Devkota, Medha creator Shanmugam, Mohana Sundaram Chairperson Shrestha, Sangam Examination Committee Natthachet Tangdamrongsub Examination Committee Thai Pipe Scholarship Scholarship Donor AIT Scholarship Scholarship Donor text series technical report th Pathum Thani, Thailand Asian Institute of Technology 2025 continuing eng

print

129 leaves : ill.+ 1 online This study presents a reservoir sedimentation assessment framework combining satellite remote sensing, machine learning (ML), and in-situ data over three major reservoirs in Thailand: Pasak, Bhumibol, and Sirikit. Using high-resolution Landsat 8 imagery, five methods were evaluated for surface water extent (SWE) classification: fixed and dynamic thresholding of water indices, and three supervised ML models namely Random Forest, Support Vector Machine, and Gradient Tree Boosting (GTB). Among these, GTB with terrain slope input (M5) consistently outperformed other methods, demonstrating superior classification accuracy, particularly in reservoirs with complex terrain. Surface-water area trends derived from GTB predictions were closely aligned with field-measured water levels, confirming the model's reliability for SWE monitoring. Volume{u2013}elevation (VE) curves generated from GTB predictions were compared against historical impoundment data to quantify storage capacity loss due to sedimentation. Results revealed significant storage reduction: approximately 5,333 MCM for Bhumibol, 1,187 MCM for Sirikit, and 323 MCM for Pasak. Yearly sedimentation rates were estimated at 97 MCM, 22 MCM, and 12 MCM respectively, correlating with catchment size, terrain ruggedness, and soil erodibility. Additionally, analysis of normalized sediment volumes revealed that while Bhumibol Reservoir showed the highest sediment yield relative to catchment area, Pasak Reservoir experienced the greatest storage capacity loss in proportion to its size, emphasizing the need for reservoir specific sediment management strategies. The performance of the GTB M5 significantly improved classification accuracy in reservoirs with complex topography. This method is replicable in other catchments provided high-quality DEM data, terrain variability, and careful model calibration are ensured, offering a practical solution for large-scale sedimentation monitoring. A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Water Engineering and Management Thesis (M. Eng.) - Asian Institute of Technology, 2025 Reservoir sedimentation Remote sensing images Machine learning Asian Institute of Technology. http://203.159.5.9/ait-thesis/detail.php?q=B22681 http://203.159.5.9/ait-thesis/detail.php?q=B22681 250820 20260817163633.0