Scale-aware and spatially-localised optimisation of kilometre-scale MLWP models
Abstract
Preserving fine-scale structure without degrading spatial accuracy is a central challenge for kilometre-scale machine-learning weather prediction (MLWP). We introduce a loss based on the dual-tree complex wavelet transform that constrains atmospheric fields jointly by scale, location, and orientation. We evaluate its performance using an experimental joint global--regional FastNet configuration, which produces \SI{1.5}{km} forecasts over the UK alongside ∼\SI{31}{km} global forecasts. Relative to both mean-squared error (MSE) and an FFT-based spectral objective, the wavelet loss substantially improves spectral fidelity. Compared with MSE training, it maintains similar pointwise performance, improves neighbourhood-based verification, and produces sharper precipitation fields with more realistic intensity distributions. These results demonstrate that spatially localised wavelet constraints can preserve multiscale atmospheric structure while largely avoiding the loss of pointwise accuracy associated with globally supported spectral objectives.
Related
- Design and trial implementation of a continental-scale, kilometre-resolution hourly precipitation analysis for Australia using satellite, radar and gauges
- A multi‐scale loss formulation for learning a probabilistic model with proper score optimisation
- Scale-sensitive river water quality indicators reveal spatially heterogeneous drivers across China
- Scale-aware emergent constraints on Tibetan Plateau warming: contrasting global and regional predictors
- SwAIther-Precip: Lead-Time-Aware Bias Correction Enables Kilometer-Scale Downscaling of Global AI Precipitation Forecasts over Switzerland
BibTeX
@misc{chappell2026scaleaware,
title = {{Scale-aware and spatially-localised optimisation of kilometre-scale MLWP models}},
author = {James Chappell and Sophie Luise Arana and Jack Bowyer and Matthew Burnand and Alejandro Coca-Castro and Evangeline Corcoran and Eric G. Daub and Tom Dunstan and Noushin Eftekhari and Anna-Louise Ellis and J. Scott Hosking and Manvendra Janmaijaya and David Salvador-Jasin and Sam Madge and Ben Shipway and Nathan Simpson and Oliver Strickson and David Walters},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.37325},
eprint = {2610.37325},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.37325},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Scale-aware and spatially-localised optimisation of kilometre-scale MLWP models
AU - James Chappell
AU - Sophie Luise Arana
AU - Jack Bowyer
AU - Matthew Burnand
AU - Alejandro Coca-Castro
AU - Evangeline Corcoran
AU - Eric G. Daub
AU - Tom Dunstan
AU - Noushin Eftekhari
AU - Anna-Louise Ellis
AU - J. Scott Hosking
AU - Manvendra Janmaijaya
AU - David Salvador-Jasin
AU - Sam Madge
AU - Ben Shipway
AU - Nathan Simpson
AU - Oliver Strickson
AU - David Walters
PY - 2026
DA - 2026/10/29
PB - arXiv
DO - 10.48550/arxiv.2610.37325
UR - https://arxiv.org/abs/2610.37325
AB - Preserving fine-scale structure without degrading spatial accuracy is a central challenge for kilometre-scale machine-learning weather prediction (MLWP). We introduce a loss based on the dual-tree complex wavelet transform that constrains atmospheric fields jointly by scale, location, and orientation. We evaluate its performance using an experimental joint global--regional FastNet configuration, which produces \SI{1.5}{km} forecasts over the UK alongside ∼\SI{31}{km} global forecasts. Relative to both mean-squared error (MSE) and an FFT-based spectral objective, the wavelet loss substantially improves spectral fidelity. Compared with MSE training, it maintains similar pointwise performance, improves neighbourhood-based verification, and produces sharper precipitation fields with more realistic intensity distributions. These results demonstrate that spatially localised wavelet constraints can preserve multiscale atmospheric structure while largely avoiding the loss of pointwise accuracy associated with globally supported spectral objectives.
ER -