A Markov renewal model for predicting tropical cyclones in Bangladesh

Bangladesh frequently suffers from devastating tropical cyclones due to its unique location. The two tropical cyclones occurred in Bangladesh in 1970 and in 1991, killed about 300,000 and 140,000 people, respectively, have been considered the world'��s deadliest cyclones ever. In this paper, we propose and validate a Markov renewal process for the evaluation of tropical cyclones occurred in Bangladesh during 1877--2012. In particular, our aim is to model tropical cyclones that are consistent with the underlying physical behavior. The non-Poisson model semi-Markovian takes into account occurrence history and some physical constraints, and is expected to be sufficiently capture the main characteristics of the storm surge process. Well-known exponential distributions are fitted for inter-occurrence times, and their parameters are estimated using the likelihood method. We validated the model using data of cyclones occurred in Bangladesh during the period 1877-2012. Our results show a good fit with the real events.