527: GCM Options

CPT always interpolates from the X data to the Y grid or station locations rather than vice versa. The interpolation options provide alternative means of performing the interpolation. The interpolate option will use a linear distance-weighted interpolation of the four X values surrounding each Y grid / station.

The GCM data can be corrected for systematic errors using the model climatology options. The following options are available:

• No correction : the raw model data are used, although an attempt will be made to convert the X to the same units as the Y data (see further details on the cpt:units tag). Because no correction is made to the model output, the contribution to the forecast-error variance is zero.
• Correct mean biases : The mean of the GCM is forced to equal the mean of the Y data, although the correction is performed in cross-validated mode, and so the mean bias may not be exactly zero. Because the correction involves only a recentering of the model output, the contribution to the forecast-error variance is the standard error of the mean.
• Correct mean and variance biases : The mean and variance of the GCM are forced to equal the corresponding values for the Y data, although the correction is performed in cross-validated mode, and so the mean bias and variance ratio may not be exactly zero and one, respectively;
• Correct for skill : A simple linear regression model is used to correct the GCM data. The mean bias will typically be close to zero (although not exactly zero, partly because of the cross-validation), but the variance ratio will tend towards zero as the skill reduces, and will generally be less than one.

The model combination option is applied when the X file contains more than one model (see further details on the cpt:model tag). The combination occurs after any correction for systematic errors, as described above. The following options are available:

• Uncalibrated average : any corrections for systematic errors specified in the Model climatology panel are first applied to each model individually, and then a simple average across the models is calculated.
• Bias-corrected average : any corrections for systematic errors specified in the Model climatology panel are first applied to each model individually, a simple average across the corrected models is calculated, and mean-bias corrections to are then made (see the discussion above on mean biases).
• Calibrated average : any corrections for systematic errors specified in the Model climatology panel are first applied to each model individually, a simple average across the corrected models is calculated, and mean and variance corrections to this model are then made (see the discussion above on mean and variance biases).
• Recalibrated average: : any corrections for systematic errors specified in the Model climatology panel are first applied to each model individually, a simple average across the corrected models is calculated, and then this model average is corrected for skill (see the discussion above on Correct for skill).
• Multi-model combination : a multiple regression model is used to combine the models. The multiple regression parameters are used to rescale the model correction parameters (described above). It would not normally be worthwhile to use anything other than the "no correction" option for the model climatology.
• Best model by location : the model with the highest correlation with observed data is used. A different model may be selected for each location. The selected model has been corrected for systematic errors depending upon the option above.
• Best overall model : only the model with the highest goodness index is used. The selected model has been corrected for systematic errors depending upon the option above.