Subseasonal Forecasts of Opportunity Identified by an Interpretable Neural Network
Subseasonal Forecasts of Opportunity Identified by an Interpretable Neural Network
复制标题
可解释的神经网络识别的次季节机会预测
DOI：
10.1002/essoar.10505448.1
复制
发表时间：
2020
影响因子：
6.8
通讯作者：
E. Barnes
中科院分区：
文献类型：
--
作者：
Kirsten J. Mayer;E. Barnes
由于大气的混沌性质，在亚季节时间尺度上进行中纬度预报是困难的，而且往往需要确定有利的大气条件，从而可能提高技术（“机会预报”）。在这里，我们证明了人工神经网络可以利用网络对给定预测的置信度提前22天识别北大西洋（40N, 325E）内热带-温带环流遥相关的机会。此外，分层相关传播是一种人工神经网络的可解释性技术，它可以精确定位人工神经网络用于做出准确预测的相关热带特征。我们发现，分层相关传播识别出与已知中纬度遥相关有利区域相对应的热带热点，并揭示了北大西洋亚季节时间尺度上预测的潜在新模式。
Midlatitude prediction on subseasonal timescales is difficult due to the chaotic nature of the atmosphere and often requires the identification of favorable atmospheric conditions that may lead to enhanced skill (“forecasts of opportunity”). Here, we demonstrate that an artificial neural network can identify such opportunities for tropical-extratropical circulation teleconnections within the North Atlantic (40N, 325E) at a lead of 22 days using the network’s confidence in a given prediction. Furthermore, layer-wise relevance propagation, an ANN explainability technique, pinpoints the relevant tropical features the ANN uses to make accurate predictions. We find that layer-wise relevance propagation identifies tropical hot spots that correspond to known favorable regions for midlatitude teleconnections and reveals a potential new pattern for prediction in the North Atlantic on subseasonal timescales.