Quality Control in Weather Monitoring with Dynamic Linear Models

TL;DR

This paper introduces an automated Bayesian-based quality control method for weather sensors, improving data reliability for agricultural decision-making by accurately detecting sensor errors using time-series models.

Contribution

It presents a novel approach combining dynamic linear models and data fusion to assess sensor data quality with uncertainty quantification, enhancing sensor validation.

Findings

Error hit rates above 80%

False negative rates below 11%

Effective on temperature, wind, and humidity data

Abstract

Decisions in agriculture are frequently based on weather. With an increase in the availability and affordability of off-the-shelf weather stations, farmers able to acquire localised weather information. However, with uncertainty in the sensor and installation quality, farmers are at risk of making poor decisions based on incorrect data. We present an automated approach to perform quality control on weather sensors. Our approach uses time-series modelling and data fusion with Bayesian principles to provide predictions with uncertainty quantification. These predictions and uncertainty are used to estimate the validity of a sensor observation. We test on temperature, wind, and humidity data and achieve error hit rates above 80% and false negative rates below 11%.