Peramalan Lokasi Sampah Laut Terapung Menggunakan Spatio-Temporal Deep Learning dan Lagrangian Particle Tracking Model

Prasetya, Riza Budi (2025) Peramalan Lokasi Sampah Laut Terapung Menggunakan Spatio-Temporal Deep Learning dan Lagrangian Particle Tracking Model. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6025232028-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (162MB)

Abstract

Sampah laut semakin mengancam ekosistem pesisir, khususnya di wilayah yang bergantung pada budidaya laut, seperti Sulawesi Selatan, Indonesia. Penelitian ini mengusulkan kerangka kerja hibrida yang menggabungkan enam model Spatio-Temporal Deep Learning (ST-DL), yaitu Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Transformer, serta tiga varian hibridanya: CNN+LSTM, Convolutional LSTM (ConvLSTM), dan CNN+Transformer, dengan Lagrangian Particle Tracking Model (LPTM). Model dilatih menggunakan data oseanografi dan atmosfer dari HYCOM dan ERA5 selama periode 2019–2020 untuk memprediksi arus permukaan laut (SSC) pada horizon 3, 15, dan 30 jam. Prediksi arus tersebut kemudian digunakan untuk menyimulasikan pergerakan sampah laut dari sepuluh muara sungai utama di Sulawesi Selatan.Hasil evaluasi menunjukkan bahwa model hibrida secara konsisten unggul dibanding model tunggal, baik dalam akurasi prediksi SSC maupun simulasi lintasan sampah laut. CNN+Transformer mencatatkan akurasi prediksi SSC terbaik dengan rata-rata Root Mean Square Error (RMSE) sebesar 0.0674 m/s. Dalam hal akurasi lintasan, CNN+Transformer menghasilkan Mean Positional Distance (MPD) terendah sebesar 0.0712◦ dan Dynamic Time Warping (DTW) sebesar 4.9092 pada bulan Juni, sedangkan CNN+LSTM menunjukkan performa terbaik pada bulan Desember dengan MPD sebesar 0.1517◦ dan DTW sebesar 23.9682. Temuan ini menyoroti pentingnya keterkaitan antara akurasi prediksi SSC dan kesesuaian lintasan partikel dengan data aktual, baik dari segi arah, posisi, maupun waktu tempuh. Selain itu, hasil ini menunjukkan potensi model hibrida ST-DL sebagai alat prediktif dalam pengawasan, sistem peringatan dini, atau perencanaan pembersihan sampah laut secara adaptif di wilayah pesisir yang dinamis.
====================================================================================================================================
Marine debris increasingly threatens coastal ecosystems, particularly in aquaculture dependent regions such as South Sulawesi, Indonesia. To address this challenge, we propose a hybrid forecasting framework that integrates six spatiotemporal deep learning (ST-DL) models—Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Transformer, and their hybrid variants CNN+LSTM, Convolutional LSTM (ConvLSTM), and CNN+Transformer—with a Lagrangian Particle Tracking Model (LPTM). These models were trained using multivariate oceanographic and atmospheric data from HYCOM and ERA5 for the 2019–2020 period to forecast sea surface currents (SSC) at 3-hour, 15-hour, and 30-hour horizons. The predicted currents were then used to simulate debris trajectories originating from ten major river mouths in South Sulawesi.Evaluation results show that hybrid models consistently outperform standalone architectures in both current forecasting and debris trajectory simulation. CNN+Transformer achieved the best SSC forecast accuracy with an average Root Mean Square Error (RMSE) of 0.0674 m/s. For trajectory accuracy, CNN+Transformer yielded the lowest Mean Positional Distance (MPD) of 0.0712° and Dynamic Time Warping (DTW) of 4.9092 in June, while CNN+LSTM performed best in December with an MPD of 0.1517° and DTW of 23.9682. These results highlight the importance of coupling SSC accuracy with realistic transport representation, and demonstrate the potential of hybrid ST-DL models as robust tools for operational marine debris forecasting and coastal management in dynamic nearshore environments

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Peramalan Lokasi Sampah Laut, Spatio-Temporal Deep Learning, Model Hibrida, Arus Permukaan Laut, Lagrangian Particle Tracking Model, Marine Debris Trajectory Forecasting, Spatio-Temporal Deep Learning, Hybrid Models, Sea Surface Currents, Lagrangian Particle Tracking
Subjects:	T Technology > T Technology (General) > T174 Technological forecasting T Technology > T Technology (General) > T57.62 Simulation
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	Riza Budi Prasetya
Date Deposited:	30 Jul 2025 04:06
Last Modified:	30 Jul 2025 04:18
URI:	http://repository.its.ac.id/id/eprint/123145

Actions (login required)

View Item