Conducting accurate air quality measurements is of critical importance for sustaining environmental and public health; however, gaps due to various reasons in respective datasets often undermine the reliability of subsequent processes.This study, therefore, aims at presenting a novel hybrid methodology that leverages the Optuna framework to optimize the hyperparameters of the Extreme Gradient Boosting (XGBoost) model for imputing missing data within one of the most significant indicators of air quality, namely PM_2.5 data. The proposed approach was systematically evaluated under varying data loss scenarios, using synthetic datasets generated under the Missing Completely at Random (MCAR) mechanism with missing rates of 5%, 10%, 20%, and 30%. Traditional interpolation methods (such as linear and spline) and widely adopted machine learning techniques (i.e., random forest, multivariate adaptive regression splines) were also utilized to not only benchmarking but also ensuring a comparative environment. In this sense, three experimental configurations were examined: (1) imputation based solely on the PM_2.5 time series, (2) integration of ERA5 reanalysis covariates and (3) inclusion of data from neighboring monitoring stations. The results indicate that the XGBoost-Optuna model outperformed its counterparts across all missing data scenarios, with R² values of 0.852, 0.874, 0.862, and 0.866 for missing rates of 5%, 10%, 20%, and 30%, respectively. These findings highlight the potential of the XGBoost-Optuna model as a robust tool for handling missing air quality data, ensuring enhanced accuracy across varying data gaps and scenarios.

How to cite: Denizoğlu, M., Sezen, İ., Deniz, A., and Ünal, A.: A Novel Hybrid Approach for Missing PM2.5 Data Imputation Using Optuna-Optimized Extreme Gradient Boosting, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12477, https://doi.org/10.5194/egusphere-egu25-12477, 2025.

PM-2.5 is a critical pollutant for air quality evaluation and public health policymaking, necessitating accurate data for reliable analysis. However, environmental data often contain missing values due to equipment malfunctions or extreme weather conditions, which undermine the credibility of analysis and predictions. In particular, the frequent fluctuations of PM-2.5 levels in Seoul highlight the importance of addressing missing data issues.

This study systematically compares the performance of various missing data imputation methods for PM-2.5 data in Seoul, aiming to identify the optimal approach for medium- and long-term predictions. By generating and evaluating missing data during high- and low-concentration periods, this research differentiates itself from prior studies and enhances practical applicability.

A range of statistical and machine learning-based methods, including FFILL, KNN, MICE, SARIMAX, DNN, and LSTM, were applied to impute missing data. The performance of each method was evaluated over 6-hour, 12-hour, and 24-hour intervals using metrics such as RMSE, MAE, and correlation coefficients. The experimental design incorporated real-world air quality conditions by selecting data from periods of significant PM-2.5 variation.

KNN demonstrated balanced performance across all time intervals and yielded the best results for medium- and long-term predictions. FFILL showed excellent accuracy over short time intervals but exhibited declining performance as the interval length increased. Conversely, deep learning-based models, such as DNN and LSTM, showed relatively poor performance, indicating the need for further optimization to account for the characteristics of time-series data.

This study confirms that KNN is the most suitable method for PM-2.5 missing data imputation due to its simplicity and computational efficiency. These findings enhance the reliability of air quality data analysis and provide a valuable foundation for effective air quality management and policymaking. Furthermore, the results underscore the importance of selecting appropriate imputation methods to improve predictive accuracy and analytical reliability.

How to cite: Lee, J.-Y., Han, S.-H., Jang, K., Wang, K.-H., Yun, H.-Y., and Choi, D.-R.: Comparison of Models for Missing Data Imputation in Environmental Data: A Case Study of PM-2.5 in Seoul, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2634, https://doi.org/10.5194/egusphere-egu25-2634, 2025.

With the advent of Machine Learning methods and the development of new techniques in data mining, knowledge representation and data extraction, new possibilities have emerged to address the shortcomings of data imperfection. In this context, there are different methods for producing synthetic time series, which vary across goals and disciplines. In certain situations, it can be challenging to obtain the relevant data required to test assumptions about the skill and performance of machine learning models. Synthetic data generation approaches provide an effective solution by enabling the testing of machine learning algorithms in the absence of real data.

Although data availability is seemingly ubiquitous these days, a paradox arises in situations where bureaucratic, practical, or technical limitations make it difficult for researchers to rely on the required data, particularly when accessing real measurements (e.g., time series data) for specific purposes.

Our preliminary study features a case in operational meteorology where synthetic data proves particularly useful, addressing challenges associated with limited or inaccessible real measurements. Specifically, we investigate the capability of machine learning algorithms to generate high-quality synthetic time series that can be applied in meteorological data processing and analysis. To achieve this, synthetic datasets were developed based on informed criteria that integrate dynamical features of near-surface temperature data, tailored to the unique geographic and environmental context of Cyprus. These criteria include key characteristics such as trends, extreme values, diurnal cycles and vertical temperature gradients, ensuring a realistic and comprehensive representation of near-surface temperature behavior. This approach facilitates the testing and validation of data-driven models in operational settings, providing a robust framework for evaluating their performance under controlled, yet realistic, conditions.

We characterized the general features of these synthetic datasets and evaluated their utility as benchmarks for data quality control purposes. Our findings underscore the potential value of synthetic datasets in operational meteorology, particularly in supporting the development and evaluation of robust, purpose-specific, machine learning algorithms. 

How to cite: Araya, J., Proestos, Y., and Lelieveld, J.: Use of synthetic time series datasets for quality control of meteorological data. , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5609, https://doi.org/10.5194/egusphere-egu25-5609, 2025.

Wastewater networks are inherently interconnected systems, yet the Shapefile model commonly used in Geographic Information Systems (GIS) fails to adequately represent their connectivity. This limitation arises from the non-topological nature of Shapefiles model, which store different components—such as manholes, pipes and pumps—in separate databases without preserving their real-world interconnections. Positional imprecision and the lack of explicit topological relationships further aggravate this issue, resulting in a representation that fails to reflect the interconnected nature of the objects. To address this problem, we propose a graph-based representation where network components are modeled as nodes and their connections as edges. This approach captures the true structure of wastewater networks while resolving disconnections and accounting for missing elements through the introduction of dummy nodes. Validation on real-world datasets demonstrates the efficacy of this method in delivering a cohesive and precise representation.

How to cite: Et-targuy, O., Delenne, C., Begdouri, A., and Benferhat, S.: Graphs as Tools for Wastewater Network Representation: Benefits and Insights, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8379, https://doi.org/10.5194/egusphere-egu25-8379, 2025.

Acquiring sufficient and reliable data for tunnel construction is challenging due to high costs, data scarcity, and the site-specific nature of geological conditions. This study introduces a Geologically Constrained Conditional Tabular GAN (CTGAN) framework to address these challenges by generating synthetic data that accurately reflects the geological characteristics of tunnels. Traditional approaches often overlook inherent geological variability, leading to synthetic data that lacks real-world relevance, particularly in industrial scenarios where each tunnel or its sections exhibit unique geological environments.

The proposed framework incorporates geological attributes defined by tunneling standards, including Face condition, Compressive strength, Weathering, and Crack/fissure characteristics. These attributes are categorized into levels that represent distinct geological states while maintaining consistency with practical engineering scenarios. A physical constraint module ensures logical relationships among these features, preserving the geological and physical validity of the generated data.

Designed for industrial applications, this approach enables the augmentation of limited real-world data with samples tailored to the geological characteristics of specific tunnels. It addresses data scarcity while avoiding the generation of artificially balanced samples, instead ensuring alignment with naturally occurring geological conditions. Initial results demonstrate that the constrained CTGAN effectively replicates field-observed patterns, providing a valuable tool for improving data-driven methodologies in tunnel construction and monitoring. This research highlights the importance of leveraging domain-specific constraints in generative models, contributing to reliable, context-aware data generation for geotechnical engineering applications.

How to cite: Xu, Y., Sato, N., Ohtomo, Y., and Kawamura, Y.: Geologically Constrained CTGAN for Reliable Prediction of Tunnel Overbreak and Blasting Variables, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14054, https://doi.org/10.5194/egusphere-egu25-14054, 2025.