Please use this identifier to cite or link to this item: http://repository.ipb.ac.id/handle/123456789/171455
Title: Downscaling Curah Hujan Pulau Jawa Berbasis Pix2Pix
Other Titles: Rainfall Downscaling over Java Island Using Pix2Pix
Authors: Faqih, Akhmad
Muttaqien, Furqon Hensan
Jannah, Salamah Zukhrufa
Issue Date: 2025
Publisher: IPB University
Abstract: Curah hujan menjadi unsur iklim dengam keragaman dan fluktuasi yang tinggi di Indonesia. Pulau Jawa memiliki topografi yang kompleks sehingga curah hujan antar wilayahnya sangat bervariasi. Peningkatan resolusi dan akurasi data curah hujan dapat dilakukan melalui metode statistical downscaling. Metode ini didasarkan pada pendekatan berbasis data dan memiliki keunggulan dalam efisiensi komputasi, akan tetapi lemah dalam sisi akurasi data. Deep learning dengan metode Pix2Pix dapat menjadi solusi karena memiliki kemampuan yang baik dalam menangani bentuk data yang beragam seperti data iklim. Evaluasi dan analisis dilakukan pada hasil downscaling Pix2Pix dengan data RegCM non-hydrostatic yang mempertimbangkan aspek topografi. Nilai model terbaik diperoleh pada epoch ke-20 dengan RMSE 9,66 mm, FSS 0,66, SSIM 0,96, dan TSS 0,72. Model menghasilkan data curah hujan yang cukup baik untuk intensitas ringan hingga sedang namun belum optimal dalam hujan ekstrim, terutama di dataran tinggi baik dalam skala harian maupun tahunan. Data hasil model dapat menggambarkan variasi curah hujan musiman secara spasial bahkan saat terjadinya fenomena iklim ENSO. Berdasarkan hasil tersebut, metode Pix2Pix dapat menjadi alternatif pendekatan downscaling curah hujan yang lebih cepat dan akurat untuk mendukung analisis variabilitas serta dampak curah hujan di berbagai bidang.
Rainfall is a climate variable with high variations and fluctuations in Indonesia, with Java island has a complex topography that contribute to rainfall disparities across regions. Improved resolution and accuracy of rainfall data can be done through statistical downscaling method. This method is based on a data-driven approach and has advantages in computational efficiency. Deep learning with Pix2Pix method can be a potential solution due to its robust performance in diverse datasets, such as climate data. Evaluation and analysis are based on Pix2Pix downscaling results with non-hydrostatic RegCM results that consider topographic aspects. The best model values were achieved at the 20th epoch, yielding an RMSE of 9,66 mm, an FSS of 0,66, an SSIM of 0,96, and an a TSS of 0,72. The model can generate accurate rainfall data for light to moderate rainfall but not optimal for extreme rainfall, particularly within highlands areas, on both daily and annual timescales. The model output can capture seasonal spatial variation in rainfall, including during ENSO climate events. Based on these results, the model can serve as an alternative approach for more efficient and accurate rainfall downscaling, thereby supporting improved assessments of rainfall impacts across various sectors.
URI: http://repository.ipb.ac.id/handle/123456789/171455
Appears in Collections:UT - Geophysics and Meteorology

Files in This Item:
File Description SizeFormat 
cover_G2401211037_ff36ffd7aeee431bb678bc0c41b5898b.pdfCover610.53 kBAdobe PDFView/Open
fulltext_G2401211037_15c8c35e87914512a73722bfad78c1d1.pdf
  Restricted Access
Fulltext2.23 MBAdobe PDFView/Open
lampiran_G2401211037_674f569d13c54d76b8ceb95f027935dd.pdf
  Restricted Access
Lampiran568.23 kBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.