3.3.5 Terminate search
The fit is stopped when either the termination criterion is fulfilled or the maximum
number of iterations is reached. The termination criterion is such that the differences
between the actual parameter values must be less than a threshold for each parameter.
Each parameter has its specific threshold, which is set to 10
–5 times the initial value. It
cannot be changed by the user. The user defines the maximum number of iterations,
which should be set high enough that a forced stop is exceptional.
4. Applications
4.1 DATA ANALYSIS
Automatic data analysis of a series of spectra is performed in the batch mode
within WASI. The spectra must be in ASCII format. The following user actions are
required:
1. Copy all spectra into the same directory
2. Specify the directories of input and output data
3. Specify file names and file format of the input data
4. Set the batch mode
5. Specify the spectrum type
6. Tune the fit procedure
7. Set the model parameters
8. Select the fit parameters
9. Start calculation
The results are stored in ASCII format in the specified output directory. A single table,
FITPARS.TXT, summarizes the fit parameters of all spectra. All calculation settings
are documented in an actual copy of the file WASI.INI. The fit curves of all spectra can
be saved as single files.
4.2 ERROR ANALYSIS
No measurement is perfect, no model is exact, and no input data set is complete.
Thus the results of data analysis are unavoidably affected by errors. Sensitivity studies
are designed to estimate the errors for a determined parameter caused by the different
error sources. The best way to perform a sensitivity study is to simulate a large number
of measurements using a reliable model, and to then analyze these subsequently as if
they were real measurements. Well-defined discrepancies between the forward and
inverse models can be introduced which cause errors in the retrieved parameters. In this
way, parameter errors can be attributed quantitatively to different error sources. The
advantages of using simulated spectra rather than measured spectra are that all studied
effects are under control, and that the entire expected parameter interval can be covered
without a gap.
4.2.1 Errors from the sensor
The number of parameters that can be derived from a measurement and the
accuracy of the estimates depends very much on the sensor and on data quality. The
following instrument characteristics are relevant: number, center wavelengths and
102
Gege and Albert
The fit is stopped when either the termination criterion is fulfilled or the maximum
number of iterations is reached. The termination criterion is such that the differences
between the actual parameter values must be less than a threshold for each parameter.
Each parameter has its specific threshold, which is set to 10
–5 times the initial value. It
cannot be changed by the user. The user defines the maximum number of iterations,
which should be set high enough that a forced stop is exceptional.
4. Applications
4.1 DATA ANALYSIS
Automatic data analysis of a series of spectra is performed in the batch mode
within WASI. The spectra must be in ASCII format. The following user actions are
required:
1. Copy all spectra into the same directory
2. Specify the directories of input and output data
3. Specify file names and file format of the input data
4. Set the batch mode
5. Specify the spectrum type
6. Tune the fit procedure
7. Set the model parameters
8. Select the fit parameters
9. Start calculation
The results are stored in ASCII format in the specified output directory. A single table,
FITPARS.TXT, summarizes the fit parameters of all spectra. All calculation settings
are documented in an actual copy of the file WASI.INI. The fit curves of all spectra can
be saved as single files.
4.2 ERROR ANALYSIS
No measurement is perfect, no model is exact, and no input data set is complete.
Thus the results of data analysis are unavoidably affected by errors. Sensitivity studies
are designed to estimate the errors for a determined parameter caused by the different
error sources. The best way to perform a sensitivity study is to simulate a large number
of measurements using a reliable model, and to then analyze these subsequently as if
they were real measurements. Well-defined discrepancies between the forward and
inverse models can be introduced which cause errors in the retrieved parameters. In this
way, parameter errors can be attributed quantitatively to different error sources. The
advantages of using simulated spectra rather than measured spectra are that all studied
effects are under control, and that the entire expected parameter interval can be covered
without a gap.
4.2.1 Errors from the sensor
The number of parameters that can be derived from a measurement and the
accuracy of the estimates depends very much on the sensor and on data quality. The
following instrument characteristics are relevant: number, center wavelengths and
102
Gege and Albert