# Evaluating PurpleAir Sensors: Do They Accurately Reflect Ambient Air Temperature?

## TL;DR

PurpleAir sensors overestimate ambient temperature significantly, but with calibration, they can help improve heat mapping and climate resilience efforts in cities.

## Contribution

First evaluation of PurpleAir sensors for ambient temperature accuracy and development of calibration methods using crowdsourced data.

## Key findings

- PurpleAir sensors overestimate temperature with an MAE of 4.71 °C and RMSE of 6.30 °C.
- Calibration methods reduced RMSE and MAE by 51% and 47%, respectively, improving accuracy significantly.
- Sensors show nonlinear behavior with seasonal and diurnal variations, making them better for trends than precise measurements.

## Abstract

What are the main findings?
PurpleAir sensors exhibit strong temperature overestimations with an MAE of 4.71 °C and RMSE of 6.30 °C.Sensor performance demonstrates nonlinear behavior with significant seasonal and diurnal variations.

PurpleAir sensors exhibit strong temperature overestimations with an MAE of 4.71 °C and RMSE of 6.30 °C.

Sensor performance demonstrates nonlinear behavior with significant seasonal and diurnal variations.

What is the implication of the main finding?
Calibrated PurpleAir sensors have the potential to advance hyperlocal heat mapping and multi-hazard vulnerability assessments.

Calibrated PurpleAir sensors have the potential to advance hyperlocal heat mapping and multi-hazard vulnerability assessments.

Low-cost sensors (LCSs) emerge as a popular tool for urban micro-climate studies by offering dense observational coverage. This study evaluates the performance of PurpleAir (PA) sensors for ambient temperature monitoring—a key but underexplored aspect of their use. While widely used for particulate matter, PA sensors’ temperature data remain underutilized and lack thorough validation. For the first time, this research evaluates their accuracy by comparing PA temperature measurements with collocated high-precision temperature data loggers across a dense urban network in a humid subtropical U.S. county. Results show a moderate correlation with reference data (r = 0.86) but an average overestimation of 3.77 °C, indicating PA sensors are better suited for identifying temperature trends but not for precise applications like extreme heat events. We also developed and compared eight calibration methods to create a replicable model using readily available crowdsourced data. The best-performing model reduced RMSE and MAE by 51% and 47%, respectively, and achieved an R2 of 0.89 compared to the uncalibrated scenario. Finally, the practical application of PA temperature data for identifying heat wave events was investigated, including an assessment of associated uncertainties. In sum, this work provides a crucial evaluation of PA’s temperature monitoring capabilities, offering a pathway for improved heat mapping, multi-hazard vulnerability assessments, and public health interventions in the development of climate-resilient cities.