Improved sea level anomaly prediction through combination of data relationship analysis and genetic programming in Singapore Regional Waters
Introduction
The understanding of the generating mechanisms and the ability to generate forecasts of tidal and non-tidal flow phenomena (also referred to sea level anomaly) and their forcing mechanisms for the highly complex Singapore Regional Waters are of both scientific and economic importance. Given its geographically constricted location, the island of Singapore, part of the Sunda Shelf, experiences a direct impact of nonlinear dynamical interactions between the South China Sea, Andaman Sea and Java Sea. The complex shallow water hydrodynamics generated due to multiple ocean currents moving into and out of its the region, combined with short term meteorological effects leads to a very high variability of the sea water level around Singapore׳s coast. In such narrow straits separating larger water bodies, it is often observed that the water currents and levels deviate significantly from their regular tidal behaviour. These deviations or residual components are generally not accounted for during ocean weather forecasting and hence seriously affect coastal planning and navigation in the region. Hence, analysis and accurate prediction of these sea level anomaly (SLA) and current anomaly becomes an important part of oceanographic modelling, especially in of such shallow water zones. A major step in their analysis and forecasting is the development of an accurate hydrodynamic model to predict the non-tidal barotropic water levels and currents in the region (Ooi et al., 2009, Ooi et al., 2011, Kurniawan et al., 2013). However, complex governing mechanisms, multi-scale, multi-dimensional, time varying, and highly non-linear dynamics of the marine systems make the oceanographic modelling efforts much more challenging. Conventional numerical models provide primary solution to this challenging task of characterizing and forecasting ocean weather (mainly water level and flow) by representing the underlying physics in terms of solvable equations. Yet, capturing the ocean dynamics in totality, accounting for the non-tidal anomaly calls for rigorous tuning of the models for further improvement. Such an exercise demands detailed domain knowledge and heavy computational effort. Hence, there is an increasing need for alternate approaches which can provide vital information leading to better domain knowledge and reduced time and effort required to tune the numerical models.
With the recent advances in measurement and information technology, there is an abundance of data available for analysis and modelling of hydrodynamic systems. Increasing spatial and temporal data coverage, better quality and reliability of data modelling and data driven techniques are becoming more favourable and acceptable by the hydrodynamic community. The data mining tools and techniques are being applied in variety of hydro-informatics applications ranging from simple data mining for pattern discovery to data driven models and numerical model error correction (Babovic et al., 2001, Babovic, 2009, Sannasiraj et al., 2005, Sun et al., 2010, Rao and Babovic, 2010, Karri et al., 2013, Karri et al., 2014, Wang and Babovic, 2014). The objectives of this paper is to explore the feasibility of applying average mutual information (AMI) theory by evaluating the amount of information contained in observed and prediction errors of non-tidal barotropic numerical modelling (i.e. assuming that the hydrodynamic model by Kurniawan et al. (2013), best represents the physics in the domain of interest given all the available data) by relating them to variables that reflect the state at which the predictions are made such as input data, state variables and model output. In addition, the present study also explores the possibility of employing ‘genetic programming’ (GP) as an offline data driven modelling tool to capture the SLA dynamics and then using them for updating the numerical model prediction in real time applications. To the best knowledge of the authors, no study has been carried out to update the numerical model SLA prediction using combination AMI and GP.
Section snippets
Data availability in the study area
The Singapore Regional Waters is defined as the area between 95°E–110°E and 6°S–11°N. It encompasses the two strategic waterways Malacca Strait and Singapore Strait, the central part of the shallow Sunda Shelf which connects the South China Sea (SCS) and the Java Sea, and part of the deep basin of the Andaman Sea. The present work is based on observations data used by Rao and Babovic (2010) and model predictions made in the year 2004 used by Kurniawan et al., 2011, Kurniawan et al., 2013. Table
Data relationship analysis
The data analysis conducted with the observed data and sea level anomaly (SLA) model simulation for the year 2004 has revealed important points regarding the time dynamics and length scales involved in the interaction between the SLA data observed at various locations around the Singapore Regional Waters, the interaction between meteorological variables and the observed SLA and SLA prediction errors. Fig. 4 shows temporal distribution of autocorrelation and AMI showing analysis of individual
Conclusions and recommendations
Data relationship analysis has been carried to evaluate the content, flow and time dynamics of information regarding the observed sea level anomaly (SLA) and its prediction errors with respect to a number of selected parameters. The analysis has been done both for the observed SLA and the SLA prediction errors. The analysis helps to establish how much information contained in the SLA prediction errors can be traced back to some of the variables. The significance test is done using average
Acknowledgements
The authors gratefully acknowledge the support and contributions of the Singapore-Delft Water Alliance (SDWA) and Deltares Strategic research funding. The research presented in this work was carried out as part of the SDWA׳s “Must-Have Box” research program (R-264-001-003-272).
References (27)
- et al.
Neuro-fuzzy and neural network techniques for forecasting sea level in Darwin Harbor, Australia
Comput. Geosci.
(2013) - et al.
Ensemble based prediction of water levels and residual currents in the Singapore regional waters for operational forecasting
Environ. Model. Softw.
(2014) - et al.
Forecasting daily lake levels using artificial intelligence approaches
Comput. Geosci.
(2012) - et al.
Comparison of genetic programming with neuro-fuzzy systems for predicting short-term water table depth fluctuations
Comput. Geosci.
(2011) - et al.
Multi-step-ahead model error prediction using time-delay neural networks combined with chaos theory
J. Hydrol.
(2010) Analysis of Observed Chaotic Data
(1996)- et al.
Information theory and neural networks for managing uncertainty in flood routing
J. Comput. Civ. Eng.
(2004) Introducing knowledge into learning based on genetic programming
J. Hydroinformatics
(2009)- et al.
Genetic programming as a model induction engine
J. Hydroinformatics
(2000) - et al.
An evolutionary approach to knowledge induction: genetic programming in hydraulic engineering
Bridging the Gap
(2001)
Evolutionary computing in hydrology
Information Theory and Reliable Comminication
Cited by (12)
Forecasting of absolute dynamic topography using deep learning algorithm with application to the Baltic Sea
2023, Computers and GeosciencesTransfer sea level learning in the Bosphorus Strait by wavelet based machine learning methods
2021, Ocean EngineeringCitation Excerpt :As regards the main focus of this research, several AI applications can be mentioned in specialized sea level prediction (Ali Ghorbani et al., 2010; Lai et al., 2019; Pashova and Popova, 2011). Performance comparison of data-driven approaches is carried out from different perspectives, i.e. monthly sea level rise prediction (Lai et al., 2019), forecast of sea level anomalies (Kurniawan et al., 2014), analyzing daily sea-level fluctuations in the coastal regions (Imani et al., 2018), and prediction of flow variables in curved channels (Gholami et al., 2018). Preliminary pre-processing techniques support the analysis and interpretation capability of data-driven models by gradually extracting multilayered information embedded into time series.
Genetic programming in water resources engineering: A state-of-the-art review
2018, Journal of HydrologyCitation Excerpt :Water level prediction in surface (rivers, lakes, reservoirs etc.) and subsurface (aquifer) water systems is crucial for various WRE applications such as irrigation, navigation, floodplain management, potable water usages. Primary examples of studies concerning modeling water level fluctuations using GP (see Table 1) include, among others, Laucelli et al., 2007; Ghorbani et al., 2010; Kisi et al., 2012, 2015; Kurniawan et al., 2014; and Zaji and Bonakdari, 2018). In a preliminary study, Laucelli et al. (2007) applied GP as a model induction engine to predict groundwater fluctuations in the Brindisi aquifer, Italy.
Generalized Cauchy model of sea level fluctuations with long-range dependence
2017, Physica A: Statistical Mechanics and its ApplicationsUnderstanding Water Level Residuals in Malacca Strait Using Genetic Programming
2016, Procedia EngineeringNumerical investigation of tide and sea level anomalies propagation in Malacca Strait
2021, Terrestrial, Atmospheric and Oceanic Sciences