A0857
Title: Extreme value analysis using semiparametric spatial zero-inflated models
Authors: Chun-Shu Chen - National Central University (Taiwan) [presenting]
Chung-Wei Shen - National Chung Cheng University (Taiwan)
Bu-Ren Hsu - National Central University (Taiwan)
Abstract: Spatial two-component mixture models are effective for analyzing spatially correlated data with zero inflation. To avoid biases from assuming a specific distribution for response variables, a semiparametric spatial zero-inflated model is utilized. This model presents significant computational challenges, especially with large datasets, due to the high dimensionality of latent spatial variables, complex matrix operations, and slow estimation convergence. To address these issues, a projection-based method is introduced that reduces dimensionality by projecting latent spatial variables onto a lower-dimensional space using selected basis functions. An efficient iterative algorithm is developed for parameter estimation within a generalized estimating equation framework. The optimal number of basis functions is determined via Akaike's information criterion, and the stability of parameter estimates is assessed using the block jackknife method. This approach is validated through simulation studies and applied to Taiwan's 2016 daily rainfall data, demonstrating its practical effectiveness.
