Two low-cost (LC) monitoring networks, PurpleAir (instrumented by Plantower PMS5003 sensors) and AirQino (Novasense SDS011), were assessed in monitoring PM_2.5 and PM₁₀ daily concentrations in the Padana Plain (Northern Italy). A total of 19 LC stations for PM_2.5 and 20 for PM₁₀ concentrations were compared vs. regulatory-grade stations during a full “heating season” (15 October 2022–15 April 2023). Both LC sensor networks showed higher accuracy in fitting the magnitude of PM₁₀ than PM_2.5 reference observations, while lower accuracy was shown in terms of RMSE, MAE and R². AirQino stations under-estimated both PM_2.5 and PM₁₀ reference concentrations (MB = −4.8 and −2.9 μg/m³, respectively), while PurpleAir stations over-estimated PM_2.5 concentrations (MB = +5.4 μg/m³) and slightly under-estimated PM₁₀ concentrations (MB = −0.4 μg/m³). PurpleAir stations were finer than AirQino at capturing the time variation of both PM_2.5 and PM₁₀ daily concentrations (R² = 0.68–0.75 vs. 0.59–0.61). LC sensors from both monitoring networks failed to capture the magnitude and dynamics of the PM_2.5/PM₁₀ ratio, confirming their well-known issues in correctly discriminating the size of individual particles. These findings suggest the need for further efforts in the implementation of mass conversion algorithms within LC units to improve the tuning of PM_2.5 vs. PM₁₀ outputs.

A pre-deployment calibration and a field validation of two low-cost (LC) stations equipped with O₃ and NO₂ metal oxide sensors were addressed. Pre-deployment calibration was performed after developing and implementing a comprehensive calibration framework including several supervised learning models, such as univariate linear and non-linear algorithms, and multiple linear and non-linear algorithms. Univariate linear models included linear and robust regression, while univariate non-linear models included a support vector machine, random forest, and gradient boosting. Multiple models consisted of both parametric and non-parametric algorithms. Internal temperature, relative humidity, and gaseous interference compounds proved to be the most suitable predictors for multiple models, as they helped effectively mitigate the impact of environmental conditions and pollutant cross-sensitivity on sensor accuracy. A feature analysis, implementing dominance analysis, feature permutations, and the SHapley Additive exPlanations method, was also performed to provide further insight into the role played by each individual predictor and its impact on sensor performances. This study demonstrated that while multiple random forest (MRF) returned a higher accuracy than multiple linear regression (MLR), it did not accurately represent physical models beyond the pre-deployment calibration dataset, so a linear approach may overall be a more suitable solution. Furthermore, as well as being less computationally demanding and generally more suitable for non-experts, parametric models such as MLR have a defined equation that also includes a few parameters, which allows easy adjustments for possible changes over time. Thus, drift correction or periodic automatable recalibration operations can be easily scheduled, which is particularly relevant for NO₂ and O₃ metal oxide sensors. As demonstrated in this study, they performed well with the same linear model form but required unique parameter values due to intersensor variability.

The strong Antarctic vortex plays a crucial role in forming an expansive region with significant stratospheric ozone depletion during austral spring, commonly referred to as the Antarctic “ozone hole”. This study examines daily ozone column behavior during this phenomenon using ERA5 reanalysis data and ground-based observations from 10 Antarctic stations collected between September and December from 2008 to 2022. A preliminary analysis of these datasets revealed smoothly varying patterns with quasi-uniform gradients in the ozone distribution within the ozone hole. This observation led to the hypothesis that average ozone columns over zones, defined as concentric areas around the South Pole, can be estimated using mean values of the measurements derived from station observations. This study aims to evaluate the validity of this hypothesis. The results indicate that the mean ozone levels calculated from daily measurements at two stations—Belgrano and Dome Concordia, or Belgrano and Arrival Heights—provide a reliable approximation of the average ozone levels over the zone spanning 70°S to 90°S. Including additional stations extended the zone of reliable approximation northward to 58°S. The approximation error was estimated to range from 5% to 7% at 1σ and from 6% to 8% at the 10th–90th percentile levels. Furthermore, the geographical distribution of the stations enabled a schematic reconstruction of the ozone hole’s position and shape. On the other hand, the high frequency of ground-based measurements contributed to studying the ozone hole variability in both the inner area and edges on an hourly time scale. These findings have practical implications for the near-real-time monitoring of ozone hole development, along with satellite observations, considering ground-based measurements as a source of information about ozone layer in the South Pole region. The results also suggest the possible role of observations from the ground in the analyses of pre-satellite-era hole behavior. Additionally, this study found a high degree of consistency between ground-based measurements and corresponding ERA5 reanalysis data, further supporting the reliability of the observations.

Gold kiwifruits from two different farms, harvested at different times, were analysed using both non-destructive and destructive methods. A computer vision system (CVS) and a portable spectroradiometer were used to perform non-destructive measurements of firmness, titratable acidity, pH, soluble solids content, dry matter, and soluble sugars (glucose and fructose), with the goal of building predictive models for the maturity index. Hyperspectral data from the visible–near-infrared (VIS–NIR) and short-wave infrared (SWIR) ranges, collected via the spectroradiometer, along with colour features extracted by the CVS, were used as predictors. Three different regression methods—Partial Least Squares (PLS), Support Vector Regression (SVR), and Gaussian process regression (GPR)—were tested to assess their predictive accuracy. The results revealed a significant increase in sugar content across the different harvesting times in the season. Regardless of the regression method used, the CVS was not able to distinguish among the different harvests, since no significant skin colour changes were measured. Instead, hyperspectral measurements from the near-infrared (NIR) region and the initial part of the SWIR region proved useful in predicting soluble solids content, glucose, and fructose. The models built using these spectral regions achieved R² average values between 0.55 and 0.60. Among the different regression models, the GPR-based model showed the best performance in predicting kiwifruit soluble solids content, glucose, and fructose. In conclusion, for the first time, the effectiveness of a fully portable spectroradiometer measuring surface reflectance until the full SWIR range for the rapid, contactless, and non-destructive estimation of the maturity index of kiwifruits was reported. The versatility of the portable spectroradiometer may allow for field applications that accurately identify the most suitable moment to carry out the harvesting.