EVALUASI PERFORMA MODEL CMIP6 BERDASARKAN LEAD TIME DALAM PREDIKSI CURAH HUJAN EKSTREM DI INDONESIA

Abstract

Pemanasan global telah meningkatkan frekuensi dan intensitas curah hujan ekstrem, terutama di wilayah tropis seperti Indonesia yang memiliki sistem iklim kompleks meliputi ENSO, MJO, dan monsun Asia-Australia. Penelitian ini mengevaluasi kinerja empat model CMIP6 DCPP, yaitu CMCC-ESM2, EC-Earth3, MOHC, dan Multi Model Ensemble (MME), dalam memprediksi curah hujan ekstrem di Indonesia pada lead time 1–5 menggunakan CHIRPS v3.0 sebagai data observasi acuan. Kinerja model dinilai menggunakan PBIAS, RMSE, MAE, dan koefisien korelasi Pearson (r). Karakteristik curah hujan ekstrem dianalisis menggunakan uji Goodness-of-Fit, yaitu Kolmogorov-Smirnov (K-S), Anderson- Darling (A-D), dan Chi-Square (C-S), yang diterapkan pada distribusi Generalized Extreme Value (GEV), Log-Normal, dan Weibull, serta diperiksa melalui analisis skewness dan excess kurtosis. Hasil penelitian menunjukkan bahwa MME secara konsisten menghasilkan kinerja terbaik curah hujan normal di seluruh lead time dengan RMSE terendah (1,756–1,886), MAE terendah (1,372–1,413), dan korelasi tertinggi (r = 0,615–0,693). GEV muncul sebagai distribusi best-fit dominan dengan persentase 57% berdasarkan evaluasi rank sum. Analisis spasial menunjukkan Papua bagian selatan secara konsisten mengalami curah hujan ekstrem tertinggi. Pendekatan MME memberikan representasi terbaik terhadap curah hujan normal, sementara EC-Earth unggul dalam mengestimasi magnitude curah hujan ekstrem karena kemampuannya menangkap nilai ekor kanan distribusi GEV.

Global warming has intensified the frequency and magnitude of extreme rainfall, particularly in tropical regions such as Indonesia, where climate variability is strongly influenced by the El Niño–Southern Oscillation (ENSO), Madden–Julian Oscillation (MJO), and the Asian–Australian monsoon. This study evaluates the performance of four CMIP6 Decadal Climate Prediction Project (DCPP) models, namely CMCC-ESM2, EC-Earth3, MOHC, and the Multi-Model Ensemble (MME), in predicting extreme rainfall over Indonesia at lead times of 1–5 years using CHIRPS v3.0 as the observational reference. Model performance was assessed using Percentage Bias (PBIAS), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Pearson's correlation coefficient (r). Extreme rainfall characteristics were analyzed using the Kolmogorov–Smirnov (K-S), Anderson– Darling (A-D), and Chi-Square (C-S) goodness-of-fit tests applied to the Generalized Extreme Value (GEV), Log-Normal, and Weibull distributions, complemented by skewness and excess kurtosis analyses. The results indicate that MME consistently provides the best performance for normal rainfall across all lead times, whereas GEV is the dominant best-fit distribution (57%) based on rank-sum evaluation. Spatial analysis identifies southern Papua as the region with the highest extreme rainfall. While MME best represents normal rainfall, EC-Earth3 more accurately estimates extreme rainfall magnitude by capturing the right tail of the GEV distribution.