Prediksi Spasial-Temporal Pola Arus Laut Di Selat Madura Menggunakan Integrasi Model Numerik dan Convolutional Long Short-Term Memory

Hakiki, Adikuasa Iman (2026) Prediksi Spasial-Temporal Pola Arus Laut Di Selat Madura Menggunakan Integrasi Model Numerik dan Convolutional Long Short-Term Memory. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5016221029_Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (11MB) | Request a copy

Abstract

Selat Madura merupakan jalur pelayaran strategis dengan dinamika arus yang kompleks akibat interaksi pasang surut, angin muson, dan variasi topografi dasar laut, sehingga dibutuhkan sistem prediksi arus yang akurat sekaligus efisien secara komputasi. Penelitian ini mengembangkan model Convolutional Long Short-Term Memory (ConvLSTM) sebagai surrogate model yang mempelajari pemetaan spasial-temporal langsung dari keluaran model numerik MIKE 21/3 untuk prediksi arus real-time di Selat Madura. Model dilatih menggunakan data simulasi MIKE 21/3 periode Januari–Desember 2024 pada grid spasial 70×50 piksel dengan input sequence 24 timestep, output sequence 6 timestep, Channel input (komponen arus U dan V, elevasi muka air, sinus-kosinus harmonik pasut M2, dan mask domain), serta fungsi loss Mean Square Error (MSE) berbasis bobot spasial selama 30 epoch. Validasi terhadap simulasi MIKE 21/3 menghasilkan RMSE = 0,051 m/s (r = 0,943) untuk komponen U, RMSE = 0,039 m/s (r = 0,957) untuk V, dan RMSE = 0,058 m (r = 0,995) untuk elevasi muka air. Validasi terhadap observasi BIG Surabaya 2025 menunjukkan akurasi sangat baik hingga periode 1 bulan (r = 0,974–0,954) dan masih dapat dipertimbangkan pada periode 3 bulan (r = 0,722). Model ConvLSTM mampu menghasilkan prediksi 1 tahun dalam 28,01 menit dibandingkan 54,52 jam pada MIKE 21/3 dengan rasio percepatan ~116,7 kali lipat. Model ConvLSTM bersifat komplementer terhadap MIKE 21/3 dan terbukti layak sebagai surrogate model operasional untuk mendukung keselamatan navigasi dan pengelolaan wilayah pesisir Selat Madura.
================================================================================================================================
The Madura Strait is a strategically important shipping lane characterized by complex current dynamics driven by tidal interactions, monsoon winds, and seabed topography variations, creating a pressing need for current prediction systems that are both accurate and computationally efficient. This study develops a Convolutional Long Short-Term Memory (ConvLSTM) model as a surrogate model capable of directly learning spatiotemporal mappings from MIKE 21/3 numerical model outputs for real-time operational current prediction in the Madura Strait. The model was trained on MIKE 21/3 hydrodynamic simulation data spanning January–December 2024, processed onto a 70×50 spatial grid with a 24 timestep input sequence, output sequence 6 timestep across six input Channels comprising U and V current velocity components, sea surface elevation, M2 tidal harmonic sine-cosine encodings, and a domain mask, using a spatially weighted MSE loss function over 30 Training epochs. Validation against MIKE 21/3 Testing period outputs yielded RMSE = 0,051 m/s (r = 0,943) for the U component, RMSE = 0,039 m/s (r = 0,957) for V, and RMSE = 0,058 m (r = 0,995) for sea surface elevation. Validation against BIG Surabaya 2025 tidal observations demonstrated excellent predictive accuracy up to a 1-month periode (r = 0,974–0,954) and acceptable performance at 3 months (r = 0,722). The ConvLSTM model generated a full-year prediction in 28,01 minutes compared to 54,52 hours for MIKE 21/3, achieving a speedup ratio of approximately 116,7×. Operating as a complement rather than a replacement to MIKE 21/3, the ConvLSTM surrogate model proves viable for operational current forecasting and offers a practical foundation for supporting navigational safety and coastal zone management in the Madura Strait.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Selat Madura, Arus Laut, Model Numerik, Convolutional Long Short-Term Memory,Madura Strait, Ocean Currents, Numerical Model, Convolutional Long Short-Term Memory
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA911 Fluid dynamics. Hydrodynamics
Divisions:	Faculty of Civil Engineering and Planning > Geomatics Engineering > 29202-(S1) Undergraduate Thesis
Depositing User:	Adikuasa Iman Hakiki
Date Deposited:	20 Jul 2026 02:02
Last Modified:	20 Jul 2026 02:02
URI:	http://repository.its.ac.id/id/eprint/135634

Actions (login required)

View Item