Integrating winter’s method and white noise for advanced air quality forecasting: a case study of Indore, India

Abstract

Air pollution poses one of the most critical challenges for urban centers globally that necessitate forecasting models to mitigate adverse impacts and inform policy decisions. This study focuses on air quality forecasting for Indore, India, employing Winter’s Method integrated with White Noise to capture both deterministic seasonal patterns and stochastic variability in air pollution data. The analysis highlights substantial seasonal and annual fluctuations in key pollutants which include PM10, PM2.5, NO, NO2, NOx, CO, SO2, NH3, and O3 using observed data from 2019 to 2023 and forecasting through 2030. The study uses a Python based algorithm to manage deterministic patterns along with stochastic fluctuations which help in providing accurate future predictions. The optimization and sensitivity analysis of the smoothing parameters level (α), trend (β), and seasonality (γ) determined the ideal combination as α=0.3, β=0.2, and γ=0.6. The model's accuracy was confirmed via error analysis, with the root mean square error for the optimal parameters determined to be 13.5. The innovative model effectively predicts long-term trends that project the average AQI to peak at 125 by 2030. Forecasted PM10 and PM2.5 levels are expected to decrease to 34 μg/m3 and 37 μg/m3, respectively, by 2030 which reflect the potential benefits of stricter emission regulations. NOx and NH3 levels are anticipated to drop significantly to 34 μg/m3 and 18 μg/m3, respectively. CO and SO2 levels are forecasted to increase to 1.5 μg/m3 and 19 μg/m3, respectively, by 2030 that emphasize the need for additional mitigation strategies. O3 concentrations are predicted to decline moderately to 19 μg/m3 by 2030. Incorporating deterministic and stochastic components in the forecasting models improves the predictive capabilities of the models and presents a flexible and applicable approach for urban air quality management.