J. Zhang, et al.
Journal of Hydrology 577 (2019) 123948
presented an approach for generating transient groundwater model
inputs to assess the long-term impacts of land reclamation on ground-
water resources. Guo and Jiao (2007) derived the analytical solution for
calculating the position of the saltwater-freshwater interface and found
that the saline wedge moved seaward after reclamation. Jiao et al.
(Adamowski and Chan, 2011; Gorgij et al., 2017; Kavousi-Fard, 2017;
Kisi and Shiri, 2011; Rajaee and Shahabi, 2016; Shoaib et al., 2016;
Wang and Jing, 2003; J. Zhang et al., 2018). Based on the theoretical
basis of wavelet transform and ANN, it can be inferred that the hybrid
model could have the advantages of both methods (Anctil and Tape,
2004; J. Zhang et al., 2018). Thus, the wavelet-ANN hybrid model (WA-
ANN) is a good option to predict the groundwater variation. To the best
of our knowledge, this method has not been applied in predicting GWL
at coastal reclamation area before. We need to compare different ANN
and WA-ANN models to find the most suitable prediction model for this
study area.
Tidal effects play an important role in coastal unconfined aquifers
and beaches (Gribovszki et al., 2015). For the reclaimed land with an
impermeable dike in our case, the tide can affect groundwater variation
with draining pipes, which were built in the dike. Therefore, the
mathematical model not only contains porous flow but also pipe flow.
Although previous studies have widely investigated for the effect of tide
on the groundwater fluctuation in the beach under natural conditions, it
is seldom studied for the response of GWL to tidal oscillations in re-
clamation land. Moreover, since the SDT is a high-frequent and non-
stationary signal, the traditional method is difficult to capture the
correlation between SDT and GWL. The wavelet coherence (WTC) is a
useful tool to analyze the correlation between two signals in the time-
frequency domain, which is particularly applicable to non-stationary
systems (Grinsted et al., 2004; Torrence and Compo, 1998). This ap-
proach was applied to many fields, such as the correlation between
GWL and climate teleconnection patterns (Gurdak et al., 2007; Holman
et al., 2011; Kuss and Gurdak, 2014), between GWL and precipitation
(Oh et al., 2017; Qi et al., 2018; Yu and Lin, 2015), and the effect of
monsoon on spring discharge (Huo et al., 2016; Zhang et al., 2017).
There is little previous research about the linkage between the tide and
GWL using wavelet analysis. In this study, the correlation structure and
resonance periodicities between SDT and GWL are analyzed in the time-
frequency domain. The resonance periodicities as the prediction periods
of ANN models can improve the accuracy of prediction and make the
analysis more reliable. Moreover, the different correlation between SDT
and GWL in reclamation and natural layers helps to analyze their dif-
ferences.
(
2006) discussed the impact of land reclamation on groundwater
system in coastal areas in Hong Kong by using two-dimensional mul-
tilayered numerical model and calibrated the parameters by comparing
the simulated and observed GWL. Previous studies focused on steady
groundwater flow or large-scale numerical simulation after land re-
clamation, but the short-term variation of groundwater flow was not
considered. The GWL can be controlled effectively by grasping the
dynamic change of GWL in a timely manner during the construction,
ensuring the safety of reclamation area construction and further for-
mulating the scheme of groundwater utilization. Therefore, real-time
prediction of GWL is fundamental to studying the impact of land re-
clamation on the groundwater system at a short-term period and to
implementing the overall planning and management of reclamation
areas. However, it is difficult to utilize either the analytical or the nu-
merical methods to predict GWL directly for the complex and diverse
engineering constructions of reclamation. In general, the change of
GWL is mainly dependent on the range of reclaimed land, hydraulic
conductivity of the filling materials, and the tide variation and rainfall
in the reclamation area (Jiao et al., 2006). On one hand, with the an-
thropogenic extension of land, the discharge of groundwater to sea-
water is weakened, leading to rising of GWL in the original coastal area
(
Guo and Jiao, 2007). On the other hand, due to the infiltration re-
duction caused by a large area of the impermeable land surface, the
reclamation could induce heavier seawater intrusion (Shiguo and Yi,
2013). The flow pattern variation of groundwater depends on the
specific hydrogeological conditions at different sites. For the land re-
clamation area, the soil is factitiously divided into the filled and the
clay layers. Different hydraulic conductivity results in the different
pattern of GWL variation, thus it is impossible to study reclamation
areas according to the general law of groundwater flow system in
coastal areas. It requires analyzing the differences between the two
layers and the different responses to the tide in order to explore the
impact of land reclamation on GWL more specifically.
To overcome the difficulty in building mathematical models in the
reclaimed land due to the complex system or lacking of detailed in-
formation, data-driven based models could be used to make hydro-
logical predictions (Anctil and Tape, 2004; Nourani et al., 2014; Shoaib
et al., 2016; Taormina and Chau, 2015; Wang and Jing, 2003). Artificial
Neural Network (ANN) is a useful method, which involves the wide
range of applications of hydrological analysis and prediction of non-
linear systems (Aqil et al., 2007; Castellano-Mendez et al., 2004;
Moosavi et al., 2013). ANNs are effective in constructing models for
nonlinear function without explicitly describing the underlying com-
plex processes in mathematical forms (Shoaib et al., 2016; Wang and
Jing, 2003). The nonlinear autoregressive network with exogenous in-
puts (NARX) was applied for forecasting the monthly GWL in several
wells in the Mississippi River Valley aquifer and southwest Germany
In this study, we use GWL, SDT and precipitation dataset from a
research site in Zhoushan Island, which reflects the impact of the pre-
cipitation and SDT variations on GWL. The objective of this study is to
develop the best data-based time-series predicting models for GWL
fluctuations, which is conducive to the research of groundwater dy-
namic system in the coastal reclamation areas. To this aim, the ANN
and WA-ANN models, using the precipitation and SDT as inputs, are
developed. Since the tidal variation is the main driving force for GWL
fluctuation in the coastal area (Gribovszki et al., 2015). We address the
specific questions of (1) Is there any difference in groundwater dy-
namics between the filled and the clay layers? (2) How does the cor-
relation between GWL and SDT differ for two layers? (3) How do dif-
ferent models, a hybrid model (WA-ANN model) and other ANN
models, perform for the same dataset? To the best of our knowledge,
few studies explore the correlation between GWL and SDT using WTC
and apply ANN and WA-ANN hybrid models to predict GWL with the
SDT as input in land reclamation area.
(
Guzman et al., 2017; Wunsch et al., 2018), and the results showed
good performance. However, ANN may not be able to cope with highly
varied non-stationary signals well unless the original data was pre-
processed (Cannas et al., 2005; J. Zhang et al., 2018). Wavelet trans-
form decomposes the original time series and extracts nontrivial and
potentially useful information from the large data sets of historical re-
cords (Moosavi et al., 2013). Therefore, it helps to better predict non-
stationary signals by decomposing them into sub-signals at different
temporal levels (Adamowski and Chan, 2011; Kisi and Shiri, 2011;
Mohanty et al., 2015; Moosavi et al., 2013; Nourani et al., 2015). It has
been reported that the data-driven technique combined ANN with
wavelet analysis performs efficiently and has been applied over a wide
range of hydrological processes, e.g., precipitation, stream-flow, rain-
fall-runoff, groundwater, evapotranspiration, and water quality
2. Materials and methods
2.1. Site description
2.1.1. Geography and climate
Zhoushan Island is the principal inhabited island of the Zhoushan
Archipelago on the mid-eastern seaboard of China. Its landform mainly
consists of the central hill range, the surrounding plains, and the tidal
flat wetlands. The tidal flats reclamation along the coast have been
started in the 1970s for mariculture, aquaculture, and harbor industry
2