<![CDATA[Abstrak Prakiraan curah hujan dari model numerik perlu diverifikasi pada skala lokal sebelum digunakan dalam layanan operasional. Penelitian ini mengevaluasi kemampuan Integrated Forecasting System (IFS) dalam mendeteksi kejadian hujan di Provinsi Bali menggunakan verifikasi kategorial. Data IFS dipasangkan dengan observasi per jam dari 22 stasiun selama 1 Januari 2021 sampai 31 Desember 2025. Kejadian hujan ditetapkan ketika curah hujan melebihi 0,1 mm jam⁻¹. Kinerja dinilai melalui Probability of Detection (POD), False Alarm Ratio (FAR), Critical Success Index (CSI), Equitable Threat Score (ETS), dan bias frekuensi, kemudian dianalisis menurut musim, topografi, bulan, dan stasiun. Sebanyak 862.562 pasangan data valid menghasilkan POD 0,602, FAR 0,842, CSI 0,143, ETS 0,074, dan bias 3,820. Model menangkap 60,2% kejadian hujan, tetapi memprakirakan hujan sekitar 3,8 kali lebih sering daripada observasi. Kinerja lebih baik pada musim hujan daripada musim kemarau, dengan CSI masing-masing 0,170 dan 0,103. Dataran tinggi memberikan CSI tertinggi sebesar 0,200 dan FAR terendah sebesar 0,773. Hasil ini menunjukkan bahwa IFS masih memiliki informasi keterampilan, tetapi memerlukan kalibrasi kejadian hujan dan evaluasi berbasis ambang sebelum digunakan sebagai dasar keputusan operasional di Bali. Kata Kunci: curah hujan, IFS, musim, topografi, verifikasi kategorial. Abstract Numerical rainfall forecasts require local verification before operational use. This study evaluates the ability of the Integrated Forecasting System (IFS) to detect rainfall events in Bali Province using categorical verification. Hourly IFS data were paired with observations from 22 stations from 1 January 2021 to 31 December 2025. Rain was defined as precipitation exceeding 0.1 mm h⁻¹. Performance was assessed using the Probability of Detection (POD), False Alarm Ratio (FAR), Critical Success Index (CSI), Equitable Threat Score (ETS), and bias frekuensi, followed by seasonal, topographic, monthly, and station-based analyses. A total of 862,562 valid pairs produced POD of 0.602, FAR of 0.842, CSI of 0.143, ETS of 0.074, and bias of 3.820. The model detected 60.2% of observed rain events but forecast rain about 3.8 times more frequently than observed. Performance was better during the rainy season than the dry season, with CSI values of 0.170 and 0.103, respectively. Highland stations produced the highest CSI of 0.200 and the lowest FAR of 0.773. The IFS retains positive skill, but event calibration and threshold-based evaluation are required before its rainfall output supports operational decisions in Bali. Keywords: categorical verification, IFS, rainfall, season, topography ]]>
