Prediction of groundwater level in seashore reclaimed land using wavelet and artificial neural network-based hybrid model

Abstract

It is challenging to predict groundwater level (GWL) accurately and reliably in reclaimed coastal areas, the dynamic of which has been greatly altered by the land reclamation activities, due to its diverse engineering boundaries. In this study, the data-driven Artificial Neural Network (ANN) models (nonlinear input-output network (NIO), nonlinear autoregressive network with exogenous inputs (NARX), and wavelet-NARX (WA-NARX)) are used to predict the GWL in a newly reclaimed land on Zhoushan Island, China. The models use semi-diurnal tide (SDT) and precipitation as input variables, and the wavelet coherence (WTC) is applied to analyze the response of GWL to SDT. The results show that the WA-NARX hybrid model provides better prediction performance, especially for short-term periods. Meanwhile, we also explore the correlations between GWL and SDT in the filled and clay layers by using the WTC and the global coherence coefficient (GCC). The results show strong correlations at 0.5-, 1-, and 15-day time scales (resonance periodicities), which are then used as prediction periods for ANN models. The correlation is stronger for the filled layer than that for the clay layer indicates GWL in the filled layer is more sensitive to SDT, especially at 0.5-day time scale. The predicted results also confirm that SDT and precipitation have great influences on GWL with better prediction in the filled layer.