Author(s): Francesco De Paola; Maurizio Giugni; Alexander Garcia-Aristizabal; Eduardo Bucchignani
Linked Author(s): Maurizio Giugni
Keywords: Hydrological processes; Stationary modeling; Non-Stationary modeling; Bayesian analysis; Climate change
Abstract: Increasing flood risk is now recognized as one of the most important threats from climate change in most parts of the world, with recent repeated severe flooding causing major loss of property and life. This has prompted public debate on the apparent increased frequency of extremes and focused attention in particular on perceived increases in rainfall intensities. Climate models predict increases in both the frequency and intensity of heavy rainfall in the high latitudes under enhanced greenhouse conditions. The United Nations Intergovernmental Panel on Climate Change (IPCC) issued a report in February 2007 stating that “It is very likely that hot extremes, heat waves, and heavy precipitation events will continue to become more frequent”. The frequencies with which physical processes exceed high thresholds are often well described by Extreme Value Distributions (EVDs). These models have specific parametric forms that follow from asymptotic arguments. Parameter estimation through likelihood-based methods allows extreme quantiles to be calculated, giving an estimate of the return level associated with a particular return period (e. g. 100 years). However, the parameters of these EVDs are typically estimated from the largest values in a dataset (e. g. annual rainfall maxima), which can make the estimates unreliable. One way to improve reliability is through the use of a Bayesian framework. This allows specifying prior information so that we can constrain the estimates according to our physical understanding of the process generating the data. In the paper, with reference to the city of Dar Es Salaam (Tanzania), a dataset of 53 years (1958-2010) of maximum daily rainfall records (24 h) is considered, showing that the EVDs well interpret the maximum daily rainfall. In particular, a comparison is made between inference models based on the Maximum Likelihood Estimation (MLE) and the Bayesian one, highlighting differences and strengths. Furthermore, a comparison between a non-stationary regression and a stationary model is done. In this case, the time series does not seem to highlight any non-stationary effects. Finally, the results achieved within CLUVA (Climate Change and Urban Vulnerability in Africa) EU project by the Euro-Mediterranean Center on Climate Change (CMCC) (at 1 km spatial resolution) according with the IPCC RCP8. 5 emission scenario are considered, extending the analysis until 2050 (93 years). In this case the process seems to be non-stationary.
Year: 2013