Modeling annual rainfall: a robust maximum likelihood approach

Let { Xt} be a zero mean, Gaussian, autoregressive process of order one with parameter α. For a realization (X1,X2,…,Xn)′ of { Xt}, we consider the transformation Yt = Xt/Xt−1, for t = 2,…, n. Then the likelihood function of (Y2,…, Yn) can be derived and is shown to be independent of the parameter σ2. Now, maximum likelihood estimator for α is derived and interval estimates are then computed and used to determine the influence of each data point. Finally, a direct application to precipitation data will be given for illustration. Copyright © 2006 John Wiley & Sons, Ltd.