Better sensor networks improve city temperature maps
Adding low-cost sensors helps capture the full range of urban temperatures.
Reference weather station networks often miss the densely built core and the cool, vegetated periphery. These networks sample a narrow portion of a city's spatial heterogeneity. This can cause city-wide predictions to underestimate temperature differences.
Researchers used Zurich, Switzerland, as a case study. They compared multiple linear regression, generalized additive models, and random forests. They added up to 117 low-cost temperature sensors to the four reference weather stations.
Land use regression models relate intra-urban air temperature to spatial predictors. Spatial predictors include building volume, tree canopy cover, and topography. These models are used to produce fine-scale temperature maps.
When trained on reference stations alone, all three models achieve comparable cross-validated performance. As low-cost sensors broaden the sampled spatial heterogeneity, multiple linear regression scores lower on both metrics. Non-linear models remain at their reference-only level.
At full coverage, only multiple linear regression leaves significant residual spatial autocorrelation. This suggests a linear form is unable to capture non-linear responses of air temperature to urban form. A third of the low-cost sensors already recovers nearly the full predicted temperature range.
Key facts
- Reference weather station networks miss the densely built core and cool, vegetated periphery.
- Adding up to 117 low-cost sensors to four reference stations was tested in Zurich, Switzerland.
- Multiple linear regression may fail to capture non-linear responses of air temperature to urban form.