IoT and Satellite Sensor Data Integration for Assessment of Environmental Variables: A Case Study on NO₂

Area of Science:

• Environmental Science
• Data Science
• Remote Sensing

Background:

• Accurate spatiotemporal monitoring of environmental variables is crucial.
• Integrating Internet of Things (IoT) sensor data with satellite imagery presents challenges.
• Existing methods may lack the desired resolution for detailed environmental analysis.

Purpose of the Study:

• To develop a novel machine learning approach for enhancing the spatiotemporal resolution of environmental variables.
• To construct high-resolution, satellite-like images from IoT sensor measurements.
• To assess the performance of different machine learning algorithms for this task.

Main Methods:

• Utilizing machine learning algorithms (1-nearest neighbor, linear regression, feed-forward neural network) to predict environmental variables.
• Gridding the study area and partitioning it into Voronoi cells based on IoT sensor locations.
• Developing separate regression models for pixels within each cell, incorporating data from central and neighboring IoT sensors.
• Applying the approach to NO2 data from Sentinel-5 Precursor satellite and IoT ground sensors.

Main Results:

• The ensemble of regression models successfully generated satellite-like images at high spatiotemporal resolution.
• Feed-forward neural networks demonstrated the highest accuracy, achieving a Root Mean Square Error (RMSE) of 15.49 ×10-6 mol/m2 for NO2 prediction.
• The approach effectively integrated data from both satellite and ground-based IoT sensors.

Conclusions:

• The proposed machine learning framework significantly improves the spatiotemporal resolution of environmental variable monitoring.
• Feed-forward neural networks are a promising tool for accurate environmental data fusion and prediction.
• This method offers a powerful approach for detailed environmental assessment using combined satellite and IoT data.