Pemodelan Urban Heat Island Menggunakan Spatial Ensemble U-Net Dan LightGBM Berbasis Citra Landsat 8/9 Dan Sentinel-2

Daniswara, Pradipta Arya (2026) Pemodelan Urban Heat Island Menggunakan Spatial Ensemble U-Net Dan LightGBM Berbasis Citra Landsat 8/9 Dan Sentinel-2. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221185-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Urban Heat Island (UHI) di Jakarta telah memperparah risiko kesehatan masyarakat dan beban energi, namun belum tersedia pemetaan resolusi tinggi yang memadai untuk mendukung kebijakan mitigasi yang efektif. Penelitian ini mengembangkan model ensemble yang mengintegrasikan Gradient Boosting Machine (GBM) dan Convolutional Neural Network (CNN) berbasis data satelit multi-sensor (Landsat 8/9 dan Sentinel-2) untuk memetakan UHI Jakarta pada resolusi spasial 30 meter. Evaluasi menunjukkan spatial ensemble (fusion) mencapai R² sebesar 0,8678, RMSE sebesar 2,3012°C, dan MAE sebesar 1,7685°C, melampaui target akurasi yang ditetapkan (R² ≥ 0,85, RMSE ≤ 2,5°C, MAE ≤ 2,0°C), dengan peningkatan R² hingga +10,08% dibandingkan CNN standalone. Peta UHI yang dihasilkan mengidentifikasi intensitas rata-rata sebesar 5,09°C, dengan lebih dari 87% wilayah mengalami anomali panas positif, sekitar 42% area terklasifikasi Very Strong hingga Extreme (>6°C). Penelitian ini menyediakan kerangka kerja yang dapat diadaptasi untuk megakota tropis di ASEAN guna mendorong kebijakan berbasis data dalam mengurangi paparan panas, risiko kesehatan, dan konsumsi energi.
==================================================================================================================================
The Urban Heat Island (UHI) effect in Jakarta has intensified public health risks and increased energy demand; however, high-resolution mapping necessary to support effective mitigation policies remains limited. This study develops an ensemble model integrating Gradient Boosting Machine (GBM) and Convolutional Neural Network (CNN) using multi-sensor satellite data (Landsat 8/9 and Sentinel-2) to map UHI in Jakarta at a 30-meter spatial resolution. Evaluation results show that the spatial ensemble (fusion) achieves an R² of 0.8678, RMSE of 2.3012°C, and MAE of 1.7685°C, surpassing the predefined performance targets (R² ≥ 0.85, RMSE ≤ 2.5°C, MAE ≤ 2.0°C), with an improvement of up to +10.08% in R² compared to the standalone CNN model. The resulting UHI map identifies an average intensity of 5.09°C, with more than 87% of the area experiencing positive heat anomalies, and approximately 42% classified as Very Strong to Extreme (>6°C). This research provides a transferable framework for tropical megacities in ASEAN to support data-driven policies aimed at reducing heat exposure, health risks, and energy consumption.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Deep Learning, Ensemble Model, Fusi Data Satelit, Pemetaan Resolusi Tinggi, Urban Heat Island, Deep Learning, Ensemble Model, High-Resolution Mapping, Satellite Data Fusion, Urban Heat Island
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Pradipta Arya Daniswara
Date Deposited:	18 Jun 2026 07:40
Last Modified:	18 Jun 2026 07:40
URI:	http://repository.its.ac.id/id/eprint/133893

Actions (login required)

View Item