Spatiotemporal Approaches for Quality Control and Error Correction of Atmospheric Data through Machine Learning.

Area of Science:

• Atmospheric Science
• Data Science
• Machine Learning

Background:

• Accurate atmospheric data is crucial for weather forecasting and climate monitoring.
• Internet of Things (IoT) sensors provide vast amounts of atmospheric data, but data quality can be a challenge.
• Existing quality control (QC) methods may not fully leverage the richness of available atmospheric data.

Purpose of the Study:

• To propose and evaluate three novel machine learning-based quality control (QC) techniques for atmospheric data.
• To assess the performance of these QC methods using various data inputs and machine learning algorithms.
• To identify the most effective QC approach for improving the accuracy of atmospheric data.

Main Methods:

• Developed three machine learning (ML) QC techniques: single time series, combined time series, and spatiotemporal.
• Applied ML algorithms, including support vector regression, to atmospheric data (e.g., temperature) from seven IoT sensor types.
• Evaluated QC performance using the root mean squared error (RMSE) metric.

Main Results:

• QC using combined weather elements showed 0.14% lower average RMSE compared to single-element QC.
• QC incorporating spatiotemporal characteristics with AWS data achieved 17% lower RMSE than using raw data alone.
• Machine learning effectively identified and corrected errors in atmospheric sensor data.

Conclusions:

• Machine learning-based QC techniques offer significant improvements in atmospheric data accuracy.
• Integrating multiple weather elements and spatiotemporal information enhances QC performance.
• The proposed methods provide a robust framework for ensuring the reliability of IoT-derived atmospheric data.