Perbandingan Metode Klasifikasi Sedimen Dasar Laut Menggunakan Angular Range Analysis (ARA) Dan Machine Learning Random Forest Berbasis Data Backscatter Multibeam Echosounder

Asmarawan, Farrel Athaya (2026) Perbandingan Metode Klasifikasi Sedimen Dasar Laut Menggunakan Angular Range Analysis (ARA) Dan Machine Learning Random Forest Berbasis Data Backscatter Multibeam Echosounder. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5016221067-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Perkembangan survei hidrografi modern semakin krusial untuk perencanaan dan pengelolaan kegiatan di perairan, khususnya dalam mendukung pemetaan detail dasar laut melalui teknologi Multibeam Echosounder (MBES). MBES tidak hanya menghasilkan data batimetri dan kolom air, tetapi juga memberikan data backscatter yang merepresentasikan sifat fisik dasar laut seperti kekasaran, kekerasan, dan tipe sedimen, di mana intensitas gelombang suara yang dipantulkan sangat dipengaruhi oleh kepadatan dan tekstur sedimen bawah laut. Inovasi metode pengolahan backscatter terus berkembang, dari Angular Range Analysis (ARA) yang memanfaatkan variasi intensitas sinyal akustik pada berbagai sudut insidensi untuk membangun kurva respons sudut, hingga penggunaan algoritma machine learning seperti Random Forest (RF) yang dapat mengotomatisasi proses klasifikasi sedimen secara efisien pada area yang luas dan data set kompleks. Penelitian ini membandingkan akurasi dan efektivitas metode ARA serta RF dalam klasifikasi sedimen dasar laut berbasis data backscatter MBES, dengan validasi menggunakan data sampel sedimen di lapangan serta analisis statistik Kappa dan MAPE untuk menguji konsistensi dan signifikansi model klasifikasi. Hasil penelitian menunjukkan bahwa metode RF mencapai Overall Accuracy 87,5% dengan Koefisien Kappa 0,808, sedangkan metode ARA hanya mencapai Overall Accuracy 45% dengan Koefisien Kappa -0,016, bahkan cenderung mengklasifikasikan seluruh sampel ke dalam satu jenis sedimen (clay) akibat keterbatasannya dalam membedakan sedimen dengan karakteristik akustik yang berdekatan namun secara geologi berbeda. Sebaliknya, RF mampu menghasilkan klasifikasi yang jauh lebih akurat. Meskipun demikian, ARA tetap memiliki keunggulan operasional karena tidak memerlukan data sampel sedimen fisik sebagai data latih, sedangkan RF memerlukan data tersebut serta kemampuan pemrograman (dalam penelitian ini menggunakan Python). Hasil penelitian ini diharapkan dapat memperkuat informasi dasar laut dan mendukung perencanaan kelautan berbasis teknologi modern, dengan model machine learning yang akurat dan skalabel untuk pemetaan sedimen di wilayah survei yang kompleks.
================================================================================================================================
The development of modern hydrographic surveys is becoming increasingly crucial for planning and managing marine activities, particularly in supporting detailed seabed mapping through Multibeam Echosounder (MBES) technology. MBES not only produces bathymetric and water column data but also provides backscatter data representing the physical properties of the seabed such as roughness, hardness, and sediment type, where the intensity of the reflected acoustic signal is strongly influenced by the density and texture of seabed sediments. Backscatter processing methods continue to evolve, from Angular Range Analysis (ARA), which utilizes variations in acoustic signal intensity across incidence angles to build an angular response curve, to machine learning algorithms such as Random Forest (RF), which can efficiently automate sediment classification over large areas and complex datasets. This study compares the accuracy and effectiveness of the ARA and RF methods in seabed sediment classification based on MBES backscatter data, validated using field ground-truth sediment data along with Kappa and MAPE statistical analyses to assess the consistency and significance of the classification models. The results show that the RF method achieved an Overall Accuracy of 87.5% with a Kappa Coefficient of 0.808, while the ARA method only achieved an Overall Accuracy of 45% with a Kappa Coefficient of -0.016, even tending to classify all samples into a single sediment type (clay) due to its limited ability to distinguish sediments with similar acoustic characteristics but different geological compositions. In contrast, RF was able to do more accurate classifications. Nevertheless, ARA still offers operational advantages as it does not require physical sediment sample data for training, whereas RF requires such data as well as programming skills (Python, in this study). The findings of this study are expected to strengthen seabed information and support marine planning based on modern technology, with an accurate and scalable machine learning model for sediment mapping in complex survey areas.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Sedimen, Angular Range Analysis (ARA), Random Forest, Sediment, Angular Range Analysis (ARA), Random Forest
Subjects:	Q Science Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QE Geology
Divisions:	Faculty of Civil Engineering and Planning > Geomatics Engineering > 29202-(S1) Undergraduate Thesis
Depositing User:	Farrel Athaya Asmarawan
Date Deposited:	16 Jul 2026 04:37
Last Modified:	16 Jul 2026 04:37
URI:	http://repository.its.ac.id/id/eprint/134930

Actions (login required)

View Item