Figure 1.

CV for semiparametric quasi-Poisson time series model selection. (A) Year forward-chaining CV: in the first iteration, the first five years’ data were taken as the training set for model fitting and the six year’s data were taken as the test set for performance assessment. This procedure rolled forward by year until the last year was used as a test set. The average RMSE of the 26 iterations was calculated as a measure of model performance, where lower values of RMSE indicate better fit. (B) Leave-one-year-out CV: in the first iteration, the first thirty years’ data were taken as the training set and the last year’s data were taken as the test set. The iterative process continued until each year’s data have been taken as a test set. The average RMSE of the 31 iterations was calculated as a measure of model performance, where lower values of RMSE indicate better fit.