T Tess Refresher Training

Working with TESS Data After a Break

If you're coming back to TESS after a few months away, the first thing to deal with is that the data products and quality flags don't stay fresh in your head. I keep meaning to run through the pipeline notes again, but the documentation is spread across MAST, the HEASARC pages, and the SPOC wiki, so it takes real effort to reconstruct the workflow from memory.

Tess Refresher Training

You can access the current and archived TESS data through MAST. Start by pulling the calibrated product files for whatever observing mode you need. Full-frame images are 2-minute cadence at about 26 megabytes per sector per camera. Target pixel files are where the short cadence data lives, typically 20 seconds per frame. Most people don't realize that short cadence is only available for about 200,000 preselected targets across all sectors, so if you're studying something that wasn't on that list you're stuck with 2-minute data or you need to request special observation time through the Guest Investigator process. The pipeline applies a number of corrections before the data reaches you. Background subtraction happens inside the SPOC pipeline. Crosstalk correction accounts for signal bleeding between adjacent CCDs. Then there's the point spread function photometry and the systematic noise removal that produces the PDCSAP flux, which is usually the better product for transit detection work compared to simple aperture photometry SAP flux. The difference matters a lot when you're looking for sub-mmag variations. I ran into a specific problem last year when I was trying to recover a transiting signal in Sector 42 data for a K-dwarf that happened to sit very close to a much brighter star on the same CCD. The standard aperture in the target pixel file was contaminated by the neighbor's flux leakage, even after crosstalk correction. The fix was straightforward once I figured it out: I pulled the raw CCD-level flux for that pixel position using a custom aperture and applied my own background subtraction mask rather than relying on the pipeline's default aperture. It took about three days to debug instead of the hour it should have taken, but it taught me to always inspect the target pixel file visually before committing to any automated product.

Quality flags are another area where people lose time. TESS uses a series of flags in the quality column of light curve files. The basic ones you need to understand include bit 0 for sun angle warnings, bit 1 for earth angle issues, and bit 3 for safe mode events. Bit 18 marks periapsis anomalies where the spacecraft had to fire thrusters. If you're doing high-precision work you should typically mask out segments flagged with bit 18 because the pointing jitter during those periods introduces correlated noise that is nearly impossible to model away. Bit 21 indicates when the target was near a bright star and could have suffered from stray light contamination. The pipeline documentation says the background estimate handles this, but it doesn't always work cleanly at the edges of the frame. One counter-intuitive thing about TESS is that more data isn't always better. The 27-day observing windows per sector mean you get roughly 3 sectors per sky region for the prime mission and up to 13 for the extended mission. But each sector has about 6 percent data gaps from telemetry downlinks and engineering observations. When I combine multiple sectors for a long-term monitoring project, I sometimes find that adding a fifth sector degrades the effective noise floor because the systematics across different pipelines and calibration states aren't perfectly consistent. In those cases I found that weighting each sector by its inverse variance during the combination step produced cleaner results than a simple average. The improvement was measurable—maybe 5 to 10 percent in noise level—which sounds small but makes the difference between a marginal detection and a confident one. Another thing the official training materials don't emphasize enough is the temperature dependence of the TESS camera detectors. The CCDs operate around 56 Kelvin and there is measurable drift over each sector. Flux variations correlated with temperature changes can mimic planetary transits if you're not careful, particularly for very shallow signals under 300 parts per million. I learned this the hard way when a candidate signal I thought was promising turned out to be a false positive caused by thermal contraction of the focal plane assembly during the first week of a sector. Running a principal component analysis on the background pixels and regressing out those variations is now standard practice in our group before we do any transit search.

The TESS software environment itself has changed since the mission started. The original pyKE package is still functional but isn't actively maintained for the latest data releases. TESSCut is the current tool for downloading custom cutouts from MAST and it handles the filtering automatically. If you need to work with full-frame images, the tpfmerge tool from the TESS Photometry package can stitch target pixel files across sectors. The learning curve for these tools is steep if you're approaching them cold, and the error messages aren't exactly helpful. For the TESS Refresher Training you'll want to focus on three areas that actually matter for daily work. First, re-read the SPOC calibration procedure notes so you know exactly which corrections have been applied to each data release. Second, spend time in the target pixel viewer examining a few sector stacks for objects you care about so you recognize normal artifacts versus real problems. Third, check the TESS Science Office announcements for any pipeline updates or bug fixes that happened since you last worked with the data. New data releases sometimes include recalibrated products that differ from older versions, and mixing products from different releases in a single analysis will introduce systematic offsets. There's a practical limitation worth stating plainly. TESS was designed as a survey instrument, not a precision photometer like Kepler or PLATO. The pixel scale is 21 arcseconds per pixel compared to Kepler's 4 arcseconds. This means blended sources are far more common and the aperture photometry you extract from TESS data contains more contaminating flux. If your science case requires parts-per-million level precision on a single star, TESS is often the wrong tool and you'd be better off applying for HST or JWST time. Knowing when to use TESS and when to pass on it is part of being effective with this dataset.

The downloads are free through MAST and the data is publicly available 6 months after obsolation. Processing a typical light curve from raw FFI to a detrended SAP and PDCSAP product takes roughly 20 minutes on a modern laptop if you already have the environment set up. Setting up the environment from scratch takes about 45 minutes including Conda package resolution and dependency installation. Budget accordingly if you're doing this for the first time.