporal trend well. (c) Fusion
SST
has similar accuracy
and spatial pattern to
MODIS
SST
. Though the accuracy is
lower than that of the
AMSR-E
SST
, the fusion
SST
has more
local detail information. The results indicated that fusion
SST
with higher accuracy, finer spatial pattern, and complete
coverage can be obtained through
HBM
based on
R-FRF
.
Index Terms: Hierarchical Bayesian Model based on
R-FRF
;
MODIS
SST
;
AMSR-E
SST
; scale transformation; local variance.
Introduction
The sea surface temperature (
SST
) is one of the important
parameters in coupling the ocean and atmosphere through ex-
changes of heat, momentum, moisture, and gases (Donlon
et
al.
2002, Zhu
et al.
2015). Currently, spatiotemporal complete
SST
dataset with higher accuracy and spatiotemporal resolu-
tion is desirable for many studies in atmospheric science and
climate change. Though satellite-derived
SST
products with
different spatiotemporal resolutions have a more completely
spatial coverage than other observations do, the spatiotem-
poral resolution and spatial coverage of satellite-derived
SST
products from a single sensor are limited. The Global Ocean
Data Assimilation Experiment has initiated a pilot project to
develop
SST
products with high spatiotemporal resolution
through integrating existing satellite products (Donlon 2001).
In recent years, the researchers have developed a series of
quantitative fusion methods for the integration of various re-
mote sensing products, such as the integration of
SST
(Guan
et
al.
2004, Guo 2010, Li
et al.
2013), leaf area index (
LAI
) (Li
et
al.
2013, Wang
et al.
2011), Normalized Difference Vegetation
Index (
NDVI
) (Busetto
et al.
2008, Hwang
et al.
2011, Rao
et al.
2015), aerosol (Chatterjee
et al.
2010, Loyola
et al.
2012, Tang
et al.
2016), snow water equivalent (Durand
et al.
2008, Foster
et al.
2011, Gao
et al.
2010, Kongoli
et al.
2007), ocean color
(Kwiatkowska
et al.
2002, Maritorena
et al.
2010, Pottier
et al.
2006), soil moisture (Yilmaz
et al.
2012), Surface Reflectance
(Gao
et al.
2015, Zhu
et al.
2016) and so on. The methods that
have been applied can mainly be categorized into two groups:
data assimilation technique and spatiotemporal fusion tech-
nique of multisource data.
Data assimilation technique is used to derive accurate
estimations of the current and future states of the system,
together with estimations of the uncertainty in the estimated
states by observations in combination with a dynamic system
model (Nichols 2010). This method has been widely used in
the parameter estimation of atmospheric science, Marine sci-
ence and land surface process. For example, Durand, Molotch
and Margulis (2008) reconstructed
SWE
in the Rio Grande
headwaters by combining time series observations from
Moderate Resolution Imaging Spectroradiometer (
MODIS
) and
Landsat Enhanced Thematic Mapper (
ETM+
) with a spatially
explicit snowmelt model. Data assimilation technique has
clear physical mechanism. It can make full use of the satellite
observations with different spatiotemporal resolution, and
can accurately simulate the evolution of the process with time
because it combines remote sensing observations with dynam-
ic process model, such as crop growth model, the ecological
Yuxin Zhu, Jinzong Zhang and Yuexiang Wang are with
School of Urban and Environmental Sciences, Institute of
land and urban-rural planning, and Jiangsu Collaborative
Innovation Center of Regional Modern Agriculture &
Environmental Protection, Huaiyin Normal University,
Jiangsu Province, 223300, China; (
);
(
); (
)
Yuxin Zhu is also with the Institute of Geographic Sciences
and Natural Resources Research, Chinese Academy of
Sciences (CAS), Beijing 100101, China
Emily Lei Kang is with the University of Cincinnati,
Cincinnati, Ohio 45221-0025 USA (
)
Yanchen Bo is with the State Key Laboratory of Remote
Sensing Science, Faculty of Geographical Science, Beijing
Normal University, Beijing 100875, China (
)
Qingxin Tang is with the School of Environment and
Planning, Liaocheng University, Liaocheng, Shandong
Province,252059, China (
)
Photogrammetric Engineering & Remote Sensing
Vol. 85, No. 2, February 2019, pp. 119–131.
0099-1112/18/119–131
© 2019 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.85.2.119
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
February 2019
119
