Efficient Uncertainty-Guided Sensor Placement for Air Quality Mapping using Gaussian Processes and Spatial Indexing

Deploying air quality sensors is expensive, and optimal placement is critical for accurate spatial mapping. We propose a greedy algorithm that sequentially selects new sensor locations by maximizing the predictive variance of a Gaussian process (GP) model. To accelerate selection and avoid redundancy placement, we integrate a Euclidean distance filter (via cKDTree) and a priority queue. Using a simulated pollution field with known ground truth, we benchmark against random and uniform grid sampling over 30 independent repeats. Our method achieves a root mean square error (RMSE) of 0.0503 ± 0.0082, significantly lower than random (0.0650 ± paired t-test p = 7 × 10⁻⁵) and uniform (0.0589 ± 0.0127, p = 4 × 10⁻⁴). The effect size (Cohen’s d = 1.03) indicates a large practical improvement. Computation time remains acceptable (∼ 5.7 seconds per greedy run). We also provide uncertainty calibration and sensitivity analyses. The approach offers a principled, reproducible, andefficient solution for designing sparse environmental monitoring networks.