Progress in Machine Learning-Based Temperature Forecast Correction

Accurate temperature forecasting is essential for responding to extreme weather events, ensuring public safety, and optimizing agriculture and energy management. The machine learning-based numerical weather prediction error correction has significantly improved the quality of temperature forecast. To explore the application effect of machine learning in temperature bias correction in recent years, this paper systematically reviews the research progress of machine learning-based temperature forecast bias correction. It also sorts out the latest application achievements of machine learning methods, such as decision tree, random forest, artificial neural network, support vector machine, clustering algorithm, and deep learning algorithms, including deep neural network, convolutional neural network, recurrent neural network, and graph neural network. The research results indicate that machine learning algorithms can effectively correct temperature forecast biases in numerical models by uncovering complex patterns and relationships in historical meteorological data. Ensemble learning methods, which combine the strengths of various machine learning models and neural networks, significantly enhance both the accuracy and stability of forecasts. Additionally, emerging technologies such as Transformer and graph neural network have shown great promise, particularly in handling complex spatiotemporal data and capturing spatial dependencies. Looking ahead, the efficient integration of multi-source data, the improved model interpretability and uncertainty analysis as well as the development of real-time learning and adaptive models may offer critical technological solutions to the further enhancement of the temperature forecasting accuracy.