NONSTATIONARY CLIMATE AND ITS IMPACT ON DECADAL-SCALE RAINFALL PREDICTION SKILL IN INDONESIA

• Main Author: Gardian Sudarman, Gian
• Format: Dissertations
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/87114
• Institution: Institut Teknologi Bandung

According to the Decadal Climate Prediction Project by the World Meteorological Organization (DCPP-WMO), the skill of decade-scale rainfall predictions in Indonesia is considered low. This is because the verification methods do not take into account the changing nature of rainfall, and the models used have not been adjusted for these changes (nonstationary). Since the Earth's climate system has evolved because of climate change or changes in climate variability, it’s crucial to consider these nonstationary conditions. This study analyses both observational data and models by looking at changes in trends, sudden shifts (step changes), and variability. Overall, rainfall in Indonesia shows significant nonstationary patterns, with a noticeable positive trend and a significant step change around 1994. Regionally, the nonstationary patterns vary, with significant trend changes in 35.8% of the area, step changes in 28.3%, and variability in 16%. Additionally, this study assesses the prediction skill of the DCPP-WMO models in nonstationary conditions. Among the nine models evaluated, MIROC6, MPI-ESM1-2-LR, and MRI-ESM2-0 scored the highest. These models were then recalibrated using both stationary and nonstationary methods, with the nonstationary method improving prediction skills by 80%. Notably, only the MPI-ESM1-2-LR model matched the nonstationary characteristics of the observational data, leading to superior performance compared to the others. Moreover this study finds that prediction accuracy tends to be higher when multiple climate variabilities occur in sequence in positive relation. Decade-scale rainfall predictions are based on the relationship between rainfall, measured by the Standardized Precipitation Index (SPI) over a 12-month period (SPI12), and the major climate factors affecting SPI12, including the Interdecadal Pacific Oscillation (IPO), El Niño Southern Oscillation (ENSO), and Dipole Mode Index (DMI). This relationship is modeled using a multivariate regression approach. According to the 1-5 year forecast data from the DCPP-WMO model, the SPI12 index is predicted to be within the normal range. Probabilistic predictions suggest there is a 14%-15% chance of extreme dryness, 21%-23% chance of very dry conditions, 30%-32% chance of moderate drought, 67%-86% chance of normal conditions, and 0-1% chance of wet conditions. This method of predicting rainfall based on climate variability offers an alternative to the DCPP-WMO's standard decade-scale rainfall predictions.