Work

Artificial neural network modeling for predicting PM10, PM2.5, NOX, and SO2 in coal mining areas

Year: 2025

Type: article

Abstract: Coal is critical to energy generation, accounting for one-third of global electricity output. However, coal mining emits pollutants into the atmosphere, which harm health. As a result, it is critical to estimate air quality in industrial locations to implement preventative measures. Artificial neural network (ANN) modelling is a popular multilayer network for predicting air pollution. The study assessed air quality in the Singrauli coalmine complex, comprising nine mines, using climatic parameters and emission rates. Concentrations of the pollutants in the region exceed permissible values in several mines, especially during dry periods. The ANN models for PM10, PM2.5, NOX, and SO2 performed best when constructed using the annual dataset that included the entire mine complex, with regression coefficient values of 0.92, 0.91, 0.89, and 0.81, respectively. The SO2 ANN model slightly underpredicts the actual SO2 values, particularly during testing, due to insufficient data diversity, imbalanced training data, and suboptimal model tuning. The sensitivity study found that cloud cover, precipitation, wind direction, wind speed, and solar radiation are crucial for accurate pollutant concentration estimates. The findings emphasize the importance of weather conditions in predicting pollution and highlight the potential of ANN models in predicting environmental risks in industrial zones. • This study aimed to assess air quality across various mining sites using Artificial Neural Networks (ANNs) to predict PM10, PM2.5, NOX, and SO2 levels. The ANN models performed best in winter & pre-monsoon seasons for individual mines, whereas for the mining complex, the annual dataset gives the optimum results (R>0.81). • Solar radiation is the most significant parameter for predicting PM10 and PM2.5 concentration, whereas wind direction has the most significant impact on the NOX and SO2 concentration prediction. • The prediction accuracy of PM10, PM2.5, and NOX is 83%, 79%, and 85%, respectively, indicating robust predictive efficacy. On the contrary, the prediction accuracy of SO2 is around 66%, showing the poor performance of the ANN model. • The study signified the ANN's potential for robust environmental monitoring in mining regions.

Source: Earth Science Informatics

DOI: 10.1007/s12145-025-01922-w

Authors: Akash Mishra, Navin Prasad, Tanushree Bhattacharya, Bindhu Lal

Institution: Birla Institute of Technology, Mesra

Language: English

Topic: Air Quality Monitoring and Forecasting

Subfield: Environmental Engineering

Field: Environmental Science

Domain: Physical Sciences

SDG: Climate action

Open Access status: hybrid

Estimated APC paid: $3,890