JWST Advanced Deep Extragalactic Survey (JADES) Data Release 5: Wisp Subtraction with the Non-negative Matrix Factorization Algorithm

TL;DR

JWST/NIRCam wisps introduce low-surface-brightness systematics that bias faint-source photometry. The authors develop a non-negative matrix factorization framework to construct multi-component, band- and detector-specific wisp templates from deep JADES data and subtract wisps via NNLS amplitudes, incorporating a persistence component when present. This approach yields lower residuals and reduced photometric bias compared with STScI templates and PCA, demonstrating robust subtraction across exposures and mosaics. The wisp templates and reduction pipeline are publicly released to enable reliable wisp mitigation for future JWST extragalactic surveys.

Abstract

Wisps are among the most prominent scattered light artifacts in JWST/NIRCam imaging. They often appear in certain regions of the detectors and contaminate observations at surface-brightness levels relevant for faint-source photometry. We introduce a new subtraction method that uses the non-negative matrix factorization (NMF) algorithm to model and remove wisps. Using deep NIRCam observations from the JWST Advanced Deep Extragalactic Survey (JADES) and other programs, we construct multi-component, filter- and detector-specific wisp templates that capture the wisp structures and their exposure-to-exposure morphological variations. Wisps in individual exposures are represented as non-negative linear combinations of these templates, consistent with their additive nature and reducing degeneracies relative to single-template scaling. Compared to existing approaches, our method delivers lower residual root mean square in wisp-affected regions and reduces photometric bias and scatter to levels consistent with clean detector areas. The NMF wisp templates are readily applicable to other datasets and are publicly released to support future NIRCam extragalactic surveys.

Figures (7)

• Figure 1: The left and middle panels are two examples of wisps in the NIRCam B4 detector in the F150W band from different exposures. The right panel shows their difference, demonstrating exposure-to-exposure morphological variations, especially in the relative strengths of wisp substructures. Other sources are masked in the image differencing.
• Figure 2: Multi-component wisp templates derived from the weighted non-negative matrix factorization (NMF) algorithm for the F150W band in the NIRCam B4 detector. The three components are combined to capture the dominant wisp morphology and its exposure-to-exposure variation. From left to right, the contribution to the wisp flux decreases. The template brightness reflects the typical flux level. For reference, $0.01\,\mathrm{MJy\,sr^{-1}}$ is $25.5\,\mathrm{mag\,arcsec^{-2}}$, and the width of the NIRCam B4 detector corresponds to 64 arcmin.
• Figure 3: Example NIRCam F150W image in the B4 detector. The left panel shows the original image, the middle panel shows the result after wisp subtraction using the NMF templates, and the right panel shows the result using the STScI v4 templates. The color bar indicates the signal-to-noise ratio (SNR).
• Figure 4: Comparison of the excess residual root mean square (RMS) in the F150W band for the NIRCam B4 detector after wisp subtraction using the NMF-based, STScI v4, and PCA templates. The color bar indicates the excess residual RMS, which quantifies errors arising from effects beyond background noise. It is defined as the difference between the measured RMS across the dataset and the RMS expected from background noise based on the formal errors, normalized by the noise level. The dashed contour delineates regions with prominent wisps. Within this contour, the mean excess RMS obtained with the NMF method is $\sim$50% of that from the STScI method, while the sharpest residual features are only $\sim$30% of the STScI excess RMS. Outside the contour, most regions are relatively clean; the structures are primarily associated with persistence, unmasked stellar diffraction spikes, and other scattered-light artifacts.
• Figure 5: Photometric differences between wisp-affected and clean regions for the same sources. The photometry is performed with a 0.5$"$-radius aperture on exposure images after wisp subtraction, using either the NMF multi-component templates or the STScI wisp templates. Results are shown for the NIRCam B4 detector in the F150W band. The field sample shows photometric differences measured outside wisp affected regions. The comparison does not include local background subtraction.