Executive Summary : | In recent decades, climate change has become substantial & astonishing phenomenon affecting life & natural resources. Due to changing climate extreme events (EE) that have witnessed low probability in past are becoming more likely (IPCC 2021). Extremes of meteorological variables rarely occur making prediction of these events more challenging. Predicting EE using traditional methods leads problems of high computational complexity & low prediction accuracy (Goerss 2000). Consequently over the year's use of Artificial Intelligence (AI) for prediction of EE have become popular. Also EE exhibit spatial variability resulting increased worry about spatiotemporal patterns of precipitation extremes. It is thus critical to pay attention to extreme climate events so as to be able to cope with the impacts. Extreme Value Theory (EVT) is a statistical method commonly used for quantification of behaviour of EE. However, EVT studies these EE in univariate series, while observations of natural phenomena viz. rainfall, temperature etc. collected from several neighbouring sites generate multivariate series in the sense of location (Diriba & Debusho 2021). Consequently Multivariate EVT has been developed as an alternative to EVT to analyse joint distribution of extremes in several series (Smith 1990). Literature survey revealed that former studies have primarily concentrated on analysis of univariate part (Diriba & Debusho 2020) while neglecting the spatial interactions & extremal dependence. However any extreme value analysis model that deals with climate extremes at point stations is insufficient to depict the regional extreme climate components spatial dependence (Zhang et al. 2014) consequently resulting in development of spatiotemporal models. Among several methods developed for spatial extreme value analysis when studying a spatial dependence structure, Max Stable Processes (MSP) has been observed to provide an ideal statistical modeling tool as they capture the dependence of spatial extremes (De Haan 1984).When modelling univariate extremes among weather stations dependency consequences are ignored. Furthermore, to account for inherent influences among weather stations estimation of extreme-value distribution poses a challenge of reliance. MSP are suited to characterise joint behaviour of maxima at all points in space under these circumstances. It avoids local oscillations that may be substantially influenced by outlier observations by constructing a model for entire set of grid cells (Shin et al. 2019).As a result, more exact extreme event return levels can be predicted. For modelling spatial dependency MSP converts marginal distribution of extreme value to Frechet margin. Proposed research will first use AI approach to identify EE & then uses several MSP models to model spatial extreme value. To our knowledge no other study has used Indian meteorological data to apply spatial MSP models considering various trend surfaces that include temporal covariates. |