基于人工神经网络的暴雨预报方法探讨

A EXPLORATION OF HEAVY RAIN FORECASTING TECHNIQUE BASED ON ARTIFICIAL NEURAL NETWORKS

探讨了基于人工神经网络模型的暴雨预报方法。该方法仿预报员的暴雨预报思路,在动力模式的降水预报产品、环流形势场和暴雨落区之间通过人工神经网络建立非线性的统计预报模型,该模型的输入是动力模式的降水预报和初始环流形势场的扩展正交分解主成份分量,输出是预报区域的暴雨落区预报。2000年的汛期试验表明该客观预报方法可明显改进数值预报模式的暴雨落区预报,因此可望在业务预报中有较好的应用前景。

A heavy rain forecasting technique is developed based on artificial neural networks. Imitating weatherman thought, the nonlinear statistical forecasting model is established between forecasting precipitation of dynamical numerical weather model, atmospheric circulation situation and heavy rain area. The model inputs are precipitation production of dynamical numerical model and principal components of extended empirical orthogonal functions(EEOF).And the model outputs are forecasting precipitation classes for heavy rain area. Operational tests during the rainy season in 2000 show that area forecasts on heavy rain can improve obviously by this statistical model, so it may have an excellent foreground in future operational forecasting on heavy rain.