Environmental event discovery through time series anomalies

Abstract

Our world has encountered numerous environmental events throughout its history, and many people have been interested in knowing more about them when they happen. Each event can be said to either be caused, or cause, some abnormal meteorological behaviour. A wildfire, for example, will always contribute to an increase of temperature due to the fires involved. A flooding will most likely be caused by a large increase in rain that happened in the vicinity of the area. Most events have a primary cause or effect that can be detected if looked for. Our system utilises environmental time series data in order to detect anomalies, which are then assumed to correlate to a specific event. It processes a dataset for the primary datapoint within it (temperature, precipitation, etc.), and then executes a discord discovery algorithm over it in order to find any anomalies. Once this is done, the discords are then linked to a news article by utilising the time span of the anomaly, and some additional details surrounding the event, such as location and type. A group of articles is requested using a news source’s respective API, of which the most relevant ones would revolve around the event that occurred. Although the current system is a prototype, it can be easily extended. One could implement APIs for further news sources, or attach the dataset to the system itself, so it can dynamically find the data, requiring even less input from the user. Observing the results obtained on temperature for wildfires, and precipitation for floodings, the system detected such events with a very high accuracy, confirming that strong anomalies in time series data heavily suggests its respective event.