Title: A semiparametric Bayesian model for large spatiotemporal Red sea surface temperature data and related hotspot estimation
Authors: Arnab Hazra - King Abdullah University of Science and Technology (Saudi Arabia) [presenting]
Raphael Huser - King Abdullah University of Science and Technology (Saudi Arabia)
Abstract: Gradually increasing sea surface temperature (SST) is a major concern for the ecosystems and we focus on pointing out the exceedance regions of SST within the Red Sea, a vital region of endangered coral reefs. Explosive growth of remote sensing and other data collection techniques are leading to large high resolution spatial datasets. We propose a Dirichlet process mixture of low-rank spatial Student-t processes for spatial analysis of large datasets where temporal replications (days, for example) are available. Our model considers the whole dataset above a very low threshold for modeling the bulk and the tail jointly. The model allows drawing extremal inference by probabilistically separating the extreme days from the moderate days and estimating the model parameters based on the cluster of extreme days. The proposed model has nonstationary mean, covariance and asymptotic dependence and under limiting conditions, it spans all possible spatial processes. Inference is drawn based on MCMC sampling where most of the parameters allow Gibbs sampling. We perform a simulation study to compare model fitting performances of the proposed model with its sub-models where our method generally outperforms its alternatives. Finally, we fit the proposed model to estimate the spatial return level maps and to identify the exceedance regions. Compared to the low-rank Gaussian processes, estimated return levels based on the proposed model are generally higher across the Red Sea.