Date of Completion


Embargo Period


Major Advisor

Jun Yan

Associate Advisor

Dipak K. Dey

Associate Advisor

Xuebin Zhang

Field of Study



Doctor of Philosophy

Open Access

Campus Access


Detection and attribution (D&A) analysis for climate extremes plays an important role in understanding human influence on the observed change in climate extremes. Recently, several D&A approaches that are based on generalized extreme value (GEV) distribution have been proposed. These approaches used signals estimated from climate model simulation under external forcing as if they were known in the D&A analysis. This practice, however, ignores the uncertainty in signal estimation from the climate model simulation, which has an effect similar to measurement errors in covariates lead to bias in the D&A analysis. Noticing that the latent signals are shared by both climate model simulation and observed data, we propose a joint model for the simulated extremes and the observed extremes, and combine the signal estimation and D&A analysis where the signal is estimated jointly from both data. We propose a point estimation algorithm in which we iteratively update the signal and its scaling factor to maximize the joint independence likelihood. The signal estimate can also be applied in D&A analysis using the combined score equation (CSE) approach. The inference is made using a block Bootstrap confidence interval to keep the spatial dependence as well as temporal dependence in the original datasets. We also propose a goodness of fit test which include test statistics for both simulation data and observation data to examine how well our model fits the data. We show that this method can reduce the bias effectively in the estimation compared to the previous method using a simulation study. The simulation study also shows that using signal estimate from the joint modeling approach can further reduce the standard error of the scaling factor estimate. We also apply the new method to the extreme temperature indices in East Asia in a perfect detection study and a real data analysis. The methodology is extended to handle multiple climate models and multiple signals.