Transfer sea level learning in the Bosphorus Strait by wavelet based machine learning methods
Introduction
The prediction of the future sea level has great importance in the planning and implementation stages of marine projects which are related to different issues such as environmental pollution (Maderich and Konstantinov, 2002), coastal management (Pethick, 2001), transportation (Whittington, 2016), renewable energy (Aboobacker, 2017) and sub-marine structures (Oda et al., 2009). Additionally, sea level records can be utilized as a reference parameter in strait hydrodynamic modeling, because surface water movement is the major operational factor that drives the coastal sea dynamics (Cucco et al., 2016). Concerns about local sea level rise and the potential impact of extreme events on coastal areas have increased the necessity of the development of the level of knowledge worldwide (Pachauri and Reisinger, 2007). However, the prediction of the future sea level for a short lead time is still limited due to the lack of in-situ records (Anderson, 2013; Hil, 2020).
Generation of site-specific information depending on the analysis of the cumulative consequence of the multiple synergic effects on a marine system needs long-term monitoring and a compact measurement network (Erol, 2011). Yet, economic and technical impossibilities can limit the number of tide gauge stations and their service life (Joseph, 2014). The absence of sufficient data encourages modeling based on information transfer obtained from one dataset to another for filling data gaps in a particular study area (Huang et al., 2003). Chang and Lin (2006) designed a neural network model to predict harmonic sea level components that depend on astronomical tide generating forces and extended the simulation to obtain future tides at six neighboring points to the location where the model is trained. Zhu et al. (2020) transferred water level fluctuation memory compiled with neural network and deep learning methods from records of multiple lakes in a region to water level prediction models of each lake in the region.
The high potential of Artificial Intelligence-based mathematical approaches in perceiving complex patterns and making rational decisions has brought them among the acceptable real-world prediction methods. The machine learning (ML) approach is a type of artificial intelligence (AI) technique that designs and develops processes based on algorithms that enable computers to learn from data. The main benefit of these prediction methods is that they infer stochastic dependency knowledge between the past and the future based on the historical data without the need of being explicitly programmed (Bontempi et al., 2012). 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. Harmonic analysis is a classical decomposition method used with the intent of determining the harmonic components (e.g. tidal effect) that contribute to sea-level changes in the frequency domain. However, due to concerns about the loss of information resulting from the steady-state assumption of classical methods, that obscure dynamic changes, wavelet has become to be used as a prominent preferred signal-processing tool in sea level time series (El-Diasty et al., 2018). Thanks to its capability in revealing time-frequency information embedded into the time series, wavelet supports the predictor algorithms to determine the meteorological events on coastal high-frequency sea surface motion (de Oliveira et al., 2009), the major tidal component (Lee, 2004), the effect of unstable driving forces and extreme environmental conditions (Ali Ghorbani et al., 2010) and, non-linear components of tidal current speed and direction (Kavousi-Fard, 2016).
The primary purpose of this study is to develop a predictive mathematical model to identify spatial and temporal relations in sea-level time series that are collected through the Bosphorus. Besides, a developed methodology using alternative model structures serves the second aim of the present study, which is to investigate the contribution and effects of pre-processing and their impact on the prediction accuracy of machine learning methods. A further objective of this study is to be able to predict the sea levels at any stations with developed models via one of the station data located in the Bosphorus. Past studies have been focused on lead-time prediction in a certain station with train and test components derived from a single data set of the station, which is used for learning and testing purposes, respectively. In this study, developed models are evaluated employing a non-dimensional metric Nash and Sutcliffe Efficiency (NSE) and a classic error metric root mean squared error (RMSE). The proposed modeling approach based on the transfer of existing system information to new situations is considered to be of practical benefit, especially in regions where there are limited available or missing data.
Section snippets
Study area and data
As the only waterway of the Black Sea to Marmara and thereby the Mediterranean Sea is Bosphorus, it has crucial importance not only for local but also for international oceanographic projects (Myers et al., 2003). How the vertical and horizontal movements of the water in the region influence the hydrodynamic behavior of Bosphorus and connected oceanographic systems is still a current interest and topic of ongoing researches (Altunkaynak and Kartal, 2019; Hossain and Meng, 2020). Though several
Model structure development
In complex system modeling, determining modeling strategy and designing an appropriate model structure are foundational to enhance the efficiency of machine learning. Since sea-level fluctuations are the cumulative responses to many hydrodynamic mechanisms varying in a wide frequency range, decomposition of sea-level data into its different components embedded into sea-level time series is the advised methodology in the literature to extract meaningful information (Cheng et al., 2017; Löfgren
Conclusion
This study uses the performances of the developed predictive mathematical models in predicting future sea levels in Bosphorus from available data that are tested in temporally and spatially different conditions. The success of the modeling approach introduced in this study, in transferring the learned modeling capabilities between stations located at different regions and/or recorded on different dates were evaluated by means of RMSE and NSE metrics.
The essential point in transferring
CRediT authorship contribution statement
Abdüsselam Altunkaynak: Conceptualization, Resources, Supervision, Writing – review & editing. Elif Kartal: Methodology, Formal analysis, Writing – original draft, Validation, Investigation, Visualization.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledge
We sincerely thank to the General Directorate of Marmaray Project of Ministry of Transportation, General Directorate of Port, Airports and Railways Construction of Turkey to provide us water level data.
References (43)
Wave energy resource assessment for eastern bay of bengal and malacca strait
Renew. Energy
(2017)- et al.
Sea water level forecasting using genetic programming and comparing the performance with Artificial Neural Networks
Comput. Geosci.
(2010) - et al.
Multi-point tidal prediction using artificial neural network with tide-generating forces
Coast. Eng.
(2006) - et al.
Analysis of tidal amplitude changes using the EMD method
Continent. Shelf Res.
(2017) - et al.
Hybrid harmonic analysis and wavelet network model for sea water level prediction
Appl. Ocean Res.
(2018) - et al.
A thematic mapping method to assess and analyze potential urban hazards and risks caused by flooding
Comput. Environ. Urban Syst.
(2020) - et al.
Development of a regional neural network for coastal water level predictions
Ocean. Eng.
(2003) - et al.
Daily sea level prediction at Chiayi coast, Taiwan using extreme learning machine and relevance vector machine
Global Planet. Change
(2018) Remote mapping of sea surface currents using HF Doppler radar networks
Meas. Ocean Curr.
(2014)
Improved sea level anomaly prediction through combination of data relationship analysis and genetic programming in Singapore Regional Waters
Comput. Geosci.
Back-propagation neural network for long-term tidal predictions
Ocean. Eng.
Sea level time series and ocean tide analysis from multipath signals at five GPS sites in different parts of the world
J. Geodyn.
Seasonal dynamics of the system sea-strait: Black Sea–Bosphorus case study
Estuar. Coast Shelf Sci.
Hydraulic calculations of postglacial connections between the mediterranean and the Black Sea
Mar. Geol.
River flow forecasting through conceptual models part I — a discussion of principles
J. Hydrol.
Comparison of wavelet and empirical mode decomposition hybrid models in drought prediction
Comput. Electron. Agric.
Daily sea level forecast at tide gauge Burgas, Bulgaria using artificial neural networks
J. Sea Res.
Coastal management and sea-level rise
Catena
Forecasting of water level in multiple temperate lakes using machine learning models
J. Hydrol.
Predicting water level fluctuations in lake van using hybrid season-neuro approach
J. Hydrol. Eng.
Cited by (9)
Transfer learning of long short-term memory analysis in significant wave height prediction off the coast of western Tohoku, Japan
2022, Ocean EngineeringCitation Excerpt :Altunkaynak and Kartala (2021) predicted the sea level at the exits of the Bosphorus Strait by applying transfer learning to a wavelet-based machine learning method. The lead time for accurate sea-level prediction increased to 7 days (Altunkaynak and Kartala, 2021). Furthermore, a combination of transfer learning and LSTM has been employed to predict the correlation between those affected by COVID-19 disease and death (Gautam, 2022) and soybean harvest (Wang et al., 2018b).
Extended lead time accurate forecasting of palmer drought severity index using hybrid wavelet-fuzzy and machine learning techniques
2021, Journal of HydrologyCitation Excerpt :Therefore, it can generate very good model results with using a simple algorithm. Disadvantages of kNN, according to Altunkaynak and Kartal (2021), the selection of suitable parameters for the development of the kNN model is pretty challenging and may have negative effects on the processing performance. The kNN tool has no mapping function that relates input to output and also makes extrapolation.
AIS-based operational phase identification using Progressive Ablation Feature Selection with machine learning for improving ship emission estimates
2024, Journal of the Air and Waste Management AssociationPredicting Sea Level Rise Using Artificial Intelligence: A Review
2023, Archives of Computational Methods in EngineeringTransfer precipitation learning via patterns of dependency matrix-based machine learning approaches
2022, Neural Computing and Applications