Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Agile Climate-Sensor Design and Calibration Algorithms Using Machine Learning: Experiments From Cape Point

Travis Barrett, Amit Kumar Mishra

TL;DR

This work presents an inexpensive, agile climate-sensor platform built around an ESP32 that co-locates with a high-accuracy reference instrument at Cape Point. It evaluates three regression-based calibration approaches—Random Forest Regression, Artificial Neural Network, and Support Vector Machine Regression—for aligning CO2 readings from low-cost sensors to the reference data. Random Forest Regression emerged as the best performer, achieving a MAE of about $0.14$ ppm and improved distribution alignment compared with raw data, suggesting potential reductions in manual calibration frequency for large-scale sensor networks. The findings indicate that ML-driven calibration can enhance the reliability of cost-effective sensor networks, though further testing in environments with greater variability is recommended to validate robustness and transferability.

Abstract

In this paper, we describe the design of an inexpensive and agile climate sensor system which can be repurposed easily to measure various pollutants. We also propose the use of machine learning regression methods to calibrate CO2 data from this cost-effective sensing platform to a reference sensor at the South African Weather Service's Cape Point measurement facility. We show the performance of these methods and found that Random Forest Regression was the best in this scenario. This shows that these machine learning methods can be used to improve the performance of cost-effective sensor platforms and possibly extend the time between manual calibration of sensor networks.

Agile Climate-Sensor Design and Calibration Algorithms Using Machine Learning: Experiments From Cape Point

TL;DR

This work presents an inexpensive, agile climate-sensor platform built around an ESP32 that co-locates with a high-accuracy reference instrument at Cape Point. It evaluates three regression-based calibration approaches—Random Forest Regression, Artificial Neural Network, and Support Vector Machine Regression—for aligning CO2 readings from low-cost sensors to the reference data. Random Forest Regression emerged as the best performer, achieving a MAE of about ppm and improved distribution alignment compared with raw data, suggesting potential reductions in manual calibration frequency for large-scale sensor networks. The findings indicate that ML-driven calibration can enhance the reliability of cost-effective sensor networks, though further testing in environments with greater variability is recommended to validate robustness and transferability.

Abstract

In this paper, we describe the design of an inexpensive and agile climate sensor system which can be repurposed easily to measure various pollutants. We also propose the use of machine learning regression methods to calibrate CO2 data from this cost-effective sensing platform to a reference sensor at the South African Weather Service's Cape Point measurement facility. We show the performance of these methods and found that Random Forest Regression was the best in this scenario. This shows that these machine learning methods can be used to improve the performance of cost-effective sensor platforms and possibly extend the time between manual calibration of sensor networks.

Paper Structure

This paper contains 16 sections, 3 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Monitoring System and Co-location
  3. System Design and Information
  4. Co-Location Site and Equipment
  5. Methodology
  6. Description of the Data
  7. Random Forest Regression
  8. Artificial Neural Network
  9. Support Vector Machine Regression
  10. Results
  11. Raw Data
  12. Random Forest Regression
  13. Artificial Neural Network
  14. Support Vector Machine Regression
  15. Discussion
  16. ...and 1 more sections

Figures (5)

  • Figure 1: Figure showing random forest regression testing metrics per the number of estimators present in the model. The training shows heavily diminished returns above 15 estimators.
  • Figure 2: Figure showing Cape Point raw test data from our cost-effective platform.
  • Figure 3: Figure showing RFR prediction of unseen Cape Point raw test data.
  • Figure 4: Figure showing ANN prediction of unseen Cape Point raw test data.
  • Figure 5: Figure showing SVR prediction of unseen Cape Point raw test data.