Data reduction method for OPTICAM multiband time series of transiting exoplanets

Abstract

We present a methodology for acquiring and reducing transiting exoplanet light curves obtained with the OPTICAM instrument in the Observatorio Astronómico Nacional en la Sierra de San Pedro Mártir (OAN-SPM). The OPTICAM sCMOS detectors generate significant warm pixels at exposures $\geq$10s, affecting both science and calibration frames. These warm pixels are not removed by standard dark subtraction because they vary unpredictably from frame to frame. We evaluate six pre-processing methods applied to science and calibration images using the transit of TOI-7149~b observed in g$^\prime$r$^\prime$i$^\prime$. A median filter with a 3$\times$3-pixel window minimizes the effect of warm pixels without affecting stellar signals. This median filter best reduces dispersion and red noise in the light curves when stellar peak counts are close to the dark current level. The improvement is less significant when the stellar peak is several thousand counts above the dark current level. We fit a multiband transit model to the light curves, measuring photometric precision, correlated noise, and retrieved planetary parameters. The transit model fitted to the light curves with pre-processing using a 3$\times$3-median filter achieves the highest Bayesian evidence. Thus, it is our recommended method for correcting warm pixels. Finally, we present a reduction pipeline that combines Python modules (PROFE) and AstroImageJ to implement our proposed method for OAN-SPM 2.1m+OPTICAM transiting planet observations.

Figures (16)

• Figure 1: 2D histogram of the standard deviation vs mean counts for each pixel in sequences of 25 darks with different exposure times (i.e., 0.01, 1, 5, 10, 20, and 30 seconds). For simplicity, all sequences are shown for OPTICAM's Channel 1, but Channels 2 and 3 behave similarly. The x- and y-axes are the same in each panel, and are chosen to optimize the comparison between the distributions with varying exposure time. A few pixels extend beyond the shown region for long exposures. As exposure time increases, the warm-pixel branches become progressively more structured, highlighting that both the average dark current and its frame-to-frame variability increase with integration time.
• Figure 2: Spatial distribution of warm pixels in a $100\times100$-pixel subframe from a single dark with 30 s exposure time, where warm pixels appear scattered across the subframe. This subframe is in Channel 1, and in the other Channels the distribution of warm pixels is similar.
• Figure 3: Examples of different types of warm pixels. We show the measured counts (left) and median-normalized counts (right) for distinct warm pixels as a function of consecutive frame number for 100 30-second darks. Top: Type A warm pixels are dominated by an offset in threshold voltage, maintaining relatively constant high mean counts with about 10--20% dispersion; the different colours correspond to different median levels (e.g., pink: 305 counts; teal: 2320 counts; orange: 9315 counts) but show the same behaviour. Middle: Type B warm pixels exhibit unpredictable, stepwise changes in count level; the green example reaches nearly the detector's saturation limit and drops to lower values in other frames, while the purple example has an average of 37214 counts and shows a mixed behaviour between types A and B. Bottom: Type C (blue) warm pixels remain near the median count value for most of the sequence but display occasional upward jumps (salt events), whereas Type D (red) warm pixels stay at high counts for most of the time series with intermittent drops to the median (pepper events). These different types of warm pixels trace the various branches seen in the 2D histograms and highlight the complexity of mitigating their effects in the light curves.
• Figure 4: A target centroid movement over a 9.5-hour time series in channel 1. Blue dots are for X-axis movement, and orange dots are for Y-axis movement. Other channels show similar centroid movement patterns, with variations in the number of displaced pixels due to differences in pixel scale across channels.
• Figure 5: OPTICAM light curves, models, and residuals of TOI-7149 for different reductions in the g$^\prime$ filter. From top to bottom: standard reduction (st), Gaussian convolution kernel with 1-$\sigma$ (g1), Gaussian convolution kernel with 3-$\sigma$ (g3), median filter with 3$\times$3 pixel window size (w3), median filter with 5$\times$5 window size (w5), and median filter with 7$\times$7 window size (w7). Top: Light curves (blue color scale) and models (black) with vertical offset. Bottom: Residuals (blue color scale) for each reduction with vertical offset. These different light curves, models, and residuals from various reductions of the same data demonstrate the impact of the reduction approach on the final results. The g$^\prime$ filter light curves are the most dispersed because in this filter, the target star achieves fewer counts than in the r$^\prime$ and i$^\prime$ filters and therefore is closer to the dark current level of the image.