EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

How Spectral Bias in Machine-Learning Forecast Emulators Propagates Through the Data-Assimilation Cycle: A Linear–Gaussian Spectral Theory

Nitin Lohan, Pankaj Goel, Sushil Kumar, R. K. Giri, Fahdah Falah Ben Hasher, Dimitris G. Kaskaoutis, Mohamed Zhran

Entropy · 19.11.2026 · 28 (11) · 1276

DOI: 10.3390/e28111276 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Entropy, online erschienen am 19.11.2026. Verfasst von Nitin Lohan und 6 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Machine-learning forecast emulators systematically under-represent small-scale variance, a spectral bias documented across architectures. Its effect inside a data-assimilation cycle has been conjectured rather than derived. Here, a closed-form theory is developed in the simplest setting that retains the following mechanism: a linear–Gaussian cycle on a periodic domain with statistically homogeneous errors, in which the covariance recursion separates into independent scalar recursions per wavenumber k. The emulator is modelled as the true propagator composed with a spectral damping d(k) ∈ [0,1]. Three results are proved. First, the filter is overconfident at damped scales by a closed-form gap that grows monotonically with the damping. Second, below an exact damping threshold, the Kalman gain falls beneath a prescribed tolerance, provided that the tolerance exceeds the gain floor set by the process noise, so that observations are assimilated with a weight below that tolerance; at weakly forced, locally growing modes, the actual error diverges. Third, without injected model-error variance, ensemble spread collapses geometrically at d(k)2m(k)2 per cycle wherever d(k) m(k) < 1, against m(k)2 for a perfect-model ensemble at decaying modes. Phase and scale-transfer errors and inflation are quantified as departures from this baseline, and localization is examined numerically. The closed forms of Theorems 1–4 are confirmed by computer algebra and reproduced in 50-digit arithmetic to better than 10−42, and the full matrix recursions agree with the closed forms to better than 10−10. Within the linear theory, the gain-miscalibration formulae bound the error of a fully emulator-driven cycle from below at decaying modes. A Lorenz-96 experiment shows that nonlinearity carries the degradation to undamped scales, where no such bound holds.

Zitieren

APA

Lohan, N., Goel, P., Kumar, S., Giri, R. K., Hasher, F. F. B., Kaskaoutis, D. G., & Zhran, M. (2026). How Spectral Bias in Machine-Learning Forecast Emulators Propagates Through the Data-Assimilation Cycle: A Linear–Gaussian Spectral Theory. Entropy, 28(11), 1276. https://doi.org/10.3390/e28111276

Vancouver

Lohan N, Goel P, Kumar S, Giri RK, Hasher FFB, Kaskaoutis DG, et al. How Spectral Bias in Machine-Learning Forecast Emulators Propagates Through the Data-Assimilation Cycle: A Linear–Gaussian Spectral Theory. Entropy. 2026;28(11):1276. doi:10.3390/e28111276

BibTeX
@article{Lohan2026,
  author = {Lohan, Nitin and Goel, Pankaj and Kumar, Sushil and Giri, R. K. and Hasher, Fahdah Falah Ben and Kaskaoutis, Dimitris G. and Zhran, Mohamed},
  title = {How Spectral Bias in Machine-Learning Forecast Emulators Propagates Through the Data-Assimilation Cycle: A Linear–Gaussian Spectral Theory},
  journal = {Entropy},
  year = {2026},
  volume = {28},
  number = {11},
  pages = {1276},
  doi = {10.3390/e28111276}
}
Kostenlos starten →