Self-supervised, physics-constrained super-resolution for wind mapping in the absence of high-resolution training references: a case study in Brazil

Abstract Planning renewable-energy expansion at the national scale requires wind maps far more detailed than sparse meteorological networks and coarse reanalysis grids can typically offer. This study introduces PC-SRCNN, a physics-constrained, self-supervised super-resolution framework that lifts ERA5 wind fields over Brazil from a native 0.25° grid to 0.025°, folding in soft regularization terms for horizontal kinematic consistency and the logarithmic vertical wind profile. The model was evaluated against observations from 190 meteorological stations from Brazil’s INMET network across all twelve months of 2024, benchmarked against IDW, Kriging, U-Net, SRCNN, and a variant constrained only by horizontal kinematics (PC-SRCNN𝜑). The full PC-SRCNN was the only model to outperform native ERA5 with statistical significance in every month, cutting MAE by 5.6–11.0%, reaching vertical-profile R² between 0.80 and 0.87, and posting the lowest error at 133 of the 190 stations (70.0%); the PC-SRCNN𝜑 variant, in contrast, performed significantly worse than ERA5, and the purely data-driven models failed to consistently beat it. These findings show that physical regularization aids wind-field super-resolution only when it respects the vertical structure of the atmospheric boundary layer, suggesting the potential of this framework as a computationally lightweight approach for super-resolution wind mapping in settings where conventional numerical downscaling is computationally demanding.

Authors

Institutions

Publication Details

Journal
GeoInformatica
Published
2026-11-17
DOI
https://doi.org/10.1007/s10707-026-00601-2
Primary Topic
Meteorological Phenomena and Simulations
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Self-supervised, physics-constrained super-resolution for wind mapping in the absence of high-resolution training references: a case study in Brazil

Lihki Rubio, Carlos Eduardo Velasquez, José Péricles Freire, Jin Yang
3 links
GeoInformatica
Meteorological Phenomena and Simulations
article

Self-supervised, physics-constrained super-resolution for wind mapping in the absence of high-resolution training references: a case study in Brazil

Lihki Rubio, Carlos Eduardo Velasquez, José Péricles Freire, Jin Yang
article en

Abstract

Abstract Planning renewable-energy expansion at the national scale requires wind maps far more detailed than sparse meteorological networks and coarse reanalysis grids can typically offer. This study introduces PC-SRCNN, a physics-constrained, self-supervised super-resolution framework that lifts ERA5 wind fields over Brazil from a native 0.25° grid to 0.025°, folding in soft regularization terms for horizontal kinematic consistency and the logarithmic vertical wind profile. The model was evaluated against observations from 190 meteorological stations from Brazil’s INMET network across all twelve months of 2024, benchmarked against IDW, Kriging, U-Net, SRCNN, and a variant constrained only by horizontal kinematics (PC-SRCNN𝜑). The full PC-SRCNN was the only model to outperform native ERA5 with statistical significance in every month, cutting MAE by 5.6–11.0%, reaching vertical-profile R² between 0.80 and 0.87, and posting the lowest error at 133 of the 190 stations (70.0%); the PC-SRCNN𝜑 variant, in contrast, performed significantly worse than ERA5, and the purely data-driven models failed to consistently beat it. These findings show that physical regularization aids wind-field super-resolution only when it respects the vertical structure of the atmospheric boundary layer, suggesting the potential of this framework as a computationally lightweight approach for super-resolution wind mapping in settings where conventional numerical downscaling is computationally demanding.

GeoInformaticaVol. 30(2)
Universidade Federal de Minas Gerais (BR), Universidad del Norte (CO), University of Glasgow (GB)
Universidade Federal de Minas Gerais
Climate action
Meteorological Phenomena and Simulations
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon