Development and Optimization of the Low-cost Optical Monitor for Real-time Monitoring of Atmospheric Black and Brown Carbon

Abstract

Black carbon (BC) is a major component of aerosols in the ambient air, adversely affecting human health and contributing to regional and global climate change. BC's light absorption, combined with its longer lifetime in the atmosphere, makes it a significant aerosol constituent that affects regional climate.(Bansal et al., 2019). Despite its climate effects, inhalation causes more than 1.8 million deaths per year. (Ezzati & Kammen, 2002). A low-cost BC monitoring system is essential for monitoring the black carbon near the surface and evaluating the mitigation action's effectiveness. To achieve this goal, we designed a low-cost, low-power, low-cost black carbon sensor named Carbon Scan. It is a unique sensor to estimate black carbon (BC) and brown carbon (BrC) concentrations in the atmosphere. Carbon Scan has an air sampler, a color sensor that captures the BC image, and a machine learning (ML) model that retrieves BC concentration from color values. Gradient Boosting Regressor (GBR) and Neural Network (NN)were trained and evaluated using a data set collected by Carbon Scan and a reference instrument AE33 Aethalometer by Magae. About 20% of the total data was used to evaluate the model’s performance without any change. For BC and BrC data, the GBR model outperforms the NN model with evaluation metrics: RMSE of 1.30, with R² of 0.95, making it the best fit for predicting black carbon concentration. The results show that the GBR model exhibits excellent predictive performance. In comparison, the NN model also demonstrates strong performance, with evaluation metrics: RMSE of 2.07, R² of 0.89. However, it performs comparatively less than the GBR model. Similarly, GBR is more efficient at predicting BrC as compared to NN. Carbon Scan can automatically compensate for the temperature. The metrics result shows that the GBR exhibits good predictive performance, benefiting from its capability to understand intricate relationships among variables. This low-cost system is low power, capable of automatic continuous air sampling, and enables air sampling and monitoring in remote areas. This low-cost sensor provides an opportunity to monitor the large temporal and spatial variation of BC and facilitates the policymakers, scientists, and health workers to take appropriate actions accordingly.