Page 106 - Demo

1st Int. Transborder Conf. of the Timor Island: Timor %u2013 Science without borderDili, 7-8 May 2025105ARIMA Modeling of ETCCDI Extreme Climate Indices in Southeast Asia: Insights from CMIP6 ProjectionsShannon M.D. ViegasUniversidade Cat%u00f3lica Timorense (UCT), Office of Research, Post-graduate & Cooperation, Dili, Timor-Leste, e-mail: shanviegas@gmail.comAbstractSoutheast Asia is one of the most climate change-exposed regions, particularly in relation to variations in extreme weather. In this study, we investigate past trends and future estimates of significant Expert Team on Climate Change Detection and Indices) climate extremes from the CMIP6 multi-model ensemble data. Specifically, we focus on four indices: maximum daily maximum temperature (TXX), maximum 1-day precipitation (RX1DAY), consecutive dry days (CDD), and consecutive wet days (CWD). Projection implies a uniform increase in TXX across the region by up to 2%u00b0C by 2050, with a general decline in RX1DAY, suggesting a potential reduction in shortduration extreme rainfalls. Concurrently, the number of CDDs is projected to rise considerably, particularly over central Indonesia, whereas CWDs will likely decline, with the greatest declines in mainland SEA, particularly in Thailand, Cambodia, and Vietnam. Overall, these trends suggest a higher risk of droughts and potentially a trend towards drier climatic conditions in some parts of the region. To investigate deeper the CWD trends, we built an ARIMA model from historical data from 1900 to 2014. The default ARIMA parameters (p=0, d=1, q=0) gave flat future forecasts for CWD, which is the reverse of the declining trend. Through hyperparameter tuning by grid search method, we had a better model (p=2, d=1, q=3) with an optimized lower RMSE of 1.956 days. This model portrayed the historical sequence more accurately but also maintained imperfections in terms of reproducing maximum values and abrupt nonlinear alterations. Seasonal decomposition of the time series pointed toward a diminishing trend for CWDs as well as the rise in the amplitude of the seasonal component, suggesting potential increased performance by the utilization of a Seasonal ARIMA approach. While ARIMA performed well in capturing short-term patterns and provided interpretable forecasts, its applicability to long-term forecasts is minimal. Model performance was much poorer than that in CMIP6 projections for SSP2-4.5 and SSP5-8.5 scenarios, particularly in not being able to simulate the amplitude and frequency of occurrence of extreme precipitation events. This highlights the ability of dynamical climate models to simulate intricate, nonlinear climate behavior under varying greenhouse gas emission scenarios. Overall, ARIMA modeling offers a low-complexity and pragmatic approach to the shortterm forecasting of extreme precipitation indices such as CWDs. However, its limitations in capturing nonlinear trends and long-term variability suggest integrating ARIMA with seasonal components or hybrid machine learning models. Higher-resolution temporal modeling (e.g., weekly) and ensemble-based approaches that combine statistical and dynamical modeling methods should be explored in future research. These advances will allow us to better forecast climate extremes and enable adaptive planning in climatevulnerable regions like Southeast Asia.