Bayesian Modeling of Long-Term Dynamics in Indian Temperature Extremes

Abstract: Annual maximum temperature data provides crucial insights into the impacts of climate change, especially for regions like India, where temperature variations have significant implications for agriculture, health, and infrastructure. In this study, we propose the Coupled Continuous Time Random Walk (CTRW) model to analyze annual maximum temperature data in India from 1901 to 2017 and compare its performance with the Bayesian Spectral Analysis Regression (BSAR) model. The CTRW model extends the standard framework by coupling temperature changes (jumps) and waiting times, capturing complex dynamics such as memory effects and non-Markovian behavior. The BSAR model, in contrast, combines a linear trend component with a non-linear isotonic function, modeled using a Gaussian Process (GP) prior, to account for smooth and flexible non-linear variations in temperature. By applying both models to the temperature data, we evaluate their ability to capture long-term trends and seasonal fluctuations, offering valuable insights into the effects of climate change on temperature dynamics in India. The comparison highlights the strengths and limitations of each approach in modeling temperature extremes and provides a robust framework for understanding climate variability.