An international team of researchers has made significant strides in enhancing the accuracy of weather forecasts by leveraging advanced machine learning techniques.
Despite the sophistication of ensemble numerical weather prediction (NWP) methods, weather forecasts often face challenges due to under-dispersion, leading to less reliable predictions. To address this, calibration tools like quantile regression (QR) have gained popularity for their flexibility and predictive performance. However, a longstanding issue with QR is quantile crossing, which can limit the interpretability of forecasts.
In their recent study, the researchers introduced a novel non-crossing quantile regression neural network (NCQRNN). This innovative model enhances the traditional QR neural network by adding a new layer that maintains the natural order of forecast values. This ensures that lower quantiles remain smaller than higher ones, thereby improving both accuracy and interpretability without sacrificing computational efficiency.
“Our NCQRNN model preserves the rank order of output nodes, ensuring more reliable and interpretable forecasts,” explained Professor Yang Dazhi from the Harbin Institute of Technology. “This advancement has the potential to significantly improve the precision of weather predictions.”
The study, published in the journal Advances in Atmospheric Sciences, is a collaborative effort among scientists from institutions across Asia and Europe, including the Chinese Academy of Sciences, Karlsruhe Institute of Technology in Germany, the National University of Singapore, and the Budapest University of Technology and Economics in Hungary.
The researchers highlight that this machine learning approach is adaptable and can be seamlessly integrated into various weather forecasting systems. “This non-crossing layer can be added to a wide range of neural network structures, ensuring the broad applicability of the technique,” noted Dr. Martin J. Mayer from the Budapest University of Technology and Economics.
Dr. Sebastian Lerch from the Karlsruhe Institute of Technology added, “The proposed neural network model is very general and can be applied to other target variables with minimal adaptations. It holds promise for other weather and climate applications beyond solar irradiance forecasting.”
Experts believe that incorporating advanced machine learning methods into numerical weather prediction models marks a significant step forward in meteorology. “This study serves as an instructive example of how machine learning can enhance the accuracy of weather forecasts and climate predictions,” said Professor Xia Xiang’ao from the Institute of Atmospheric Physics of the Chinese Academy of Sciences.
The advancements not only benefit meteorologists and climate scientists but also have broader implications for industries and communities worldwide that rely on precise weather information. As weather patterns become increasingly unpredictable due to climate change, such innovations are crucial for preparedness and risk management.
Reference(s):
Scientists enhance weather forecasts reliability with machine learning
cgtn.com
