Soil Moisture Retrieval Using Modified
Particle Swarm Optimization and
Back-Propagation Neural Network
Liangliang Tao, Guojie Wang, Xi Chen, Jing Li, and Qingkong Cai
Abstract
In order to eliminate the influences of surface roughness and
vegetation on radar signals in the vegetation-covered soil
moisture estimation, the present paper proposes a combin-
ing method based on modified particle swarm optimization
(
MPSO
) and back-propagation (
BP
) neural network algorithm.
This method combines optical and radar data at the field
scale and uses
MPSO
to optimize the weight of the neural
network. An effective inertia weight is introduced in the
MPSO
and an implicit relationship between backscatter coefficient
and soil moisture is established. Experimental results show
that the combining method produces better accuracy than
other inversion methods with R
2
of 72.2% and Root Mean
Square Error (
RMSE
) of 0.033 cm
3
/cm
3
, respectively. Mean-
while, the estimated accuracy of surface soil moisture using
radar and optical data simultaneously is much higher than
that using only a single data source as input with R
2
of 0.827
and
RMSE
of 0.029 cm
3
/cm
3
. Therefore, the combining method
can effectively improve the accuracy of soil moisture retrieval
and provide support for large-scale agricultural monitoring.
Introduction
Soil moisture plays an important role in the exchange of
energy between land and atmosphere. It is a key factor in
land surface evapotranspiration, water migration, and carbon
cycle. Disasters caused by water shortage have become the
main factors in restricting the agricultural production and
economic development. Meanwhile, it
dation, such as soil desertification, salin
degradation, and soil erosion (Dorigo
et
development of multitemporal, multispectral, hyperspectral,
high-resolution, and active-passive microwave remote sensing
technology, acquisition accuracies of soil moisture retrieval
have been greatly improved (Hassan-Esfahani
et al.
2015; Li
et
al.
2018; Oltra-Carrió
et al.
2015; Zhu
et al.
2019).
Many effective methods, including empirical and physical
models, have been proposed to retrieve surface soil moisture
using remote sensing data (Attema and Ulaby 1978; De Roo
et al.
2001; Dubois
et al.
1995; Fung
et al.
1992; Oh 2004; Shi
et al.
1997; Ulaby
et al.
1990; Wu
et al.
2001). However, some
studies have shown that the consistency between the predict-
ed microwave signal and the measured value is not satisfying
(Baghdadi
et al.
2006; Baghdadi and Zribi 2006; Zribi
et al.
1997). Thus, soil moisture retrieval using multisource remote
sensing data is becoming one of the research hotspots in
recent years (Dorigo
et al.
2012; Hao
et al.
2015; O’Neill
et al.
2006; Peng
et al.
2010). Combining Synthetic Aperture Radar
(
SAR
) and optical data, active and passive microwave data
are the two most widely used methods (Babaeian
et al.
2019;
Baghdadi
et al.
2016; Li
et al.
2018; Liu
et al.
2012; Santi
et al.
2016; Zhang
et al.
2018; Zhao
et al.
2017). Studies illustrate
that the combined model can effectively improve the esti-
mated accuracy of soil moisture, which is significantly better
than the result of single model (Bousbih
et al.
2018; Shi
et al.
2017a; Kolassa
et al.
2017b).
The combined model can provide a reasonable description
of the surface scattering mechanism, but the inverse operation
of soil moisture from radar observations is a typical nonlin-
ear problem. Thus, many inversion algorithms are used to
establish the nonlinear relationship between soil moisture
and backscatter coefficient, which include a mathematical
statistics algorithm (Chen
et al.
2009; Mao
et al.
2007; Shi
et
al.
2005), a forward model inversion algorithm (Njoku and Li
ork and data assimilation algorithms
uwen 1996; Nagarajan
et al.
2011; Qin
nd Feyen 2009). The neural network al-
gorithm simulates the relationship between the backscattering
coefficient and the surface parameters by designing the hid-
den layer. It can approximate any complex nonlinear relation-
ship without involving any algorithm with specific expres-
sions to omit the analytical work on complex problems such
as physical model principles (Bacour
et al.
2006; Kimes
et
al.
2002). This method requires training data to continuously
train and learn its network in order to obtain the suitable net-
work structure through the adjustment of input parameters.
It is widely used in soil moisture retrieval based on active or
passive spaceborne sensors (Hassan-Esfahani
et al.
2015; Lu
et
al.
2017; Rodríguez-Fernández
et al.
2015; Rodríguez-Fernán-
dez
et al.
2016; Santi
et al.
2016; Yao
et al.
2017).
However, it is difficult to meet the actual needs by us-
ing only heuristic methods such as a neural network. Thus,
a swarm intelligence algorithm, as an important branch of
Liangliang Tao, and Guojie Wang are with the Collaborative
Innovation Center on Forecast and Evaluation of Meteor-
ological Disaster, School of Geographical Sciences,
Nanjing University of Information Science & Technology,
No. 219, Ningliu Road, Nanjing, Jiangsu, 210044, China
(
;
).
Xi Chen is with the School of Earth and Space Science,
Peking University, No. 5 Yiheyuan Road, Haidian District,
Beijing 100871, China (
).
Jing Li is with the Beijing Key Laboratory for Remote Sensing
of Environment and Digital Cities, Faculty of Geographical
Science, Beijing Normal University, No. 19, XinJieKouWai St.,
HaiDian District, Beijing 100875, China (
.
Qingkong Cai is with the Institute of Civil Engineering, Henan
Institute of Engineering, No. 1 Xianghe Road, Longhu Town, Xin-
zheng, Zhengzhou, Henan, 451191, China (
).
Photogrammetric Engineering & Remote Sensing
Vol. 85, No. 11, November 2019, pp. 789–798.
0099-1112/19/789–798
© 2019 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.85.11.789
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
November 2019
789
