In this study, we have used the MAGIC model together with data from the Birkenes catchment in Norway, at which 27 years of data (1974–2000) are available. We calibrated the MAGIC model to the five year observed average chemistry around 1990, and then used the data from the five year period around 1980 to refine the calibration. From 1990, forecasts were run for the different sets of inputs and parameters, and the sets of inputs and parameters were further refined using observations for the period 1996–2000. Through an automatic calibration routine, the model was calibrated a large number of times with different sets of input data to account for the uncertainties in the observed data using a Monte Carlo set-up. The results show that the uncertainty in the model predictions decreases as more observed data from different points in time are used in the model calibration. The results also show that when using the time series data in calibration, the distribution of the forecast changed. The distribution of the predicted Acid Neutralisation Capacity (ANC) in the future is lower for the more refined model calibration. The 10 and 90 percentiles of predicted ANC in 2010 are –3 to 21 μeq L−1when only a five-year average is used for calibration, but are −;7 to 9 μeq L−1when data from the three different time periods are used.