Radiation damage to the Hubble Space Telescope during two Solar cycles, and correction of Charge Transfer Inefficiency using ArCTIc

From 2002 to 2025, the Hubble Space Telescope’s Advanced Camera for Surveys has suffered in the harsh radiation environment above the protection of the Earth’s atmosphere. We track the degradation of its image quality, as Solar protons and galactic cosmic rays have damaged its photosensitive charge-coupled device (CCD) imaging sensors. The rate of damage in low Earth orbit is modulated by $18.5^{+4.5}{-0.5}$ per cent during an 11 year Solar cycle, peaking $430^{+11}{-5}$ days after Solar minimum as recorded in the number of sunspots. The type of damage is consistent with defects in the silicon lattice that have all stabilised into one of three configurations. We also present the open-source Algorithm for Charge Transfer Inefficiency correction (ArCTIc) v7. This models the (instantaneous or gradual) capture of photoelectrons into lattice defects, and their release after (a discrete set or continuum of) characteristic time delays, which creates spurious trailing in an image. Calibrated using the trailing of hot pixels, and applied during post-processing of astronomical images, ArCTIc can correct 99.5% of Charge Transfer Inefficiency trailing averaged over the camera’s lifetime, and 99.9% of trailing in the worst-affected recent data.

💡 Research Summary

**
This paper presents a comprehensive study of radiation‑induced degradation of the Hubble Space Telescope’s Advanced Camera for Surveys/Wide Field Channel (ACS/WFC) CCDs over the period 2002–2025, and introduces a new open‑source correction tool, ArCTIc (Algorithm for Charge Transfer Inefficiency correction) version 7. The authors first quantify the evolution of charge transfer inefficiency (CTI) by analysing a homogeneous set of in‑orbit calibration data. They find that the rate of damage is modulated by the 11‑year solar cycle, varying by 18.5 % ± 4.5 % and peaking 430 ± 10 days after solar minimum, as measured by sunspot number. The damage manifests as three distinct trap species—surface, intermediate, and deep traps—each characterised by a specific capture cross‑section and release time constant.

CTI arises because electrons (or holes) are temporarily captured by lattice defects during the many parallel and serial transfers that occur during CCD readout. Captured charge is released after a characteristic delay, producing a faint trailing behind every source. This effect is highly non‑linear: the amount of trailing depends on source brightness, morphology, and the illumination history of neighboring pixels, making simple convolution‑based de‑convolution impossible. Existing correction schemes (e.g., Anderson & Bedin 2010, CALACS) rely on empirical trap parameters and iterative de‑trailing, achieving at best ~98 % correction for recent data.

ArCTIc v7 addresses these limitations with a physically motivated “volume‑driven” model. The electron packet occupying a pixel is described by a volume‑filling factor V(nₑ)=((nₑ−d·w−d)/w)^β, where d is the notch depth, w the full‑well capacity, and β the well‑fill exponent. Trap species i are characterised by an effective density ρᵢ and a release time τ_relᵢ; capture can be treated as instantaneous (τ_cap→0) or as a finite‑time process (τ_cap>0), the latter allowing the code to reproduce the gradual capture observed in missions such as Gaia. The algorithm tracks trap occupancy using “watermarks” that mark the current fill level of traps within the electron cloud. This abstraction eliminates the need to simulate each individual trap, dramatically reducing memory usage and computational cost. Implemented in parallelised C++ with a Python wrapper, ArCTIc processes a full ACS/WFC quadrant in roughly one second—orders of magnitude faster than previous approaches.

Calibration of the model is performed using the trailing of hot (warm) pixels, which provide a clean, point‑like source of known charge. By fitting the observed trails across the mission timeline, the authors derive year‑by‑year values for ρᵢ, τ_relᵢ, and (when needed) τ_capᵢ. The resulting correction removes 99.5 % of CTI‑induced trailing on average over the entire 23‑year span, and reaches 99.9 % for the most recent, heavily damaged data. This level of correction satisfies the stringent <1 % systematic error budget required for high‑precision photometry, astrometry, and morphology studies (e.g., supernova cosmology, weak‑lensing shear measurements, stellar proper motions).

The paper also discusses limitations. The assumption of spatially uniform trap density ignores Poisson‑level variations that can be important for very low‑signal regimes. The treatment of noise‑induced asymmetry (positive vs. negative read‑noise spikes) is a design choice that may affect ultra‑faint analyses. Moreover, temperature and clock‑speed variations, which alter τ_relᵢ, are not yet incorporated into a real‑time adaptive scheme. The authors propose future work including dynamic temperature‑dependent parameter updates, extension of the model to other space‑based imagers (e.g., JWST NIRCam, Euclid VIS), and integration of machine‑learning techniques to refine trap species identification.

In summary, the study provides a robust, physically grounded description of HST CCD radiation damage, demonstrates that the damage follows the solar cycle, and delivers a high‑performance, open‑source correction tool that restores image quality to the level required for contemporary precision cosmology and astrophysics. The availability of ArCTIc on GitHub, together with extensive documentation and unit tests, invites community adoption and further development, ensuring that the legacy of Hubble’s imaging data remains scientifically valuable for years to come.