MODELLING NONSTATIONARITY IN DESIGN RAINFALL ESTIMATE UNDER THE CLIMATE CHANGE

Abstract

Industrialization and excessive environmental degradation cause change in the earth's climate system, disrupting hydrologic balance and leading to extreme events. Potential modification of hydrological events associated with anthropogenic climate change emerged as strategic concerns for human lives, especially in coastal urban areas. The effects of climate change have resulted in many extreme rainfall events recently. Thus strategic planning and mitigation of those events require a timely update of design rainfall estimates. More often, incorporating the climate change effects on the estimate of design rainfall is rarely considered. Rainfall statistical properties change over time as climate changes, increasing randomness and unpredictability, resulting in nonstationarity. Consequently, an imperfect model representation of the real data distribution due to rainfall's nonstationary behavior would lead to over/underestimation of design rainfall values. In this thesis, the effects of nonstationarity in rainfall data are analyzed. Daily gridded rainfall data collected from Indian Meteorological Department for nine selected major coastal cities in the southern India peninsula was used to demonstrate the proposed methodology. The projected climate model data of RCP 4.5 and RCP 8.5 is used to analyze the effect of climate change for each duration of rainfall in each city. Regional climate model data is bias-corrected using Distribution Mapping. The Bartlet Lewis Rectangular Pulse clustering mechanism is used to disaggregate the coarse resolution daily gridded rain into fine resolution hourly rain. The 180 (1901 to 2080) years long hour and day basis annual maximum rainfall dataset is grouped into three timeframes the 1940s (1901-1960), the 1990s (1961-2020), and the 2050s (2021-2080), and each timeframe annual maximum rainfall in each duration is examined for any significant nonstationary conduct using statistical tests. A stationary and six time-varying Generalized Extreme Value distribution models are fitted to the 1,2,4,6, 8, and 12 hours and 1-day, 2-days, 3-days, 4-days, 5-days, 6-days, and 7-days annual maximum rainfall of each city, and best fit is selected based on criteria's. The parameters for calculating return levels are derived by the Bayesian analysis and modeled through the Markov Chain Monte Carlo (MCMC) simulation technique using the Metropolis-Hastings algorithm. The return periods, along with the different magnitude of rainfall of various iv

v

durations, are utilized to derive the IDF curve, and for which the change in design rainfall estimate between the present and future period is analyzed. While stationary models have been found to fit well in the extended period rainfall, nonstationary models have been often observed to best fitted the short duration rainfalls with significant nonstationary behavior in most of the cities. Estimated return levels in each timeframe show an increasing trend with time in most of the cities.