National Conference on Water Resources & Flood Management with special reference to Flood Modelling October 14-15, 2016 SVNIT Surat WRF-25-1 COVARIATE GENERATION USING MULTI-OBJECTIVE GENETIC ALGORITHM FOR DEVELOPING NON-STATIONARY RAINFALL IDF CURVE V Agilan 1 and N V Umamahesh 2 1 Project Associate, Department of Civil Engineering, National Institute of Technology, Warangal, Telangana, India-506004, Email: agilanvensiv@gmail.com. 2 Professor, Department of Civil Engineering, National Institute of Technology, Warangal, Telangana, India-506004. ABSTRACT The infrastructure design and stormwater management are primarily based on rainfall Intensity- Duration-Frequency (IDF) curves and the existing IDF curves are based on the concept of stationary extreme value theory. But, the extreme precipitation events are increasing due to global climate change and questioning the reliability of our current infrastructure design. Based on recent theoretical developments in the Extreme Value Theory (EVT), researchers have developed a non-stationary rainfall IDF curve by incorporating linear trend in the location parameter of the Generalized Extreme Value Distribution (GEVD). However, upon detecting a significant trend in the extreme rainfall series, directly applying the linear trend (Time as a covariate) may increase the bias of the non-stationary GEVD. Hence, it is important to use covariate which will produce a non-stationary GEVD which has less bias than the stationary model. Therefore, in this study, Multi-Objective Genetic Algorithm (MOGA) based methodology has been developed to generate covariate which will produce good quality non-stationary GEVD with less bias and the proposed methodology is demonstrated with a case study. Results of the case study reveal that the proposed MOGA based method is able to build the less bias and good quality non-stationary GEVDs. In addition, it is also noted that the usage of a linear trend (Time as a covariate) for modelling non-stationary in the time series sometimes increases the bias of the non-stationary model. Keywords: Covariate, Extreme rainfall, Genetic Algorithm, IDF curves, Non-stationary. 1. INTRODUCTION The rainfall Intensity-Duration-Frequency (IDF) curves are commonly used in storm water management and other engineering design applications across the world (Endreny & Imbeah, 2009) and these curves are developed based on historical rainfall time series data by fitting an appropriate theoretical probability distribution to annual maximum rainfall series or partial duration series (Cheng & AghaKouchak, 2014). The existing IDF curves are based on the concept of stationary extreme value theory (i.e. occurrence probability of extreme precipitation event is not expected to change significantly over time (Jakob, 2013)). However, in recent years, the extreme precipitation events are intensifying due to global climate change (Allen & Ingram, 2002; Trenberth, et al., 2003; Emori & Brown, 2005; Tramblay, et al., 2012; Cavanaugh, et al., 2015; Xu, et al., 2015). As an approximation, based on recent theoretical advancements in the Extreme Value Theory (EVT), the linear trend is often used to represent a long-term trend (Katz, et al., 2002; Cheng, et al., 2014; Cheng & AghaKouchak, 2014; Yilmaz & Perera, 2014). In particular, Cheng & AghaKouchak (2014) constructed a non- stationary rainfall IDF curves by introducing linear trend in the location parameter of the Generalized Extreme Value (GEV) distribution. Yilmaz & Perera (2014) investigated the non- stationary in the IDF curves of Melbourne, Australia by introducing linear trend in location and shape parameters of the GEV distribution. The previous studies on the non-stationary IDF curves have used only linear trend. In the wake of detecting a significant trend in the extreme