Kinerja Model S2S dalam Memprediksi Kejadian Hujan Ekstrem 7 Maret 2024 di Sumatra Barat

Abstract

Peristiwa hujan ekstrem di Sumatra Barat pada 7 Maret 2024 yang memicu banjir dipengaruhi oleh interaksi antara Madden-Julian Oscillation (MJO) fase 4, Westerly Wind Burst (WWB), dan gelombang Kelvin, serta pengaruh orografis Pegunungan Bukit Barisan. Penelitian ini bertujuan mengevaluasi kinerja model S2S ECMWF dalam memprediksi peristiwa tersebut berdasarkan variasi initial date dan lead day. Data yang digunakan meliputi prakiraan real-time S2S ECMWF, reanalisis ERA5, observasi IMERG, OLR dari NOAA CDR, serta data curah hujan stasiun BMKG. Evaluasi dilakukan terhadap lima variabel atmosfer menggunakan pattern correlation dan Root Mean Square Error (RMSE). Hasil penelitian menunjukkan bahwa model S2S mampu merepresentasikan pola spasial curah hujan, tetapi secara konsisten underestimate dibandingkan observasi pada seluruh lead day. Kinerja model S2S menurun seiring bertambahnya lead day, dengan kecepatan angin menjadi variabel yang diprediksi paling baik (r = 0,91 pada lead day 2) dan VIMFC sebagai variabel dengan performa paling rendah (r < 0,5 untuk seluruh lead day). WWB menunjukkan performa prediksi yang tinggi dengan korelasi maksimum sebesar r = 0,80 meskipun waktu kemunculannya cenderung bergeser dibandingkan pada data observasi.

The extreme rainfall event in West Sumatra on 7 March 2024, which triggered flooding, was influenced by the interaction of phase 4 Madden-Julian Oscillation (MJO), a Westerly Wind Burst (WWB), Kelvin waves, and the orographic effects of the Bukit Barisan mountain range. This study aims to evaluate the performance of the ECMWF Subseasonal-to-Seasonal (S2S) model in predicting this event based on variations in initial date and lead day. The datasets used include ECMWF S2S real-time forecasts, ERA5 reanalysis, IMERG observations, NOAA Climate Data Record (CDR) outgoing longwave radiation (OLR), and rainfall observations from BMKG stations. Model evaluation was conducted for five atmospheric variables using pattern correlation and Root Mean Square Error (RMSE). The results show that the S2S model is able to reproduce the spatial pattern of rainfall, but consistently underestimates rainfall compared with observations across all lead days. Model performance generally decreases with increasing lead day, with wind speed being the best-predicted variable (r = 0,91 at lead day 2) and VIMFC exhibiting the lowest performance (r < 0,5 for all lead days). WWB demonstrates relatively high predictive skill, reaching a maximum correlation of r = 0,80, although its timing tends to shift relative to observations.