Analyzing SED Data in Iris

An SED data file for an object of interest can be loaded into Iris simply by starting the application and selecting the Load File icon on the Iris dekstop. If the file comes in a format which is not natively supported by Iris - so, neither of IVOA-compliant FITS or VOTable, or the XML/CSV format of files downloaded from NED - then it must first be converted to a supported format in the SED Builder interface before it can be displayed in the Iris Visualizer.

The SED Builder accepts as 'non-standard' any binary or text-based file that does not conform to the IVOA SED, Spectrum and Photometry Data Model; these include FITS and VOTable files which do not conform to the IVOA standards. You can learn more about the file formats recognized by the SED Builder on the Iris Supported File Formats page, and refer to the Building and Managing SEDs section of the Iris How-to Guide to learn how to use the SED Builder to convert data for display in Iris.

In this thread we consider the example ASCII data files 3c273.hut.ascii and 3c273.wuppe.ascii, which contain tables of ultraviolet photometric data observed by the Hopkins Ultraviolet Telescope and Wisconsin Ultraviolet Polarimeter Experiment, respectively. The first column in each file contains wavelength values in Angstrom units, and the second column the flux density values in ergs/s/cm²/Angstrom.

As an alternative to loading data from file on disk, we could choose to import SED data directly from NED via the Get an SED from the NED Service option in the Iris Load File window. This feature allows you to search the NED photometry database by object name and/or position, to return the available SED data for that object.

For example, to access SED data for object 3c273 using this option, we would enter the identifier "3c273" in the Target Name field, and then click Import NED SED to automatically display the SED available for this object in the Iris Visualizer.

In order to view both of the SED data segments of 3c273 simultaneously in the Iris display, we simply load and configure both segments in the SED Builder, as outlined above; multiple segments of an aggregate SED (those listed in the "Segments" field of the SED Builder) always display together.

Separate SEDs may also be maintained in a given Iris session, where only one is displayed at a time, by loading each new SED via the the "New" button at the top of the SED Builder window, as opposed to adding new segments in the "Segments" area (see the "Building and Managing SEDs" section of the Iris How-to Guide for details).

It is important to note that when multiple SED data segments are plotted together, the spectral data from each segment is not combined, coadded, or spliced in memory in any way. The raw data from the multiple input spectra is completely preserved in the resulting combination, which is why the two spectra of 3c273 shown in the figure above are offset in the y-direction (flux density) at the same x value (wavelength).

Finally, note that one of the loaded data segments displays in magenta-colored symbols; this indicates that the data have associated zero-value errors. Data points with zero-value errors are ignored in the model fitting in Iris (see the Iris FAQs for details).

Clicking the "Metadata" button in the upper-right corner of the Iris display opens a window containing two separate tabs of information, one displaying the metadata associated with each SED data point in the plot, and the second the independent and dependent data arrays defining the plot.

When available, the Metadata tab includes many useful pieces of information about the data points currently displayed. When a SED is imported from NED, for example, the metadata includes such things as the bibliographic reference code for each data point, the spectral range covered by instrument used to obtain data point, data point uncertainty and flux values as they are published, data point significance values, among others (find the full list of metadata items returned in the "Visualizing SED Data" section of the Iris How-to Guide). For our input data, only the data point ID is returned since our data files were loaded without associated metadata.

The Data tab shows the X (Spectral Axis) and Y (Flux Axis)coordinate values of each SED data point in the Iris display, reflecting the values as they were imported into Iris from NED or uploaded from a file on your hard disk - not what is currently plotted in the Iris display (in the event that you changed the units of the data plot within the display). For example, if the data were uploaded in Jansky flux units versus frequency in Hertz, but then you change the display units to ergs/s/cm2/Angstrom versus Angstrom, the data point values returned will be in Jansky flux units and Hertz.

The Metadata function in the Iris main display also allows us to filter the current SED so that selected (or unselected) points appear suppressed or 'grayed out'. This can be done by selecting the rows corresponding to these points in the Metadata tab, and then clicking "View un-selected" at the bottom of that window. In the display, the points selected in the Metadata window will appear fainter than the unselected points.

Here, we have truncated one of our data segments by suppressing the data points at one end, using this option.

Selected data points may even be extracted into a whole new SED by clicking "Extract" after making the desired row selections in the Metadata window. This will open a new SED in the SED Builder window named "FilterSED" - an ID which you can change - which will display in the Iris Visualizer.

For our analysis we choose to fit only a subset of one of our loaded segments of 3c273, the ~1370-1830 Angstroms range within the HUT segment (recalling that the WUPPE segment has zero-value errors, and therefore would not be included in the fit had we left it in the fitting range). We set this as the range of data to be fit by selecting the Define Range option in the Fit window. This prompts us to click on the data display twice, once to define the lower endpoint of the data range, and once for the upper endpoint.

New in Iris v1.1: the new Custom Model Manager interface allows you to import into Iris your custom table, template, and Python user models, for use with the Iris Fitting Tool. Refer to the "Modeling and Fitting SED Data" section of the Iris How-to Guide to learn how to load your own models into Iris and use them to fit SED data in Iris.

We decide to fit our chosen subset of data with a multi-component, 1-D source model consisting of the sum of one polynomial and two Gaussians, to model the observed continuum emission and two emission lines, respectively.

Before defining our model expression, we delete the default power-law component which is automatically added to the list of model components in the Fit window, since we do not wish to include this component in our model expression for fitting (it is not required that we delete this component if we do not wish to fit with it; it may simply be ignored when defining the model expression). Then, we select "Add" to browse the full list of optical and X-ray models available, to find a 1-D polynomial and Gaussian model. From this list, we make three selections: "polynomial1d" once, and "gauss1d" twice, in order to define these as the components with which to build our full model expression. The components will appear with identifiers "c1", "c2", and "c3" in the Fit window.

Having selected our model components, we combine them into the appropriate arithmetic expression in the "Model Expression" field: "c1+c2+c3"; this is the composite model which will be used to fit our specified data range of the HUT SED of 3c273.

Next, we set some starting model parameter values for the fit by selecting the desired model component in the Fit window, and then "Edit". Parameter values may be set in the fields of the editing window which opens, by entering the desired number and then selecting "Return" on your computer keyboard.

For our analysis, we choose to allow all model parameters to vary in the fit except the position (in Angstroms) of the two Gaussian models. Wet set this preference by keeping the "Fit" box checked in the parameter editing window for all model parameters but the 'pos' gauss1d model parameters. Finally, we measure the position (in Angstroms) of each emission line in the Iris main display - aligning the crosshairs of the mouse cursor with the center of each line and reading the associated X-axis value - and set these values in the parameter editing window.

Our final step in the preparation for fitting involves choosing a fit statistic and optimization method appropriate for our analysis. Selecting "Fit" in the Fit window launches a new window containing our statistic and method options.

We see that the fit can be done with a least-squares statistic, or chi-squared (with various methods for estimating the variances used by chi-squared), or with either of two maximum likelihood statistics that are useful when the data have low numbers of counts. We choose to fit with the least-squares statistic and an optimization method we know to be very robust, Nelder-Mead.

We are now ready to fit the data using our defined data range and model expression, initial model parameters values, and chosen fit statistic and method. Selecting "Start" in the window displaying the fit statistics, initiates the fitting process.

When the fit completes, we see the fitted model overlaid on the SED data in the Iris main display; a data-to-model residuals plot in dex units ( log₁₀(data/model)); and the following fit statistics returned: the final fit statistic value, the number of data points used in the fit, and the number of degrees of freedom in the fit (the null hypothesis probability value of the fit, and the reduced statistic, are not calculated with least-squares).

Changing the statistic to Cash and switching over to the Confidence tab in the window displaying fit statistics, we enter "1.6" in the "sigma" field and click "Start" to calculate and display the 90% confidence limits on the best-fit model parameter values resulting from the fit.

Not completely satisfied with this initial fit - as many parameter bounds have been found to lie outside their hard limit boundaries - we adjust the model parameter values to better match the data, thaw the Gaussian position parameters, experiment with different chi-squared fit statistics, and then re-fit. We continue on in this fashion, iterating through the fit - adding or deleting models from the model expression; changing parameter values or ranges; extending or shrinking the fitted data range - until we find a satisfactory model that best describes our 3c273 SED segment. Finally, we opt to write our custom model expression and best-fit parameters to file via the "Save" option in the Iris Fit window, in order to resurrect this particular fit in a future Iris analysis session.

The SED data open in an Iris session may be written to one or multiple FITS (.fits) or VOTable (.vot) format files (you can save each segment to a separate file, or save the multi-segment SED to one file). For example, to save our 3c273 multi-segment SED to a FITS format file, excluding model data values, we would select the "Save" option at the top of the Iris SED Builder window, to create file 3c273_hut_wuppe.fits. We could then access our saved SED data in a future session of Iris simply by selecting the "Load File" icon on the Iris desktop, and browsing our disk for file 3c273_hut_wuppe.fits.

When writing output data to a file in Iris, the metadata from the input SED file is copied to the output file. When multiple SED data segments are coplotted in Iris, the output file contains the merged metadata from all input files together.

Our Iris fitting session may be saved separately by being sure to select the "Save" option in the Iris Fit window after we have finalized our fit, and prior to exiting the fitting session with the "Dismiss" button in the Fit window.

This is equivalent to selecting the File->Write to file option, which saves the model to a CDB format file which can be read back into Iris at a later time to restore the fitting session; the File->Read from File menu option may be used for this purpose.

We may also use the File -> Write to text file options to save the spectral model to file in a human-readable text format, either all model components, or only those active in the fit. Example output is shown below:

File: Sed0

Thu Jul 12 23:30:18 EDT 2012  Iris 1.1 

TARGNAME: 3c273

Fit parameters:
        Final fit statistic:      0.0090508836113682416
        Reduced statistic:        nan
        Probability [Q-value]:    nan
        Degrees of freedom:       904.0
        Data points:              911
        Last function evaluation: 648

Component 1: polynomial
       c0     = 0.018123535                                               
   F   c1     = 0.0                                                       
   F   c2     = 0.0                                                       
   F   c3     = 0.0                                                       
   F   c4     = 0.0                                                       
   F   c5     = 0.0                                                       
   F   offset = 0.0                                                       

Component 2: gauss1d
       fwhm = 23.508135                                                   
       pos  = 1407.886                                                    
       ampl = 0.04278606                                                  

Component 3: gauss1d
       fwhm = 27.478205   
       pos  = 1790.7229                                                   
       ampl = 0.017182805