UVIS calibration update Greg Holsclaw, Bill McClintock Jan 7, 2008 Outline • • • • Detection of Spica variability by UVIS FUV evil pixel response FUV absolute calibration Differences between independent determinations of UVIS sensitivity UVIS detection of Spica flux variability Introduction • Motivation: Spica is known to be a variable star, but the quantitative impact of this characteristic on the systematic variation observed with UVIS is unknown • Approach: Search for correlation of UVISFUV photometry with Spica flux variation models Background on Alpha Vir (Spica) • Spica is a non-eclipsing double-lined spectroscopic binary system – – – Though not spatially resolvable, each component is detectable through measurements of out-of-phase Doppler shifts in the constituent spectral lines Non-eclipsing due to large apparent orbital inclination of ~70 degrees Both stars are of a similar spectral class: • • • Spica is the brightest rotating ellipsoidal variable star – – – – – • Primary: B1V Secondary: B4V The stars have a distorted ellipsoidal shape due to mutual gravitation effects As the components revolve, the visible area (and thus the observed flux) changes with orbital phase Since this is a geometric effect, it should be roughly wavelength-independent Orbital period is 4.01454 days Amplitude of flux variation in V-filter ~3% http://observatory.sfasu.edu The primary of Spica is a Cepheid variable – – – – Periodic variation in the pulsating primary star is much shorter than the system’s orbital period and about a factor of 2 less in magnitude Period is 4.17 hours Amplitude of flux variation in V-filter ~1.5% This short-term variation, identified in 1968, became undetectable in the early 1970’s (but may return again due to precession of the primary’s rotation axis relative to the orbital plane, which has a period of 200 years [Balona, 1986]) Ellipsoidal variation model Variation in flux is given by [Shobbrook, 1969; Sterken et al, 1986]: dE = A M2/M1 (R/D)3 (1+e cos(TA+Φ))3 (1-3cos2(TA+TA0+Φ) sin2i ) Where: A=0.822 (wavelength dependent “photometric distortion”) M2/M1 = 1/1.59 (ratio of masses) R = 7.6 Rsun = 5.2858e6 km (polar radius of primary) D = 1.92916e7 km (mean separation between stars) e = 0.14 (orbital eccentricity) TA = time/T0·2π (true anomaly, approx as mean) T0 = 4.01454 days (orbital period) TA0 = 242 degrees (apparent angle to line of apsides in year 2005, has precession period of 133 years) i = 65.9 degrees (orbital inclination) Φ = empirical phase shift, a free parameter to match with data One period of the expected variation in flux from Spica for the year 2005 UVIS measurements Spatial profile of a single scan • We now have several drift observations of Spica • This is where the spacecraft is slewed slowly, such that the star moves along the entrance slit while many scans are acquired UVIS measurements Spatial profile of a single scan • • • Spatial profiles of all scans in one observation We now have several drift scans of Spica This is where the spacecraft is slewed slowly, such that the star moves along the entrance slit while many scans are acquired These plots show a spectral summation across the array, yielding a spatial profile UVIS measurements Total FUV signal vs. star location • • • Spica is the bright UV calibration star of choice for UVIS Many observations span over the years since launch, but there are only nine easily comparable along-slit slew scans There is modest variation in absolute sensitivity along the slit, so the instrument response is averaged over the region where the star is located between rows 10 and 20 UVIS measurements vs. time Total FUV signal vs. star location • • • Total FUV signal vs. time Spica is the bright UV calibration star of choice for UVIS Many observations span over the years since launch, but there are only nine easily comparable along-slit slew scans There is modest variation in absolute sensitivity along the slit, so the instrument response is averaged over the region where the star is located between rows 10 and 20 Predicted and UVIS-measured variation measured model The predicted variation has a period of only ~4 days, so the eight UVIS observations sample this curve sporadically. The correlation between the measured values and the sampled predicted variation is good. Predicted and UVIS-measured variation measured model The predicted variation has a period of only ~4 days, so the eight UVIS observations sample this curve sporadically. The correlation between the measured values and the sampled predicted variation is good. Predicted and UVIS-measured variation measured model The predicted variation has a period of only ~4 days, so the nine UVIS observations sample this curve sporadically. The correlation between the measured values and the sampled predicted variation is good. Dividing out the predicted variation leaves a more obvious slow decline in sensitivity with time, which is not surprising. Predicted and UVIS-measured variation time modulo T0 This view shows how well a single period of the ellipsoidal variation curve has been sampled by UVIS so far. Predicted and UVIS-measured variation time modulo T0 This view shows how well a single period of the ellipsoidal variation curve has been sampled by UVIS so far. Predicted vs measured relative values shows a high correlation. Dividing out the slow linear decline improves the r value to 0.86. Conclusion • In spite of the challenges of the UVIS-FUV detector, fewpercent photometric variations in the flux from Spica are detectable • Removing this systematic trend from the data will enable more accurate tracking of the relative instrument sensitivity • Residual discrepancies from the model could be due to: – contribution from the high frequency (~4 hr) Cepheid variation – spectral variability in the magnitude of the ellipsoidal fluctuation Continued Work • Future routine calibration observations of Spica will provide better sampling of the ellipsoidal variation • A search for the transient Cepheid variation of the Spica primary could be conducted with the HSP or FUV channel – For the HSP this would require staring at the star for an entire period of 4 hours, resulting in a well-sampled light curve. – For the FUV, in order to avoid starburn, we could perform multiple along-slit slew zigzag scans across half the slit length. The typical duration of a single along-slit slew scan is ~1 hour. This method would acquire eight samples across the 4 hour period. • Validation of these results as well as an independent search for the Cepheid variation can be gained by analyzing the observations of Spica by the Cassini Imaging Science Subsystem (ISS) References • Balona, On the amplitude decrease of the beta Cephei stars SPICA and 16 Lacertae, Royal Astronomical Society, Monthly Notices, vol. 217, Nov. 15, 1985, p. 17P-21P • Shobbrook et al, Light variations in Spica, Monthly Notices of the Royal Astronomical Society, Vol. 145, p.131 (1969) • Sterken et al, The variability of Spica, Astronomy and Astrophysics, vol. 169, no. 1-2, Nov. 1986, p. 166-170 Evil pixel characterization Evil pixel definition • Evil pixels are those FUV detector elements which exhibit an anomalously low response relative to their immediate neighbors • Evil pixels have been identified by smoothing an image where all pixels are illuminated (such as a LISM scan) and setting a statistical threshold below this mean local value; pixels below this value are considered ‘evil’ • This threshold is somewhat arbitrary (different UVIS team members have chosen a variety of defendable values - 3 sigma, 5 sigma, etc low) and therefore evil pixel identification is not unique Example of evil pixels • This plot shows a single column from a aggregate (summed) image of Spica as it was slewed along the entrance slit • Evil pixels are seen to have a significantly lower response than “good” pixels Distribution of evil pixels Evil pixels comprise ~15% of the detector Fraction of evil pixels per column is highly variable Fraction of evil pixels per row is fairly constant FUV evil pixel characterization • It remains unknown if the low-response ‘evil’ pixels present any useful information • One test is to see if they respond linearly to a change in incident flux • We have many observations of several UV bright stars • However, to do this experiment properly, the star should be slewed along the slit, thereby illuminating each column uniformly Comparison of FUV stellar fluxes • • These plots show the spectra of all the stars observed thus far The comparison star should fill most of the detector, be of a significantly different flux level than Spica, but not be too dim Comparison of FUV stellar fluxes • • Plot of the total count rate from each star, normalized and sorted Choose three candidate stars, ~4x lower in irradiance than Spica – – – • Alp Pav Del Sco Eta UMa Comparison with absolute irradiance measurements by SORCE-SOLSTICE will provide an additional path to validating the calibration Selected data for analysis • • • • • • Spica: Eta Uma: We now have identical along-slit slews of Alpha Vir (Spica) and Eta UMa This plot shows the total counts acquired from both stars Alp Vir spectral class: B1III Eta Uma spectral class: B3V As the stars are not the same spectral class, we cannot expect the spectral ratio to be a flat multiplicative factor (though Spica is roughly 4.3x brighter) In addition, the stars exist in different directions in the sky and are located at different distances, thus the scattering and absorption effects by the interstellar medium are not shared FUV2007_140_16_20_53_UVIS_045IC_ALPVIR001_PRIME FUV2007_192_02_04_18_UVIS_048IC_ETAUMA001_PRIME Analysis procedure • Assume the star was slewed in a direction parallel to the entrance slit • Sum the images where the star is located comfortably on the detector • Ratio the summed image from Eta Uma (Eta Ursae Majoris, Alcaid) to the summed image from Alp Vir (Alpha Virginis, Spica) • Assume all pixels in a column should exhibit the same flux ratio (i.e. the absolute irradiance spectrum from either star need not be known) • Normalize each column vector in the ratio image to the mean of the good pixels in that column • Overplot every normalized column vector – We expect good pixels to be distributed about 1, with a spread determined by the residual flat field and counting statistics – Are the evil pixels also distributed about 1, or are they consistently higher or lower? Image sums and ratio Eta UMa Alp Vir Image sums and ratio Eta UMa Alp Vir Eta UMa divided by Alp Vir Relative pixel response • • • This plot shows the ratio of the pixel values in a single column to the goodpixel average Evil pixel values seem to be somewhat lower on the average In order to bring out a trend, overplot columns normalized in the same way on one plot Relative pixel response • • • • Evil pixel values (blue) have been shifted 0.5 pixels to the right for clarity This plot shows the ratio of each column to the column good-pixel average If both good and evil pixels respond the same to a multiplicative change in flux, both ratio values should be centered about 1 Good pixels are seen to be distributed about 1, while evil pixels are distributed about 0.8 To see this a different way, try computing a histogram for the good and evil pixels separately Relative pixel response • Another useful view of this data is to calculate a histogram for the good and evil pixel ratio values • Here the evil pixel mode (most frequent value) is seen to be about 20% lower than the good-pixel mode Relative pixel response • • Normalizing the histograms to the peak value shows that there is a much wider distribution of evil pixel values If Eta Uma is exactly 4x fainter than Spica, the evil pixels would show it to be ~4.8x fainter Relative pixel response across detector • • Evil pixels This is a plot of the average value of each column (in the ratio image) for good and evil pixels, normalized to the good pixel average Evil pixel averages are seen to be consistently lower Conclusion • Evil pixels are defined to be anomalously low-response detector elements • This investigation has shown that these pixels respond to a change in incident flux differently than “good” pixels • If we assume that good pixels are linear, then the evil pixels can be considered to exhibit a nonlinear response • This provides additional motivation for the exclusion of evil pixels in the analysis of UVIS-FUV data Absolute calibration of the FUV channel Introduction • Given the inherent difficulty in calibrating FUV spectrometers, previous measurements of stellar sources are of uncertain utility for use as a reference for UVIS • SOLSTICE is a UV monochromator onboard the SORCE spacecraft, and was developed at LASP under the direction of Bill McClintock • It’s objective is to measure the absolute spectral irradiance of the Sun in the FUV and MUV, and tracks degradation through the measurement of stellar irradiance • SOLSTICE was rigorously calibrated on the ground using a stable and well-known UV source at a NIST facility (SURF), and may be the best calibrated spaceborne FUV spectrometer • Spica is one of the brightest FUV sources in the sky, and has been observed often by both UVIS and SOLSTICE • Recent access to the SOLSTICE stellar data allows for a new validation of the UVIS-FUV absolute sensitivity Aggregation of various measurements of the FUV irradiance from Spica • • • This plot shows the calibrated spectral irradiance of Spica from several previous investigations Most measurements agree fairly well above ~140nm The minimum wavelength for SOLSTICE stellar irradiance is ~130nm, due to contamination by Lyman-alpha emission from the Earth’s geocorona SOLSTICE, Rocket, and UVIS comparison • • • A rocket measurement by Brune et al (1979) has been previously used as the Spica irradiance standard Rocket and SOLSTICE data are 1nm resolution, so UVIS has been smoothed to 1nm On the average, we find that UVIS-FUV is within ~10% of the SOLSTICE spectrum SOLSTICE, Rocket, and UVIS comparison • • • A rocket measurement by Brune et al (1979) has been previously used as the Spica irradiance standard Rocket and SOLSTICE data are 1nm resolution, so UVIS has been smoothed to 1nm On the average, we find that UVIS-FUV is within ~10% of the SOLSTICE spectrum Conclusion • The current FUV absolute calibration is consistent to within ~10% of the best measurement to date of the stellar flux of Spica Continued Work • Compare other UVIS-measured FUV stellar fluxes with the SOLSTICE measurements Comparison of Independently derived FUV sensitivities Introduction • At the June 2007 team meeting in Goslar, it was noted that inconsistent FUV results were obtained when combining calibration tools provided by Don and LASP • The origin of this problem lies in the details of how the two independent instrument calibrations were derived Objectives • Quantitatively compare the two independently derived FUV calibrations • Understand the differences between the derivation techniques • Later: – Highlight the strengths and weaknesses of the two approaches – Determine through team discussion if the disagreement can be resolved through: • • • • A new perspective new laboratory results a re-analysis of the original data new observations – Reach a team consensus on the instrument calibration FUV Calibration comparison between Don and LASP Lores, filled slit • • • • The 1999 LASP sensitivity is based on the laboratory measurements Don’s sensitivity corresponds to an independent derivation based on the laboratory measurements and modified by lunar results Significant shape differences are present, on the order of 20 – 30% Largest discrepancies at 115nm, 135nm, 170nm Adjustment of absolute magnitude of Don’s calibration • • • • The LASP 1999 curve refers to the state of the on-ground laboratory calibration Don’s calibration has been adjusted for results from the Moon, and so involves a red-response correction (significant above ~173nm) Increasing the magnitude of Don’s calibration curve by 9% will equate the areas between the LASP 1999 lab cal and Don’s curve within the wavelengths of 114 and 173nm This will minimize the relative deviation from the LASP calibrated results. Fundamental differences between the two calibration approaches • The LASP calibration is based on the measured instrument sensitivity using standard detectors. These values were validated through a comparison of the UVIS measurements of H2 and N2 emission with Joe Ajello’s laboratory measurements of H2 and N2 with a known excitation source • Don’s calibration is based on a comparison of the same UVIS measurements of H2 with a theoretical model of the H2 emission • These two different references produce significant differences in the derived calibration using the same laboratory dataset Conclusions • Significant differences between the two independently derived FUV calibrations have been known since at least 2000 – Spectral variations of 20 – 30% • Fundamental differences in interpretation of the laboratory calibration results have led to the two disparate calibration vectors Continued Work • Produce absolute stellar fluxes using Don’s calibration, adjusted for time variations in the sensitivity, and compare with SOLSTICE