Werner, Abdullah Luthfi (2026) Analisis Spasial-Temporal Perubahan Tutupan Lahan Berbasis Deep Learning untuk Pemodelan Sebaran Banjir di DAS Maluka Kabupaten Tanah Laut. Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 5014221004-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
5014221004-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (17MB) | Request a copy |
Abstract
DAS Maluka di Kabupaten Tanah Laut mengalami peningkatan frekuensi banjir lebih dari 30% selama periode 2014–2023, dengan debit puncak melampaui debit rencana Q25 hingga 40% tanpa perubahan curah hujan yang signifikan, yang mengindikasikan perubahan respons hidrologi akibat dinamika tutupan lahan. Pemodelan banjir yang umum digunakan masih mengandalkan peta tutupan lahan statis dan belum mengkuantifikasi kontribusi tiap kelas tutupan lahan terhadap debit banjir maupun luas genangan. Penelitian ini bertujuan memodelkan dinamika spasial-temporal tutupan lahan periode 2016–2035, respons hidrologi dan genangan banjir yang ditimbulkannya, serta besaran pengaruh masing-masing kelas tutupan lahan melalui pendekatan marginal effect di DAS Maluka. Klasifikasi tutupan lahan periode 2016–2025 dilakukan menggunakan citra Landsat 8 OLI dan Sentinel-2 MSI melalui arsitektur Deep Learning U-Net dan DeepLabV3, sedangkan proyeksi 2026–2035 menggunakan ANN-CA. Peta tutupan lahan tahunan digunakan sebagai input model SWAT untuk mensimulasikan neraca air dan hidrograf debit, kemudian diintegrasikan ke HEC-RAS untuk mensimulasikan genangan banjir tahunan. Pengaruh perubahan tutupan lahan terhadap indikator hidrologi dan banjir dikuantifikasi menggunakan marginal effect berbasis korelasi Spearman bivariat-parsial dengan CN-Routing, yang dinyatakan per penambahan 10 ha tiap kelas tutupan lahan. U-Net mengungguli DeepLabV3 dengan Overall Accuracy 89,73%, Mean IoU 83,37%, dan Macro-F1 90,65%, sedangkan ANN-CA mencapai akurasi validasi 96,10% (Kappa 0,9357). Luas Forest menurun 10,1% dan kawasan Urban meningkat 36,2%, mendorong Curve Number meningkat dari 90,474 menjadi 93,059, serta meningkatkan debit puncak dari 1.589,2 m³/s menjadi 1.647,4 m³/s selama periode 2016-2035. Luas genangan meningkat 19,1% pada tahun 2025 dibandingkan tahun 2016, kemudian menjadi 19,2% lebih rendah dibandingkan kondisi tahun 2016 pada tahun 2035, sementara kedalaman genangan tetap meningkat dan terkonsentrasi pada zona inti DAS. Analisis marginal effect menunjukkan bahwa setiap penambahan 10 ha Forest menurunkan Curve Number sebesar 0,0019 dan debit puncak sebesar 0,244 m³/s, sedangkan Urban meningkatkan debit puncak sebesar 0,856 m³/s dan luas genangan sebesar 1,47 ha. Hasil ini menunjukkan bahwa kehilangan hutan dan ekspansi kawasan perkotaan merupakan faktor utama intensifikasi banjir di DAS Maluka.
===================================================================================================================================
The Maluka Watershed in Tanah Laut Regency has experienced an increase in flood frequency of more than 30% during 2014–2023, with peak discharge exceeding the 25-year design flood (Q25) by up to 40% despite the absence of significant changes in rainfall. This indicates that changes in hydrological response are primarily driven by land cover dynamics. However, conventional flood modeling generally relies on static land cover maps and has not quantified the contribution of individual land cover classes to flood discharge and inundation extent. This study aims to model the spatiotemporal dynamics of land cover during 2016–2035, evaluate their impacts on hydrological response and flood inundation, and quantify the contribution of each land cover class using a marginal effect approach in the Maluka Watershed. Historical land cover maps for 2016–2025 were generated from Landsat 8 OLI and Sentinel-2 MSI imagery using Deep Learning-based U-Net and DeepLabV3 architectures, while land cover projections for 2026–2035 were produced using an Artificial Neural Network–Cellular Automata (ANN-CA) model. The annual land cover maps were used as inputs to the SWAT model to simulate the watershed water balance and flood hydrographs, which were subsequently integrated into the HEC-RAS model to simulate annual flood inundation. The effects of land cover changes on hydrological and flood characteristics were quantified using a CN-Routing-based marginal effect approach with bivariate-partial Spearman correlation, expressed per 10-ha increase in each land cover class. U-Net outperformed DeepLabV3, achieving an Overall Accuracy of 89.73%, a Mean Intersection over Union (Mean IoU) of 83.37%, and a Macro-F1 score of 90.65%, while the ANN-CA model achieved a validation accuracy of 96.10% (Kappa = 0.9357). Between 2016 and 2035, forest cover decreased by 10.1%, whereas urban areas expanded by 36.2%, resulting in an increase in the watershed Curve Number from 90.474 to 93.059 and an increase in peak discharge from 1,589.2 m³/s to 1,647.4 m³/s. Flood inundation extent increased by 19.1% in 2025 relative to 2016 but became 19.2% lower than the 2016 baseline by 2035, while flood depth continued to increase and became more concentrated within the watershed core. Marginal effect analysis showed that every additional 10 ha of forest reduced the Curve Number by 0.0019 and peak discharge by 0.244 m³/s, whereas every additional 10 ha of urban area increased peak discharge by 0.856 m³/s and flood inundation extent by 1.47 ha. These findings demonstrate that forest loss and urban expansion are the primary drivers of flood intensification in the Maluka Watershed.
Actions (login required)
 |
View Item |