Forecasting the future levels of air pollution provides valuable information that holds importance for the general public, vulnerable populations, and policymakers. High-quality data are essential for precise and reliable forecasts and investigations of air pollution. Missing observations arise when the sensors utilized for assessing air quality parameters experience malfunctions, which result in erroneous measurements or gaps in the dataset and hinder the data quality. This research paper presents a novel approach for imputing missing values in air quality data in a univariate approach. The algorithm employs the random forest (RF) algorithm to impute missing observations in a bi-directional (forward and reverse in time) manner for air quality (particulate matter less than 2.5 μm (PM_2.5)) data from the Republic of Serbia. The algorithm was evaluated against simple methods, such as the mean and median imputation methods, for missing observations over durations of 24, 48, and 72 h. The results indicate that our algorithm yielded comparable error rates to the median imputation method for all periods when imputing the PM_2.5 data. Ultimately, the algorithm’s higher computational complexity proved itself as not justified considering the minimal error decrease it achieved compared with the simpler methods. However, for future improvement, additional research is needed, such as utilizing low-code machine learning libraries and time-series forecasting techniques.

The process of manually checking, validating, and excluding data in an ionospheric very-low-frequency (VLF) analysis during extreme events is a labor-intensive and time-consuming task. However, this task can be automated through the utilization of machine learning (ML) classification techniques. This research paper employed the Random Forest (RF) classification algorithm to automatically classify the impact of solar flares on ionospheric VLF data and erroneous data points, such as instrumentation errors and noisy data. The data used for analysis were collected during September and October 2011, encompassing solar flare classes ranging from C2.5 to X2.1. The F1-score values obtained from the test dataset displayed values of 0.848; meanwhile, a more detailed analysis revealed that, due to the imbalanced distribution of the target class, the per-class F1-score indicated higher values for the normal data point class (0.69–0.97) compared to those of the anomalous data point class (0.31 to 0.71). Instances of successful and inadequate categorization were analyzed and presented visually. This research investigated the potential application of ML techniques in the automated identification and classification of erroneous VLF amplitude data points; however, the findings of this research hold promise for the detection of short-term ionospheric responses to, e.g., gamma ray bursts (GRBs), or in the analysis of pre-earthquake ionospheric anomalies.

Following solar flares (SF), the abrupt increase in X-radiation and EUV emission generates additional ionization and higher absorption of, e.g., electromagnetic waves in the sunlit hemisphere of the Earth’s ionosphere. The modeling of the ionosphere under solar flares are motivated by new observations with spacecrafts, satellites, and ground-based measurements. The estimation of modeling parameters for the ionospheric D-region during SF events poses a significant challenge, typically requiring a trial-and-error approach. This research presents a machine learning (ML) methodology for modeling the sharpness (β) and reflection height (H′) during SF events occurred from 2008 to 2017. The research methodology was divided into two separate approaches: an instance-based approach, which involved obtaining SF parameters during the peak SF, and a time-series approach, which involved analyzing time-series data during SFs. The findings of the study revealed that the model for the instance-based approach exhibited mean absolute percentage error (MAPE) values of 9.1% for the β parameter and 2.45% for the H′ parameter. The findings from the time-series approach indicated that the model exhibited lower error rates compared to the instance-based approach. However, it was observed that the model demonstrated an increase in β residuals as the predicted β increased, whereas the opposite trend was observed for the H′ parameter. The main goal of the research is to develop an easy-to-use method that provides ionospheric parameters utilizing ML, which can be refined with additional and novel data as well as other techniques for data pre-processing and other algorithms. The proposed method and the utilized workflow and datasets are available at GitHub.

This study presents the first comprehensive spatial accessibility assessment of essential urban services in Belgrade, Republic of Serbia, conducted entirely with open-source tools and data. The analysis focused on six facility categories: primary healthcare centers, public pharmacies, primary and secondary schools, libraries, and green markets. Spatial accessibility was modelled using OpenRouteService (ORS) isochrones for walking travel times of 5, 10, and 15 min, combined with population data from the Global Human Settlement Layer (GHSL). Results indicate that 79% of residents live within a 15-min walk of a healthcare facility, 74% of a pharmacy, 89% of an elementary school, 52% of a high school, 60% of a library, and 62% of a green market. Central administrative units such as Vračar, Zvezdara, and Stari Grad demonstrated nearly complete service coverage, while peripheral areas, including Resnik, Jajinci, and Višnjica, exhibited substantial accessibility deficits, often coinciding with lower-income zones. The developed workflow provides a transparent, replicable approach for identifying underserved neighborhoods and prioritizing investments in public infrastructure. This research emphasizes the role of spatial accessibility analysis in advancing Sustainable Development Goals (SDGs), contributing to the creation of more inclusive, walkable, and sustainable urban environments, while on the other hand, it offers practical insights for improving urban equity, guiding policy formulation, and supporting necessary planning decisions. Subsequent research will focus on alternative facilities, other cities such as Novi Sad and Niš, and the disparity between urban and rural populations.

We demonstrate the use of Facebook's Prophet (usually just called Prophet) model for short-term air quality forecasting at Belgrade-Zeleno brdo monitoring station. To address missing data, we applied minimally-altering data distribution imputation techniques. Linear interpolation proved effective for short-term gaps (1–3 hours), hourly mean method for mid-term gaps (24–26 hours), and Hermite interpolation polynomial for long-term gaps (132–148 hours). The most significant data change was a 3.4% shift in skewness. Partitioning the time series enabled a detailed quality assessment of the Prophet model, with PM_2.5 predictions being more precise than PM₁₀. Using the longest time series for forecasting yielded absolute errors of 6.5 μg/m³ for PM₁₀ and 2.7 μg/m³ for PM_2.5. Based on 173 forecasts, we anticipate Prophet model root-mean-square values under 6.26 μg/m³ and 9.99 μg/m³ for PM_2.5 and PM₁₀ in 50% of cases. The Prophet model demonstrates several advantages and yields satisfactory results. In future research, the results obtained from the Prophet model will serve as benchmark values for other models. Additionally, the Prophet model is capable of providing satisfactory air quality forecasting results and will be utilized in future research.