Perbandingan Metode Random Forest dan XGBoost untuk Statistical Downscaling Presipitasi Di Pulau Jawa Pada Eksperimen G6sulfur (GEOMIP)

Risnanto, Dhia Evan Rafif Pratama (2026) Perbandingan Metode Random Forest dan XGBoost untuk Statistical Downscaling Presipitasi Di Pulau Jawa Pada Eksperimen G6sulfur (GEOMIP). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221072_Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Eksperimen Stratospheric Aerosol Injection (SAI) dalam kerangka Geoengineering Model Intercomparison Project (GeoMIP), khususnya eksperimen G6sulfur, dirancang untuk mensimulasikan respons sistem iklim terhadap intervensi aerosol stratosfer sebagai upaya mitigasi pemanasan global. Meskipun SAI berpotensi menekan kenaikan suhu global, implementasinya dapat memengaruhi pola presipitasi regional secara tidak seragam. Pulau Jawa, sebagai wilayah dengan kepadatan penduduk tinggi dan aktivitas ekonomi yang dominan, menjadi area yang penting untuk dikaji terkait perubahan presipitasi akibat scenario geoengineering tersebut. Namun, keterbatasan resolusi spasial keluaran model iklim global mengharuskan adanya proses statistical downscaling agar informasi presipitasi lebih representatif pada skala regional. Penelitian ini bertujuan untuk membandingkan kinerja metode Random Forest dan Extreme Gradient Boosting (XGBoost) dalam melakukan statistical downscaling presipitasi melalui proses bias correction terhadap data hasil eksperimen G6sulfur di Pulau Jawa. Data presipitasi model iklim global terlebih dahulu diregridding menggunakan metode konservatif agar sesuai dengan resolusi spasial data reanalisis ERA5 yang digunakan sebagai data referensi. Selanjutnya, model Random Forest dan XGBoost dibangun dengan optimasi hiperparameter dan dievaluasi menggunakan metrik Root Mean Square Error (RMSE), Mean Absolute Error (MAE), serta koefisien determinasi (R²) berdasarkan data uji. Hasil evaluasi menunjukkan bahwa metode Random Forest menghasilkan nilai RMSE sebesar 3,441, MAE sebesar 2,579, dan R² sebesar 0,614, sedangkan metode XGBoost memberikan kinerja yang lebih baik dengan nilai RMSE sebesar 3,225, MAE sebesar 2,432, dan R² sebesar 0,660. Perbandingan kedua metode menunjukkan bahwa XGBoost memiliki kemampuan yang lebih unggul dalam merepresentasikan variasi presipitasi bulanan di Pulau Jawa pada skenario eksperimen G6sulfur. Penelitian ini diharapkan dapat memberikan kontribusi dalam pengembangan metode statistical downscaling berbasis machine learning untuk kajian iklim geoengineering di Indonesia.
=======================================================================================================================================
The Stratospheric Aerosol Injection (SAI) experiment within the framework of the Geoengineering Model Intercomparison Project (GeoMIP), particularly the G6sulfur experiment, is designed to simulate the response of the climate system to stratospheric aerosol intervention as a potential strategy for mitigating global warming. Although SAI has the potential to suppress the increase in global temperature, its implementation may lead to spatially heterogeneous impacts on regional precipitation patterns. Java Island, as a region with high population density and dominant economic activities, represents a critical area for assessing precipitation changes under such geoengineering scenarios. However, the coarse spatial resolution of global climate model outputs necessitates the application of statistical downscaling to obtain precipitation information that is more representative at the regional scale. This study aims to compare the performance of Random Forest and Extreme Gradient Boosting (XGBoost) methods in performing statistical downscaling of precipitation through bias correction using output from the G6sulfur experiment over Java Island. Prior to model development, precipitation data from the global climate model were regridded using a conservative method to match the spatial resolution of ERA5 reanalysis data, which were employed as the reference dataset. Subsequently, Random Forest and XGBoost models were developed with hyperparameter optimization and evaluated using the Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and coefficient of determination (R²) based on the testing dataset. The evaluation results indicate that the Random Forest method produces an RMSE of 3.441, an MAE of 2.579, and an R² of 0.614, while the XGBoost method demonstrates superior performance with an RMSE of 3.225, an MAE of 2.432, and an R² of 0.660. The comparison of both methods suggests that XGBoost has a stronger capability in representing monthly precipitation variability over Java Island under the G6sulfur experiment scenario. This study is expected to contribute to the development of machine learning–based statistical downscaling methods for geoengineering climate studies in Indonesia.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	G6sulfur, Presipitasi, Random Forest, Statistical Downscaling, XGBoost. G6sulfur, Precipitation, Random Forest, Statistical Downscaling, XGBoost.
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Mathematics and Science > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Dhia Evan Rafif Pratama Risnanto
Date Deposited:	29 Jan 2026 03:26
Last Modified:	29 Jan 2026 03:26
URI:	http://repository.its.ac.id/id/eprint/130948

Actions (login required)

View Item