Unravelling a black box: an open-source methodology for the field calibration of small air quality sensors
<p>The last 2 decades have seen substantial technological advances in the development of low-cost air pollution instruments using small sensors. While their use continues to spread across the field of atmospheric chemistry, the air quality monitoring community, and for commercial and private u...
Guardado en:
| Autores principales: | , , , , , , , , |
|---|---|
| Formato: | article |
| Lenguaje: | EN |
| Publicado: |
Copernicus Publications
2021
|
| Materias: | |
| Acceso en línea: | https://doaj.org/article/c674128a208b467aae11c974e8f3ae39 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Sumario: | <p>The last 2 decades have seen substantial technological advances
in the development of low-cost air pollution instruments using small
sensors. While their use continues to spread across the field of atmospheric chemistry, the air quality monitoring community, and for commercial and private use, challenges remain in ensuring data quality and
comparability of calibration methods. This study introduces a seven-step
methodology for the field calibration of low-cost sensor systems using
reference instrumentation with user-friendly guidelines, open-access code,
and a discussion of common barriers to such an approach. The methodology has been developed and is applicable for gas-phase pollutants, such as for the measurement of nitrogen dioxide (NO<span class="inline-formula"><sub>2</sub></span>) or ozone (O<span class="inline-formula"><sub>3</sub></span>). A full
example of the application of this methodology to a case study in an urban
environment using both multiple linear regression (MLR) and the random
forest (RF) machine-learning technique is presented with relevant <span class="inline-formula"><i>R</i></span> code
provided, including error estimation. In this case, we have applied it to
the calibration of metal oxide gas-phase sensors (MOSs). Results reiterate
previous findings that MLR and RF are similarly accurate, though with
differing limitations. The methodology presented here goes a step further
than most studies by including explicit transparent steps for addressing
model selection, validation, and tuning, as well as addressing the common
issues of autocorrelation and multicollinearity. We also highlight the need
for standardized reporting of methods for data cleaning and flagging, model
selection and tuning, and model metrics. In the absence of a standardized
methodology for the calibration of low-cost sensor systems, we suggest a
number of best practices for future studies using low-cost sensor systems to ensure greater comparability of research.</p> |
|---|