The Colorado Water Watch’s real-time monitoring data that is transmitted from the Colorado Water Watch monitoring stations to the Colorado Water Watch database is continuously analyzed by a sophisticated contamination detection software called CANARY. CANARY was developed by the USEPA and is designed to look for any anomalies or events in groundwater quality based on historical data patterns. This contamination detection software uses three complex statistical and mathematical algorithms to scan the raw data and to identify any changes in groundwater water quality. The three event detection algorithms used by CANARY include linear prediction coefficient filter (LPCF), multivariat e near-neighbor (MVNN), and set-point proximate algorithm (SPPE).
The LPCF algorithm identifies a pattern in the monitoring data under normal conditions and predicts the future data in accordance to the historical pattern. Any deviation from the predicted pattern triggers an anomaly alert. The probability of an event occurrence is based on how variant the registered anomaly is from the predicted pattern.
The MVNN algorithm identifies a pattern between the various parameters in the monitoring dataset under normal conditions and compares it to the latest monitoring dataset. Any change in the normal pattern within the different monitoring parameters triggers an anomaly alert. The probability of an event occurrence is based on how variant the registered anomaly is from the historical pattern.
The SPPE algorithm is coded to scan the monitoring data and identified if the data set registers within the given minimum or maximum set point limit. Any deviation from the specified range triggers an anomaly alert. The probability of an event occurrence is based on how variant the registered anomaly is from the specified range.
The results of these three algorithms are compiled and published in terms of maximum probability of a contamination event (CMAX). This CMAX trend displays the probabilities of the monitoring data, stating the extent to which it could be anomalous. The closer the probability is to zero, there is no or minimal chance of a contamination, meaning that groundwater quality has not changed. The closer the probability is to one, there is a higher chance of a contamination.
For more details about the anomaly detection system, please go to here