Summary: | Code incorporating revisions made during discussion of GMD manuscript. Three main changes from v1.0.0-alpha: Corrected a <code>for</code> loop that inadvertently left out several catchments of the Greenland Ice Sheet. The default now includes training data for 260 catchments. Corrected our implementation of <code>statsmodels.tsa.ar_model.AutoReg</code> to use the optional argument <code>hold_back</code>. This argument, set to the maximum number of lags intended to be tested, ensures that AR models of different lags are computed on the same data and their BIC can be directly compared. This change truncates the length of the residuals used to compute the empirical correlation matrix. Corrected the test range of autoregressive lags to include 0 (white noise). See the manuscript by Ultee, Robel, and Castruccio in Geoscientific Model Development (accepted 2023) for further discussion.
|