In this study, a method that integrates convolutional neural networks (CNNs) with ensemble numerical weather prediction (NWP) models is proposed. This method enables surface temperature forecasting with lead times beyond the short-range, extending up to five days. Due to limited computational resources, operational medium-range temperature forecasts typically rely on low-resolution NWP models, which are prone to systematic and random errors. To resolve these limitations, the proposed method applies CNN-based post-processing (bias correction and spatial super-resolution) to an ensemble NWP system. First, the post-processing is applied to each ensemble member to reduce systematic errors and reconstruct high-resolution temperature fields from low-resolution model outputs. This approach reduces the systematic and random errors in NWP model outputs and outperforms operational post-processing. Second, the CNN is applied to all ensemble members to construct a new ensemble forecasting system, in which deterministic forecast accuracy, probabilistic reliability, and representation of ensemble spread are improved compared with those of the original system. We demonstrate that this CNN-based post-processing is fundamentally different from the artificial error reduction caused by smoothing inherent in ensemble averaging because the post-processing reduces forecast errors without degrading the forecast information. These results indicate that the proposed method provides a practical and scalable solution for improving medium-range temperature forecasts and is particularly valuable for use in operational centers with limited computational resources.

翻译：本研究提出了一种将卷积神经网络（CNN）与集合数值天气预报（NWP）模型相结合的方法。该方法能够实现短时预报以外、最长至五天的地表温度预报。由于计算资源有限，业务化的中期温度预报通常依赖于低分辨率的NWP模型，这些模型容易产生系统性误差和随机误差。为解决这些局限，所提方法对集合NWP系统应用了基于CNN的后处理（偏差订正与空间超分辨率）。首先，对每个集合成员进行后处理，以减少系统性误差，并从低分辨率模式输出中重建高分辨率温度场。该方法降低了NWP模式输出的系统性和随机性误差，其表现优于业务化后处理方法。其次，将CNN应用于所有集合成员，构建了一个新的集合预报系统，该系统在确定性预报精度、概率可靠性以及集合离散度表征方面均优于原始系统。我们证明，这种基于CNN的后处理与集合平均中固有的平滑效应所导致的人为误差减少有本质区别，因为后处理在减少预报误差的同时并未损害预报信息。这些结果表明，所提方法为改进中期温度预报提供了一种实用且可扩展的解决方案，对于计算资源有限的气象业务中心尤其具有应用价值。