Convolutional Neural Network untuk Prediksi Kedalaman Dasar Laut Berdasarkan Prediksi TSS dan Citra Satelit Multispektral

Hudaya, Ahmad Ilmi (2026) Convolutional Neural Network untuk Prediksi Kedalaman Dasar Laut Berdasarkan Prediksi TSS dan Citra Satelit Multispektral. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6016241017-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Metode Satellite-Derived Bathymetry (SDB) berbasis citra optis umumnya mengalami penurunan akurasi pada perairan keruh akibat pengaruh Total Suspended Solids (TSS) terhadap reflektansi kolom air. Penelitian ini bertujuan mengembangkan model Convolutional Neural Network (CNN) berbasis citra Sentinel-2 dengan integrasi TSS untuk meningkatkan akurasi estimasi kedalaman di perairan Pantai Pekalongan, Jawa Tengah. Data yang digunakan meliputi citra Sentinel-2 12 band, 1.486 titik data batimetri, dan 26 sampel TSS lapangan. Penelitian dilakukan menggunakan dua tahap model, yaitu CNN-TSS untuk memprediksi TSS dan CNN-SDB+TSS untuk estimasi kedalaman dengan tambahan kanal TSS sebagai input. Pengujian dilakukan pada empat variasi ukuran patch yaitu 9×9, 12×12, 15×15, dan 16×16. Hasil penelitian menunjukkan bahwa model CNN-TSS mampu memprediksi TSS dengan akurasi sangat tinggi, yaitu R2=0,999–1,000, RMSE sebesar 0,135–0,194 mg/L, dan MAPE kurang dari 0,65%. Pada estimasi kedalaman, model terbaik diperoleh pada CNNSDB+TSS ukuran patch 15×15 dengan nilai R2=0,934, RMSE sebesar 0,271 m, MAE sebesar 0,137 m, dan MAPE sebesar 2,94%. Hasil ini lebih baik dibandingkan metode Stumpf (R2=0,586; RMSE = 0,677 m) dan Lyzenga (R2=0,813; RMSE = 0,443 m). Model CNN mampu mereduksi RMSE sebesar 59,97% terhadap Stumpf dan 38,83% terhadap Lyzenga. Penelitian ini menunjukkan bahwa CNN berbasis patch multispektral dengan integrasi TSS mampu menjadi metode yang akurat dan efektif untuk pemetaan batimetri di perairan keruh.
===================================================================================================================================
Optical image-based Satellite-Derived Bathymetry (SDB) methods generally experience decreased accuracy in turbid waters due to the influence of Total Suspended Solidss (TSS) on water column reflectance. This study aims to develop a Convolutional Neural Network (CNN)-based model using Sentinel-2 imagery integrated with TSS information to improve bathymetric estimation accuracy in the coastal waters of Pekalongan, Central Java, Indonesia. The data used in this study consisted of 12-band Sentinel-2 imagery, 1,486 bathymetric data points, and 26 in situ TSS samples. The proposed approach employed a two-stage modeling framework, namely CNN-TSS for TSS prediction and CNN-SDB+TSS for depth estimation by incorporating TSS as an additional input channel. Model evaluation was conducted using four patch-size variations: 9×9, 12×12, 15×15, and 16×16. The results demonstrated that the CNN-TSS model achieved very high prediction accuracy, with R2=0.999–1.000, RMSE values ranging from 0.135 to 0.194 mg/L, and MAPE values below 0.65%. For bathymetric estimation, the best performance was achieved by the CNN-SDB+TSS model with a 15×15 patch size, yielding R2=0.934, RMSE = 0.271 m, MAE = 0.137 m, and MAPE = 2.94%. These results outperformed the Stumpf method (R2=0.586; RMSE = 0.677 m) and the Lyzenga method (R2=0.813; RMSE = 0.443 m). The proposed CNN model reduced RMSE by 59.97% compared to the Stumpf method and by 38.83% compared to the Lyzenga method. This study demonstrates that a patch-based multispectral CNN integrated with TSS information can serve as an accurate and effective approach for bathymetric mapping in turbid coastal waters.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Convolutional Neural Network, Deep Learning, Total Suspended Solids, Satellite-Derived Bathymetry, Sentinel-2.
Subjects:	G Geography. Anthropology. Recreation > G Geography (General) > G109.5 Global Positioning System G Geography. Anthropology. Recreation > G Geography (General) > G70.217 Geospatial data G Geography. Anthropology. Recreation > G Geography (General) > G70.5.I4 Remote sensing G Geography. Anthropology. Recreation > GC Oceanography
Divisions:	Faculty of Civil Engineering and Planning > Geomatics Engineering > 29101-(S2) Master Thesis
Depositing User:	Ahmad Ilmi Hudaya
Date Deposited:	27 Jul 2026 06:14
Last Modified:	27 Jul 2026 06:14
URI:	http://repository.its.ac.id/id/eprint/138014

Actions (login required)

View Item