Testing and Validation of the Updated Pixel-Based Non-Linearity Calibration File for WFC3/IR

The WFC3\IR channel has an innate non-linear response to incident photons, which is corrected for in the calwf3 pipeline with the NLINFILE reference file. The 2009 solution is based on an average polynomial correction for each IR quadrant and is found to be poorly constrained at high fluence levels (e-) approaching the saturation limit. Using a variety of image types, sources, and sample sequences, we test a new pixel-based linearity correction developed by Shenoy et al. (2025). In nearly all cases, the new correction improves the linearity at fluence levels higher than 50,000 e-, with improvements up to 7% for pixels with fluences approaching the saturation limit (80,000 e-) in the last ima reads. The pixel-based solution also significantly decreases the number of cosmic rays erroneously flagged (due to non-linearity correction errors) during ramp fitting in calwf3, leading to improved photometric accuracy in the calibrated flt data and higher signal-to-noise ratios, particularly in Quad 1 (upper-left detector quadrant). Because the new solution tends to make sources brighter, we recalibrate the five HST flux standards used to compute the IR zeropoints and find a negligible impact (0.1-0.2%) on the published values by Calamida et al. (2024), smaller than the RMS dispersion (0.5%) in the observed to synthetic flux ratios for all five flux standards. The new NLINFILE 9au15283i lin.fits was delivered to CRDS in October 2025 and will be used to reprocess all WFC3/IR imaging and grism observations in the MAST archive. An updated reference file a2412448i lin.fits was delivered in February 2026, improving the results at the highest fluence levels by a few tenths of a percent. Please consult the Addendum for details.

💡 Research Summary

The Wide Field Camera 3 Infrared (WFC3/IR) detector exhibits an intrinsic non‑linear response that must be corrected in the calwf3 pipeline using a reference NLINFILE. The long‑standing 2009 NLINFILE (u1k1727mi.lin.fits) applies a third‑order polynomial that is averaged over the four detector quadrants and was derived from ground‑based STEP50 flat‑field data. Because STEP50 samples the early reads densely but the later reads sparsely, the polynomial is poorly constrained at high fluence (≈ 80 000 e⁻), i.e., near the full‑well limit.

Shenoy et al. (2025) introduced a new pixel‑by‑pixel non‑linearity correction file (NLINFILE 9au15283i.lin.fits) derived from on‑orbit SPARS25 internal flat‑fields. The file retains the original NODE extension that stores per‑pixel saturation limits but replaces the quadrant‑averaged coefficients with individual pixel coefficients. An updated version (a2412448i.lin.fits) was delivered in February 2026, refining the highest‑fluence performance by a few tenths of a percent.

To evaluate the new correction, the authors processed a suite of data covering five sample sequences (SPARS50, SPARS25, SPARS10, STEP25, RAPID), internal flat‑fields, two globular clusters (NGC 1851 and 47 Tuc), and CALSPEC spectrophotometric standards observed both in imaging and grism modes. The analysis focused on three diagnostics: (1) the ratio of instantaneous count rates measured in the individual “ima” reads to the count rate derived from the up‑the‑ramp fit in the calibrated “flt” product, (2) the corresponding fluence level (total accumulated electrons) for each read, and (3) the incidence of CRHIT flags (DQ = 8192) that indicate a pixel was rejected as a cosmic‑ray hit during ramp fitting.

For internal flats, the new correction reduced the peak‑to‑peak variation of the count‑rate ratio across the full fluence range from ~3 % to ~1 % for SPARS50, from ~6–7 % to ~2 % for SPARS25 (the largest improvement), and from ~5 % to ~2 % for STEP25. In the low‑fluence SPARS10 and RAPID sequences the variation remained at ~1 % and ~0.4 % respectively, indicating both corrections perform similarly where the detector is well within its linear regime.

CRHIT flag statistics showed a dramatic decrease: the 2009 correction flagged 23 % of all pixels in a SPARS50 flat and 27 % in a SPARS25 flat, whereas the new correction flagged only 11 % and 9 % respectively. The over‑flagging was especially pronounced in Quadrant 1, reflecting the inadequacy of a quadrant‑averaged polynomial to capture local pixel‑to‑pixel variations.

Star‑cluster tests reinforced these findings. In a 900‑second F110W exposure of NGC 1851 (STEP100 sequence), the peak pixel count‑rate of bright stars (≈ 150–200 e⁻ s⁻¹) deviated from linearity by up to 4 % with the old correction, while the new correction limited the deviation to 1–2 %. Aperture photometry (3‑pixel radius) showed that sources brighter than ~100 e⁻ s⁻¹ appeared up to 0.5 % brighter after applying the new correction. A similar analysis of a 350‑second F160W exposure of 47 Tuc (SPARS25) demonstrated that the old correction could be off by as much as 7 % at the highest fluences, whereas the new correction kept the error below ~2 %.

Finally, the impact on absolute flux calibration was assessed by re‑measuring the count rates of the five HST flux standards (used by Calamida et al. 2024 to derive IR zeropoints) in all 15 IR filters. The new correction made the standards brighter by only 0.1–0.2 %, well below the 0.5 % RMS dispersion of observed‑to‑synthetic flux ratios reported in the zeropoint study. Consequently, the published zeropoints remain essentially unchanged.

In summary, the pixel‑based non‑linearity calibration delivers substantial improvements at high fluence: (i) tighter linearity (1–2 % residuals versus up to 7 % with the old file), (ii) a factor‑two to three reduction in spurious CRHIT flags, (iii) modest but measurable gains in signal‑to‑noise for bright sources, and (iv) negligible impact on the established photometric zeropoints. The NLINFILE 9au15283i was ingested into CRDS in October 2025 and will be applied to all WFC3/IR imaging and grism data in the MAST archive; the February 2026 update (a2412448i) further refines the correction at the very highest fluences. These results underscore the importance of pixel‑level calibration for precision infrared astronomy with HST.