Making a Forecast
Before a forecast can be made, a file containing the values of the X variables to be used to make the forecast must be specified. The X input file is set as the default forecast file under the assumption that the latest data are stored in the same file as the historical data. However, it is possible to specify an alternative file and to adjust the default settings for the years and the number of years to be forecasted. This file and its settings can be saved in the Project File so that they are automatically set for next time. The input file is set using the browse button on the CPT Window. The file should be in an identical format to the X input file. Once specified, CPT will identify the first year of data available in the file, and will try to set a start-at date for the first forecast. By default, CPT will attempt to make forecasts for all dates after the end of the training period, but will use the last date of the training period if there are no additional dates in the input file. You can change the start-at date from Edit ~Start at ~ New Predictors (Z) , and may then need to adjust the number of forecasts to be made. CPT will generate one forecast for each year of data it reads from the forecast data file. In most operational cases it will be appropriate to make only one forecast: a forecast for the target season from the current year's data.
Once the forecast input file has been set, it is possible to generate and view the forecast(s) using the Forecast option under the Tools menu item. It may be desirable to generate forecasts only for those locations with at least a minimum level of skill; for example, you may wish to forecast only at locations with positive skill. A Skill Mask can be applied if desired, and any location with insufficient skill will be given a climatological forecast.
By selecting Series , a Forecast Window will be opened, showing tables of the forecasts for individual gridpoints/stations/series and a graph of the historical and current forecasts. In the top left of the window, values of the upper and lower thresholds that define the three categories are indicated, together with the climatological probabilities and odds of each of the categories. These values can be adjusted, as described later in Customising the Results .
Below the thresholds table, the forecasts table shows the year for which the forecast applies, the value of the forecast, and the forecast probabilities and odds for the three categories. These probabilities are derived from the prediction error of the forecast. The odds can be expressed relative to the climatological odds using the Options ~ Forecast Settings menu item. If the normal category has the lowest probability and the categories are climatologically equiprobable, then the data for that station is probably not normally distributed. There is an option in CPT to transform the Y data to normal (see Options ~ Data ~ Transform Y Data ), although the low probability for the normal category may not always be resolved by selecting this option.
The forecast ranges table indicates the year and forecast, just as in the forecast table, but shows upper and lower prediction intervals as alternatives to forecast probabilities for indicating forecast uncertainty. The confidence level for the prediction intervals can be adjusted, as described in Customising the Results .
The graph shows the history of the cross-validated forecasts (green line) and observations (red line), and the current forecast(s) shown by green crosses. The forecast category for the unadjusted forecasts can be determined simply from the category divisions shown on the graph. The graph can be exported as graphics files by right-clicking anywhere in the child window. A default name is given to the graphics files, but this name can be changed using the browse button. The image quality for all the graphics formats can be improved by increasing the size of the image (see Options ~ Graphics ~ Graph Scaling ).
By selecting Map , options to view either the forecast values or the probabilities are given. The values map option provides a set of three maps (if the data are gridded or station) or bar charts (otherwise), one of which indicates the best estimate forecasts, and the other two indicate the lower and upper prediction intervals. The probabilities map option provides a set of three maps, each of which indicates the forecast probability for the three categories. The maps are shaded from blue for low probabilities, to red for high probabilities. If more than one forecast is produced, the forecasts can be cycled using the up and down arrows (which will work better than attempting to type in the year).
The titles of the maps or graphs can be customised by right-clicking in the child window and selecting Customise . The colour-scaling of the maps can be adjusted using Customise ~ Contour Levels .
The maps and graphs can be saved as graphics files, individually or as a montage, by right-clicking anywhere in the child window and selecting Export from the pop-up menu. A default name is given to the graphics file, but this name can be changed using the browse button. The image quality for all the graphics formats can be improved by increasing the size of the image (see Options ~ Graphics ~ Graph Scaling ).
Forecasts can be saved using the File ~ Output Results menu item. It is possible to save the forecast probabilities, odds, the actual forecast values and the prediction intervals.