An Artificial Neural Network (ANN) was trained to recognize significant weather events, generally defined as having a threshold of a parameter exceeding +/-2 standard deviations from normal using a database of climatological standardized anomalies from the North American Global Reanalysis data. The ANN learns by associating real, targeted information with the desired output. During the learning process some of its neurons are excited while others are inhibited, similar to how learning occurs in the human brain. By adjusting the weights of each neuron, via repetitive error correction techniques, the predicted output eventually comes closer to the targeted output. Once the ANN has been trained to an acceptable level of error it can be used with input from real-time data to generalize and pattern match for significant weather events. A description of the ANN and how it was trained with standardized anomalies of various significant weather events will be presented. Its application to real-time forecast challenges, using case studies of heat and cold events, heavy precipitation, and severe weather will also be presented to show its usefulness for operational forecasting.
Supplementary URL: http://nws.met.psu.edu/NN