Related Experiment Video
Updated: Jan 4, 2026
09:47
Author Spotlight: Advancing Alzheimer's Research – Exploring Early Detection and Multi-Omics Approaches
Published on: December 15, 2023
1.6K
An artificial neural network ensemble approach to generate air pollution maps
S Van Roode1, J J Ruiz-Aguilar2, J González-Enrique3
1Intelligent Modelling of Systems, Department of Computer Science Engineering, University of Cádiz, Polytechnic School of Engineering, 11202, Algeciras, Spain. steffanie.vanroode@gm.uca.es.
Environmental Monitoring and Assessment
|November 9, 2019
Summary
An artificial neural network (ANN) ensemble effectively estimates hourly nitrogen dioxide (NO2) concentrations at unmonitored locations. This advanced model outperforms traditional spatial interpolation and regression methods for air quality mapping.
Area of Science:
- Environmental Science
- Data Science
- Atmospheric Chemistry
Background:
- Industrialized and high-traffic areas like the Bay of Algeciras face significant air pollution challenges.
- Accurate estimation of nitrogen dioxide (NO2) concentrations is crucial for public health and environmental monitoring.
- Existing spatial interpolation and regression models have limitations in precisely estimating pollutant levels at unsampled locations.
Purpose of the Study:
- To develop and evaluate an artificial neural network (ANN) ensemble model for estimating hourly NO2 concentrations.
- To compare the performance of the ANN ensemble against Inverse Distance Weight (IDW) and Least Absolute Shrinkage and Selection Operator (LASSO) models.
- To generate accurate air pollution maps and provide reliable NO2 concentration estimates for unmonitored areas.
Main Methods:
- An ensemble approach using an ANN with backpropagation learning was developed.
- Spatial interpolation (IDW) and regularized linear regression (LASSO) were employed to generate initial pollutant concentration maps.
- Cross-validation and leave-one-out strategies were used to compare model performance using metrics like R, MSE, MAE, and the d index.
Main Results:
- The ANN ensemble model demonstrated superior performance compared to individual IDW and LASSO models.
- The ANN ensemble achieved an average R correlation coefficient of 0.77 and a minimum MSE of 54.05.
- IDW and LASSO models showed average R values of 0.72 and 0.76, respectively, indicating the ensemble's enhanced predictive capability.
Conclusions:
- The proposed ANN ensemble is a robust and effective tool for estimating hourly NO2 concentrations at unsampled locations.
- This approach significantly improves air quality mapping accuracy in complex urban and industrial environments.
- The system offers potential applications in data imputation and identifying errors within air quality monitoring networks.
