Can Low-Cost Sensor Networks Improve Analysis of Extreme Pollution Episodes in Delhi During Stubble Burning?

Abstract Extreme post-monsoon PM₂.₅ episodes across the Indo-Gangetic Plain are often underestimated because fire-emission uncertainties coincide with sparse reference-grade monitoring in crop-residue-burning source regions. We assess whether a low-cost sensor network deployed across source regions can improve atmospheric composition forecasts during November 2024. Twice-daily PM₂.₅ observations from 33 CUPI-G sensors across rural source areas and downwind transport pathways were assimilated into WRF-Chem using 3D-Var. DA-EXP1, using a realistic 10% observation error, reduced mean fractional bias over Haryana from − 116.7% to − 56.8% and correctly predicted the observed Severe AQI category in approximately 95% of Delhi cases and 100% of Punjab and Haryana cases; CNTL predicted no Severe cases at these locations. DA-EXP2, using a 1.5 µg m⁻³ observation error representative of reference-grade monitoring stations, provided further improvements, highlighting the importance of observation-error specification. These results demonstrate the operational value of assimilating source-region low-cost observations into AIRWISE for extreme-pollution forecasting and early warning.