Sep 2026· Artificial Intelligence for the Earth Systems· Vol 5· 0 citations
TL;DR
This work shows that uneven accuracy is strongly associated with local overfitting: optimal model complexity and training schedule diverge across time-, space-, and spatiotemporal cross-validation splits, so that no single hyperparameter setting is optimal for all.
Abstract
Atmospheric and environmental monitoring relies on observation networks with highly uneven spatial density, yet machine-learning models trained on such networks prioritize data-dense regions, producing systematic uneven accuracy precisely where reliable predictions are most needed. Existing fairness methods from machine learning address such gaps but require explicit group definitions and domain expertise. We show that uneven accuracy is strongly associated with local overfitting: optimal model complexity and training schedule diverge across time-, space-, and spatiotemporal cross-validation splits, so that no single hyperparameter setting is optimal for all. Using a range of environmental datasets, AI models tend to memorize dense regions while others generalize better but lose local structure. These dynamics motivate our Overfit-to-Generalization framework, which integrates into existing pipelines. Across nine model configurations, OG improves
R
2
by 10.8–29.7% over baselines, largest in data-sparse regions. A complementary diagnostic predicts model accuracy at unsampled locations from local density and spatial proximity.
Ground-level air quality affects human health, well-being, and social outcomes. Gold-standard monitoring relies on calibrated, ground-based sensors that can be costly to install and maintain, limiting spatial coverage especially in resource-constrained regions that face higher pollution burdens. Recent advances in sa...
Amanda Murray, Alexander S. Kolker, Kory Kirshenbaum et al.· ACS ES&T Air· 0 citations
Accurate pressure is required in high-precision geodetic applications, including global navigation satellite systems (GNSS) and very long baseline interferometry (VLBI). Widely used empirical models provide pressure estimates from location and time alone, but their predefined formulations leave residual spatiotemporal...
Li-Ying Cao, Jun-Yu Li, Bao-Shuang Zhang et al.· IEEE Journal of Selected Top...· 0 citations
Frozen Earth-observation embeddings are judged almost entirely by spatially blocked cross-validation inside one study region; it is shown that this number does not predict accuracy in a new region; it is shown why; and the one setting in which such a model does keep working is shown.
Philip Stark, A. Sopasakis, Ola Hall· arXiv.org· 0 citations
Diffusion models have shown strong potential for kilometer-scale precipitation downscaling, but their performance in geographically unseen regions and event regimes remains insufficiently understood. Building on the wavelet diffusion model (WDM) framework, this study evaluates cross-region and cross-event generalizatio...
Wei-Kang Qian, Yi-Xin Wen, Chu-Gang Yi et al.· arXiv.org· 0 citations
In recent years, data-driven approaches emerged as alternatives to traditional physics-based retrievals, taking advantage of machine learning techniques such as learnable pseudoinverse, random forests, or deep learning architectures. Classical data-driven models generalize poorly to out-of-sample regimes, as they optim...
Michele Martinazzo, C. Sgattoni, Marco Menarini et al.· Remote Sensing of Clouds and...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.