Peramalan Curah Hujan Menggunakan Extreme Learning Machine Dan Analisis Risiko Curah Hujan Ekstrem Menggunakan Extreme Value Theory Di Kota Samarinda

Zaman, Saifan Nur (2026) Peramalan Curah Hujan Menggunakan Extreme Learning Machine Dan Analisis Risiko Curah Hujan Ekstrem Menggunakan Extreme Value Theory Di Kota Samarinda. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5006221072-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Curah hujan ekstrem merupakan fenomena hidrometeorologis yang jarang terjadi, tetapi dapat memberikan dampak besar terhadap risiko banjir dan bencana terkait. Kota Samarinda memiliki kerentanan terhadap kejadian curah hujan ekstrem karena kondisi geografis, sistem hidrologi, dan perkembangan wilayah perkotaan. Penelitian ini bertujuan untuk menganalisis karakteristik curah hujan harian, membangun model peramalan curah hujan jangka pendek menggunakan Extreme Learning Machine (ELM), serta mengestimasi risiko curah hujan ekstrem jangka panjang menggunakan Extreme Value Theory (EVT). Data yang digunakan merupakan data curah hujan harian dan kelembapan relatif udara Kota Samarinda periode 1 Januari 2000 hingga 31 Desember 2025 yang diperoleh dari NASA POWER. Pemodelan ELM dilakukan menggunakan kombinasi input lag berdasarkan Partial Autocorrelation Function dan dievaluasi menggunakan Mean Square Error. Analisis EVT dilakukan melalui pendekatan Block Maxima dengan distribusi Generalized Extreme Value dan pendekatan Peak Over Threshold dengan distribusi Generalized Pareto Distribution. Sebelum pemodelan EVT, dilakukan uji tren menggunakan Mann-Kendall Test dan analisis korelasi menggunakan Spearman’s Correlation terhadap data ekstrem dan kovariat. Pemilihan model terbaik dilakukan menggunakan Akaike Information Criterion, Bayesian Information Criterion, dan Likelihood Ratio Test. Hasil penelitian menunjukkan bahwa data curah hujan harian memiliki distribusi menceng ke kanan dengan nilai skewness sebesar 3,17 dan kurtosis sebesar 19,22, sehingga mengindikasikan adanya kejadian ekstrem. Model ELM terbaik diperoleh pada kombinasi lag 1, 2, 3, 5, 7, 8, dan 16 dengan 50 hidden neuron, menghasilkan MSE testing sebesar 94,3056. Hasil peramalan 31 hari ke depan menunjukkan curah hujan berada pada rentang 8,31 mm hingga 9,64 mm. Pada pemodelan EVT, model GEV terbaik adalah model dengan parameter scale sebagai fungsi kelembapan relatif. Model GPD terbaik juga diperoleh pada model dengan parameter scale sebagai fungsi kelembapan relatif dengan threshold sebesar 42,4304 mm/hari. Hasil return level menunjukkan bahwa estimasi curah hujan ekstrem meningkat seiring bertambahnya return period. Pada tahun 2025, return level periode ulang 100 tahun sebesar 167,7091 mm/hari berdasarkan model GEV dan 127,8849 mm/hari berdasarkan model GPD. Model GEV memberikan estimasi yang lebih konservatif, sedangkan model GPD memberikan estimasi yang lebih stabil. Dengan demikian, ELM dapat digunakan untuk menggambarkan dinamika curah hujan jangka pendek, sedangkan EVT memberikan informasi kuantitatif mengenai risiko curah hujan ekstrem jangka panjang di Kota Samarinda.
=================================================================================================================================
Extreme rainfall is a hydrometeorological phenomenon that occurs infrequently but may cause substantial impacts on flood risk and related disasters. Samarinda City is vulnerable to extreme rainfall events due to its geographical characteristics, hydrological system, and urban development. This study aims to analyze the characteristics of daily rainfall, develop a short-term rainfall forecasting model using Extreme Learning Machine (ELM), and estimate long-term extreme rainfall risk using Extreme Value Theory (EVT). The data used in this study consist of daily rainfall and relative humidity records in Samarinda City from January 1, 2000 to December 31, 2025, obtained from NASA POWER. The ELM model was developed using lag combinations selected based on the Partial Autocorrelation Function and evaluated using Mean Square Error. The EVT analysis was conducted using the Block Maxima approach with the Generalized Extreme Value distribution and the Peak Over Threshold approach with the Generalized Pareto Distribution. Prior to EVT modeling, trend analysis using the Mann-Kendall Test and correlation analysis using Spearman’s Correlation were performed on the extreme data and covariates. The best models were selected based on the Akaike Information Criterion, Bayesian Information Criterion, and Likelihood Ratio Test. The results show that the daily rainfall data are right-skewed, with a skewness value of 3.17 and a kurtosis value of 19.22, indicating the presence of extreme events. The best ELM model was obtained using lags 1, 2, 3, 5, 7, 8, and 16 with 50 hidden neurons, resulting in a testing MSE of 94.3056. The 31-day ahead forecast indicates that rainfall is expected to range from 8.31 mm to 9.64 mm. In the EVT modeling, the best GEV model was obtained by allowing the scale parameter to depend on relative humidity. Similarly, the best GPD model was obtained by allowing the scale parameter to depend on relative humidity with a threshold of 42.4304 mm/day. The return level results show that extreme rainfall estimates increase as the return period becomes longer. In 2025, the 100-year return level was estimated at 167.7091 mm/day using the GEV model and 127.8849 mm/day using the GPD model. The GEV model provides a more conservative estimate, while the GPD model provides a more stable estimate. Therefore, ELM can be used to describe short-term rainfall dynamics, while EVT provides quantitative information on long-term extreme rainfall risk in Samarinda City.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Block Maxima, Extreme Learning Machine, Generalized Extreme Value, Generalized Pareto Distribution, Peak Over Threshold, Return Level.
Subjects:	Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry)
Divisions:	Faculty of Mathematics, Computation, and Data Science > Actuaria > 94203-(S1) Undergraduate Thesis
Depositing User:	Saifan Nur Zaman
Date Deposited:	17 Jul 2026 07:39
Last Modified:	17 Jul 2026 07:39
URI:	http://repository.its.ac.id/id/eprint/135337

Actions (login required)

View Item