CMStatistics 2021: Start Registration
View Submission - CMStatistics
B1780
Title: Bivariate deep kriging for large-scale spatial interpolation of wind field Authors:  Pratik Nag - King Abdullah University of Science and Technology (Saudi Arabia) [presenting]
Ying Sun - KAUST (Saudi Arabia)
Brian Reich - North Carolina State University (United States)
Abstract: High spatial resolution wind data are essential for a wide range of applications in climate, oceanographic and meteorological studies. However, these often tend to be nonGaussian with high spatial variability and heterogeneity. In spatial statistics, cokriging is commonly used for predicting bivariate spatial fields but it is not optimal except for Gaussian processes also cokriging is computationally prohibitive for large datasets. We propose a method, called bivariate DeepKriging, which is a spatially dependent deep neural network (DNN)with an embedding layer constructed by spatial Radial basis functions for bivariate spatial data prediction. We then develop a distribution-free uncertainty quantification method based on bootstrap and ensemble DNN. The proposed approach outperforms the traditional cokriging predictor with commonly used covariance functions, such as the linear model of co-regionalization and flexible bivariate Matern covariance. We show that the proposed DNN model is computationally efficient and scalable, with twenty times faster computations on average. We apply the bivariate DeepKriging method to the wind data over the Middle East region at 506771 locations. The prediction performance of the proposed method is superior over the cokriging predictors and dramatically reduces the time of computation and the large-scale computational complexity.