Investigating indoor CO2 heterogeneity: The impact of sensor location on CO2-based IAQ classification and presence detection in Brazilian educational buildings

Indoor carbon dioxide (CO2) distributions are spatially non-uniform, yet most monitoring and occupancy sensing applications rely on single-point measurements. This study examined the impact of sensor placement on CO2-based Indoor Air Quality (IAQ) assessments and data-driven presence detection in two Brazilian educational buildings. Multi-point CO2 measurements were combined with Computational Fluid Dynamics (CFD) simulations to characterize spatial CO2 distributions. Stacked ensemble and Long Short-Term Memory (LSTM) presence detection models were trained repeatedly for each sensor location and feature combination and statistically compared using Welch's ANOVA and Games-Howell tests.In the purely naturally ventilated open-plan office, low CO2 levels prevailed at all positions, and sensor placement had a negligible practical effect on IAQ assessment. In rooms with natural ventilation combined with air conditioning, persistent vertical CO2 gradients made sensor height the dominant factor. This produced differences of up to 18 percentage points between heights. For presence detection, sensor location had a statistically significant effect in all rooms. Top-mounted sensors outperformed lower-mounted ones in the room with purely natural ventilation. The most important feature was the vertical CO2 gradient, and dual-sensor configurations improved model performance, consistent with prior observations in a German school. However, these advantages did not generalize to the air-conditioned rooms. Active recirculation disrupted vertical stratification, reversed the height advantage, and diminished the value of gradient features.These findings demonstrate that the practical relevance of sensor placement depends on whether the ventilation configuration produces spatial CO2 heterogeneity and that placement guidance must be adapted accordingly.