Time Series Matching Using Weather Stations to Predict Power Outages

Associating the impact of severe weather on infrastructure (i.e. number of blocked roads by trees) and business (i.e. revenue lost due to power outages) is an important task for government agencies, insurance companies and critical infrastructure network operators. As a complement to existing analog methods in the weather forecasting community that relate large-scale circulation patterns to local weather conditions, in this session we show how a time series of regularly recorded weather station observations can be leveraged, segmented and joined with recorded power outage impacts in an associated radius of influence to create an early warning system. In addition, we demonstrate how adding supplementary weather stations where observations are sparse can enhance the early warning system by providing more localized information. This methodology may be especially useful where gridded historical weather data is unavailable.