Machine Learning-Assisted, Process-Based Quality Control for Detecting Compromised Environmental Sensors.

Area of Science:

• Environmental Science
• Data Science
• Sensor Technology

Background:

• High-quality environmental data is crucial for research and management.
• Manual quality assurance and control (QAQC) of sensor data is labor-intensive and hinders scalability.
• Existing automated QAQC methods struggle with the complex noise profiles of environmental sensors.

Purpose of the Study:

• To develop a machine learning-assisted QAQC methodology robust to low signal-to-noise ratio data.
• To enable automated detection of compromised environmental sensors.
• To increase the volume of high-quality data from sensor networks.

Main Methods:

• Embedding sensor measurements into a dynamical feature space.
• Training a binary classification algorithm (Support Vector Machine) to identify deviations from expected dynamics.
• Applying the methodology to diverse environmental sensor datasets (stream level, pH, electroconductivity).

Main Results:

• The ML-assisted QAQC methodology achieved accuracy scores up to 0.97.
• The approach effectively detects a wide range of nonphysical signals from compromised sensors.
• Performance consistently surpassed state-of-the-art anomaly detection techniques.

Conclusions:

• The proposed ML-assisted QAQC methodology offers a scalable and robust solution for environmental sensor data.
• This approach significantly enhances the reliability and volume of environmental monitoring data.
• It addresses a key challenge in leveraging sensor networks for environmental research and management.