Accurate air pollution monitoring is critical to understand and mitigate the impacts of air pollution on human health and ecosystems. Due to the limited number and geographical coverage of advanced, highly accurate sensors monitoring air pollutants, many low-cost and low-accuracy sensors have been deployed. Calibrating low-cost sensors is essential to fill the geographical gap in sensor coverage. We systematically examined how different machine learning (ML) models and open-source packages could help improve the accuracy of particulate matter (PM) 2.5 data collected by Purple Air sensors. Eleven ML models and five packages were examined. This systematic study found that both models and packages impacted accuracy, while the random training/testing split ratio (e.g., 80/20 vs. 70/30) had minimal impact (0.745% difference for R²). Long Short-Term Memory (LSTM) models trained in RStudio and TensorFlow excelled, with high R² scores of 0.856 and 0.857 and low Root Mean Squared Errors (RMSEs) of 4.25 µg/m³ and 4.26 µg/m³, respectively. However, LSTM models may be too slow (1.5 h) or computation-intensive for applications with fast response requirements. Tree-boosted models including XGBoost (0.7612, 5.377 µg/m³) in RStudio and Random Forest (RF) (0.7632, 5.366 µg/m³) in TensorFlow offered good performance with shorter training times (<1 min) and may be suitable for such applications. These findings suggest that AI/ML models, particularly LSTM models, can effectively calibrate low-cost sensors to produce precise, localized air quality data. This research is among the most comprehensive studies on AI/ML for air pollutant calibration. We also discussed limitations, applicability to other sensors, and the explanations for good model performances. This research can be adapted to enhance air quality monitoring for public health risk assessments, support broader environmental health initiatives, and inform policy decisions.

Accurate air pollution monitoring is critical to understand and mitigate the impacts of air pollu-tion to human health and ecosystems. Due to the limited number and geographical coverage of advanced high accurate sensors to monitor air pollutants, many low-cost but low accuracy sensors have been deployed. Calibrating the low-cost sensors is essential to leverage them for filling this geographical gap in sensor coverage. We examine systematically how different Machine Learning (ML) models and open-source packages could help improve the accuracy of particulate matter (PM) 2.5 data collected by Purple Air sensors. Eleven ML models and five packages are examined. This systematic study found that both models and packages impact the accuracy while the choice of random training/testing split ration (e.g., 80/20 vs 70/30) has minimal impact (0.745% difference for R2). Long Short-Term Memory (LSTM) models trained in RStudio and TensorFlow excelled, with high R2 scores of 0.856 and 0.857 and low Root Mean Squared Errors (RMSEs) of 4.25 µg/m3 and 4.26 µg/m3, respectively. However, LSTM may be too slow or computation intensive (1.5 hours) in applications with fast response requirements. Tree-boosted models including XGBoost (0.7612, 5.377 µg/m3) in RStudio and Random Forest (RF) (0.7632, 5.366 µg/m3) in TensorFlow, offer good performance at shorter training times (< 1 minute), and may be suitable for such ap-plications. These findings suggest that AI/ML models, particularly LSTM, can effectively calibrate low-cost sensors to produce precise, localized air quality data. This research is the most com-prehensive in comparison to existing literature on AI/ML for air pollutants calibration. We also discussed the limitation, applicability to other sensors and the reason for well performed models. The research can be adapted to enhance air quality monitoring for public health risk assessments and support broader environmental health initiatives and inform policy decisions.

The world evolves continuously across space and time. Massive volumes of data are generated through sensing, simulation, remote observation, and human activities, capturing dynamic processes in environmental, social, economic, and engineered systems. Critical insights are embedded within these large-scale spatiotemporal datasets. Spatiotemporal Data Science provides a conceptual and methodological framework for analyzing such data by integrating spatiotemporal thinking, computational infrastructure, artificial intelligence, and domain knowledge. The field advances methods for data acquisition, harmonization, modeling, visualization, and decision support, enabling applications in natural disaster response, public health, climate adaptation, infrastructure resilience, and geopolitical analysis. By leveraging emerging technologies—including generative Artificial Intelligence (AI), large-scale cloud platforms, Graphics Processing Unit (GPU) acceleration, and digital twin systems—Spatiotemporal Data Science enables scalable, interoperable, and solution-oriented research and innovation. It represents a critical frontier for scientific discovery, engineering advancement, technological innovation, education, and societal benefit. Spatiotemporal Data Science is a transdisciplinary field that studies and models dynamic phenomena across space and time by integrating spatial theory, temporal reasoning, artificial intelligence, and scalable computational infrastructure. It enables the development of adaptive, predictive, and increasingly autonomous systems for understanding and managing complex real-world processes.

In recent years, air quality has decreased substantially because of increasing levels of harmful pollutants from various industrial practices, vehicle emissions, and wildfires. Particulate matter with diameters less than 2.5 µm or PM2.5 is considered a leading air pollutant, causing over 8 million deaths annually worldwide. Hence, accurate monitoring and prediction of air pollution, specifically PM2.5, is essential for public health protection; however, current air quality monitoring methods face limitations. Currently, air quality data is collected from ground-level monitoring stations, which utilize Environmental Protection Agency (EPA) sensors. While these sensors provide accurate data, their high cost prevents them from widespread use. On the other hand, low-cost air quality sensors can be distributed across different areas and fill geographic gaps in sensor coverage, but they are not very accurate. This trade-off between cost and accuracy presents a challenge in effectively monitoring widespread air quality. To address this challenge, this study presents a new approach to improving the accuracy of air quality measurements of low-cost sensors by calibrating them using various machine learning (ML) and deep learning (DL) techniques, allowing them to attain the high accuracy of EPA sensors while still remaining inexpensive. ML/DL models were trained on spatiotemporal data, including PM2.5 levels, temperature, and humidity from the low-cost sensors and PM2.5 levels from the EPA sensors. Then, using these models, the temperature, humidity, and PM2.5 levels from the low-cost sensors were used to predict their corresponding highly accurate PM2.5 measurements from EPA sensors with RMSEs as low as 5.53. Better accuracy in low-cost sensors can lead to more extensive and reliable air quality monitoring networks. This approach can be applied globally, enabling better response strategies to air pollution and its associated health risks.

Low-cost air quality sensors can expand PM₂.₅ monitoring networks but require calibration against regulatory-grade monitors to correct systematic bias. Although interest in predictive uncertainty for air quality estimation has increased, uncertainty reliability under spatial sparsity and wildfire-induced distribution shift remains poorly understood. This study develops a transformer-based PM₂.₅ calibration model and evaluates two uncertainty quantification methods, GeoConformal Prediction (GCP) and Monte Carlo Dropout (MCD), using sensor pairs in California and the Northeast United States. The calibration model achieved strong performance at short spatial distances, with test R² values of 0.89 and 0.91 at the 1 km threshold in California and the Northeast, respectively, with accuracy declining as sensor separation increased. GCP generally produced calibration curves closer to the ideal diagonal, while MCD generated tighter prediction intervals under normal conditions. During wildfire events, uncertainty performance depended on sensor separation. At short distances, MCD expanded its uncertainty intervals and captured PM₂.₅ spikes more effectively than GCP (71% vs. 61% coverage). At larger separations, MCD captured only 44% of elevated observations, whereas GCP widened its intervals and achieved 83% coverage. These results demonstrate that uncertainty reliability is strongly influenced by spatial separation and environmental conditions, highlighting the need for uncertainty-aware calibration of low-cost PM₂.₅ sensors.

Low-cost air quality sensors can expand PM_2.5 monitoring networks but require calibration against regulatory-grade monitors to correct systematic bias towards trustworthy observations. Although interest in predictive uncertainty for air quality estimation has increased, the reliability of uncertainty under spatial sparsity and wildfire-induced distribution shift remains poorly understood. With the increasing use of low-cost sensors during wildfire-smoke episodes, uncertainty quantification during calibration is essential to determine when corrected PM_2.5 estimates are reliable enough to support near-real-time alert, public-health, and exposure-related decisions. This study presents a spatial uncertainty evaluation framework for trustworthy AI-enabled environmental sensing. The framework is implemented with a transformer-based PM_2.5 calibration model and two uncertainty quantification methods, GeoConformal Prediction (GCP) and Monte Carlo Dropout (MCD), using sensor pairs in California and the Northeast United States. The calibration model achieved strong performance at short spatial distances, with test R² values of 0.89 and 0.91 at the 1 km threshold in California and the Northeast, respectively, with accuracy declining as sensor separation increased. GCP generally produced calibration curves closer to the ideal diagonal, while MCD generated tighter prediction intervals under normal conditions. During wildfire events, the uncertainty performance depended on sensor separation. At short distances, MCD expanded its uncertainty intervals and captured PM_2.5 spikes more effectively than GCP (71% vs. 61% coverage). At larger separations, MCD captured only 44% of elevated observations, whereas GCP widened its intervals and achieved 83% coverage. These results demonstrate that uncertainty reliability is strongly influenced by spatial separation and environmental conditions, highlighting the need for uncertainty-aware and spatially adaptive calibration of low-cost PM_2.5 sensors for trustworthy actionable air quality observation.