[A-TT35] Machine Learning Techniques in Weather, Climate, Ocean, Hydrology and Disease Predictions

[ATT35-P07] Bias correction for numerical weather prediction with artificial neural networks using fine-scale preserving loss

*Viktor Artemovich Golikov^1,2, Mikhail Krinitskiy^2,3, Alexander Gavrikov³, Vladimir Vanovskiy^1,2 (1.Skolkovo Institute of Science and Technology, 2.Moscow Institute of Physics and Technology, 3.Shirshov Institute of Oceanology, Russian Academy of Sciences)

Keywords:statistical bias correction, numerical weather prediction, mesoscale dynamics, artificial neural networks, Arctic ocean, Barents sea

Numerical weather prediction (NWP) models in the Arctic often suffer from systematic biases due to imperfect initial conditions, parameterization schemes, and resolution limitations. To address this, we propose a statistical bias correction framework that combines ERA5 reanalysis data, meteorological station observations, and satellite scatterometer measurements. As a vivid example case, we applied our approach in the region of the Kara and Barents Seas. A hybrid deep learning architecture, incorporating a convolutional U-net with a Transformer in the latent space, is trained to correct WRF model outputs. The proposed method introduces an additional term in the loss function that minimizes the difference between local deviations in the original high-resolution forecasts and their corrected counterparts, ensuring the preservation of small-scale dynamics. Results show a significant reduction in root-mean-square error (RMSE) while maintaining fine-scale atmospheric structures. Furthermore, the integration of diverse observational data during training improves forecast quality across all data sources, demonstrating the framework’s robustness and applicability in the Arctic region.
The framework of statistical correction with artificial neural networks exploiting the proposed loss terms and their variations may be further applied in various regions, where climatic biases are typically observed, e.g., other regions of the Arctic, large metropolitan areas, South-Eastern Tropical Atlantic, Subtropical Eastern North Pacific (SENP), soil moisture-limited regions, etc.