Logit-Normal Mixed Model for Indian Monsoon Rainfall Extremes

• Activity Number: 469
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics and the Environment
• Author(s): Lindsey Dietz*+ and Snigdhansu Chatterjee
• Companies: University of Minnesota and University of Minnesota
• Keywords: Climate Change ; Indian Monsoon ; Generalized Linear Mixed Model ; Logit-Normal Mixed Model

Abstract

Describing the nature and variability of Indian monsoon precipitation is a topic of much debate in current literature. We suggest the use of a generalized linear mixed model (GLMM), specifically, the logit-normal mixed model, to describe the underlying structure of this complex climatic event. Several GLMM algorithms - approximate likelihood, method of moments, and Bayesian paradigms- were vetted in simulations before application to Indian rainfall data. Logit-normal models were fit with fixed covariates of latitude, longitude, elevation, minimum and maximum temperature, tropospheric temperature difference, Nino-3.4 index, and Indian dipole mode index (DMI) with a random intercept by weather station. The output indicated the model structure aligned with understood monsoon physics and found a non-negligible random effect by weather station. This work provides a valuable starting point for extending GLMM for use in climate data analysis.