RESUMEN
Low-cost sensors can provide inaccurate data as temperature and humidity affect sensor accuracy. Therefore, calibration and data correction are essential to obtain reliable measurements. This article presents a training and testing method used to calibrate a sensor module assembled from SO2 and NO2 electrochemical sensors (Alphasense B4 and B43F) alongside air temperature (T) and humidity (RH) sensors. Field training and testing were conducted in the industrialized coastal area of Quintero Bay, Chile. The raw responses of the electrochemical (mV) and T-RH sensors were subjected to multiple linear regression (MLR) using three data segments, based on either voltage (SO2 sensor) or temperature (NO2). The resulting MLR equations were used to estimate the reference concentration. In the field test, calibration improved the performance of the sensors after adding T and RH in a linear model. The most robust models for NO2 were associated with data collected at T < 10 °C (R2 = 0.85), while SO2 robust models (R2 = 0.97) were associated with data segments containing higher voltages. Overall, this training and testing method reduced the bias due to T and HR in the evaluated sensors and could be replicated in similar environments to correct raw data from low-cost electrochemical sensors. A calibration method based on training and sensor testing after relocation is presented. The results show that the SO2 sensor performed better when modeled for different segments of voltage data, and the NO2 sensor model performed better when calibrated for different temperature data segments.
