A1088
Title: Spatio-temporal rainfall forecasting by novel Kumaraswamy-Teissier distribution and seasonal VARMA models
Authors: Kamana Mishra - Indian institute of technology Mandi, India (India) [presenting]
Neeraj Poonia - Cytel Pune (India)
Tanmay Kayal - Indian institute of technology Mandi (India)
Sarita Azad - Indian institute of technology Mandi India (India)
Abstract: A novel three-parameter Kumaraswamy-Teissier distribution is proposed, extending the classical Teissier family, offering enhanced flexibility in modeling skewed data. The key statistical properties of the distribution are derived, including its moments, quantile function, moment generating function, and order statistics. Parameter estimation is performed via maximum likelihood, maximum product spacing, and a Bayesian framework with the Metropolis-Hastings algorithm, which allows posterior inference with credible intervals and highest posterior density regions. Monte Carlo simulations are conducted to compare the efficiency of the estimator. The proposed model is applied to rainfall data from five stations in the Northwest Himalaya and demonstrates a superior fit relative to conventional models. To perform a spatiotemporal analysis, seasonal ARIMA and VARMA models are fitted to KTD-transformed data and raw data. Comparative forecasting reveals that incorporating spatial structure leads to improved predictive performance. Moreover, it is observed that applying these time series models directly to raw rainfall data often yields implausible negative forecasts and prediction intervals. In contrast, applying the VARMA model to KTD-transformed data ensures physical realism and statistical consistency, emphasizing the necessity of proper marginal transformations for modeling. Results highlight the model's potential for applications in hydrology and climate-related studies.
