Online Software Help Manual

BookmarkIndexPrint
Contents
Display Legacy Contents

Calibration Model Wizard - Step 5 - Range Selection/Data extraction

Step 5 assists the user in selecting a significant spectral range for the calibration. Finding the significant spectral ranges is the most challenging part of the calibration model setup since the chosen area strongly influences the quality of the resulting model. The software assists the user in finding the optimal spectral regions to create a good calibration model.

The range selection / data extraction will differ depending on the selected calibration model. Step 5 of the Calibration Model Wizard will show the appropriate method for chosen calibration model. The options are:

  • Data extraction for the univariate model

  • Spectral range selection for PLS models

  • Spectral range selection for MLR models

The three different options will in explained in detail in the following section.

Data extraction - Univariate Model

The data extraction for the Univariate Model is described in detail in the topic Data extraction - Univariate Calibration. Please refer to this topic in the chapter Chemometric Analysis.

Spectral range selection for PLS models

Spectral range selection for PLS calibration models does not be as accurate as for MLR the models, because the algorithm will find the optimal spectral regions by iteration on its own. The user is advised to limit investigations and modelling to spectral regions of interest in order to avoid misinterpretation. Thus the user starts with a wide spectral range and will refine selection during subsequent recalculation of the model.

The initial selection is by default the complete spectral range:

Modifying spectral region

Refinement of the spectral region requires unchecking the "Use Full Spectral Range" checkbox. A spectral range table is shown with a default range in the center of the applicable spectral range. User defined spectral ranges can be adjusted easily either graphically by adjusting the range selector with the mouse or by directly entering numerical values in the table.

This example uses the automatically proposed ranges.

MLR spectral range selection

MLR is a very sensitive calibration model providing excellent and reliable results if parameters are properly adjusted. There are several and mostly  time consuming automatic algorithms available for this purpose, but the most reliable results will be obtained by manual selection.

The wizard will assist the user during spectral range selection which is an interactive process strongly related to cross correlations between highly correlated spectral regions. Initially the software proposes the optimal and highest correlated spectral range. After fine tuning by the user the next spectral range is proposed. These steps are repeated until no more cross correlations are detected.

This procedure will prevent the user from over fitting the calibration model or selecting non-correlated spectral ranges. Furthermore spectral regions are very narrow and can be precisely defined according to the quality of spectral data.

The corresponding dialog looks like this:

Contents

In the data view, the average spectrum (red line), the standard deviation shape (light red shape) and the correlation spectrum (grey line) are displayed as described before. Additionally, the initial spectral range proposed by the software is shown as vertical light blue rectangle shape (at approx. 1550 wave numbers in the picture above).

In the table below details of the first proposed spectral range are shown.

Fine tuning of spectral ranges

After proposal of a spectral range by the software, the user must tune the precise position of the range to fit the peak position of interest best. For this purpose, the user should zoom into the region around the selected spectral range:

The center of the selected spectral region is emphasized by a vertical red line. Start and end markers are indicated by little blue squares. All items (the spectral range, start marker and end marker) can be moved to desired positions within the spectrum using drag&drop with the Left mouse button.

Alternatively, the precise values can be directly entered into the particular fields of the spectral range table below the data view.

Automatic proposal of spectral ranges!

The software always proposes a symmetrical spectral range with an absolute size of five data points. The user may change the number of data points or select different start and end markers, so that an asymmetric range is applied. This strongly depends on the peak requirements.

Adding new spectral ranges

After proper adjustment of the first spectral range, further ranges can be added by clicking the New button.

The next spectral range is proposed automatically and the correlation spectrum is updated accordingly. The slope of correlation spectrum changes, because already selected spectral ranges are no longer considered in calculation. The actual correlation spectrum will help the user to identify the next significant range and it also helps to prevent selection of cross correlated regions.

The new proposed range is shown as light blue rectangle, whereby the color of the last one has changed into light green.

Again, the position and the size of the actual spectral region can be changed by the user.

Changing previously selected spectral ranges!

Previously adjusted ranges affect all subsequent ranges. The software will display a warning if the user tries to modify the previously selected ranges. An earlier range may be modified if the later ranges are discarded.

Over fitting warning

The user can add as many new spectral ranges as he likes, but adding too many of them will over fit the model. In this case cross correlations will reduce the quality of the final result. The user will be warned by the software automatically, if this limit is reached.

The warning limit depends on the Overfitting Threshold. The parameter can be adjusted in the corresponding box in the dialog.

After confirming the message by clicking the OK button, the new spectral range is added anyway to give the user the freedom of choice. It should be deleted again to ensure integrity of the model.

Deleting spectral ranges

Deletion of spectral ranges is performed as described in the following:

  1. Select the desired range either in the data view or in the spectral ranges table.

  2. Click on the Delete-button or press the Del-key on your keyboard to delete it.

Deletion must only be applied to the last / actual range in the list.

Can I remove intermediate spectral ranges?

If any intermediate range is removed from the list, the whole selection process will collapse and any successors must be removed as well in order to avoid over fitting or misinterpretation.

Resetting spectral ranges

To completely remove all selected spectral ranges and restart the selection process, the user may click the Reset button.

Range selection for manual background noise correction

When using the manual background noise correction the range selection differs slightly from the automatic workflow. If the Manual Noise Correction checkbox has been activated, the range selection dialog will look like this:

Again, the software will propose a spectral region for the calibration (vertical red line with blue border) and the user may change the position and extent of this region. Additionally a second region for the background noise correction (vertical red line with yellow border) will be proposed. The position and extent of this selection may also be edited by the user. Ideally, the select region should lie in a preferably non-correlated area with similar background noise as the calibration region. Regions close to the selected calibration region should be preferred for noise correction.

Navigation

Just click the Next > button to proceed to the next step.

Clicking the Cancel button will abort creating a new calibration.