Zhang et al. (2026) Integrating Machine Learning with Adaptive Kalman Filtering to Downscale GFS Air Temperature Forecasts in Mountainous Areas

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Remote Sensing
• Year: 2026
• Date: 2026-06-03
• Authors: Guixin Zhang, Jingpeng Liang, Shanyou Zhu, Yan Xu
• DOI: 10.3390/rs18111829

Research Groups

Not specified

Short Summary

The study proposes a two-stage "DOWN + BC" framework using Random Forest and an Adaptive Kalman Filter to downscale and bias-correct GFS air temperature forecasts for high-resolution applications in mountainous terrain.

Objective

• To improve the spatial resolution and forecast accuracy of Global Forecast System (GFS) air temperature data in areas with complex terrain.

Study Configuration

• Spatial Scale: Downscaling from 0.25° (coarse) to 30 m (high resolution).
• Temporal Scale: 3-hourly forecast intervals; evaluation conducted for January 2020 and July 2023.

Methodology and Data

• Models used: Random Forest (RF) for geographical downscaling (DOWN), first-order adaptive Kalman filter (AKF) for bias correction (BC), and Extreme Gradient Boosting (XGB) for generating reference temperature fields.
• Data sources: Global Forecast System (GFS) forecasts and Automatic Weather Station (AWS) observations.

Main Results

• The DOWN stage improves the spatial detail of temperature distribution but offers limited improvement in absolute accuracy.
• The combined DOWN + BC framework significantly reduces the Root Mean Square Error (RMSE) compared to raw GFS forecasts at AWS locations: by 37.84% in January 2020 and 41.16% in July 2023.
• Compared to XGB-derived temperature distributions, the RMSE was reduced by 47.27% (January 2020) and 33.79% (July 2023).

Contributions

• Introduces a hybrid processing pipeline that combines machine learning-based downscaling with adaptive filtering to provide high-resolution, bias-corrected temperature forecasts specifically tailored for complex mountainous environments.

Funding

Not specified

Citation

@article{Zhang2026Integrating,
  author = {Zhang, Guixin and Liang, Jingpeng and Zhu, Shanyou and Xu, Yan},
  title = {Integrating Machine Learning with Adaptive Kalman Filtering to Downscale GFS Air Temperature Forecasts in Mountainous Areas},
  journal = {Remote Sensing},
  year = {2026},
  doi = {10.3390/rs18111829},
  url = {https://doi.org/10.3390/rs18111829}
}