Re: Transit Fitting plus referencing software in paper
Posted by karenacollins on Feb 13, 2015; 11:03am
URL: http://astroimagej.170.s1.nabble.com/Transit-Fitting-plus-referencing-software-in-paper-tp199p247.html
Hi David,
The 'Host Star Parameters' on the AIJ fit panel are used only to estimate the actual planet radius, 'Rp (Rjup)', when combined with the value of (Rp/R*)^2 from the light curve model fit. The only parameter that is actually used in the calculation is 'R* (Rsun)'. So, if you have prior knowledge of 'R* (Rsun)', then enter that number directly and ignore all of the other host star parameter values that show up. The other host star values will not affect the determination of the actual planet radius or anything else about the transit model. Entering one of the other values is useful only if R* is not known directly. These values are all tied together based on tables in Allen's Astrophysical Quantities, 4th edition. The connection between the values is only a very rough estimate which is valid only for "Normal" main sequence stars (as defined in Allen's Astrophysical Quantities). As an example, if the only prior information that is known about the host star is the J-K color, enter that value in the corresponding box, and the corresponding value of R* from the Allen's tables will be displayed in the 'R* (Rsun)' box and that value of R* will be used to calculate Rp.
Just to clarify, the host star parameters, and in particular 'R* (Rsun)' (either entered direct or estimated from one of the other host star parameters) are not used at all in the light curve model fit. The only way that the resulting 'R* (Rsun)' value is used is to calculate 'Rp (Rjup)' that is displayed in the lower-right corner of the 'Transit Parameters'.
For your case, enter R* directly and ignore the other host star parameters. They other parameters will not affect the light curve model fit in any way.
Detrending requires a little experimentation with your data, and will become easier once you understand what trend parameters tend to affect your data. In the fit example I showed previously, airmass, time (BJD_TDB), FWHM in the image (Width_T1), and total comparison star counts (tot_C_Cnts) were used for detrending. Normally four detrend parameters is too many, but that light curve has good out-of-transit baseline data on both side of the transit. One or two detrend parameters is usually more appropriate.
Airmass is generally the most important parameter to use for detrending since it affects all observations for which the target and comparison stars are not of the same color (and they are generally not of the same color). If you use more than airmass as a detrending parameter, you will generally need good pre-ingress and post-egress baseline data. If you don't have good baseline line on both sides of the transit, using multiple detrend parameters will likely modify the light curve inappropriately. For partial transits, even airmass as a single detrending parameter can introduce more problems in the light curve than it corrects. Detrending using your time base column can remove a linear trend in time from the light curve (for instance if the host star or one of the comparison stars is actually changing brightness slowly throughout the transit). I have found that changes in FWHM from image to image also introduces noise in the light curve that can often be detrended out of the data if good baseline data are available. Finally, detrending with total comparison star counts sometimes reduces systematics in data, particularly if your target and comparison stars are not of equal brightness, and particularly if you don't correct for CCD non-linearity as part of the data calibration process.
It is tempting to throw a bunch of detrending parameters at your light curves to make them as clean as possible. However, multiple detrending varaibles can combine to cause undesirable effects in the data (over fitting). So, it is best to find a good balance between adding more detrending parameters and retaining the integrity of the data. One way to approach that is to observe the Bayesian Information Criterion (BIC) value, which is circled in the figure below. You can read about the details of BIC on Wikipedia if you interested. The main thing you need to know is that if the BIC value reduces by more than 2.0 when a new detrend parameter is added, you might consider keeping that detrend parameter. If the BIC value reduces by less than 2.0 or increases, then that detrending parameter is likely not appropriate to keep as part of the fit.
I normally apply an airmass detrend to the comp stars for display purposes to help gauge if they are flat after airmass detrending. If you are fitting rel_flux_Txx though, the displayed detrending of the comp star does not actually affect the target star light curve or model fit. Only detrending of the target star light curve affects its model fit.
Meridian flip detrending can be used any time there is a discontinuity in the data. For instance, if you lose guiding and then regain guiding and the field is positioned differently on the CCD, then Meridian flip detrending might be appropriate. Meridian flip detrending allows for an arbitrary offset in the light curve at the marked "meridian flip" time, so unless you have good out-of-transit data, Meridian flip detrending can definitely introduce unwanted effects in the light curve. So, only use Meridian flip detrending if you suspect that there is an offset in the lightcurve at the discontinuity in the data (i.e. an instantaneous increase or decrease in the light curve from before and after the discontinuity), and if you have good out-of-transit data.
In general, lots of out-of-transit data make detrending much more effective and reduces the chance that detrending will incorrectly modify your light curve and light curve model fit.
Karen
URL: http://astroimagej.170.s1.nabble.com/Transit-Fitting-plus-referencing-software-in-paper-tp199p247.html
Hi David,
The 'Host Star Parameters' on the AIJ fit panel are used only to estimate the actual planet radius, 'Rp (Rjup)', when combined with the value of (Rp/R*)^2 from the light curve model fit. The only parameter that is actually used in the calculation is 'R* (Rsun)'. So, if you have prior knowledge of 'R* (Rsun)', then enter that number directly and ignore all of the other host star parameter values that show up. The other host star values will not affect the determination of the actual planet radius or anything else about the transit model. Entering one of the other values is useful only if R* is not known directly. These values are all tied together based on tables in Allen's Astrophysical Quantities, 4th edition. The connection between the values is only a very rough estimate which is valid only for "Normal" main sequence stars (as defined in Allen's Astrophysical Quantities). As an example, if the only prior information that is known about the host star is the J-K color, enter that value in the corresponding box, and the corresponding value of R* from the Allen's tables will be displayed in the 'R* (Rsun)' box and that value of R* will be used to calculate Rp.
Just to clarify, the host star parameters, and in particular 'R* (Rsun)' (either entered direct or estimated from one of the other host star parameters) are not used at all in the light curve model fit. The only way that the resulting 'R* (Rsun)' value is used is to calculate 'Rp (Rjup)' that is displayed in the lower-right corner of the 'Transit Parameters'.
For your case, enter R* directly and ignore the other host star parameters. They other parameters will not affect the light curve model fit in any way.
Detrending requires a little experimentation with your data, and will become easier once you understand what trend parameters tend to affect your data. In the fit example I showed previously, airmass, time (BJD_TDB), FWHM in the image (Width_T1), and total comparison star counts (tot_C_Cnts) were used for detrending. Normally four detrend parameters is too many, but that light curve has good out-of-transit baseline data on both side of the transit. One or two detrend parameters is usually more appropriate.
Airmass is generally the most important parameter to use for detrending since it affects all observations for which the target and comparison stars are not of the same color (and they are generally not of the same color). If you use more than airmass as a detrending parameter, you will generally need good pre-ingress and post-egress baseline data. If you don't have good baseline line on both sides of the transit, using multiple detrend parameters will likely modify the light curve inappropriately. For partial transits, even airmass as a single detrending parameter can introduce more problems in the light curve than it corrects. Detrending using your time base column can remove a linear trend in time from the light curve (for instance if the host star or one of the comparison stars is actually changing brightness slowly throughout the transit). I have found that changes in FWHM from image to image also introduces noise in the light curve that can often be detrended out of the data if good baseline data are available. Finally, detrending with total comparison star counts sometimes reduces systematics in data, particularly if your target and comparison stars are not of equal brightness, and particularly if you don't correct for CCD non-linearity as part of the data calibration process.
It is tempting to throw a bunch of detrending parameters at your light curves to make them as clean as possible. However, multiple detrending varaibles can combine to cause undesirable effects in the data (over fitting). So, it is best to find a good balance between adding more detrending parameters and retaining the integrity of the data. One way to approach that is to observe the Bayesian Information Criterion (BIC) value, which is circled in the figure below. You can read about the details of BIC on Wikipedia if you interested. The main thing you need to know is that if the BIC value reduces by more than 2.0 when a new detrend parameter is added, you might consider keeping that detrend parameter. If the BIC value reduces by less than 2.0 or increases, then that detrending parameter is likely not appropriate to keep as part of the fit.
I normally apply an airmass detrend to the comp stars for display purposes to help gauge if they are flat after airmass detrending. If you are fitting rel_flux_Txx though, the displayed detrending of the comp star does not actually affect the target star light curve or model fit. Only detrending of the target star light curve affects its model fit.
Meridian flip detrending can be used any time there is a discontinuity in the data. For instance, if you lose guiding and then regain guiding and the field is positioned differently on the CCD, then Meridian flip detrending might be appropriate. Meridian flip detrending allows for an arbitrary offset in the light curve at the marked "meridian flip" time, so unless you have good out-of-transit data, Meridian flip detrending can definitely introduce unwanted effects in the light curve. So, only use Meridian flip detrending if you suspect that there is an offset in the lightcurve at the discontinuity in the data (i.e. an instantaneous increase or decrease in the light curve from before and after the discontinuity), and if you have good out-of-transit data.
In general, lots of out-of-transit data make detrending much more effective and reduces the chance that detrending will incorrectly modify your light curve and light curve model fit.
Karen
| Free forum by Nabble | Edit this page |