Task sdbaseline fits and/or subtracts a baseline from single-dish spectra in MS format. Given parameters that define the baseline to be fit (function type, order or the polynomial, etc.), sdbaseline computes the best-fit baseline for each spectrum using the least-squares fitting method and, if you want, subtracts it. The best-fit baseline parameters (including baseline type, coefficients of basis functions, etc.) and other values such as residual rms can be saved in various formats including ascii text (in human-readable format or CSV format) or a baseline table (a CASA table). Task sdbaseline also has a mode to 'apply' a baseline table to MS data.  For each spectrum in the MS, the best-fit baseline is reproduced from baseline parameters stored in the specified baseline table, and subtracted. Putting the "fit" and "subtract" into separate processes can be useful for pipeline processing of huge datasets.

Baseline Model Functions 

The user can specify the function to be used for the baseline with the blfunc keyword (e.g. blfunc = 'poly'). In general, polynomial fitting is stable. Sinusoid fitting is a special mode that could be useful for data that clearly shows a standing wave in the spectral baseline.

The sdbaseline procedure gives the user the opportunity to specify unique baseline fitting parameters for each spectrum, using the setting blfunc='variable'. Note this is an expert mode! The fitting parameters should be defined in a text file for each spectrum in the input MS. The text file should store comma separated values in the following order: row ID, polarization, mask, clipniter, clipthresh, use_linefinder,  thresh, left edge, right edge, avg_limit, blfunc, order, npiece, nwave. Each row in the text file must contain the following keywords and values:

• 'row': row number after selection
• 'pol': polarization index in the row
• 'clipniter': maximum iteration number for iterative fitting
• 'blfunc': function name.  available options are 'poly', 'chebyshev', 'cspline',and 'sinusoid'
• 'order': maximum order of the polynomial. Needed when blfunc='poly' or 'chebyshev'
• 'npiece': number of piecewise polynomials. Needed when blfunc='cspline'
• 'nwave': a list of sinusoidal wave numbers. Needed when blfunc='sinusoid'

Output Files 

The task outputs the baseline-subtracted MS data set.  Users should specify the output data file name with the outfile keyword. 
Also, the fit parameters, terms, and rms of the baseline can be saved into an ascii text file (in human-readable format or CSV format) or a baseline table (a CASA table). By default, a text file named  + '_ blparam.txt' is output. The saved baseline table can be used later to subtract the baselines from an MS.

Fitting and Clipping

In general, least-squares fitting is strongly affected by extreme data points, making the resulting fit poor. Sigma clipping is an iterative baseline fitting method that clips data based on a certain threshold. The threshold is set as a certain factor times the rms of the resulting (baseline-subtracted) spectra. If sigma clipping is on, baseline fit/removal is performed several times, iteratively. After each baseline subtraction, data whose absolute value is above the threshold are excluded from the next round of fitting. By using sigma clipping, extreme data are excluded from the fit so the resulting fit is more robust.

The user can control the rms multiplication factor using the parameter clipthresh, for the clipping threshold. The actual threshold for sigma clipping will then be (clipthresh) x (rms of spectra). Also, the user can specify the maximum number of iterations with the parameter clipniter.

In general, sigma clipping will make the procedure slower since it increases the number of fits per spectra. However, it is strongly recommended to turn on sigma clipping unless you are sure that the data is free from any kind of extreme values that may affect the fit.