Pemodelan Probabilitas Kekeringan Multi-Skala di Jawa Barat Menggunakan K-Means Clustering dan LASSO Logistic Regression

Abstract

Kekeringan merupakan bencana hidrometeorologi yang berdampak pada sektor pertanian, sumber daya air, dan kondisi sosial-ekonomi masyarakat. Provinsi Jawa Barat memiliki tingkat kerentanan kekeringan tinggi akibat variabilitas curah hujan (P), evapotranspirasi (ET0), dan pengaruh El Niño–Southern Oscillation (ENSO). Penelitian ini bertujuan menganalisis karakteristik, dinamika spasial-temporal serta memodelkan probabilitas kekeringan multi-skala di Jawa Barat menggunakan Standardized Precipitation Evapotranspiration Index (SPEI) pada skala 1, 3, 6, dan 12 bulan pada periode 1996-2025. Data curah hujan diperoleh dari CHIRPS, evapotranspirasi potensial dihitung dengan metode Penman–Monteith berbasis data iklim ERA5, dan ENSO direpresentasikan menggunakan indeks Niño3.4. Analisis dilakukan melalui statistik deskriptif, run theory, analisis spasial, uji Mann-Kendall, cross-correlation, K-Means Clustering, dan LASSO Logistic Regression. Hasil penelitian menunjukkan bahwa kekeringan di Jawa Barat bersifat heterogen secara spasial dengan kerentanan tertinggi di wilayah Pantura. Curah hujan menjadi faktor dominan pada skala pendek, sedangkan ENSO dan efek lag lebih berpengaruh pada skala menengah hingga panjang. Model LASSO Logistic Regression menghasilkan nilai AUC 0,86-0,99 yang menunjukkan kemampuan klasifikasi kekeringan sangat baik. Pendekatan zonasi berbasis K-Means mampu meningkatkan pemodelan kekeringan sesuai karakteristik wilayah.

Drought is a hydrometeorological disaster that affects the agricultural sector, water resources, and the socio-economic conditions of communities. West Java Province has a high level of drought vulnerability due to rainfall (P) variability, evapotranspiration (ET0), and the influence of the El Niño–Southern Oscillation (ENSO). This study aims to analyze the characteristics, spatial-temporal dynamics, and multi-scale drought probability modeling in West Java using the Standardized Precipitation Evapotranspiration Index (SPEI) at 1-, 3-, 6-, and 12-month timescales during the 1996–2025 period. Rainfall data were obtained from CHIRPS, potential evapotranspiration was calculated using the Penman–Monteith method based on ERA5 climate data, and ENSO variability was represented using the Niño3.4 index. The analyses were conducted using descriptive statistics, run theory, spatial analysis, the Mann–Kendall test, cross-correlation, K-Means Clustering, and LASSO Logistic Regression. The results show that drought in West Java is spatially heterogeneous, with the highest vulnerability found in the Pantura region. Rainfall was identified as the dominant factor at short timescales, while ENSO and lag effects had stronger influences at medium to long timescales. The LASSO Logistic Regression model produced AUC values ranging from 0,86 to 0,99, indicating excellent drought classification performance. The K-Means based zoning approach improved drought modeling according to regional characteristics.