EOF和人工神经网络方法相结合预测中国华北地区向外长波辐射通量

Combining EOF and artificial neural network to predict the outgoing longwave radiation flux in the Northern China

利用搭载在Aqua卫星上的CERES产品,探讨了2005年6月-2009年6月华北地区向外长波辐射的时空变化特征.从经验正交分解结果来看,第一模态华北地区向外长波辐射变化具有很好的一致性,整个区域向外长波辐射有增大趋势.第二模态中呈现出南北反位相的空间分布特征,尤其在2005年6月、2006年6月、2009年6月表现得尤为显著.采用后向传播神经网络方法,利用华北地区历史时段向外长波辐射场经验正交分解展开的前6个主分量作为训练样本,预测北京地区2010年6月向外长波辐射日变化特征.研究表明:经验正交分解和神经网络相结合的方法对预测向外长波辐射会取得较好的效果.尤其当训练样本中加入北京地区云量后,预测效果提高1/3.

The temporal and spatial variability of outgoing longwave radiation in the Northern China in June of 2005-2009 based on the empirical orthogonal function was analyzed. The results showed that the trend of outgoing longwave radiation in the Northern China during the historical period was consistent as a whole and increased in the most areas of the region. In the second mode, the outgoing longwave radia- tion showed a decreasing trend in the southern part and an increasing trend in the northern part, especially in June of 2005, 2006 and 2009. The first six principal components of the empirical orthogonal function from the outgoing longwave radiation anomaly were longwave radiation in June 2010 in Beijing based on used as training samples to predict the outgoing the artificial neural network. It indicated that the combination of the empirical orthogonal function and neural network can predict the spatial distribution of outgoing longwave radiation well. Especially, when the cloud factor in Beijing was added to the fore- casting factors, the forecasting precision further increased by one third.