Ekstraksi Kedalaman Perairan Berbasis Satelit Sentinel-2 Menggunakan Metode Extreme Gradient Boosting (XGBoost)

Salma, Ika Bella Feby (2025) Ekstraksi Kedalaman Perairan Berbasis Satelit Sentinel-2 Menggunakan Metode Extreme Gradient Boosting (XGBoost). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5016211043-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Kemajuan teknologi penginderaan jauh telah membuka peluang untuk meningkatkan metode pemetaan perairan, terutama di wilayah yang sulit dijangkau oleh survei konvensional. Teknik Satellite-Derived Bathymetry (SDB) memungkinkan estimasi kedalaman perairan secara efisien menggunakan citra satelit. Model seperti Decision Tree (DT), Random Forest (RF), dan Gradient Boosting Machine (GBM) telah digunakan untuk estimasi SDB, namun masing-masing memiliki keterbatasan, seperti overfitting pada DT, kompleksitas komputasi pada RF, dan sensitivitas terhadap parameter pada GBM. Penelitian ini bertujuan mengembangkan model estimasi kedalaman perairan dangkal di sekitar Pulau Bawean menggunakan algoritma Extreme Gradient Boosting (XGBoost) pada citra multispektral Sentinel-2. XGBoost dipilih karena kemampuannya dalam menangani dataset besar, mengurangi risiko overfitting melalui regularisasi, serta efisiensi komputasi yang tinggi. Dataset yang digunakan terdiri dari citra Sentinel-2 Level-2A dan data kedalaman in-situ dari survei Singlebeam Echosounder (SBES), dengan total 3.809 titik data hasil filter spasial dan temporal. Model dilatih menggunakan nilai reflektansi dari band 1–12 dan dievaluasi dengan metrik Root Mean Square Error (RMSE) dan R² (Koefisien Determinasi). Penelitian ini mendukung pencapaian Tujuan Pembangunan Berkelanjutan (SDGs), khususnya SDG 14: "Kehidupan di Bawah Air", dengan menyediakan data dasar laut yang akurat untuk pengelolaan sumber daya laut secara berkelanjutan. Hasil menunjukkan bahwa XGBoost mampu memetakan kedalaman laut dengan rentang 0 hingga -12 meter terhadap referensi Low Water Spring (LWS) dengan performa terbaik pada kedalaman -2 hingga -1 meter (RMSE = 3,52 m). Secara keseluruhan, XGBoost menghasilkan RMSE sebesar 1,946 meter dan R² sebesar 0,777. Model ini menunjukkan akurasi yang setara dengan GBM dan lebih unggul dibandingkan RF dan DT. Analisis sensitivitas fitur menunjukkan bahwa Band 12 (SWIR-2), Band 11 (SWIR-1), dan Band 3 (Green) merupakan input paling berpengaruh dalam proses prediksi.
================================================================================================================================
Advancements in remote sensing technology have opened up opportunities to enhance water mapping methods, especially in areas that are difficult to access through conventional surveys. The Satellite-Derived Bathymetry (SDB) technique enables efficient estimation of water depth using satellite imagery. Models such as Decision Tree (DT), Random Forest (RF), and Gradient Boosting Machine (GBM) have been applied for SDB estimation; however, each has its limitations, such as overfitting in DT, computational complexity in RF, and sensitivity to parameters in GBM. This study aims to develop a shallow-water depth estimation model around Bawean Island using the Extreme Gradient Boosting (XGBoost) algorithm applied to Sentinel-2 multispectral imagery. XGBoost was chosen for its ability to handle large datasets, reduce overfitting risk through regularization, and provide high computational efficiency. The dataset comprises Sentinel-2 Level-2A imagery and in-situ depth data from Singlebeam Echosounder (SBES) surveys, totaling 3,809 data points after spatial and temporal filtering. The model was trained using reflectance values from bands 1–12 and evaluated using Root Mean Square Error (RMSE) and R² metrics. This research supports the achievement of the Sustainable Development Goals (SDGs), particularly SDG 14: "Life Below Water", by providing accurate seafloor baseline data for the sustainable management of marine resources. Results show that XGBoost can map seabed depths ranging from 0 to -12 meters relative to the Low Water Spring (LWS) reference, with its best performance observed in the -2 to -1 meter depth range (RMSE = 3.52 m). Overall, XGBoost achieved a RMSE of 1.946 meters and R² of 0.777. The model demonstrated comparable accuracy to GBM and outperformed RF and DT. Feature sensitivity analysis indicated that Band 12 (SWIR-2), Band 11 (SWIR-1), and Band 3 (Green) were the most influential inputs for prediction.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Survei Batimetri, Remote Sensing, Machine Learning, Satellite Derived Bathymetry, Kehiidupan di Bawah Air, Bathymetric Survey, Remote Sensing, Machine Learning, Satellite Derived Bathymetry, Life Below Water.
Subjects:	G Geography. Anthropology. Recreation > G Geography (General) > G70.212 ArcGIS. Geographic information systems. G Geography. Anthropology. Recreation > G Geography (General) > G70.217 Geospatial data G Geography. Anthropology. Recreation > G Geography (General) > G70.5.I4 Remote sensing
Divisions:	Faculty of Civil, Planning, and Geo Engineering (CIVPLAN) > Geomatics Engineering > 29202-(S1) Undergraduate Thesis
Depositing User:	Ika Bella Feby Salma
Date Deposited:	23 Jul 2025 06:21
Last Modified:	23 Jul 2025 06:21
URI:	http://repository.its.ac.id/id/eprint/120784

Actions (login required)

View Item