LightRCAN: A Lightweight Super-Resolution Method for ERA5 Temperature Fields with Cross-Pressure-Level Transferability

Temperature data are fundamental to meteorological and climatological research. High-Resolution (HR) fields are essential for regional climate, model validation, and extreme weather assessment. This study proposes a Lightweight Residual Channel Attention Network (LightRCAN) for spatial super-resolution (SR) reconstruction of ERA5 temperature fields, which is the fifth-generation global atmospheric reanalysis produced by the European Centre for Medium-Range Weather Forecasts (ECMWF). We reduce the resolution 0.25° ERA5 to 0.5°, create HR and Low-Resolution (LR) data pairs via bicubic interpolation, and train the network to reconstruct HR spatial structures from LR inputs. Results show LightRCAN accurately recovers spatial details. Compared with bicubic and bilinear interpolation, LightRCAN achieves lower reconstruction errors. Compared with the Residual Channel Attention Network (RCAN), LightRCAN reduces the number of parameters by 85% and floating-point operations (FLOPs) by 83%, while incurring only a 0.16 dB drop in Peak Signal-to-Noise Ratio (PSNR), thereby achieving a favorable accuracy–efficiency trade-off. We also evaluate cross-pressure-level transferability by freezing the backbone and fine-tuning the upsampling module and the reconstruction convolutional layer (3.6% of parameters). This strategy achieves the best balance between reconstruction performance and parameter efficiency, outperforming direct transfer. LightRCAN offers a computationally efficient solution for reconstruction.

Authors

Institutions

Publication Details

Journal
Applied Sciences
Published
2026-11-19
DOI
https://doi.org/10.3390/app162211462
Primary Topic
Advanced Image Processing Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

LightRCAN: A Lightweight Super-Resolution Method for ERA5 Temperature Fields with Cross-Pressure-Level Transferability

Chanseng Wong, Hoiio Kong, Ping Yu, Zijun Li
Applied Sciences
Advanced Image Processing Techniques
article

LightRCAN: A Lightweight Super-Resolution Method for ERA5 Temperature Fields with Cross-Pressure-Level Transferability

Chanseng Wong, Hoiio Kong, Ping Yu, Zijun Li
article en

Abstract

Temperature data are fundamental to meteorological and climatological research. High-Resolution (HR) fields are essential for regional climate, model validation, and extreme weather assessment. This study proposes a Lightweight Residual Channel Attention Network (LightRCAN) for spatial super-resolution (SR) reconstruction of ERA5 temperature fields, which is the fifth-generation global atmospheric reanalysis produced by the European Centre for Medium-Range Weather Forecasts (ECMWF). We reduce the resolution 0.25° ERA5 to 0.5°, create HR and Low-Resolution (LR) data pairs via bicubic interpolation, and train the network to reconstruct HR spatial structures from LR inputs. Results show LightRCAN accurately recovers spatial details. Compared with bicubic and bilinear interpolation, LightRCAN achieves lower reconstruction errors. Compared with the Residual Channel Attention Network (RCAN), LightRCAN reduces the number of parameters by 85% and floating-point operations (FLOPs) by 83%, while incurring only a 0.16 dB drop in Peak Signal-to-Noise Ratio (PSNR), thereby achieving a favorable accuracy–efficiency trade-off. We also evaluate cross-pressure-level transferability by freezing the backbone and fine-tuning the upsampling module and the reconstruction convolutional layer (3.6% of parameters). This strategy achieves the best balance between reconstruction performance and parameter efficiency, outperforming direct transfer. LightRCAN offers a computationally efficient solution for reconstruction.

Applied SciencesVol. 16(22)
City University of Macau (MO)
Advanced Image Processing Techniques
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