Performance assessment and deployment of a low-cost device for urban air quality monitoring in a developing country

Abstract

Air pollution poses a significant threat to public health in rapidly developing countries like Bangladesh, necessitating robust and cost-effective monitoring solutions. This study validates the performance of the CUPI-G device, a low-cost air quality monitoring device, in Dhaka, Bangladesh. The CUPI-G, equipped with electrochemical sensors for PM<inf>2.5</inf>, CO, NO, NO<inf>2</inf>, and O<inf>3</inf>, was deployed across multiple sites representing diverse urban environments, including residential, near-road, and educational areas. Data from the CUPI-G was validated with a collocated reference instrument using statistical (single and multiple linear regression) and machine learning (random forest, RF) approaches. The RF model, particularly when incorporating relative humidity, demonstrated superior performance in predicting pollutant concentrations, with high correlation coefficients (e.g., O<inf>3</inf>: R²=0.798) and low error metrics (RMSE=3.594 ppb, MAPE=4.812 %). However, model accuracy decreased when applied outside the training humidity range, highlighting the need for broader validation datasets. Despite this, the CUPI-G device was validated without using the relative humidity as a factor and was found to still perform adequately. A two-month spatial analysis across three different areas revealed that the hourly average of PM<inf>2.5</inf> and O<inf>3</inf> concentrations peaked in the near roadways as 89 μg/m³ and 66.50 ppb, respectively. NO<inf>2</inf> levels were highest in the residential area at 63.49 ppb. The results demonstrate that the CUPI-G device provides a reliable and cost-effective solution for expanding air quality monitoring networks, offering detailed spatial and temporal data essential for public health advisories and policy interventions, particularly in resource-limited settings.