Ratih Prasetya, Ratih
Stasiun Meteorologi Sam Ratulangi Manado

Published : 2 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 2 Documents
Search

PENERAPAN TEKNIK DATA MINING DENGAN ALGORITMA CLASSIFICATION TREE UNTUK PREDIKSI HUJAN Prasetya, Ratih
Jurnal Widya Climago Vol 2 No 2 (2020): Adaptasi Kebiasaan Baru
Publisher : Pusdiklat BMKG

Show Abstract | Download Original | Original Source | Check in Google Scholar

Abstract

Classification is a data mining technique used to predict the relationship between data in a dataset. Prediction is done by classifying data into several different classes by considering certain factors. Classification is one of the empirical approaches that can be used for shortterm weather prediction. The classification algorithm used in this study is the Classification Tree utilizing software of Orange Data Mining 3.3.12. Furthermore, the algorithm is used to predict rain with the Confusion Matrix test parameters. The input data is a synoptic data from the Kemayoran Meteorological Station, Jakarta (96745) for 10 years (2006 - 2015) as many as 3528 datasets and consists of 8 attributes. Based on a series of processing, selection and testing of the model shows that the accuracy of the Classification Tree algorithm is 74.7% with a fair classification category where the number of correct predictions is 818 datasets out ofthe total amount of data tested that is 1095 datasets. The dominant weather attributes in the formation of rain respectively are humidity (RHavg), minimum temperature (Tmin), maximum temperature (Tmax), average temperature (Tavg) and wind direction (ddd).
Penerapan Model Arsitektur UNet untuk Peningkatan Resolusi Spasial Curah Hujan di Wilayah Pulau Jawa Berbasis Data MSWEP Putri, Nurulita Purnama; Saputro, Adhi Harmoko; Prasetya, Ratih; Soebroto, Arief Andy
Jurnal Teknologi Informasi dan Ilmu Komputer Vol 13 No 1: Februari 2026
Publisher : Fakultas Ilmu Komputer, Universitas Brawijaya

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.25126/jtiik.2026131

Abstract

Pemodelan curah hujan dengan resolusi tinggi sangat penting untuk berbagai aplikasi meteorologi dan hidrologi, termasuk peringatan dini bencana, manajemen sumber daya air, dan perubahan iklim. Namun, data curah hujan dengan resolusi tinggi sering kali tidak tersedia atau terbatas dalam cakupan wilayah dan periode waktu tertentu. Oleh karena itu, metode downscaling berbasis deep learning dapat menjadi solusi untuk meningkatkan resolusi data curah hujan dengan akurasi yang lebih baik. Penelitian ini berfokus pada evaluasi arsitektur Convolutional Neural Network (CNN) yaitu U-Net dalam melakukan downscaling data curah hujan Multi-Source Weighted-Ensemble Precipitation (MSWEP) untuk wilayah Pulau Jawa. Tujuannya adalah untuk mengevaluasi efektivitas model U-Net dalam meningkatkan resolusi data curah hujan dari 0.2° ke 0.1°. Hasil evaluasi pada data testing menunjukkan bahwa U-Net memiliki performa lebih baik jika dibandingkan ResNet. U-Net menghasilkan RMSE 0.0168, MAE 0.0107, MSE 0.00028, dan R² 0.9919, sementara ResNet memiliki RMSE 0.0188, MAE 0.0122, MSE 0.00035, dan R² 0.9899. Dengan nilai kesalahan yang lebih kecil dan akurasi lebih tinggi, U-Net terbukti lebih unggul dalam menangkap pola data curah hujan. Penelitian ini menyimpulkan bahwa U-Net lebih unggul dalam meningkatkan resolusi data curah hujan dan lebih efisien dalam menangkap pola data, menjadikannya pilihan yang lebih baik untuk aplikasi downscaling curah hujan wilayah Pulau Jawa.   Abstract High-resolution rainfall modeling is crucial for various meteorological and hydrological applications, including disaster early warning systems, water resource management, and climate change analysis. However, high-resolution rainfall data are often unavailable or limited in spatial coverage and time periods. Therefore, deep learning-based downscaling methods can serve as a promising solution to enhance the resolution of rainfall data with improved accuracy. This study focuses on evaluating the performance of a Convolutional Neural Network (CNN) architecture, specifically U-Net, for downscaling Multi-Source Weighted-Ensemble Precipitation (MSWEP) data over the island of Java. The objective is to assess the effectiveness of the U-Net model in increasing the spatial resolution of rainfall data from 0.2° to 0.1°. Evaluation on the testing dataset shows that U-Net outperforms the ResNet model, achieving an RMSE of 0.0168, MAE of 0.0107, MSE of 0.00028, and R² of 0.9919, compared to ResNet’s RMSE of 0.0188, MAE of 0.0122, MSE of 0.00035, and R² of 0.9899. With lower error values and higher accuracy, U-Net demonstrates superior capability in capturing rainfall patterns. The findings of this study conclude that U-Net is more effective in enhancing rainfall data resolution and more efficient in learning spatial patterns, making it a better choice for rainfall downscaling applications over the Java region.