NMFwisp

Overview

Wisps are scattered-light artifacts that appear in JWST/NIRCam short-wavelength images. NMFwisp models this structure with a compact template basis and returns both the best-fit wisp model and its uncertainty.

Because wisps exhibit significant morphological variation (see examples here), traditional subtraction methods based on a single template can leave residuals, especially when the wisp morphology differs substantially from the reference template. The NMF method captures this variation using a multi-component wisp basis.

NMFwisp is designed for simple implementation. In most cases, users only need to call fit_wisp. Wisp templates are downloaded automatically with the package, which are learned from the deep, extensive JADES data. The wisp subtraction should be applied in the Stage 2 JWST calibration before 1/f correction.

For questions about the code, please contact Zihao Wu (zihao.wu@cfa.harvard.edu).

Installation

Install the released package from PyPI:

pip install nmfwisp

For a development install from source:

git clone https://github.com/zihaowu-astro/NMFwisp.git
cd NMFwisp
pip install -e .

NMFwisp requires NumPy, Astropy, and SciPy.

Quickstart

The following example loads a NIRCam exposure, estimates the wisp model, and subtracts it from the science image.

from astropy.io import fits
import matplotlib.pyplot as plt
import numpy as np
from nmfwisp import fit_wisp

filter_name = "F150W"
detector_name = "nrcb4"

filename = "./data/jw01286001001_07201_00003_nrcb4_rate.fits"
maskfile = "./data/jw01286001001_07201_00003_nrcb4_cal_bkgsub_tweak_smask-full.fits"

data = fits.open(filename)["SCI"].data
err = fits.open(filename)["ERR"].data
mask = fits.open(maskfile)[0].data.astype(bool)

wisp, wisp_e = fit_wisp(
    data,
    err,
    mask,
    detector_name=detector_name,
    filter_name=filter_name,
    correct_1f=False,
)

corrected = data - wisp

Example data are available here: example-data.tar.gz.

You may also download the example data with the following commands:

curl -L -O https://github.com/zihaowu-astro/NMFwisp/releases/download/v1.0/example-data.tar.gz
tar -xzf example-data.tar.gz

The source mask can be built from long-wavelength (LW) NIRCam images, which are usually obtained together with the short-wavelength (SW) images. We provide a source-mask script, adapted from the JADES DR5 reduction pipeline. Please see the Source Mask page for details.

API Reference

fit_wisp(data, err, mask, wisp_path=None, 
         detector_name=None, filter_name=None, correct_1f=False)

NMF Components

The NMF template library represents wisps as a combination of multiple components. These components capture the main variations while keeping the fit constrained to recurring detector features.

Templates

Bundled templates are available for the detector/filter combinations distributed in nmfwisp/templates. The package checks the following detectors and filters.

Advanced Usage

Custom template path

We recommend using the default wisp templates bundled with the package. If you want to use wisp templates built from your own data, set wisp_path to the custom template directory as shown below. The custom directory should follow the same detector/filter layout as the bundled templates. Code for building NMF wisp templates is documented on the Developer API page.

wisp, wisp_e = fit_wisp(
    data,
    err,
    mask,
    wisp_path="/path/to/templates",
    detector_name="nrcb4",
    filter_name="F150W",
)

Iterative 1/f correction

Set correct_1f=True to fit the wisps and 1/f noise simultaneously. We recommend enabling this option because residual 1/f noise can bias the wisp fit. A prior 1/f correction is often imperfect due to wisp contamination, so the iterative correction can further reduce 1/f residuals and improve the inferred wisp flux.

We use a fast, median-based 1/f correction algorithm. The correction is applied only internally during the wisp fit; it improves the wisp-flux inference but does not directly subtract 1/f noise from the data. We recommend applying additoinal 1/f correction using more advanced methods after subtracting wisps.

wisp, wisp_e = fit_wisp(
    data,
    err,
    mask,
    detector_name="nrcb4",
    filter_name="F150W",
    correct_1f=True,
)

Wisp Morphology Variation

Wisp morphology is not perfectly fixed between observations. A traditional single-template subtraction cannot adapt to these shape changes, so it can leave residual wisp structure or subtract the artifact imperfectly.

Citation

If you use NMFwisp, please cite the associated paper. The paper PDF is available at arXiv:2601.15958.

@ARTICLE{2026arXiv260115958W,
       author = {{Wu}, Zihao and {Johnson}, Benjamin D. and {Eisenstein}, Daniel J. and {Cargile}, Phillip and {Hainline}, Kevin and {Hausen}, Ryan and {Rinaldi}, Pierluigi and {Robertson}, Brant E. and {Tacchella}, Sandro and {Williams}, Christina C. and {Willmer}, Christopher N.~A.},
        title = "{JWST Advanced Deep Extragalactic Survey (JADES) Data Release 5: Wisp Subtraction with the Non-negative Matrix Factorization Algorithm}",
      journal = {arXiv e-prints},
     keywords = {Instrumentation and Methods for Astrophysics, Astrophysics of Galaxies},
         year = 2026,
        month = jan,
          eid = {arXiv:2601.15958},
        pages = {arXiv:2601.15958},
          doi = {10.48550/arXiv.2601.15958},
archivePrefix = {arXiv},
       eprint = {2601.15958},
 primaryClass = {astro-ph.IM},
       adsurl = {https://ui.adsabs.harvard.edu/abs/2026arXiv260115958W},
      adsnote = {Provided by the SAO/NASA Astrophysics Data System}
}