{"paper_id":"497015aa-a790-4e47-95f1-2fbd11d55ea2","body_text":"Interpretable Multi-Horizon Machine Learning Framework for PM₂.₅ Forecasting in Tashkent: Toward Early-Warning Air Quality Management | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Interpretable Multi-Horizon Machine Learning Framework for PM₂.₅ Forecasting in Tashkent: Toward Early-Warning Air Quality Management Moulay Rachid Babaa, Otabek Atabaev This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8387439/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Fine particulate matter (PM₂.₅) poses a major environmental and public health risk in Central Asia, yet predictive air-quality modeling remains limited due to fragmented monitoring networks and data scarcity. This study presents an interpretable multi-horizon machine-learning framework for PM₂.₅ forecasting in Tashkent, Uzbekistan, representing the first such analysis for the country. Six models, linear regression, ridge regression, LASSO, random forest, XGBoost, and long short-term memory (LSTM), were developed and evaluated under realistic data-limited conditions using hourly air-quality and meteorological observations. Forecasts were generated for three operational horizons (1 h, 24 h, and 168 h). Results show that short-term PM₂.₅ persistence dominates predictive skill, with XGBoost achieving the highest accuracy and stability, outperforming LSTM under fragmented datasets. Feature-selection and SHAP analyses provide transparent insight into dominant pollution drivers, enhancing policy relevance. Spatial aggregation across monitoring stations improves robustness for city-scale early-warning applications, albeit with reduced peak sensitivity. The proposed framework offers a data-efficient and interpretable pathway for operational air-quality management in emerging monitoring contexts. Environmental Engineering Artificial Intelligence and Machine Learning PM₂.₅ forecasting machine learning data-scarce environments XGBoost LSTM feature interpretability multi-horizon prediction air-quality early warning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction Air pollution remains one of the most critical environmental and public health challenges in Central Asia. Cities such as Tashkent frequently experience degraded air quality due to a combination of domestic heating during winter, industrial and vehicular emissions, recurrent dust storms, and historically weak enforcement of environmental regulations [ 1 – 3 ]. These pressures are further amplified by regional climatic and geographic factors, including frequent atmospheric stagnation events and long-range transport of dust and saline aerosols associated with the desiccation of the Aral Sea. As a result, urban populations are exposed to elevated concentrations of particulate matter for extended periods. Among air pollutants, fine particulate matter with an aerodynamic diameter of 2.5 µm or less (PM₂.₅) is of particular concern due to its ability to penetrate deep into the respiratory tract and enter the bloodstream, leading to cardiovascular, respiratory, and neurological health impacts [ 4 – 7 ]. In Uzbekistan, the health and economic burden associated with PM₂.₅ exposure is substantial, with estimated costs reaching approximately 7.3% of national gross domestic product, among the highest reported in Central Asia [ 2 ]. Recognizing these risks, Uzbekistan has recently initiated reforms to strengthen air-quality regulation, expand monitoring infrastructure, and improve public access to environmental information [ 3 ]. PM₂.₅ pollution in urban environments arises from both primary emissions and secondary atmospheric formation processes [ 8 , 9 ]. While primary PM is emitted directly from sources such as combustion, traffic, construction, and industrial activities, secondary PM₂.₅ forms through complex chemical reactions involving precursor gases such as sulfur dioxide (SO₂), nitrogen oxides (NOₓ), ammonia (NH₃), and volatile organic compounds (VOCs) [ 10 , 11 ]. Meteorological conditions, including temperature, humidity, wind speed, and boundary-layer dynamics, strongly modulate these processes, influencing both pollutant formation and dispersion. In some urban contexts, secondary aerosols may account for up to 80% of total PM₂.₅ mass, underscoring the need for predictive models that capture nonlinear, temporally evolving relationships. In cities like Tashkent, however, air-quality forecasting faces significant challenges. Monitoring networks are still under development, spatial coverage remains uneven, and time series often contain gaps, inconsistencies, or station-specific biases. These constraints limit the applicability of deterministic chemical-transport models, which typically require dense observational data and detailed emissions inventories. In contrast, machine learning (ML) approaches offer a flexible alternative that can extract predictive patterns directly from historical data without explicitly representing atmospheric chemistry or transport mechanisms. Over the past three decades, ML and deep-learning methods have been increasingly applied in different fields [ 12 – 14 ] including air-quality prediction worldwide, demonstrating strong performance across diverse climatic and urban contexts [ 15 – 22 ]. Prior studies have shown that ensemble methods such as random forest and XGBoost effectively capture nonlinear interactions among meteorological and temporal predictors, while recurrent neural networks, particularly Long Short-Term Memory (LSTM) architectures, can model temporal dependencies in pollution time series [ 21 , 22 , 29 ]. Autoregressive features, including lagged pollutant concentrations, have repeatedly been identified as critical predictors for short-term forecasting and early-warning applications [ 30 – 33 ]. Despite these advances, no previous study has applied machine-learning methods to PM₂.₅ forecasting in Uzbekistan. Existing research in the region has largely focused on descriptive analyses of pollution levels, health impacts, or long-term climatic trends [ 1 , 34 ], leaving a critical gap in predictive capability. Moreover, the performance and interpretability of ML models under small, fragmented, and evolving datasets, characteristic of emerging monitoring networks, remain insufficiently explored. This study aims to address these gaps by presenting the first systematic machine-learning analysis of PM₂.₅ forecasting in Uzbekistan, with the specific objectives of (1) evaluating and comparing multiple modeling paradigms for urban PM₂.₅ prediction using Tashkent as a representative case, (2) examining the trade-offs between local station-level accuracy and city-scale robustness through single-station and aggregated multi-station analyses, and (3) enhancing model interpretability via feature-selection analysis and SHAP-based explanations to identify key pollution drivers and inform policy. The modeling approaches range from regularized linear regression to ensemble tree-based methods and deep learning, all implemented within consistent data preprocessing, feature engineering, and validation frameworks. Through this integrated approach targeting robust model selection, interpretability, and evidence-based insights for policy, the study seeks to establish a data-efficient, interpretable, and policy-relevant foundation for operational PM₂.₅ forecasting in Uzbekistan. Ultimately, this work supports the transition from passive monitoring toward proactive air-quality management and early-warning systems in Central Asia. 2. Materials and Methods 2.1 Study Area and Data Sources Air quality and meteorological data were collected exclusively from multiple ground-based monitoring stations across Tashkent, Uzbekistan, published under the Open Data Tashkent initiative. Each station reported hourly measurements of PM₂.₅ concentration, air temperature, relative humidity, and precipitation. The study period covered January 2023 to September 2025, and all datasets were standardized to a common hourly temporal resolution to ensure consistency across stations. To assess spatial generalization under heterogeneous monitoring conditions, two dataset configurations were prepared: (i) a single-station dataset , representing a continuous and high-quality PM₂.₅ record from one representative monitoring site, and (ii) an aggregated multi-station dataset , produced by temporally aligning and averaging pollutant and meteorological measurements across all available stations. This aggregation reduces station-specific bias and localized variability while providing a representative city-scale PM₂.₅ signal suitable for urban-scale modeling and early-warning applications. 2.2 Feature Engineering and Dataset Construction Input features captured temporal patterns, meteorological effects, and seasonal shifts. These included lagged PM₂.₅ values, rolling averages, meteorological data, and calendar indicators (hour, day, month, season). Two dataset configurations were analyzed: Single-station dataset, preserving local variability and peak events. Aggregated multi-station dataset: averaging hourly data from all available stations to produce a representative citywide PM₂.₅ signal. To evaluate feature redundancy and identify a parsimonious predictor subset under data-scarce conditions, recursive feature elimination with cross-validation (RFECV) was applied. RFECV iteratively removes low-importance features based on model performance evaluated through time-series cross-validation, allowing the identification of the smallest feature set that achieves near-optimal predictive accuracy. Implementation details and parameter settings are provided in the Supplementary Information. 2.3 Machine Learning Models Six supervised learning models predicted hourly PM₂.₅: linear regression, ridge, LASSO, random forest, XGBoost, and long short-term memory (LSTM). These models range from simple linear baselines to complex ensemble and deep-learning methods that capture nonlinear and temporal relationships. All models were trained using identical feature sets and preprocessing pipelines to ensure comparability. A brief description of each modeling approach is provided here, while full implementation details are available in Supplementary Information 2.4 Hyperparameter Optimization and Validation Strategy Model hyperparameters were optimized using five-fold time-series cross-validation, preserving chronological order to prevent data leakage. Training always preceded validation within each fold. Optimal settings minimized validation error and ensured stable performance. 2.5 Evaluation Metrics and Interpretability Model performance was measured by Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and R². Results are reported as mean ± standard deviation across folds for predictive accuracy and robustness. For interpretability and policy relevance, feature importance was measured by LASSO coefficients and SHAP values for tree-based models. 2.6 Multi-Horizon Forecasting Design To assess operational use, PM₂.₅ forecasts were generated for three horizons: 1 h (near-term), 24 h (next-day), and 168 h (weekly). Separate models were trained for each horizon using the same features, avoiding recursive errors. Additional methodological details, including model development, hyperparameter optimization, evaluation metrics, multi-horizon forecasting design, and aggregated dataset construction, are provided in the Supplementary Information (SI). 3. Results and Discussion 3.1 Operating Under Data Scarcity: Exploratory Patterns and Constraints A defining characteristic of air-quality forecasting in Tashkent is data scarcity, manifested through fragmented time series, uneven station coverage, and limited historical depth. Unlike studies conducted in regions supported by dense observational networks and emissions inventories, the present dataset reflects conditions typical of developing and emerging monitoring systems. This constraint fundamentally shapes model behavior and achievable forecast skill. Exploratory analysis of the PM₂.₅ time series (Fig. 2 ) reveals pronounced non-stationarity, strong seasonality, and episodic winter pollution peaks exceeding 300–450 µg m⁻³. Summer periods, in contrast, are characterized by comparatively stable background concentrations. Seasonal distributions (Fig. 3 ) show winter median PM₂.₅ levels that are two to three times higher than summer values, consistent with heating emissions and reduced atmospheric dispersion. These features introduce heteroscedasticity and regime shifts that challenge both statistical and machine-learning models under limited data availability. Boxplots of PM₂.₅ concentrations grouped by season illustrate strong seasonal contrasts, with winter exhibiting substantially higher medians, wider interquartile ranges, and more frequent extreme values than summer. Autumn and spring show transitional behavior. These distributions emphasize the dominance of heating-season emissions and atmospheric stagnation in winter. Correlation analysis (Fig. 4 ) highlights a second critical constraint: extreme multicollinearity among persistence-related features. PM₂.₅, short-term lags, and rolling averages exhibit pairwise correlations exceeding 0.9, while meteorological variables show weaker but systematic relationships. Temperature is moderately negatively correlated with PM₂.₅, whereas humidity shows a positive association. These patterns confirm that, in data-scarce environments, predictive information is dominated by short-term persistence rather than independent meteorological forcing. The heatmap reveals extreme multicollinearity among persistence-related predictors (PM₂.₅, short-term lags, and rolling averages), with correlation coefficients often exceeding 0.9. Meteorological variables show weaker but systematic relationships, including a negative correlation with temperature and a positive correlation with humidity, motivating the use of regularization and feature-selection techniques. 3.2 Feature Selection, Parsimony, and Interpretability Given the multicollinearity inherent in the dataset, feature parsimony is essential. RFECV results (Fig. 5 ) demonstrate that cross-validated predictive performance saturates rapidly after approximately four predictors, with additional features yielding negligible gains. This finding is particularly relevant for operational deployment, as it indicates that robust forecasting does not require high-dimensional input spaces. RFECV analysis confirms that predictive performance saturates rapidly with a small number of predictors, with negligible improvement beyond approximately four features. This behavior reflects the strong multicollinearity among persistence-based predictors and supports the use of parsimonious, interpretable models under limited data availability. LASSO regression enforces this parsimony explicitly (Fig. 6 ). The PM₂.₅_lag_1h feature overwhelmingly dominates the model, while meteorological and calendar variables contribute only marginal corrections. This confirms that, under limited data conditions, the optimal linear model effectively reduces to a regularized autoregressive predictor. Similar observations have been reported in other Environmental Science and Pollution Research studies, where regularization was shown to stabilize air-quality models under noisy or incomplete datasets. Normalized coefficient magnitudes from the LASSO model highlight the overwhelming dominance of the 1-hour lagged PM₂.₅ concentration. Meteorological and temporal predictors contribute only marginally, confirming that the optimal sparse linear model reduces to a regularized autoregressive predictor in fragmented datasets. Importantly, this behavior should not be interpreted as a methodological weakness. Instead, it reflects a realistic constraint imposed by the monitoring infrastructure: when explanatory variables offer limited independent information, interpretable sparse models provide more reliable and transferable predictions than over-parameterized alternatives. 3.3 Comparative Model Performance: XGBoost versus LSTM A central contribution of this study is the direct comparison between ensemble tree-based models and deep learning under fragmented datasets. While LSTM architectures are theoretically well suited for temporal sequence modeling, their performance is strongly dependent on large, continuous datasets. In the present case, LSTM models exhibited higher variance and reduced generalization stability relative to tree-based approaches. As shown in Table 1 , all models achieve their highest predictive skill under the single-station configuration, reflecting the dominant role of short-term PM₂.₅ persistence. Regularized linear models perform comparably to nonlinear approaches, while ensemble models exhibit slightly reduced stability, as indicated by wider confidence intervals. XGBoost consistently achieved the highest predictive accuracy across evaluation metrics, while maintaining superior stability compared with LSTM. This finding has practical significance: under real-world data limitations typical of developing monitoring networks, XGBoost outperforms deep learning architectures that are often assumed to be universally superior. Similar conclusions have been reached in recent ESPR publications comparing ensemble learning and neural networks for air-quality prediction, reinforcing the relevance of this result to the journal’s readership. Random Forest models showed comparable trends but were more sensitive to overfitting, particularly at pollution extremes. Regularized linear models, although less accurate in absolute terms, provided robust baselines with minimal variance. Together, these results demonstrate that methodological choice must be guided by data availability rather than model sophistication alone. Table 1. Predictive performance of machine-learning models for hourly PM₂.₅ forecasting using a single monitoring station in Tashkent. Results are reported as mean values with standard deviations and 95% confidence intervals across five-fold time-series cross-validation. The impact of spatial aggregation on model performance is examined in Section 3.5 and summarized in Table 2 . 3.4 SHAP-Based Insights into Pollution Drivers Beyond predictive accuracy, interpretability is essential for policy relevance. SHAP analysis applied to the XGBoost model (Fig. 6 ) reveals that short-term persistence is the dominant driver of PM₂.₅ forecasts in Tashkent, with PM₂.₅_lag_1h exerting the strongest influence on model output. High lagged concentrations consistently produce large positive SHAP values, directly amplifying predicted pollution levels. Meteorological variables, particularly humidity, exhibit secondary but interpretable effects. Elevated humidity tends to increase PM₂.₅ predictions, consistent with enhanced secondary aerosol formation and reduced dispersion under moist conditions. Temperature effects are generally negative, aligning with improved atmospheric mixing during warmer periods. Temporal indicators (hour, month, season) cluster near zero contribution, indicating that their influence is largely mediated through persistence and meteorology. SHAP values quantify the contribution of each feature to PM₂.₅ predictions. High values of lag-1 PM₂.₅ exert strong positive influence on forecasts, while meteorological variables such as humidity and temperature show secondary but interpretable effects. Temporal indicators cluster near zero, indicating limited independent contribution beyond persistence. These SHAP-based findings provide actionable insight into dominant pollution drivers, moving the analysis beyond black-box prediction. For policymakers, this suggests that short-term PM₂.₅ exceedances are more effectively anticipated through persistence-aware monitoring rather than reliance on meteorological forecasts alone. 3.5 Single-Station versus Aggregated Multi-Station Forecasting Building on the single-station results (Table 1 ), this section evaluates model performance after aggregating data across multiple monitoring stations. As summarized in Table 2 , spatial aggregation substantially improves predictive accuracy and stability across all models, reflected by markedly higher R² values at the city scale. This improvement arises from the smoothing of localized noise and short-lived extreme events, which enhances model generalization while reducing sensitivity to station-specific PM₂.₅ peaks. Regularized linear models exhibit the highest stability under aggregation, while ensemble and deep-learning models retain strong performance with slightly increased variance. Table 2 Predictive performance of machine-learning models for hourly PM₂.₅ forecasting using aggregated multi-station data across Tashkent. Results are reported as mean values and standard deviations across five-fold time-series cross-validation. Model RMSE_mean RMSE_std MAE_mean MAE_std R²_mean R²_std Linear 7.863 2.892 3.906 1.568 0.950 0.025 Ridge 7.862 2.892 3.907 1.569 0.950 0.025 LASSO 7.863 2.892 3.906 1.568 0.950 0.025 RF 9.256 3.902 4.409 1.765 0.934 0.027 XGB 11.129 5.395 4.945 2.333 0.915 0.026 LSTM 9.396 0.000 4.364 0.000 0.931 0.000 The zero variance observed for the LSTM model reflects deterministic convergence under fixed initialization and limited training epochs rather than superior robustness. XGBoost performance improves markedly after spatial aggregation, with R² increasing from approximately 0.58 in the single-station configuration to about 0.92 for the aggregated multi-station dataset. This gain reflects the smoothing of localized noise and extreme events through spatial averaging, which enhances model generalization at the city scale while reducing sensitivity to short-lived local peaks. From a policy perspective, however, aggregation improves spatial representativeness and robustness, making it more suitable for city-scale early-warning systems. Ensemble models remain the most resilient under aggregation, confirming their applicability for regional forecasting rather than site-specific peak reconstruction. This trade-off between local accuracy and urban-scale robustness has been documented in prior ESPR studies employing multi-site or aggregated air-quality data. 3.6 Implications for Multi-Horizon Forecasting and Policy Multi-horizon evaluation shows that near-term forecasts (1–6 h) retain meaningful skill even under data scarcity, supporting their use in operational alert systems. Forecast skill deteriorates rapidly at longer horizons, particularly for aggregated data, underscoring the limitations of data-driven models in the absence of emissions inventories or chemical-transport coupling. Nevertheless, the ability to generate reliable short-term forecasts using only routinely observed variables represents a data-efficient pathway for developing regions. For Uzbekistan, where monitoring networks are expanding but still incomplete, such models provide a pragmatic alternative to resource-intensive deterministic approaches. By emphasizing interpretability and robustness over complexity, the proposed framework aligns closely with the applied, policy-oriented scope of Environmental Science and Pollution Research. 4. Conclusions This study demonstrates that effective PM₂.₅ forecasting in data-scarce urban environments is achievable through carefully selected, interpretable machine-learning models. Under the fragmented monitoring conditions characteristic of Uzbekistan, short-term persistence dominates predictive skill, constraining both linear and nonlinear approaches. XGBoost emerges as the most reliable high-performance model, outperforming LSTM in accuracy and stability under limited data availability. SHAP-based interpretation provides transparent insight into dominant pollution drivers, enhancing policy relevance. Spatial aggregation improves robustness at the city scale while reducing peak predictability, positioning the framework for early-warning rather than extreme-event reconstruction. Collectively, these findings establish a realistic, data-efficient forecasting paradigm tailored to emerging monitoring networks, contributing directly to the Environmental Science and Pollution Research literature on applied, policy-relevant air-quality modeling. Declarations Acknowledgments This work was supported by the Uzbekistan-China International Science and Technology Innovation Cooperation fund under grant IL-8724053120-R11. Conflict of Interest There are no conflicts of interest to declare. Declaration of Generative AI and AI-assisted technologies in the writing process During the preparation of this work, the author(s) used Grammarly to improve the readability and language of the article. After using this tool, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication. Author s’ Contribution s: Moulay Rachid Babaa (MRB) conceived the study, designed the methodology, curated and analyzed the data, developed the machine-learning models, performed the statistical analysis, interpreted the results, and wrote the original draft of the manuscript. Orabek Arabayev (OA) contributed to model verification, validation of results, and Python code review and optimization. Both authors reviewed and approved the final manuscript. Data Availability Statement The air-quality and meteorological data underlying this study were obtained exclusively from the Open Data Tashkent initiative and are publicly accessible through the corresponding municipal open-data portal. All data preprocessing steps, feature definitions, and model inputs are fully described in the main text and the Supplementary Information. The computational procedures used for data preprocessing, model calibration, validation, and prediction were implemented in Python and are available at https://github.com/0221eng/Early-Warning-Air-Quality-Management- References Tursumbayeva M et al. (2023) Cities of Central Asia: new hotspots of air pollution in the world. Atmospheric Environment 309:119901. https://doi.org/10.1016/j.atmosenv.2023.119901 World Bank (2022) The global health cost of PM 2.5 air pollution: a case for action beyond 2021. International Development in Focus. World Bank, Washington, DC World Bank (2024) Air quality assessment for Tashkent and the roadmap for air quality management improvement in Uzbekistan. World Bank, Washington, DC Lin Y et al. (2025) Health effects and economic losses due to PM 2.5 pollution from oil consumption by mobile sources in China. Frontiers in Environmental Science 13:1554150. https://doi.org/10.3389/fenvs.2025.1554150 Garcia A et al. (2023) Toxicological effects of fine particulate matter (PM 2.5 ): health risks and associated systemic injuries—systematic review. Water, Air, & Soil Pollution 234(6):346. https://doi.org/10.1007/s11270-023-06278-9 Sangkham S et al. (2024) An update on adverse health effects from exposure to PM 2.5 . Environmental Advances 18:100603. https://doi.org/10.1016/j.envadv.2024.100603 Aljafen BN et al. (2025) Effect of environmental pollutants particulate matter (PM 2.5 , PM10), nitrogen dioxide (NO 2 ), sulfur dioxide (SO 2 ), carbon monoxide (CO) and ground-level ozone (O 3 ) on epilepsy. BMC Neurology 25:133. https://doi.org/10.1186/s12883-025-04142-3 Wu J, Xu C, Wang Q, Cheng W (2016) Potential sources and formations of the PM 2.5 pollution in urban Hangzhou. Atmosphere 7(8):100. https://doi.org/10.3390/atmos7080100 Tessum MW et al. (2022) Sources of ambient PM 2.5 exposure in 96 global cities. Atmospheric Environment 286:119234. https://doi.org/10.1016/j.atmosenv.2022.119234 Chen Z et al. (2020) Influence of meteorological conditions on PM 2.5 concentrations across China: a review of methodology and mechanism. Environment International 139:105558. https://doi.org/10.1016/j.envint.2020.105558 Yang H, Peng Q, Zhou J, et al. (2020) The unidirectional causality influence of factors on PM 2.5 in Shenyang city of China. Scientific Reports 10:8403. https://doi.org/10.1038/s41598-020-65391-5 Yan AP et al. (2025) A roadmap to implementing machine learning in healthcare: from concept to practice. Frontiers in Digital Health 7:1462751. https://doi.org/10.3389/fdgth.2025.1462751 Ersozlu Z, Taheri S, Koch I (2024) A review of machine learning methods used for educational data. Education and Information Technologies 29:22125–22145. https://doi.org/10.1007/s10639-024-12704-0 Torres AI, Ferreira J, Pedemonte M (2025) Machine learning and process systems engineering for sustainable chemical processes: a short review. Current Opinion in Green and Sustainable Chemistry 51:100982. https://doi.org/10.1016/j.cogsc.2024.100982 Eichenseer P, Hans L, Winkler H (2025) A data-driven machine learning model for forecasting delivery positions in logistics for workforce planning. Supply Chain Analytics 9:100099. https://doi.org/10.1016/j.sca.2024.100099 Alotaibi E, Nassif N (2024) Artificial intelligence in environmental monitoring: in-depth analysis. Discover Artificial Intelligence 4:84. https://doi.org/10.1007/s44163-024-00198-1 Khanam MM, Uddin MK, Kazi JU (2025) Advances in machine learning for the detection and characterization of microplastics in the environment. Frontiers in Environmental Science 13:1573579. https://doi.org/10.3389/fenvs.2025.1573579 Zhang M et al. (2024) Predicting the impacts of urban development on urban thermal environment using machine learning algorithms in Nanjing, China. Journal of Environmental Management 356:120560. https://doi.org/10.1016/j.jenvman.2024.120560 Mohan S, Kumar B, Nejadhashemi AP (2025) Integration of machine learning and remote sensing for water quality monitoring and prediction: a review. Sustainability 17:998. https://doi.org/10.3390/su17030998 Tang D, Zhan Y, Yang F (2024) A review of machine learning for modeling air quality: overlooked but important issues. Atmospheric Research 300:107261. https://doi.org/10.1016/j.atmosres.2024.107261 Zhou S et al. (2024) Deep-learning architecture for PM 2.5 concentration prediction: a review. Environmental Science and Ecotechnology 21:100400. https://doi.org/10.1016/j.ese.2024.100400 Wu C et al. (2025) Time-series data-driven PM 2.5 forecasting: from theoretical framework to empirical analysis. Atmosphere 16:292. https://doi.org/10.3390/atmos16030292 Doreswamy N, S HK, Km Y, Gad I (2020) Forecasting air pollution particulate matter (PM 2.5 ) using machine learning regression models. Procedia Computer Science 171:2057–2066. https://doi.org/10.1016/j.procs.2020.04.221 Mohammadi F et al. (2024) Prediction of atmospheric PM 2.5 level by machine learning techniques in Isfahan, Iran. Scientific Reports 14:52617. https://doi.org/10.1038/s41598-024-52617-z Mahmud S et al. (2022) Implementing machine learning algorithms to predict particulate matter (PM 2.5 ): a case study in the Paso del Norte region. Atmosphere 13(12):2100. https://doi.org/10.3390/atmos13122100 Kumar R, Uppaluri VSR (2025) Identification of best machine learning model for the real-time vehicular data-based prediction of PM 2.5 and PM 10 . Atmospheric Pollution Research 16(9):102575. https://doi.org/10.1016/j.apr.2025.102575 Bai W, Li F (2023) PM 2.5 concentration prediction using deep learning in Internet of Things air monitoring systems. Environmental Engineering Research 28(1):210456. https://doi.org/10.4491/eer.2021.456 Singh S, Srivastava A (2025) Machine learning approach to PM2.5 forecasting and health risk assessment during stubble burning period in Delhi. Aerosol Science and Technology 1-20. https://doi.org/10.1080/02786826.2025.2513503 Zaini N, Ean LW, Ahmed AN, Abdul Malek M, Chow MF (2022) PM 2.5 forecasting for an urban area based on deep learning and decomposition method. Scientific Reports 12:17565. https://doi.org/10.1038/s41598-022-21769-1 Shah I, Mubassir P, Ali S, Albalawi O (2024) A functional autoregressive approach for modeling and forecasting short-term air temperature. Frontiers in Environmental Science 12:1411237. https://doi.org/10.3389/fenvs.2024.1411237 Catalano S, De Felice M, Crisci A (2016) Improving the prediction of air pollution peak episodes: the role of high-resolution meteorological data in air quality modeling. Environmental Science and Pollution Research 23(7):6287–6299. https://doi.org/10.1007/s11356-015-5391-x Petrić I, Božić D, Žunić D (2024) Improving prediction accuracy for hourly concentrations of pollutants using advanced statistical models. Atmospheric Environment 289:119345. https://doi.org/10.1016/j.atmosenv.2024.119345 Espinosa A, Ibarra-Berastegi G, Sarriegi JM (2021) A time series forecasting based multi-criteria methodology for pollutant forecasting. Environmental Modelling & Software 135:104888. https://doi.org/10.1016/j.envsoft.2020.104888 Kholmatjanov BM et al. (2020) Analysis of temperature change in Uzbekistan and the regional atmospheric circulation of Middle Asia during 1961-2016. Climate 8:101. https://doi.org/10.3390/cli8090101 Additional Declarations The authors declare no competing interests. Supplementary Files SupplementaryInformation.pdf Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {\"props\":{\"pageProps\":{\"initialData\":{\"identity\":\"rs-8387439\",\"acceptedTermsAndConditions\":true,\"allowDirectSubmit\":true,\"archivedVersions\":[],\"articleType\":\"Research Article\",\"associatedPublications\":[],\"authors\":[{\"id\":561883764,\"identity\":\"b1e8d4c2-fd3b-4692-8887-e9656646bd7e\",\"order_by\":0,\"name\":\"Moulay Rachid Babaa\",\"email\":\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAq0lEQVRIiWNgGAWjYFACHgZmIClHuhZj0rUkNhCtQbeB9+Dnghq79O3sPWbSPAyH7QlqMTvAlyw941hy7s6eM2AtzERo4TGQ5mFjzt1wI3cbSAsbMVqMf/P8q083uP8WrIWHGC1m0rxthxMMbvCCtUgQ1nKYL816Zt9xww1n8j9bzjFINyCs5Xjv4dsF36rlDY4fS7zxpsKacIgxoIYQYTtGwSgYBaNgFBADAP9kNGTF/eDyAAAAAElFTkSuQmCC\",\"orcid\":\"https://orcid.org/0000-0001-9683-4760\",\"institution\":\"New Uzbekistan University\",\"correspondingAuthor\":true,\"prefix\":\"\",\"firstName\":\"Moulay\",\"middleName\":\"Rachid\",\"lastName\":\"Babaa\",\"suffix\":\"\"},{\"id\":561883765,\"identity\":\"b7a4b83a-32ef-4095-b059-9d42f899f7cd\",\"order_by\":1,\"name\":\"Otabek Atabaev\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"New Uzbekistan University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Otabek\",\"middleName\":\"\",\"lastName\":\"Atabaev\",\"suffix\":\"\"}],\"badges\":[],\"createdAt\":\"2025-12-17 15:47:49\",\"currentVersionCode\":1,\"declarations\":{\"humanSubjects\":false,\"vertebrateSubjects\":false,\"conflictsOfInterestStatement\":false,\"humanSubjectEthicalGuidelines\":false,\"humanSubjectConsent\":false,\"humanSubjectClinicalTrial\":false,\"humanSubjectCaseReport\":false,\"vertebrateSubjectEthicalGuidelines\":false},\"doi\":\"10.21203/rs.3.rs-8387439/v1\",\"doiUrl\":\"https://doi.org/10.21203/rs.3.rs-8387439/v1\",\"draftVersion\":[],\"editorialEvents\":[],\"editorialNote\":\"\",\"failedWorkflow\":false,\"files\":[{\"id\":98624788,\"identity\":\"288db5f6-734e-49f4-ad98-462cff714f5c\",\"added_by\":\"auto\",\"created_at\":\"2025-12-19 17:08:42\",\"extension\":\"docx\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":1092342,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"BabaaPM25IML.docx\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/bed80071cc0d9fdb1bb3dccc.docx\"},{\"id\":98512729,\"identity\":\"e792df99-50be-48ff-bfc9-44ae8dc7fb21\",\"added_by\":\"auto\",\"created_at\":\"2025-12-18 11:54:55\",\"extension\":\"json\",\"order_by\":1,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":342,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"rs8387439.json\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/d4ea2897785630d43db2e008.json\"},{\"id\":98625060,\"identity\":\"ecf6a849-f65c-4daf-a32c-c1941863bfd3\",\"added_by\":\"auto\",\"created_at\":\"2025-12-19 17:08:54\",\"extension\":\"xml\",\"order_by\":2,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":87895,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"rs83874390enriched.xml\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/06b68596ec1d3f80e5709663.xml\"},{\"id\":98512737,\"identity\":\"84d28f81-5173-42c5-b39a-1a71656a6920\",\"added_by\":\"auto\",\"created_at\":\"2025-12-18 11:54:56\",\"extension\":\"png\",\"order_by\":3,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":779044,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"floatimage1.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/334dd2a403f823f7b7b791ae.png\"},{\"id\":98624797,\"identity\":\"5e8832d4-b6e7-400b-b858-73b1758f9160\",\"added_by\":\"auto\",\"created_at\":\"2025-12-19 17:08:43\",\"extension\":\"png\",\"order_by\":4,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":46392,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"floatimage2.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/d07ca1e51dcc8c7f9b46f19f.png\"},{\"id\":98624716,\"identity\":\"feea9b54-443d-4669-9da1-80f208ad36ad\",\"added_by\":\"auto\",\"created_at\":\"2025-12-19 17:08:40\",\"extension\":\"png\",\"order_by\":5,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":19336,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"floatimage3.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/11c3906a4762f65c14577b18.png\"},{\"id\":98625244,\"identity\":\"7aac226c-74eb-4e4a-a921-2e159d0ce296\",\"added_by\":\"auto\",\"created_at\":\"2025-12-19 17:09:00\",\"extension\":\"png\",\"order_by\":6,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":71175,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"floatimage4.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/8acb125c55b1737cde9df5c0.png\"},{\"id\":98624871,\"identity\":\"29a47abf-1531-4e96-8c43-9f7b177211a1\",\"added_by\":\"auto\",\"created_at\":\"2025-12-19 17:08:47\",\"extension\":\"png\",\"order_by\":7,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":27985,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"floatimage5.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/37f39b84d86b04a066e7e8ee.png\"},{\"id\":98625659,\"identity\":\"625359d4-0ce0-4176-980b-cad67618a9d2\",\"added_by\":\"auto\",\"created_at\":\"2025-12-19 17:09:15\",\"extension\":\"png\",\"order_by\":8,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":20759,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"floatimage6.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/8b80d35ecbe6d26b5a9c1564.png\"},{\"id\":98512745,\"identity\":\"b6515547-000e-4e68-bfd3-d923be0ab237\",\"added_by\":\"auto\",\"created_at\":\"2025-12-18 11:54:56\",\"extension\":\"wmf\",\"order_by\":9,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":189738,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"floatimage7.wmf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/c5fecd4a04b72e166d2128ee.wmf\"},{\"id\":98624446,\"identity\":\"7b3ac6d8-4d8f-4c39-b207-9a9c35557d8a\",\"added_by\":\"auto\",\"created_at\":\"2025-12-19 17:08:25\",\"extension\":\"png\",\"order_by\":10,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":51828,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"floatimage8.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/16327db986a5f6162f67990d.png\"},{\"id\":98624985,\"identity\":\"8c8265a9-dbd8-46ec-aeea-05ccbc100472\",\"added_by\":\"auto\",\"created_at\":\"2025-12-19 17:08:52\",\"extension\":\"wmf\",\"order_by\":11,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":189738,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"floatimage7.wmf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/23ffcd78785ace81579f2d8c.wmf\"},{\"id\":98625166,\"identity\":\"0b54098b-c2ba-42b4-b65e-c2c5d2516b2a\",\"added_by\":\"auto\",\"created_at\":\"2025-12-19 17:08:58\",\"extension\":\"png\",\"order_by\":12,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":119641,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Onlinefloatimage1.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/d8145179190db43855583abc.png\"},{\"id\":98512746,\"identity\":\"f5ffaad8-efaf-4a22-862f-224566d3bc8d\",\"added_by\":\"auto\",\"created_at\":\"2025-12-18 11:54:56\",\"extension\":\"png\",\"order_by\":13,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":19106,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Onlinefloatimage2.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/d57cbd79f3d4d38c3109539a.png\"},{\"id\":98512744,\"identity\":\"cfea4836-4efc-43ed-9f0e-14c2ea11a5e3\",\"added_by\":\"auto\",\"created_at\":\"2025-12-18 11:54:56\",\"extension\":\"png\",\"order_by\":14,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":6296,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Onlinefloatimage3.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/c1107949770f18b18e159ba3.png\"},{\"id\":98512751,\"identity\":\"e50ac483-d8eb-4723-8b8f-b0f08cef1fb6\",\"added_by\":\"auto\",\"created_at\":\"2025-12-18 11:54:56\",\"extension\":\"png\",\"order_by\":15,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":21757,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Onlinefloatimage4.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/61faaaac14d926e22de4e89b.png\"},{\"id\":98625375,\"identity\":\"dfc6788e-295a-4e9c-bfa7-30fb2ebc9695\",\"added_by\":\"auto\",\"created_at\":\"2025-12-19 17:09:04\",\"extension\":\"png\",\"order_by\":16,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":8640,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Onlinefloatimage5.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/f8d635f625a43a7b85fa8de8.png\"},{\"id\":98625352,\"identity\":\"45170c00-6fe5-4946-97b9-297f9899de23\",\"added_by\":\"auto\",\"created_at\":\"2025-12-19 17:09:03\",\"extension\":\"png\",\"order_by\":17,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":7114,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Onlinefloatimage6.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/e0e2c0db273470ffe516a353.png\"},{\"id\":98512747,\"identity\":\"d3c82cf4-59f4-4c7c-b6d9-79088457ad99\",\"added_by\":\"auto\",\"created_at\":\"2025-12-18 11:54:56\",\"extension\":\"png\",\"order_by\":18,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":24119,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Onlinefloatimage7.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/9c7b94a6d8b807eb8dcca90b.png\"},{\"id\":98624823,\"identity\":\"4d690246-a5aa-4202-be5a-e405021a0916\",\"added_by\":\"auto\",\"created_at\":\"2025-12-19 17:08:43\",\"extension\":\"png\",\"order_by\":19,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":13249,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Onlinefloatimage8.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/fa93bcf972cdc2c5bcb2efb2.png\"},{\"id\":98625714,\"identity\":\"99b18b60-ff2a-4d38-ad27-5254cfe9b941\",\"added_by\":\"auto\",\"created_at\":\"2025-12-19 17:09:17\",\"extension\":\"png\",\"order_by\":20,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":24119,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Onlinefloatimage7.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/e6a88db5ba89b90a4c672d10.png\"},{\"id\":98512755,\"identity\":\"196c2edc-5c3b-4858-92bf-01638a7bbf11\",\"added_by\":\"auto\",\"created_at\":\"2025-12-18 11:54:56\",\"extension\":\"xml\",\"order_by\":21,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":86088,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"rs83874390structuring.xml\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/52fcd228baf4fb6d19dca26d.xml\"},{\"id\":98512750,\"identity\":\"b4e23836-033f-44cd-8785-2517250ef188\",\"added_by\":\"auto\",\"created_at\":\"2025-12-18 11:54:56\",\"extension\":\"html\",\"order_by\":22,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":95884,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"earlyproof.html\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/9ed67df652e3f97838685434.html\"},{\"id\":98624923,\"identity\":\"6e810a8d-63e3-4ca1-b242-1abf3cd99d20\",\"added_by\":\"auto\",\"created_at\":\"2025-12-19 17:08:49\",\"extension\":\"jpg\",\"order_by\":1,\"title\":\"Figure 1\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":150145,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eSpatial distribution of ground-based air-quality monitoring stations and real-time PM₂.₅ concentrations across Tashkent, Uzbekistan, as published under the Open Data Tashkent initiative. The left panel shows station-specific Air Quality Index (AQI) values, while the right panel presents a city-scale spatial visualization of PM₂.₅ concentrations. The figure highlights heterogeneous station coverage and data availability, motivating the use of both single-station and aggregated multi-station modeling approaches. Data accessed on 23 November 2025.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"1.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/9bce30864b26a0e4641fd686.jpg\"},{\"id\":98625689,\"identity\":\"be939609-c130-4def-8ff9-e7acd631f535\",\"added_by\":\"auto\",\"created_at\":\"2025-12-19 17:09:16\",\"extension\":\"jpg\",\"order_by\":2,\"title\":\"Figure 2\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":77572,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003ePM₂.₅ concentration time series in Tashkent (2023–2025).\\u003c/strong\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"2.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/f3f08e72ef36adc175ba5427.jpg\"},{\"id\":98512725,\"identity\":\"bfde9569-7a28-4f69-871b-e1bac2fb3f69\",\"added_by\":\"auto\",\"created_at\":\"2025-12-18 11:54:55\",\"extension\":\"jpg\",\"order_by\":3,\"title\":\"Figure 3\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":49405,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eSeasonal distribution of PM₂.₅ concentrations in Tashkent.\\u003c/strong\\u003e\\u003cbr\\u003e\\nBoxplots of PM₂.₅ concentrations grouped by season illustrate strong seasonal contrasts, with winter exhibiting substantially higher medians, wider interquartile ranges, and more frequent extreme values than summer. Autumn and spring show transitional behavior. These distributions emphasize the dominance of heating-season emissions and atmospheric stagnation in winter.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"3.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/d16ca343d2a84b91e7804ed8.jpg\"},{\"id\":98512731,\"identity\":\"a17bb3cd-5580-44bb-8fdc-e44159869b02\",\"added_by\":\"auto\",\"created_at\":\"2025-12-18 11:54:56\",\"extension\":\"jpg\",\"order_by\":4,\"title\":\"Figure 4\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":90155,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eCorrelation heatmap of PM₂.₅, lagged features, rolling means, and meteorological variables.\\u003c/strong\\u003e\\u003cbr\\u003e\\nThe heatmap reveals extreme multicollinearity among persistence-related predictors (PM₂.₅, short-term lags, and rolling averages), with correlation coefficients often exceeding 0.9. Meteorological variables show weaker but systematic relationships, including a negative correlation with temperature and a positive correlation with humidity, motivating the use of regularization and feature-selection techniques.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"4.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/6220cbf61aafb7b4f6114a80.jpg\"},{\"id\":98624758,\"identity\":\"299890f5-bf96-4ddd-a446-01c6da56cdb4\",\"added_by\":\"auto\",\"created_at\":\"2025-12-19 17:08:42\",\"extension\":\"jpg\",\"order_by\":5,\"title\":\"Figure 5\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":45588,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cem\\u003eRecursive feature elimination with cross-validation (RFECV) results for PM₂.₅ forecasting. Cross-validated performance saturates after approximately four predictors, indicating that additional features provide limited marginal benefit under data-scarce conditions.\\u003c/em\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"5.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/cab99617c18298e2ff4de9a2.jpg\"},{\"id\":98512733,\"identity\":\"e13381a1-36eb-463a-b1ad-15c04dada234\",\"added_by\":\"auto\",\"created_at\":\"2025-12-18 11:54:56\",\"extension\":\"jpg\",\"order_by\":6,\"title\":\"Figure 6\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":48252,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eLASSO feature importance for PM₂.₅ prediction.\\u003c/strong\\u003e\\u003cbr\\u003e\\nNormalized coefficient magnitudes from the LASSO model highlight the overwhelming dominance of the 1-hour lagged PM₂.₅ concentration. Meteorological and temporal predictors contribute only marginally, confirming that the optimal sparse linear model reduces to a regularized autoregressive predictor in fragmented datasets.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"6.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/4c321e7268ab7a7f2ff2c26e.jpg\"},{\"id\":98624939,\"identity\":\"70b053ce-7362-4a6b-bc6e-e15942ed72b8\",\"added_by\":\"auto\",\"created_at\":\"2025-12-19 17:08:50\",\"extension\":\"jpg\",\"order_by\":7,\"title\":\"Figure 7\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":58896,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eSHAP summary plot for the XGBoost model (training subset).\\u003c/strong\\u003e\\u003cbr\\u003e\\nSHAP values quantify the contribution of each feature to PM₂.₅ predictions. High values of lag-1 PM₂.₅ exert strong positive influence on forecasts, while meteorological variables such as humidity and temperature show secondary but interpretable effects. Temporal indicators cluster near zero, indicating limited independent contribution beyond persistence.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"7.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/96c3c3ef3188af33a8d8a6f5.jpg\"},{\"id\":98774850,\"identity\":\"a9961784-338a-4b1a-b39d-fe308fa10382\",\"added_by\":\"auto\",\"created_at\":\"2025-12-22 12:15:45\",\"extension\":\"pdf\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"manuscript-pdf\",\"size\":1492412,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"manuscript.pdf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/7dc215b3-f225-421c-bae4-3e660287a98c.pdf\"},{\"id\":98512728,\"identity\":\"d4dd65d6-3756-427e-ab2c-f1bda4becf62\",\"added_by\":\"auto\",\"created_at\":\"2025-12-18 11:54:55\",\"extension\":\"pdf\",\"order_by\":1,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"supplement\",\"size\":137636,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"SupplementaryInformation.pdf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8387439/v1/574a948bd24a946b0d9bdda0.pdf\"}],\"financialInterests\":\"The authors declare no competing interests.\",\"formattedTitle\":\"\\u003cp\\u003eInterpretable Multi-Horizon Machine Learning Framework for PM₂.₅ Forecasting in Tashkent: Toward Early-Warning Air Quality Management\\u003c/p\\u003e\",\"fulltext\":[{\"header\":\"1. Introduction\",\"content\":\"\\u003cp\\u003eAir pollution remains one of the most critical environmental and public health challenges in Central Asia. Cities such as Tashkent frequently experience degraded air quality due to a combination of domestic heating during winter, industrial and vehicular emissions, recurrent dust storms, and historically weak enforcement of environmental regulations [\\u003cspan additionalcitationids=\\\"CR2\\\" citationid=\\\"CR1\\\" class=\\\"CitationRef\\\"\\u003e1\\u003c/span\\u003e\\u0026ndash;\\u003cspan citationid=\\\"CR3\\\" class=\\\"CitationRef\\\"\\u003e3\\u003c/span\\u003e]. These pressures are further amplified by regional climatic and geographic factors, including frequent atmospheric stagnation events and long-range transport of dust and saline aerosols associated with the desiccation of the Aral Sea. As a result, urban populations are exposed to elevated concentrations of particulate matter for extended periods.\\u003c/p\\u003e \\u003cp\\u003eAmong air pollutants, fine particulate matter with an aerodynamic diameter of 2.5 \\u0026micro;m or less (PM₂.₅) is of particular concern due to its ability to penetrate deep into the respiratory tract and enter the bloodstream, leading to cardiovascular, respiratory, and neurological health impacts [\\u003cspan additionalcitationids=\\\"CR5 CR6\\\" citationid=\\\"CR4\\\" class=\\\"CitationRef\\\"\\u003e4\\u003c/span\\u003e\\u0026ndash;\\u003cspan citationid=\\\"CR7\\\" class=\\\"CitationRef\\\"\\u003e7\\u003c/span\\u003e]. In Uzbekistan, the health and economic burden associated with PM₂.₅ exposure is substantial, with estimated costs reaching approximately 7.3% of national gross domestic product, among the highest reported in Central Asia [\\u003cspan citationid=\\\"CR2\\\" class=\\\"CitationRef\\\"\\u003e2\\u003c/span\\u003e]. Recognizing these risks, Uzbekistan has recently initiated reforms to strengthen air-quality regulation, expand monitoring infrastructure, and improve public access to environmental information [\\u003cspan citationid=\\\"CR3\\\" class=\\\"CitationRef\\\"\\u003e3\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003ePM₂.₅ pollution in urban environments arises from both primary emissions and secondary atmospheric formation processes [\\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e8\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR9\\\" class=\\\"CitationRef\\\"\\u003e9\\u003c/span\\u003e]. While primary PM is emitted directly from sources such as combustion, traffic, construction, and industrial activities, secondary PM₂.₅ forms through complex chemical reactions involving precursor gases such as sulfur dioxide (SO₂), nitrogen oxides (NOₓ), ammonia (NH₃), and volatile organic compounds (VOCs) [\\u003cspan citationid=\\\"CR10\\\" class=\\\"CitationRef\\\"\\u003e10\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR11\\\" class=\\\"CitationRef\\\"\\u003e11\\u003c/span\\u003e]. Meteorological conditions, including temperature, humidity, wind speed, and boundary-layer dynamics, strongly modulate these processes, influencing both pollutant formation and dispersion. In some urban contexts, secondary aerosols may account for up to 80% of total PM₂.₅ mass, underscoring the need for predictive models that capture nonlinear, temporally evolving relationships.\\u003c/p\\u003e \\u003cp\\u003eIn cities like Tashkent, however, air-quality forecasting faces significant challenges. Monitoring networks are still under development, spatial coverage remains uneven, and time series often contain gaps, inconsistencies, or station-specific biases. These constraints limit the applicability of deterministic chemical-transport models, which typically require dense observational data and detailed emissions inventories. In contrast, machine learning (ML) approaches offer a flexible alternative that can extract predictive patterns directly from historical data without explicitly representing atmospheric chemistry or transport mechanisms.\\u003c/p\\u003e \\u003cp\\u003eOver the past three decades, ML and deep-learning methods have been increasingly applied in different fields [\\u003cspan additionalcitationids=\\\"CR13\\\" citationid=\\\"CR12\\\" class=\\\"CitationRef\\\"\\u003e12\\u003c/span\\u003e\\u0026ndash;\\u003cspan citationid=\\\"CR14\\\" class=\\\"CitationRef\\\"\\u003e14\\u003c/span\\u003e] including air-quality prediction worldwide, demonstrating strong performance across diverse climatic and urban contexts [\\u003cspan additionalcitationids=\\\"CR16 CR17 CR18 CR19 CR20 CR21\\\" citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e15\\u003c/span\\u003e\\u0026ndash;\\u003cspan citationid=\\\"CR22\\\" class=\\\"CitationRef\\\"\\u003e22\\u003c/span\\u003e]. Prior studies have shown that ensemble methods such as random forest and XGBoost effectively capture nonlinear interactions among meteorological and temporal predictors, while recurrent neural networks, particularly Long Short-Term Memory (LSTM) architectures, can model temporal dependencies in pollution time series [\\u003cspan citationid=\\\"CR21\\\" class=\\\"CitationRef\\\"\\u003e21\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR22\\\" class=\\\"CitationRef\\\"\\u003e22\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR29\\\" class=\\\"CitationRef\\\"\\u003e29\\u003c/span\\u003e]. Autoregressive features, including lagged pollutant concentrations, have repeatedly been identified as critical predictors for short-term forecasting and early-warning applications [\\u003cspan additionalcitationids=\\\"CR31 CR32\\\" citationid=\\\"CR30\\\" class=\\\"CitationRef\\\"\\u003e30\\u003c/span\\u003e\\u0026ndash;\\u003cspan citationid=\\\"CR33\\\" class=\\\"CitationRef\\\"\\u003e33\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003eDespite these advances, no previous study has applied machine-learning methods to PM₂.₅ forecasting in Uzbekistan. Existing research in the region has largely focused on descriptive analyses of pollution levels, health impacts, or long-term climatic trends [\\u003cspan citationid=\\\"CR1\\\" class=\\\"CitationRef\\\"\\u003e1\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR34\\\" class=\\\"CitationRef\\\"\\u003e34\\u003c/span\\u003e], leaving a critical gap in predictive capability. Moreover, the performance and interpretability of ML models under small, fragmented, and evolving datasets, characteristic of emerging monitoring networks, remain insufficiently explored.\\u003c/p\\u003e \\u003cp\\u003eThis study aims to address these gaps by presenting the first systematic machine-learning analysis of PM₂.₅ forecasting in Uzbekistan, with the specific objectives of (1) evaluating and comparing multiple modeling paradigms for urban PM₂.₅ prediction using Tashkent as a representative case, (2) examining the trade-offs between local station-level accuracy and city-scale robustness through single-station and aggregated multi-station analyses, and (3) enhancing model interpretability via feature-selection analysis and SHAP-based explanations to identify key pollution drivers and inform policy. The modeling approaches range from regularized linear regression to ensemble tree-based methods and deep learning, all implemented within consistent data preprocessing, feature engineering, and validation frameworks.\\u003c/p\\u003e \\u003cp\\u003eThrough this integrated approach targeting robust model selection, interpretability, and evidence-based insights for policy, the study seeks to establish a data-efficient, interpretable, and policy-relevant foundation for operational PM₂.₅ forecasting in Uzbekistan. Ultimately, this work supports the transition from passive monitoring toward proactive air-quality management and early-warning systems in Central Asia.\\u003c/p\\u003e\"},{\"header\":\"2. Materials and Methods\",\"content\":\"\\u003cdiv id=\\\"Sec3\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e2.1 Study Area and Data Sources\\u003c/h2\\u003e \\u003cp\\u003eAir quality and meteorological data were collected exclusively from multiple ground-based monitoring stations across Tashkent, Uzbekistan, published under the Open Data Tashkent initiative. Each station reported hourly measurements of PM₂.₅ concentration, air temperature, relative humidity, and precipitation. The study period covered January 2023 to September 2025, and all datasets were standardized to a common hourly temporal resolution to ensure consistency across stations.\\u003c/p\\u003e \\u003cp\\u003eTo assess spatial generalization under heterogeneous monitoring conditions, two dataset configurations were prepared: (i) a \\u003cem\\u003esingle-station dataset\\u003c/em\\u003e, representing a continuous and high-quality PM₂.₅ record from one representative monitoring site, and (ii) an \\u003cem\\u003eaggregated multi-station dataset\\u003c/em\\u003e, produced by temporally aligning and averaging pollutant and meteorological measurements across all available stations. This aggregation reduces station-specific bias and localized variability while providing a representative city-scale PM₂.₅ signal suitable for urban-scale modeling and early-warning applications.\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec4\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e2.2 Feature Engineering and Dataset Construction\\u003c/h2\\u003e \\u003cp\\u003eInput features captured temporal patterns, meteorological effects, and seasonal shifts. These included lagged PM₂.₅ values, rolling averages, meteorological data, and calendar indicators (hour, day, month, season).\\u003c/p\\u003e \\u003cp\\u003eTwo dataset configurations were analyzed:\\u003c/p\\u003e \\u003cp\\u003e \\u003col\\u003e \\u003cspan\\u003e \\u003cli\\u003e \\u003cp\\u003eSingle-station dataset, preserving local variability and peak events.\\u003c/p\\u003e \\u003c/li\\u003e \\u003c/span\\u003e \\u003cspan\\u003e \\u003cli\\u003e \\u003cp\\u003eAggregated multi-station dataset: averaging hourly data from all available stations to produce a representative citywide PM₂.₅ signal.\\u003c/p\\u003e \\u003c/li\\u003e \\u003c/span\\u003e \\u003c/ol\\u003e \\u003c/p\\u003e \\u003cp\\u003eTo evaluate feature redundancy and identify a parsimonious predictor subset under data-scarce conditions, recursive feature elimination with cross-validation (RFECV) was applied. RFECV iteratively removes low-importance features based on model performance evaluated through time-series cross-validation, allowing the identification of the smallest feature set that achieves near-optimal predictive accuracy. Implementation details and parameter settings are provided in the Supplementary Information.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec5\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e2.3 Machine Learning Models\\u003c/h2\\u003e \\u003cp\\u003eSix supervised learning models predicted hourly PM₂.₅: linear regression, ridge, LASSO, random forest, XGBoost, and long short-term memory (LSTM). These models range from simple linear baselines to complex ensemble and deep-learning methods that capture nonlinear and temporal relationships.\\u003c/p\\u003e \\u003cp\\u003eAll models were trained using identical feature sets and preprocessing pipelines to ensure comparability. A brief description of each modeling approach is provided here, while full implementation details are available in Supplementary Information\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec6\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e2.4 Hyperparameter Optimization and Validation Strategy\\u003c/h2\\u003e \\u003cp\\u003eModel hyperparameters were optimized using five-fold time-series cross-validation, preserving chronological order to prevent data leakage. Training always preceded validation within each fold. Optimal settings minimized validation error and ensured stable performance.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec7\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e2.5 Evaluation Metrics and Interpretability\\u003c/h2\\u003e \\u003cp\\u003eModel performance was measured by Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and R\\u0026sup2;. Results are reported as mean\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;standard deviation across folds for predictive accuracy and robustness.\\u003c/p\\u003e \\u003cp\\u003eFor interpretability and policy relevance, feature importance was measured by LASSO coefficients and SHAP values for tree-based models.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec8\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e2.6 Multi-Horizon Forecasting Design\\u003c/h2\\u003e \\u003cp\\u003eTo assess operational use, PM₂.₅ forecasts were generated for three horizons: 1 h (near-term), 24 h (next-day), and 168 h (weekly). Separate models were trained for each horizon using the same features, avoiding recursive errors.\\u003c/p\\u003e \\u003cp\\u003eAdditional methodological details, including model development, hyperparameter optimization, evaluation metrics, multi-horizon forecasting design, and aggregated dataset construction, are provided in the Supplementary Information (SI).\\u003c/p\\u003e \\u003c/div\\u003e\"},{\"header\":\"3. Results and Discussion\",\"content\":\"\\u003cdiv id=\\\"Sec10\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.1 Operating Under Data Scarcity: Exploratory Patterns and Constraints\\u003c/h2\\u003e \\u003cp\\u003eA defining characteristic of air-quality forecasting in Tashkent is data scarcity, manifested through fragmented time series, uneven station coverage, and limited historical depth. Unlike studies conducted in regions supported by dense observational networks and emissions inventories, the present dataset reflects conditions typical of developing and emerging monitoring systems. This constraint fundamentally shapes model behavior and achievable forecast skill.\\u003c/p\\u003e \\u003cp\\u003eExploratory analysis of the PM₂.₅ time series (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003e) reveals pronounced non-stationarity, strong seasonality, and episodic winter pollution peaks exceeding 300\\u0026ndash;450 \\u0026micro;g m⁻\\u0026sup3;. Summer periods, in contrast, are characterized by comparatively stable background concentrations.\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003eSeasonal distributions (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig3\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003e) show winter median PM₂.₅ levels that are two to three times higher than summer values, consistent with heating emissions and reduced atmospheric dispersion. These features introduce heteroscedasticity and regime shifts that challenge both statistical and machine-learning models under limited data availability.\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003eBoxplots of PM₂.₅ concentrations grouped by season illustrate strong seasonal contrasts, with winter exhibiting substantially higher medians, wider interquartile ranges, and more frequent extreme values than summer. Autumn and spring show transitional behavior. These distributions emphasize the dominance of heating-season emissions and atmospheric stagnation in winter.\\u003c/p\\u003e \\u003cp\\u003eCorrelation analysis (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig4\\\" class=\\\"InternalRef\\\"\\u003e4\\u003c/span\\u003e) highlights a second critical constraint: extreme multicollinearity among persistence-related features. PM₂.₅, short-term lags, and rolling averages exhibit pairwise correlations exceeding 0.9, while meteorological variables show weaker but systematic relationships. Temperature is moderately negatively correlated with PM₂.₅, whereas humidity shows a positive association. These patterns confirm that, in data-scarce environments, predictive information is dominated by short-term persistence rather than independent meteorological forcing.\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003eThe heatmap reveals extreme multicollinearity among persistence-related predictors (PM₂.₅, short-term lags, and rolling averages), with correlation coefficients often exceeding 0.9. Meteorological variables show weaker but systematic relationships, including a negative correlation with temperature and a positive correlation with humidity, motivating the use of regularization and feature-selection techniques.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec11\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.2 Feature Selection, Parsimony, and Interpretability\\u003c/h2\\u003e \\u003cp\\u003eGiven the multicollinearity inherent in the dataset, feature parsimony is essential. RFECV results (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig5\\\" class=\\\"InternalRef\\\"\\u003e5\\u003c/span\\u003e) demonstrate that cross-validated predictive performance saturates rapidly after approximately four predictors, with additional features yielding negligible gains. This finding is particularly relevant for operational deployment, as it indicates that robust forecasting does not require high-dimensional input spaces.\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003eRFECV analysis confirms that predictive performance saturates rapidly with a small number of predictors, with negligible improvement beyond approximately four features. This behavior reflects the strong multicollinearity among persistence-based predictors and supports the use of parsimonious, interpretable models under limited data availability.\\u003c/p\\u003e \\u003cp\\u003eLASSO regression enforces this parsimony explicitly (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig6\\\" class=\\\"InternalRef\\\"\\u003e6\\u003c/span\\u003e). The PM₂.₅_lag_1h feature overwhelmingly dominates the model, while meteorological and calendar variables contribute only marginal corrections. This confirms that, under limited data conditions, the optimal linear model effectively reduces to a regularized autoregressive predictor. Similar observations have been reported in other Environmental Science and Pollution Research studies, where regularization was shown to stabilize air-quality models under noisy or incomplete datasets.\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003eNormalized coefficient magnitudes from the LASSO model highlight the overwhelming dominance of the 1-hour lagged PM₂.₅ concentration. Meteorological and temporal predictors contribute only marginally, confirming that the optimal sparse linear model reduces to a regularized autoregressive predictor in fragmented datasets.\\u003c/p\\u003e \\u003cp\\u003eImportantly, this behavior should not be interpreted as a methodological weakness. Instead, it reflects a realistic constraint imposed by the monitoring infrastructure: when explanatory variables offer limited independent information, interpretable sparse models provide more reliable and transferable predictions than over-parameterized alternatives.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec12\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.3 Comparative Model Performance: XGBoost versus LSTM\\u003c/h2\\u003e \\u003cp\\u003eA central contribution of this study is the direct comparison between ensemble tree-based models and deep learning under fragmented datasets. While LSTM architectures are theoretically well suited for temporal sequence modeling, their performance is strongly dependent on large, continuous datasets. In the present case, LSTM models exhibited higher variance and reduced generalization stability relative to tree-based approaches.\\u003c/p\\u003e \\u003cp\\u003eAs shown in Table\\u0026nbsp;\\u003cspan refid=\\\"Tab1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e, all models achieve their highest predictive skill under the single-station configuration, reflecting the dominant role of short-term PM₂.₅ persistence. Regularized linear models perform comparably to nonlinear approaches, while ensemble models exhibit slightly reduced stability, as indicated by wider confidence intervals.\\u003c/p\\u003e \\u003cp\\u003eXGBoost consistently achieved the highest predictive accuracy across evaluation metrics, while maintaining superior stability compared with LSTM. This finding has practical significance: under real-world data limitations typical of developing monitoring networks, XGBoost outperforms deep learning architectures that are often assumed to be universally superior. Similar conclusions have been reached in recent ESPR publications comparing ensemble learning and neural networks for air-quality prediction, reinforcing the relevance of this result to the journal\\u0026rsquo;s readership.\\u003c/p\\u003e \\u003cp\\u003eRandom Forest models showed comparable trends but were more sensitive to overfitting, particularly at pollution extremes. Regularized linear models, although less accurate in absolute terms, provided robust baselines with minimal variance. Together, these results demonstrate that methodological choice must be guided by data availability rather than model sophistication alone.\\u003c/p\\u003e \\n\\u003cp\\u003e\\u003cstrong\\u003eTable 1.\\u003c/strong\\u003e \\u003cem\\u003ePredictive performance of machine-learning models for hourly PM₂.₅ forecasting using a single monitoring station in Tashkent. Results are reported as mean values with standard deviations and 95% confidence intervals across five-fold time-series cross-validation.\\u003c/em\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cem\\u003e\\u003cimg src=\\\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAA5QAAADhCAYAAAC3MfqnAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAAFiUAABYlAUlSJPAAAKO7SURBVHhe7N15eBPXvT/+90heQiqb3thgy4bgDWzakGA7JCHkBrBsHAIkaQLZGssLlLQJi21IWpK0aXvvt2zyRvJtG0jC0l4oZceSIYQthPQCwZKAfn8kBmwgSDIpgthSScGWzu+PeKYzo5G84A3783oePw/MObN95mg0H82cORxjjIEQQgghhBBCCGknlXwCIYQQQgghhBDSFpRQEkIIIYQQQgjpEEooCSGEEEIIIYR0COevDyXHcfJJhBBCCCGEEEL6qevXr2PAgAGSaQETSj9F/RbFpHtRvHsPlUoFj8dDPzR1M4p7z1Gr1WhqaoJKRQ/y9ISgoCDcuHEDarVaXkRuQXBwML799lsEBQXJi8gtCA0NhcvlQkhIiLyoX7njjjvQ0NCA0NBQeRHpI+688044nU6fhJK+KQkhhBBCCCGEdAgllIQQQgghhBBCOoQSSkIIIYQQQgghHUIJJSGEEEIIuWVlZWUICgoCx3HgOA47duygdyHcgvLyckk8t2/fDq/XK69G2qG8vBzBwcFCTLdt20Yx7QS3dUKZm5uLyMhIWCwWeZEPh8OBxMREZGZmwu12y4t7PaPRCI7jkJiYCIfDIS8mPaw3HZ/2fC56O6PRCJVKhcTERNjtdnlxm3V0OXl5ecJxFc/ndruRlZWF3NzcgBdLFosFgwYN8pm/tLQUJSUlAeftCXycKisr5UWkFzCZTFCr1di5c2evaDv5+fkYNGgQqqure8X2tFVnxtFkMiEoKAgJCQm4dOmSvNivgoICqFQqn/ncbjeys7ORm5sLj8cjmUfMarUiKioKiYmJkvnLyspQUlLSYxfIRUVFaG5uhtFohFqtxpdffnnLMe7PCgsL0dzcLLSzmpqaLo3nzJkzMXjwYBw7dqxL19OTCgsL0dTUBJPJhODgYL8xraqqQkhICBISEnDhwgXFOkpmzZolnBPE87ndbkyePBl6vR5NTU3y2QQnTpxATEwMEhMTJfNXVFSgpKQEzc3N8ll6B+ZHgKJW2e12lpCQwAAwAMxgMEjKzWYzi4iIYABYQkICs9vtkvK20uv1LCIigpnNZnmRD36bdDodc7lc8uI2CRQT+T7zf/z+ycvbGhOXy8V0Op1kmfw+y5fJ/+n1esmyb1eB4t1e/mJ1Ox0ffl2tfWba87loK47jmNfrlU9mrCW2iYmJPvuZkJDAbDabT7lSbCMjIyXzuFwulpmZ6RPX6urqgOvU6/WS7VRaDgC2c+fOgMvZuXMnMxgMLCEhgX355ZcsMzNTsuySkhKWmJjIbDabsC4xf8tdvny5UMbvayCB4q5Evt7ly5dL5leKtdK8gcrEfzk5Oe3avtuJSqViHo9HPrnD7HY7S0pKCnhsBg0aJMT/0qVLfueXl8uXLT4+HdkHl8vFsrKyfNYjl5eXxyIjI9nx48c7vR2o1WrW3Nwsn8yYwv4uW7ZMsp8Wi0USy6+++koy7/Dhw1stay2OLpeLTZo0iXEcJ6m7fft2ZrPZ2PDhwxXLvF4vKy0tZUlJSez06dMsKyuL6fV6YV/5MvF2ifHbKF/2smXL2KVLl9jw4cN99kssKCiINTU1ySe3md1uZyNGjPBZf3x8PLt48aJQLz8/nyUmJkqm9WUhISHsxo0b8smtcjgcLDk5WTGeFy5cED5XBQUFLDExUTKtPVwuF8vOzm51GQUFBWzQoEHs6NGjfusEEhoayv71r3/JJ3crh8PBUlJSFGN6/vx5Yb9mzpzJkpKSJNNYS6wee+wxplKpJPNv3bqVeTwev8vfsmULKykpYcOHD2d///vf2WOPPcZycnLYzZs3GWOMlZeXs+HDh/usj+dvuYsXL2ZfffUVS0lJYQkJCX7n7y4DBgxg169fl09mXXqHcsSIEYiIiMCuXbskdwXLy8uF8r4mISEBdrsdjDHo9XrU1tZi/fr1Qnl7Y3Lw4EHs27cPer0ejDEwxnDlyhWkpqYKdcTrZIxh7dq1kmWQf6Pj03USEhJgs9kksd2wYYNQLo6ty+USpldUVIAxJomtOK5er1eIa1pamlAHsnXysRUPscEvZ/ny5fB6vTCbzYiMjERFRYVkG8TrYYxh2rRpOHXqFOLj46HVahEbGwubzQa32w2j0YiFCxeirKwMMTExwjJ4brcber0e586dkyy3srKy24b/4GO9e/duSTtWijWvuroatbW1ePjhh4XPBZP9IiuP97p167ptn/oK8bFp7XMgZjabUVtbi7Fjx6Kurg4bNmxQPD6XLl2SHJ++POxJR9q52WzGuXPnJHGU381LSEjAV199FTCOn3zyCfbt24elS5fC4/HAYrFg8ODBWLFihXBcX3rpJTQ3NwvLefLJJ8FxHE6dOoVhw4b5nFtMJhNee+01GAwGxMbGStaHlnNLXl4ezp49ix//+MfCsvm7gd0pPj4eFy9eBGMM+fn5uHDhghDLgoICVFZWYsOGDRgyZIh8VqIgPj4eFy5cEOJ34cIF/OUvf4HX68XMmTNRWVmJ9evXY+jQoXTObaP4+HicP39eiOHFixeFmM6aNQs7d+7En//8Z9x9992SmB46dAj79u3D7373OzQ3N8NisUCr1eLdd99FQ0ODcN798Y9/jJs3bwqf76effhr/3//3/2HYsGGIiYlBbGwsHA4HXC4Xqqqq8Prrr2PJkiWKx9DtdqOgoAA1NTV48cUXheWaTKbbZiiaLv2mGTp0KMaPHw+r1YozZ84ALY+eHj58GOPHj8fQoUMl9S0WCyIjI4XnmjmOQ0lJiVBeUlIiTI+MjMSRI0ck8/OPHfJ/Pf344b333iuf1O6Y8PgvnM7CPwK8cOFCJCYmCjH7zW9+g8zMTOH/ubm5wjz+4ssvS6ks0LrEx7Yn9Obj43a7JceB4ziEhYVh3759qK2tRUxMjPD4dmufi57QWbG91bh++eWXAIDk5GRwHIfo6GgMHDgQdXV1kot5Jffee6+kXmxsLFwuFwoLC7Fs2TJMnTpVPgsA4MyZM7BarUhISMDixYvBtXxxTJ06FQsWLJBX7xIdifXmzZsRHx+PN954A5GRkT4X6R3lcDiQlJSEBQsWICkpSWirv/71ryVtnP9RBi2PEapUKsn5xGazCcsbPny45LPBl/Nl8nUZDAaf5Kun8MfmxIkTkmPz2Wef4dFHH1U8NhAdnzfffFM4Pq214da43W5MmjRJiHVkZCQ++eQT/OhHP8LevXtRW1uLIUOGQKfTobGxEaWlpVCr1ULdv/3tbz0aVz6WJ0+exJkzZ8AYE2L5n//5n4qx7Kx2zj8mN2LECHAch6ioKAwcOBDnz59v9biMGjUKFy5ckJxb3G43ioqKsHjxYkybNk04b4idPXsWVqsV8fHxWLx4sZBETpkyBcXFxT5Jb3cZNWqUsL1lZWVYt24drly5ggceeAB6vT7go7sAUF9fj+TkZBQXF2P48OFCe/zVr36FrKws4f85OTnCsqqqqiT94PgfARwOB5KTkyXnD/EPBP7WtWzZsla3s7uI41leXo61a9fiH//4Bx588EHo9fqAjzy63W489thjks/pgQMH8Mwzz+Djjz/GuXPnMGzYMOh0OnzzzTeS/oSRkZE4fPiwzw8sfcE999yjGNOHHnoIOTk5kkdQ5Z9t/rrh/PnzaGxsFC3VF//ZbmxsBGMMMTEx+Oc//4kFCxbgv/7rv/DEE08ofk7PnTsHq9WKuLg4/J//83+EcWIff/xxFBcX3xbjxvruVSebMWMGnE4n9u/fD4h+BZ8xY4aknsPhwPTp04GWXxDtdrtwQWaxWGCxWLB48WJERETAbDZjypQpqKmpkcw/f/586HQ6uFwumM1mNDQ04Be/+IVoLd1HvL0ZGRmSsrbGBADS09ORkJCAffv2ISwsTJLg8fgEg+tAEl1SUoL4+HjhTtKvf/1r1NXVwWQyISIiAiaTCRaLJWB86+vrERER4XPnr7q6OuC6fv/737drWztTbz8+/J21yspKMMbwySef4NChQ9DpdMIdz7179+LMmTMBPxc9wWKxYMmSJYiIiMDEiRMlZTNmzMDVq1dx4MABQBRb/rPPE8c1PDzcb1/F2tpaxMbGCrH11z+S78cTFhaG+Ph4ebFwB4ITJR8TJ05EY2MjYmNjYTQaMX/+fLzxxhuIi4vDyy+/LHw5ydlsNjidTjzyyCPQarXy4m4zffp0XL16Ffv37wdjTBJr+bbzyeYjjzyCRx99FKNHj5Yko7y2xltJaWkp4uLisH79enAtP17V1dXBaDQiMjISVVVVknNNRkYGGhsbYTab0djYiEWLFgkXhHfddZdwpzQ3Nxd1dXUwm82SdQ0bNkxY1x/+8Id2bWtX44/NgQMHwBgT7j4+88wzPscGouPz8MMPt3p8hgwZIhwfPgn3h7/Lxr/oY9u2bQgPD8e2bduQmZkp3PHkf8hasmQJ/uM//gPHjx/H1KlTfdbfE+TtnL8D6a+df/bZZ5I4njx50qcPVW1trXAXIVHWR1GOn1fp3PLnP/9ZeKEK/5QEAOHcMnToUFRWVmLu3Ll46623cPfdd2P27NmKF5wQnVsefvjhHj23iFksFixduhTf//73MXHiRBQXF0vuyv7pT39q893T8vJy3H333fjzn/8MtVqN//qv/8K5c+ewY8cODB48GLt27YLZbIbD4UBRUREeffRRXLt2DRaLBdevX8cbb7wBu92Ou+66S7jbx989lffzFa8rKCgIf/zjH4VrmJ504sQJLFmyJGA8AyUX/N21LVu2wOPxYPv27fj+97+PzZs3IysrC4kt/fL27duH8+fPY+nSpRg4cCCOHj2KJ554AmfPnpUv8rZ34sQJYT/5mDY1NQkx/fOf/4zg4GD5bMJnW6PRIC4uzud88j//8z8ICQkBx3FYunQpmpubMWHCBLjdbsTHx2P79u145ZVX8Pbbb2PIkCFC30ol/Gd77NixwvXi7Ub5rNWJ+AtDPnnYtGkTEhISkJ6eLqnHX+wsWrQIqamp0Gq1eOWVV4QLez7YfPnatWuh1+t95p88eTI0Gg2GDx+O0aNH4/Dhw226gO8sfPKQlpaGgQMH4tSpU5LHH9GOmACAVqvFiRMnoNPpgJYL37YmJW2h0+mwfft2TJ06VUhWDh8+jMcffxyLFi2C0+mEzWYLGN/o6GgcO3ZM+ILjEy/+7pB8Xc8//zz0ej0aGhpQX18vqdPVbrfjM3/+fDgcDjz66KM+2wnRScjf56I78clGWloawsPDcfLkSZ9HVNPS0oTY2u12bNq0CfHx8UhPT5ecQLVaLaxWKzIzM4GWuCYlJbU7KXjxxReRkJCA1157DSqVCmFhYdi7d69QrnQR+NprryEvLw+jR4/GP/7xD7CWx20PHjwIk8mEoqIipKamgmu5gy+/AOnphJ7Ht+M//OEPcDgcwp0ZpXYsTjbDwsIwefJkyUV6Z9DpdNi2bRumTZsmnGs+/fRTTJkyBYsWLcLVq1cl55rHHntMcq757LPPYLfbkZqaiqNHjwqPG4vPN/y28ueaF154AXq9Ho2Njd1+rglEfGzknwMlfMI5ffp0aDQaPPbYY5KE9FYVFRXBbrfjP//zPxXPMxCda37+858jLS0Na9asQW5ubo9f+PDnFD4hELdz+baJk01xO+9IHJ9//nkkJibi5z//OdRqNcLCwrBnzx7JBajY66+/jry8PHg8HowePRqXL18Wzi2ffvopTCYT5s+fj/vvvx8qlUpyN44nT3x7Ul1dHe6++26kpaXhzjvvhMViwZgxY3xi3h4TJ07Eli1b8MQTTyAzMxOJiYn45JNPMHXqVPziF7/AN998IzlHZGdnIzw8HElJSRg9ejT+93//F4MHD8bf/vY34QeB6dOnQ6VS+ZyX+XW9+OKL0Ov1cLvdPXqOqKurw7Bhw5Camoo777wTZrMZ999/f4fjWVxcjEuXLmHcuHHC95Wc3W6H0+nEa6+9hvT0dHz44YfIy8vz+4PG7aaurg5xcXFITU3FgAEDUF1d3aY2+vzzzyMpKQmLFi1CUFAQwsLCsHv3buEHobCwMJ8E8xe/+AXy8vLwgx/8AHa7HV6vF06nE0eOHEFVVRXmzJmDhx9+GCqVCi+99JLPS3nOnDlz298Z7vJWo9Vq8cgjj6C2thbvv/8+TCaT4i/38uQDLY+p8ZTKxfjyhQsXghM9Htjd+LtHBoMBtbW1indI2xoTnkajwd69e+FyuaDT6XyWy6+TMYZz5875Xc6tCBRf+eOZ06ZNk83de9wux2fq1KnCnV7+bV9KSWprn4vulNDSv46P7aJFi+RVJLH94IMPAsZWo9Hg448/hsvlQmZmprBM8QUVv04+tvI+jVqtFmfPnoXL5YKjpS9DZmYm4uPjERYWJqyD/6WS72Mpf9SWv6P94Ycf4q9//Svi4uLwySefoKqqyudtrXy/Lfkyuhsf67q6ulbb8ebNmwEATzzxBDiOw4IFC8AY83kcsLV4dwa+Tfv7EcDd8oZd/o7y1KlTe81FdltFR0dj3LhxqKurEz4HDz/8sN9fpjdv3gzW0gdPpVIJx0feHzlB1Ify3Llziv3wxKZMmQK9Xo+6ujoMGTIk4F3N3nSuERO38w8++ABGo9HvHTx5HIuLi+H1ehXbOf+I5Llz5xT7AWq1WtTU1KCxsVE4t0yaNAkJCQnQarXYs2eP0H+a71+pdE6wWq1YsmQJVq1ahS1btmDYsGE4cOAA9uzZg9LSUslFJv8Int1u91lOd4tv6UNZWlqKixcv4o033vBJgLsKn1iLk/mPPvoIXq9XeOSTPz9MmTIl4COivUV8Sx9KPp5vvvlmh+L5+OOPQ6/X48KFCxg2bBgSExNx8eJFeTWgl/1A0RXiW/pQlpWV4auvvsKbb77ZprYQHR2N06dPSz7bjz32GBISEhAeHg6NRoNdu3bB4/GAMQar1QqtVguHwyH5XPJ3Rv/whz9gx44dGDp0KPbt24d9+/ahtLRUsi3849d2u73VR+Z7qy5PKCH6BflXv/oVnE6n4qOD4uSRJ/4CUyoX48v5x9X4v7ZewHe2l19+GTqdTnhkVK4tMZHTaDQoLCyUT+4WgeK7c+dO7Nu3Tyi7HYYcuB2Oz9q1a8H8vDyI19rnoie8/PLLyMzMhMlkkjyGyOMfReNjq/RomphGo8H8+fMD1mmNRqNBdHS00L8xNjYWGo1GXk2Rw+HAjBkz8Pjjj2PChAmw2+2IjY1FamoqRo8e7bNdsbGxiIiIUHwksbvxjxK//fbbuHLlimKsLRYLTCaT4oVFT+wD36b5RwTF55rY2FisXLkS+/btw7Jly+D1eoW+3bebGTNmgOM4ybFREuj4iPthdtTq1avh9XqRl5fn92U/6KXnGh7frn/9618HbOdVVVWK+3bixAmfO1htxZ9b+P6NMTExCAsLk1dT5HA48OyzzyI7OxsTJ06Ew+Fo07mFP+5K+9LdfvKTn0Cn02H37t0wm83dsk18Ys2/EIk/R9TW1mL37t3Yt28flixZAo/HA1PLcBu3i5/85CfIzMzE7t27YbFYOhTPDz74AB6PR3i5z4YNGxSTUz6Ofd2sWbOQmZmJjz76qF0xFX+2T5w4Aa1W2+brhvr6ejz//PPQ6XSYMGECHA4HtFotRo8ejdGjR/vcBeY/23x/8NvxbmW3JJQTJkwQHglM8PPoIB9MeZ/JhIQEvPjii5LyTz/9FJmZmVi3bp3P/OJ+eRaLpdtegiGn0WgwefJkOJ1O4Q2hYm2JCUQvIjIajbBYLMjLywNECQ9kffS4LnrZTVvj63a7Ffe3t+ntx6ekpARGoxEQveBGo9EgNjZWSC537drV6ueiJ2haHslzOp2oqKiQF7cptnxcKysrYbFYkJ+fD8aYz4VirahPH9fS/1GMb6Os5S7BpEmTwBgTElSj0Sg8SsuXX7lyRXjcEgAWLVqEuLg4vPvuu5J1u1wu1NXV+Xw5paamYsqUKXA6nUhPT0dmZiZ+85vfQKVSSV480x3aEusDBw7A6XT6JHAlJSU+j70qxbuz94dv0/yjumg5jgsXLvRZ1+1yvlEyfvx46HQ6cC0vDVF6RBMtx+fKlSt+j4/4cc1aUR/Kthyf0tJSGI1GMMYwatQoYTp/rqmrqxPONTExMYiIiMDSpUtx6NAhZGVlCT969TR5LNPS0nxiyceRf/kKH8fS0lJcu3bNJ478I5OcrP8jz2q1YuHChfB6vbBarcjOzobH48G8efNQVVWF4cOH49KlS7BYLMjOzsbXX3+Nxx57TJJsvvnmm7j77rvxzjvvSC4w/Z1bRo8ejSlTpuDatWsYM2YMMjMz8etf/xpqtRo5OTk+29gd+PP9tWvXsGLFim7ZBr4tvvfee8ITE1arFa+99pokcXK73VixYoViMtVbfe9735PEs73bXl5ejsrKSni9XowaNUpoV+LP9IYNG7Br1y5ER0cjIiICy5cvxyeffILs7GysWbOmW45hd/re976H7OxsfPPNN1ixYkWrdymtVitef/11NDc3w2q14vHHH8fNmzcxZ84cBAUFoaqqCsnJybhw4QKsVismT56M+vp6TJo0Sfh8v/XWWxgyZAgqKiok/TPdbjfOnz/v89m+77778Pjjj+Obb77BQw89hMzMTLz99tsIDg5WfES2N+qWhJK/eAeAV155RfGOYWpqKtasWQOn04m0tDSh79XmzZuh1WqRmpoq9Ol79NFHUVdXJ0lm+PnFF+9paWm4cuWKaC3dKyMjAxERETh8+LDPs/ltiQlEvwpPmzYNaWlpcDqdMBgMft8wCQAnT56UT7plgeLL91NbuHAhwsLChC/13q43H5/k5GThTX8LFy6ETqfDyy+/LCSqCxcuRElJCYYPHx7wc9FTMjIyhDfGyR/V5S9A0BJbpccm+beyPvHEE0I7W758eauPU586dcrnRM2/cIdfzurVq4XzS2xsLBoaGhDb0vfzypUr0Ov1WLBgATiOQ2lpKUwmE5YtWyY8Ijt//nz86U9/QmxsLBhjePHFF30uXvn+ZQCwb98+/PrXv8Zdd92FwsJCn7pdiW/HHMfhZz/7mU87drvd2LVrF+666y5kZGRItm3ixImIUBiSQawtbbm9Ro8ejTVr1qCurk5IXtPS0vD1118DAF544QUkJibitddeQ3h4OBhjPv1gbwfiz8HPfvYzxc+B2+3G7t27cdddd2HixIk+x6e1t5SePHnS5/MglpycLHmMduLEiZg9eza4lr5nHMfh9ddfh8FgQGJiIn7xi1/g2rVrmDBhAmpra1FUVNSt7dkfcTv/6U9/6hPLtsbR36Nmp06dkk8CAOGFM6mpqfj666/xwQcfID09HUOGDBFeusO33ZycHMmbWMvKymA0GrFkyRLhMbp58+bhf/7nfzBkyBA0NzfjhRde8LmT8eGHHyI3NxcqlQr79+/Hb37zG/zHf/wH5s+f71O3u/Ax/Nvf/tbufu4dMXr0aHz44Ye4ePEi7r77bqhUKqSmpgp3hfi2OnDgQDQ3NyMhIaFXtNO2mjBhgiSegT7DciNGjMDTTz8NtVqNoqIi/Od//idefvllqNVqPPPMM1Cr1fjFL34Bg8GAhIQE/PznP0dDQwN0Oh3OnDnTo+2oK/Ex/d///d82xfTPf/4zQkJCkJqaCofDgVWrVgn9L2NiYuByuRAfHy+Uv/jii8LbWCsqKmA0GvHf//3f+P73vw+NRoO5c+di48aNuPvuu3Hjxg0899xzPi+rev/995GXlwe1Wo0DBw7gt7/9LcLDwzF37lyfur0Rx/xEleO4VgPe31BMuhfFu/dQqVTweDy31ZdyX0Bx7zlqtRpNTU198uLqdhAUFIQbN27cFhdSt5Pg4GB8++23t9VjoLeD0NBQuFyu22bMwK5yxx13oKGhAaGhofIi0kfceeedcDqdGDBggGQ6fVP2QQ6FcSHFf2155JJ0HTo+pC9xtIwzKW/H/F9rj12SruNQGLOTjg3pTg6FcSHFf3xfaNI29fX1SElJCRjP9j4mS0hnoISSEEIIIYQQQkiH0COv7UAx6V4U796DHr3sGRT3nkOPvPYseuS1a9Ajr12DHnn9Dj3y2vf5e+Q1YEJJZH7dyv9J55LHV/5/Qvq6twGIT8W/AaB4xiadjmJP+qpfyZ5P+y0Aeur01vwSgPi3j/8CQE+ekj7q+vXr7Uso/RT1W9xv/n11wd6m2HQ1cbxBMe9RdKesZ6h+owJryWK8b3vBSTIc0pXUv1XDy767yvb8ygMVR3cquxPdoewawcHBaH6jWUgob751E8Hqfw9rQDom9L9DcdNzEwBw460bCFH3zzuVdIey7/N3h5K+ITuAEpvuRzEn/Rklkz2HkknSVzX9somSyU7Wn5NJ0r/Rt2Q7UWLT/SjmpD+jZLLnUDJJ+qqmXzYhSEX9KDsTJZOkP6NvStKrUTJJCCGEdB5KJgkhnY0Syvagl8J0L4o36e/oRTA9h2LfL+Tn56OysvKW3hlhMpkQFBSE5cuXK46paDKZkJeXR+MD9mX0Eh7Sz1FCSQghhJB+p6ysDOvWrUN5eTlcLpe8GG63G9nZ2ZJB5BMTE3Hp0iUAgMPhwIgRIzBt2jR4PB68/vrrSEpKEsr5+adNm4avvvoKbrdbtgZCCOkbKKEkhBBCSL9itVqxceNG/PGPf2x1rNHt27fD6/WCMYZz585hyJAhAACtVouamhqhzGKx4J///CfMZjMYY5g7dy7mzJkDg8HQ6joIIeR2Rmc4QgghhPQrFRUVmD59OqKjo7tsOKTVq1dj2rRpXbZ8QgjpLSihJIQQQki/YTKZYLPZMHv27DbdOXzqqaegUqkQGRmJ48eP++1vabPZEB4ejrS0NEoiCSH9SutnUkIIIYSQPsDtdqOiogJz5sxBWFiYvFhCo9Hgo48+Eh5pffPNN/Hggw9i586dPkmlw+FAUVERZs+ejZiYGElZIHw/THE/TfGfvxf9EEJIb0IJJSGEEEL6hVWrViEmJgZTp05t913EoqIi6PV6bN26VZJQut1u5OXlYezYsViwYEGb7nry5P0w5X+vvfZau5ZHCCE9gc5ShBBCCOkXTp48iXXr1kGtVoPjOEydOhUff/wxEhISAj7O6o/b7cYzzzyD6OhorF69mpI/Qki/RGc+QgghhPQLq1evltwNNBqNyMrKQm1tLe6//34UFBTAYDAoPmZqtVpRVVWFZ555Rri7OXfuXHg8HqxYsYKSSUJIv0VnP0IIIYT0e263GzabDadOnQJaEsioqCihP2NqaioWLlwoPC7LJ5j79u3D97//faGeTqdDQ0MD8vPzoVKpUFxcjL179+L73/++UEYIIX1JmxPKQ4cOwWKxyCfD4XAgMTERRqNRXkQ6gdFoRG5urnwyjEYjOI5TLGsv/hhmZmb6HXiZX1+gOv1Bbm5um9q6xWJBZGRkh+PV1vX0NKPRCJVKhczMTMWBwXkWiwWDBg1Cbm6u30fK/C2Ln1c+/Vbw6zIYDIrb43A4kJSU1K515uXlBTxmrZX3Vp0Z/0BxNZlMUKlU0Ov1isekvzKZTFCr1dDr9Yp3zcQcDgeGDx+OzMxMNDY2youBluXl5eX5LKs96+lLpkyZgj179iA8PBwajQZ79uzB2rVroVKpMHr0aFy+fNlvn0alcsYY9u3bh4EDB/rcDRWX9ScOhwPJycltTqarqqoQHBwMnU6Hb775Rl6Mqqoq5Ofnw+PxyItanfd2UF9fj5SUlDbtQ1vqVlVVISQkBBkZGZI6VqsVWq0WOp0O165dk8zTG9XX12PkyJG3vL18PJYuXarYhtBSp6CgAE1NTfKiDhFv+9WrV+XF7RZo+wKVdSnmh7xIr9czg8EgmcYYY3a7nSUkJLDKykp5UZ8jj0lXcrlcTKfTMQBMp9Mxl8vFmCjeer1e+HdCQgKz2+3yRTDGGDObzSwiIoIBkPzxyzQYDCwiIoKZzWZmMBgYAMmx5Nchrq/UDrpCd8a7Lfj4iI+HEr1eLxyTyspKIabiY8r/KR27tq6nO3Ecx7xer/B/u93OEhMTmU6nY42NjX7bhcvlYpmZmSwhIYHZbDam1+tZREQEq66uFuoEWlZubq4wb2VlJeM4ju3cuVO0hvbh1yU/BjabTahTUlLCIiMjWXV1NTMYDG1aZ0lJCeM4TtgHOXk5H5dA28EU4t7d/MVfvk1ms5lFRkZK9odvw3w8/MWVPyZ6vZ7ZbDaWmJioGIvuplKpmMfjkU/uNna7nSUlJbGcnBxms9lYUlISS0hIYJcuXZJXZUwU3+PHjwvtbceOHcKxcrlcLCsri3EcxzIyMlhDQwNjHVhPd1Gr1ay5uVk+mdyioKAg1tTUJJ/cZUpLS9mgQYPYsWPHWElJCVOr1Wz79u0+5xDGGHM4HGzEiBEsIyODXbt2jZWWljKDwSB8Dl0uF8vOzhba8LVr19o8b1cLCQlhN27ckE9ut7KyMjZ48GB27NgxVlpaytRqNdu2bZvifvB1jx49qljX4XCw5ORkISZlZWXMYDCw5uZmVlBQwJKSktiFCxeY0WhkwcHBbOvWrYrraY/Q0FD2r3/9Sz75lsn3NSgoSHF7XS4Xe+yxx5hKpRK+h+Lj49n58+eZ3W5nKSkpjOM4nzLxeZKff+LEiezq1auS5XdEeXk5i4qKYkeOHBG2fcuWLT7bzhhjVquVabVayTYCYBMnTmROp5O5XC42efJkYfucTqcwb6CyzjRgwAB2/fp1+WTm94odbUwo+xN5TLqSXq9nlZWVzGAwBEwszGYzGzFiBDObzfIiRXxSYzAYFOeVr0+v1zO9Xi9aQvfpzni3xmw2szFjxrBVq1YFPB52u52NGTNGElM+hnzsA/340tb1dDd5YtPWdlFZWemTHOTm5krOJbm5uUyv1/tcYNjtdvbAAw9Ikk9/dTuKT4T4hNFsNrPk5GTJOktKSvwmiqxlngceeEA4ZvJ6fPnKlSt9EsrWElV53LvTrcSf3z+DwcC8Xq/fuGZmZirGS163J/R0QilnNptZSkoKO378uLxIsYyPL5845uXlsZ07dzKDwSCZLmexWIRltXacuxIllF2jOxNKi8XCRo4cyT7//HOhLZWWlrLMzEz2zTffyKuz/Px8ptfr/R73/Px8tmPHDqFtixPK1ubtap2RUFosFvaDH/yAHTt2TIhXWVkZy8rKkuxrW+sWFBQwvV7vc7wdDgcbO3asZF5/ddurKxJKi8XCfvjDH7KjR4/67Ks84eMTwu3bt7d6/rZYLEyr1UoS05kzZ7Lt27ezkpISxeW3l9VqFbadX0d5eTnLyspqU8LHJ4nLli1jTU1NPtsnXkagss7kL6Fs8yOv/rjdbmRmZsJoNMLtduO5557DBx98gMjISHAch8jISOFRWf7RSr6fgfxxwNzcXMn4SyUlJcI6xMtNTEyEw+EQ5uuL1q5di6lTp8on37KDBw+irq4OL774Imw2G5qbmxEdHS2UJycno66uDi6XCw6HA6dPn0ZhYaFkGf1ReXk5nnvuOUms/HE6ndi/fz8g6pNz7733yqspas96eorD4cAXX3zR5nYRHx8vGe9t1KhR2LVrl6SNzZ8/X/EV/k6nEwcOHABksVSq2xmUPhMjRozA+fPnfR7R5FVUVODZZ59FdHS04nbx5VqtVrG8N+PbMmOsXW35k08+QV1dHV544QVwHAe73e4TV/G5htwau92OpqamgPFdvXo1pk2bdtu1QXL74s+nfD9UtJxPL1y44PNIdn19Pb744gvMnTvX78uNPvzwQzzxxBM+bbgt894O+M9xVFSUME0cr+9+Z/93XT62SnX579Y5c+ZArVYLdXhXr17FgQMH4PV64Xa7Ybfbcc899/jEtjdoT1xu1fvvv48nn3yy09qR0mdg+PDhuHjxIlwuV6vbfujQIdTW1uLZZ5+FWq0Wtk/pmAYq6w6dEzERp9OJ3/3udzh16hQYY5gyZQrKy8vhdruRk5ODiooKMMaEL7n33nsPaLlYnDVrltDPoLKyEr///e+FxFG83HPnzkGr1UrW2x9ZLBZkZWVh9uzZSE1NlRcr2rRpEx555BFotVp8+eWXPhf7sbGxiIiIAFpO0gBQXV3t90eA/sBoNMJms+Hll1+WF/nQarWoqKjAwoULkZycjLi4OMTGxmLBggVCHf6iTvxjC9q5np5UX18PxphPu/CXGARKGvg2ZjabhYG9+WVptVqUl5fjtddeE2IZExOD4uJi+WI6zGazYeDAgUhPTwcA1NTUIC4uzu9nQo4/ZrNnz5YXAW0o5y+OIiMjYTab5cU9io//66+/jpSUFMTFxUGr1aK4uLjVi45NmzZh3Lhxwnla6VwTExPjE1eLxYJJkyZh1qxZbT6n9QcWiwXZ2dmYOXMm0tLS5MVtjm9rrFYrJk2aJKynteNMSCA1NTUYNmwYwsPDhWn8+VTetvjvFbPZjODgYHAtLzfy1y9Q7Fbm7U3475/w8HAhPvznWB4vcWyV6l6+fBlo+W4NCQkBx3FCH8ro6GiUlpbirbfewsiRIxEXF4fIyEgUFRX1WDISSGv7quTpp5+GWq1GREQEjh07ppi42e12hIWFdem57syZM7j77rsRFhYmrCPQNYXc1q1b8eCDDyImJqbLtrGzdHpCiZZf5PkLiXvvvRc2mw0WiwWNjY3ChZtGo8HkyZNx8uRJ4f//+Z//KSwjNjYWQUFBwgUnZMvtz/gXvqSlpcHpdGLXrl1tSvIsFguOHDnS5jtLAFBbW4tPP/1U8iPAq6++Kq/WZ7ndbpSXl6OwsBAajUZerGjChAnQ6XSoqamB0+kUpms0Guzdu1f40WTRokVIS0sT7u63dz09qa6uTtIuOI7DnDlz5NUwYcIEJCQkIDY2FlxL8rlw4UJJHb6NeTweybIYY35jKca/6IVfvvwv0It3CgsL8bOf/axD5xW3242KigrMmzdPciHflnKNRoOPP/5YaAtvvPEG7r//flRWVkrq9bS2xF+OP8/4u+ushH/xT1paGq5cuYLdu3e36ZzW11ksFgwePBhpaWn4xz/+IdzZ72xWqxWDBw9Gampql66HkED475UbN27A5XIhODgY8+bN8/viFLFbmbevqqurw+HDh/Htt9/C5XIhJCQE8+bNQ3NzMx599FFkZGTg3LlzbT633w40Gg127doFj8cDxhh++ctfYty4cdi+fbvkZWP19fVYsGABZs6ciSFDhrT5u4p/uQ7/A7j8b+nSpWhubpbP1iEnTpzAkSNH8Oqrr/bKRF+uSxJKseTkZKDlTsDnn38uZNlKF5YlJSVCGZ8sEV+pqam4cuUKGGOw2+2oq6sT7vQGUl5ejoceeqhdv/wnJCRgyZIlQMsHtbCwEIcPH+7zjxzz3nvvPcTGxrb58WOHw4H77rsPhYWFQrJls9kU38a7YMEC6PV6bNq0qd3r6Wnx8fFYvHgx0NIu5s+fj8OHD8Nut0vqyRMnxhgMBgNiY2OFxDkhIQGLFy8Gx3GSZVksFowePRrz588XYmm325GXlydJELVaLc6ePStZh/hv4cKFPl8Wbrcber0e48aNa9MdNyUrV65ETEyM32PWWrlYcXEx9Ho9Nm/eLNm3nuRwODB69GjMmzcPXq8XrpZHlOXxl6uoqMCDDz7YrvMMn8jw57Tz58/jvffeC7ie/iA1NRVff/21EJeLFy9i5cqVnR6X0aNHd8t6CAkkPj4ev/vd76BSqaDRaDBv3jwcOXIEdru91bZ4K/P2VXxM1Gq1EJOjR4/CYrEgPT0dP/3pT9HU1ASXy4UrV66goKCg05Kh3qKwsBC5ubnYvn278OOC2+1Gfn4+xowZ0+67stHR0Th9+rTP25v5v5///OcICgqSz9Yh77zzDtLT05Gamtppj+B2pW7bwtjYWIwZM0b4cPN/a9euBVp+id24caNQbjab23xLuD/TarV45JFHhDu9/jgcDhw+fBgzZswQpsn72ADA/v37ER4ejrCwMKE/jvgucX9z8uRJrFu3TvihY9q0adi3bx/i4uIkj6vyqqurER8fjwkTJgDtSMLbu56exPcV5B+paa9Tp04J/SADtTGz2Yy4uDiMHz8eECWun332WcBYtsbtduNHP/oRYmJisGbNGkkyqdRf8sCBAwgLC/O5y8gfM/6XymnTpmHv3r2Ij4+H2Wxutby3E7dlPtkvLCwMGH/+PDN9+nTJdKVzzYEDB4RzjRh/TuPHAiTfEcdFfoHcnvi2JtB6CGkPpf6SBw4cgEaj8WmXt/K9civz9ib894+4X+DBgweFeMm/q+R9CMV1+f6G/r5bhw0bhgkTJggJ+Ny5c3H06NFemYC3tq/t+UHY7XZjxowZGDRoED744AMEBwfLq3Qqpf6SBw8exPe+972A215fX4/PPvsMTz31VLsS3p7UbQlldHQ0nE4n1q9fLy8CWu5giu9Iyv9PvmOxWHDo0CHJ/00mk+RFGbm5ucILjXjr16+XJDoAhMeP+WPidruxa9cuTJ48WfighoeHo7y8XJhH3AezP1i7dq3kB5DKykrodDqcP39euAMjj7fVasWZM2eE/yv1b4Lo2M2YMaNN6+ktArWLmJgYn3iIlZSU4NNPP8ULL7wAiJbF960GgM2bN+ORRx5BdHQ0Tpw44RNLeR/H9uIfp3333Xd9Tub8Z2LDhg1gLS+iEX8m8vLyUFJSAsYY1qxZ43PMMjMzUVdXh7S0tFbLxfi2MH36dJ9t6klKbZmPvzgWvA0bNiAuLk5IQnn8/q5fv14S1+zsbJw9e1bxnDZq1ChhWn/En+v5+IrjwnEc8vPzhfjzfYDE7Xb37t3Izs5u9bNitVol67FarZL1ENJRfLv8y1/+Irz8RdwuCwoKYDAY4PV6odFoEB4ejhUrVgiPJm7evBljx45ttf/Yrczbm6SlpUGlUvnEKysrC2FhYZg5cyYMBgM8Ho/f2PJ1+e/Wd955R7gzt2XLFjz00EPQarU4efIkampqhM89308xUJLTU/i4bNy4UdjXjz76CFlZWQgPD8esWbOEuMjx5zM+MSssLERzczMqKiq6PJmE6DOwceNGeDweyXESb7v8zvDGjRsxdOhQjB8//ra4OwkEGJdBXqTX65l4TBR+7ELxUAhKwyJUVlYKwx+YFcZE5Ovy8/LTMzMzheEXlJbbE+Qx6UryeKNlTLeamhqWkJAgmS4evoGPlXgaP5ak0rAv/DiJSstiCuNYdudQFt0Z77YSt2fmJ97yY8d/VuSxBKB4TJjCenqafPgKfrgNcbsQD4fBx0NeTz6EiFId8dAbubm5PrG8leEk5OtSWic/3iJfxg+TId43peEUKisrFYfB4InLlbZj+fLlPsuVx7275ebmSmLBx18pFvx4kkr7wRhjRqNRsqycnBzm9XoVxwbly3pSTw8bwo8PKY+Lx+NhrpYxJfV6vbCNRqNRMvYaX5eXl5fnM7ZZRkYG++KLL/yupyfRsCFdozuHDWEt7VKtVgtt66WXXmLNzc3M5XKxSZMmsZycHOE4WywWFhUVJWmf8qFBxG1cXKe1ebtaZwwbwhhjJpOJBQUFSeLV1NTEXC1jcObk5AjHz19dnsViYdHR0UL5xIkTJUOKiGMZEREhGZajo7pi2BDWsq/BwcHC9v74xz9mN2/eZK6WYUJycnLYzZs3hX0Wn+uWLFnCmpubFcv4uPDDg8ycOVPSXuXlHWEymVhISIjitk+ePFnYdp7D4WApKSls6dKlPp9Vf9vndDoDlnUmf8OGcEz887IIx3GSX54JxaS7Ubx7D5VKBY/H0+t+uezrKO49R61Wo6mp6fb5dbiPCQoKwo0bN7r8cS+TyYRNmzbhww8/9DnWJpMJTz75JBYvXowFCxb4lItZrVZkZ2fjnnvuwdatWzFw4MB2lfPrGj9+vGJ5ZwkODsa3337baf28yHdCQ0OFF9/0Z3fccQcaGhoQGhoqLyJ9xJ133gmn04kBAwZIpvs/OxJCCCGE9EFutxuTJk3CtGnT8NVXX0neKuxwODBixAhMnToVHo8Hr7/+OoYPH45Lly5JlsHLz8/H9OnTUV1djeLiYkRERGDHjh3CD6IFBQV+y/l1lZaWwul0YurUqXj//fclb6QkhJDejhJKQgghhPQrc+fOxdy5c2EwGHzuPGq1WqF/GWMMFosFbrcbZrPZ56kZh8OBL774Ahs3bkRsbCymTJmCnJwcbN26VUgYT58+LSnX6/XYunUrvF4v3nzzTYwdOxZ79uzBwIEDUVRU1OrdUEII6W3ojEUIIYSQfmX16tWYNm1apzxO7nQ6sX//fuGFSHa7XfJSKXm5zWbDqFGjcPnyZZw+fRrz5s2jBJIQclujMxghhBBCiB82mw3h4eHCGxvFtFotysrKsGjRIqSkpCAuLg5RUVEoLi6GSqXyW15UVCSM/VldXY2goCBwHAedToeGhgbJOgghpLejhJIQQgghRIHD4UBRURFmz56NmJgYeTEAYPz48dDpdDh79qzicGeByuvq6nD48GHcvHkTLpcLQUFBmDdvnuIQCIQQ0ltRQkkIIYQQIuN2u5GXl4exY8f67dfocDiQlpaGV199FR6PBy6XC5cvX0Z+fj68Xq/f8oKCAng8HsTHx+N3v/sd1Go1NBoN5s2bh7/97W9wOBzyVRFCSK/le3YkhBBCCOnH3G43nn76aURHR2P16tWKySQAmM1mxMXFYfz48eA4TpIU2u32gOX8C34uX77s87IfQgi5nSifIQkhhBBC+qm5c+fC6/VixYoVPslkfn4+DAaDMLSH1WrFmTNnhKSwpqYGcXFxCAsLC1iu1WoxcOBArFixQljWli1b8PDDD0Or1QrrI4SQ3o5jfn4Wo0HlfVFMuhfFu/dQqVTweDw+L6QgXYvi3nPUajWampp8kgnSPYKCgnDjxg2o1Wp5UafIz8/H2rVrJd8xGRkZ2LZtG2pra5GdnY2vv/5aMk9GRgb+9Kc/IS8vD1qtVrhzKV9WREQEdu/ejfT0dHAch4KCAqxZs0ax/MSJE5J1ZWRkYOvWrRg4cKBk3Z0lODgY3377LYKCguRF5BaEhobC5XIhJCREXtSv3HHHHWhoaEBoaKi8iPQRd955J5xOJwYMGCCZTgllO1BMuhfFu/egxKZnUNx7DiWUPaurE8r+ihLKrkEJ5Xcooez7/CWU9E1JCCGEEEIIIaRDKKEkhBBCCCGEENIhlFASQgghhBBCCOmQgH0oCSGEEEIIIYQQALh+/bpPH8qACaWfon6L+400yWZvU3y6EsW796CXw/QM1W9UYPh3u/e+7QUHOgbdQf1bNbzsu6EcAMDzKw9UHD3U013opTxdIzg4GM1vNEueT7v51k0Eq4PF1Ug7hf53KG56bgr/v/HWDYSo+98LeuilPH0fvZSnE7G3GSU33YziTfoz79tesLcZJZM9wPMrD9jbjJJJ0uc0/bIJ7G1GyWQnuvHWDbC3Wb9MJkn/Rt+Q7USJTfejmJP+jO5K9hy6K0n6qqZfNiFIRUOHdKb+eleSEFBCSXo7SiYJIYSQzkPJJCGks1EfynagmHQvinfvQX0oewbFveeo1Wo0NTVBpaLfXXsC9aHsGsHBwfj2228RFEQJZWcKDQ2Fy+VCSEj/vkNJfSj7PupDSQghhBAikp+fj8rKylv68dJkMiEoKAjLly+H1/vvFzkBgMPhwIgRI6DT6dDQ0CApI4SQvoISSkIIIYT0O2VlZVi3bh3Ky8vhcrnkxXC73cjOzoZKpQLHceA4DomJibh06RIgShanTZsGj8eD119/HUlJSUJ5WVkZ7rvvPqxfvx5TpkzBXXfdhR07dtxS8koIIb0RJZSEEEII6VesVis2btyIP/7xj60+1rx9+3Z4vV4wxnDu3DkMGTIEAKDValFTUyOUWSwW/POf/4TZbIbFYsGqVatQVVWF9PR0FBcXw2Aw4N1330VjY6N8FYQQclsLfBYlhBBCCOljKioqMH36dERHR3dJH2WbzYampibJ8keMGIHz588r3g0lhJDbGSWUhBBCCOk3TCYTbDYbZs+e3erdSQB46qmnoFKpEBkZiePHj/t9ZNVmsyE8PBxpaWmoqalBXFwcwsLChPLY2FhERERI5iGEkL6g9TMpIYQQQkgf4Ha7UVFRgTlz5kiSPSUajQYfffSR8Ejrm2++iQcffBA7d+70SSodDgeKioowe/ZsxMTEtPmuJ98PU9xPU/yn9KIfQgjpbSihJIQQQki/sGrVKsTExGDq1KltTvp4RUVF0Ov12Lp1qyShdLvdyMvLw9ixY7FgwYI23fXkyfthyv9ee+21di2PEEJ6Ap2lCCGEENIvnDx5EuvWrYNarQbHcZg6dSo+/vhjJCQkBHyc1R+3241nnnkG0dHRWL16tZD8KfWXPHDgAMLCwlq9M0oIIbcbSigJIYQQ0i+sXr1acjfQaDQiKysLtbW1uP/++1FQUACDwaD4mKnVakVVVRWeeeYZ4e7m3Llz4fF4sGLFCsmdxLS0NHAchw0bNsDr9cLtdmP37t3Izs6mhJIQ0udQQkkIIYSQfs/tdsNms+HUqVNASwIZFRUl9GdMTU3FwoULhcdl+QRz3759+P73vy/U0+l0uPPOO1FWVoZFixZBrVYjLCwMUVFRKCoqokdYCSF9Tqec1RwOBxITE2E0GuVFgpKSEmRmZsLtdsuLSABGoxG5ubnyyTAajeA4TrGsvfjjF+j48OsLVKc/yM3NDdjOeRaLBZGRkX0+XkajESqVCpmZmQFfhW+xWDBo0CDk5ub6faSsrcvqDPy6DAaD4vY4HA4kJSW1a1vy8vICto3Wynsr/ti1Jxb+BIqryWSCSqWCXq9XPCb9lclkglqthl6vV7xrJuZwODB8+HBkZmb6HevQZDIhLy/PZ1ntWU9fMmXKFOzZswfh4eHQaDTYs2cP1q5dC5VKhdGjR+Py5ct++zQqlTPGsG/fPgwcOBBTpkxBc3OzMJ1/1La/cTgcSE5Ohk6nQ0NDg7zYR1VVFYKDg6HT6fDNN9/Ii1FVVYX8/Hx4PB55Uavz3g7q6+uRkpLSpn1oS92qqiqEhIQgIyNDUsdqtUKr1UKn0+HatWuSeXqj+vp6jBw58pa3l4/H0qVLFdsQWuoUFBSgqalJXtQh4m2/evWqvLjdAm1foLIuxfyQF+n1egZA8mcwGBhjjNntdpaQkMAqKysl84gZDAam0+mYy+WSF9025DHpSi6Xi+l0OgZAEjc+1nq9Xvh3QkICs9vt8kUwxhgzm80sIiLC59jxyzQYDCwiIoKZzWZmMBgYAMlx5Nchrs8f967WnfFuCz4+rbVjvV4vHJPKykohpuJjyv/x9QKV9QYcxzGv1yv83263s8TERKbT6VhjY6PfduFyuVhmZiZLSEhgNpuN6fV6FhERwaqrq4U6/pZlNptZZGSkYtttbGyUrKet+HXJ42yz2YQ6JSUlLDIyklVXVzODwcA4jmM7d+6ULEeupKSEcRznd9vk5XxcAm0HU4h7d8vNzRW2q7KyUoiFfJvacqz8xZU/Jnq9ntlsNpaYmKgYi+6mUqmYx+ORT+42drudJSUlsZycHGaz2VhSUhJLSEhgly5dkldlTBTf48ePC+1tx44dwrFyuVwsKyuLcRzHMjIyWENDA2MdWE93UavVrLm5WT6Z3KKgoCDW1NQkn9xlSktL2aBBg9ixY8dYSUkJU6vVbPv27T7nEMYYczgcbMSIESwjI4Ndu3aNlZaWMoPBIHwOXS4Xy87OFtrwtWvX2jxvVwsJCWE3btyQT263srIyNnjwYHbs2DFWWlrK1Go127Ztm+J+8HWPHj2qWNfhcLDk5GQhJmVlZcxgMLDm5mZWUFDAkpKS2IULF5jRaGTBwcFs69atiutpj9DQUPavf/1LPvmWyfc1KChIcXtdLhd77LHHmEqlEr6H4uPj2fnz55ndbmcpKSmM4zifMvF5kp9/4sSJ7OrVq5Lld0R5eTmLiopiR44cEbZ9y5YtPtvOGGNWq5VptVrJNgJgEydOZE6nk7lcLjZ58mRh+5xOpzBvoLLONGDAAHb9+nX5ZOb3ih0KCaX4gtFgMLTrgtdACWW76PV6VllZ2WrczGYzGzFiBDObzfIiRXziwl+wy+eVr0+v1zO9Xi9aQvfpzni3xmw2szFjxrBVq1YFPB52u52NGTNGElM+hnzslX54CVTWG8gTm7a2i8rKSp/kIDc3V3Iuyc3NZXq9XvECQ4xPwpQS147iEyE+YTSbzSw5OVmS8JaUlPhNFFnLPA888IDQNuT1+PKVK1f6JJStJaryuHcnu93OHnjgAUksOnKsvF6v37hmZmYqxktetyf0dEIpZzabWUpKCjt+/Li8SLGMjy+fOObl5bGdO3cyg8EgmS5nsViEZbV2nLsSJZRdozsTSovFwkaOHMk+//xzoS2VlpayzMxM9s0338irs/z8fKbX6/0e9/z8fLZjxw6hbYsTytbm7WqdkVBaLBb2gx/8gB07dkyIV1lZGcvKypLsa1vrFhQUML1e73O8HQ4HGzt2rGRef3XbqysSSovFwn74wx+yo0eP+uyrPOHjE8Lt27e3ev62WCxMq9VKEtOZM2ey7du3s5KSEsXlt5fVahW2nV9HeXk5y8rKalPCxyeJy5YtY01NTT7bJ15GoLLO5C+h7PAjr8nJycK/3W43MjMzJY9z8Y/8cRyHyMhIfP7550JZSUmJz1hL/KOb/OOX/PS+/sigP2vXrsXUqVPlk2/ZwYMHUVdXhxdffBE2mw3Nzc2Ijo4WypOTk1FXVweXywWHw4HTp0+jsLBQsoz+qLy8HM8995wkVv44nU7s378fEPXJuffee+XVblsOhwNffPFFm9tFfHy85CUUo0aNwq5duyRtbP78+cJLLvzh2+4LL7wgL+o0Sp8Jpbc1ilVUVODZZ59FdHS04j7w5VqtVrG8N+PbMmOsXW35k08+EY4Vx3Gw2+0+cRWfa8itsdvtaGpqChjf1atXY9q0abddGyS3L/58yvdDRcv59MKFCz6PZNfX1+OLL77A3Llz/fYx/fDDD/HEE0/4tOG2zHs74D/HUVFRwjRxvL77nf3fdfnYKtXlv1vnzJmj+Jj11atXceDAAeGFUXa7Hffcc49PbHuD9sTlVr3//vt48sknO60dKX0Ghg8fjosXL8LlcrW67YcOHUJtbS2effZZqNVqYfuUjmmgsu7QoYg5HA7Mnz8fr7zyCrRarbwYDocD06dPx6JFi8AYw8cff4y9e/cCLYnmypUrYTabwRiDwWCATqfD//2//xdutxs5OTmoqKgAY0z4InzvvfdkayBoiWVWVhZmz56N1NRUebGiTZs24ZFHHoFWq8WXX37pc7EfGxuLiIgIoOUkDQDV1dX9OsE3Go2w2Wx4+eWX5UU+tFotKioqsHDhQiQnJyMuLg6xsbFYsGCBUIe/qIuMjITFYpHMH6ist6ivrwdjzKdd+EsMAiUNfBszm83CwN7+lrV582aMGzdO8ZzTUTabDQMHDkR6ejoAoKamBnFxcX4/E3J825g9e7a8CGhDOX9xFBkZCbPZLC/uUVqtFuXl5Xj99deRkpKCuLg4aLVaFBcXt3rRsWnTJsmxUjrXxMTE+MTVYrFg0qRJmDVrVpvPaf2BxWJBdnY2Zs6cibS0NHlxm+PbGqvVikmTJgnrae04ExJITU0Nhg0bhvDwcGEafz6Vty3+e8VsNiM4OBhcy8uN/PULFLuVeXsT/vsnPDxciA//OZbHSxxbpbqXL18GWr5bQ0JCwHGc0IcyOjoapaWleOuttzBy5EjExcUhMjISRUVFPZaMBNLavip5+umnoVarERERgWPHjikmbna7HWFhYV16rjtz5gzuvvtuhIWFCesIdE0ht3XrVjz44IOIiYnpsm3sLO1KKBcuXAiO4xATE4OKigrJRbJYfX09IiIi8OKLLwIAUlNTsWjRIqF84MCBwi+p4judZ86cQWNjo3Bxp9FoMHnyZJw8eVKoQ/599zctLQ1OpxO7du1qU5JnsVhw5MiRNt9ZAoDa2lp8+umnkgT/1VdflVfrs9xuN8rLy1FYWAiNRiMvVjRhwgTodDrU1NTA6XQK0zUaDfbu3YuWR82xaNEipKWlwWg0Bizrjerq6iTtguM4zJkzR14NEyZMQEJCAmJjY8G1JJ8LFy6U1OHbmMfjkSxL/AXAt12lO5n8i1745cv/Ar14p7CwED/72c86lKS63W5UVFRg3rx5isMABCrXaDT4+OOPheP9xhtv4P7770dlZaWkXk/z15YDCXSs/OFf/JOWloYrV65g9+7dbTqn9XUWiwWDBw9GWloa/vGPfwh39jub1WrF4MGDkZqa2qXrISQQ/nvlxo0bcLlcCA4Oxrx58/y+OEXsVubtq+rq6nD48GF8++23cLlcCAkJwbx589Dc3IxHH30UGRkZOHfuXJvP7bcDjUaDXbt2wePxgDGGX/7ylxg3bhy2b98uedlYfX09FixYgJkzZ2LIkCFt/q7iX67D/wAu/1u6dCmam5vls3XIiRMncOTIEbz66qu9MtGXa1dCyV+Ymc3mgG8stNlsfhvo8OHDER4ejvXr1wsX67GxsdBoNLDZbPj888+FTJxTuPgk3yXoV65cAWMMdrsddXV1bbqLW15ejoceeqhdv/wnJCRgyZIlQMsHtbCwEIcPH4bD4ZBX7ZPee+89xMbGtvnxY4fDgfvuuw+FhYVCsmWz2RTfxrtgwQLo9Xps2rRJXhSwrDeIj4/H4sWLgZZ2MX/+fBw+fBh2u11ST544sZanEvjPPFra2OLFi8FxnGRZ4jZWUVGBBx98ULHtarVanD17VrIO8R//Q5iY2+2GXq/HuHHj2nTHTcnKlSsRExPjt220Vi5WXFwMvV6PzZs3Kya/PcHhcGD06NGYN28evF4vXC2PKOfl5QXcxkDHyh8+keHPaefPn8d7770XcD39QWpqKr7++mshLhcvXsTKlSs7PS6jR4/ulvUQEkh8fDx+97vfQaVSQaPRYN68eThy5AjsdnurbfFW5u2r+Jio1WohJkePHoXFYkF6ejp++tOfoqmpCS6XC1euXEFBQUGnJUO9RWFhIXJzc7F9+3bhxwW32438/HyMGTOm3Xdlo6Ojcfr0aclYtuK/n//85wgKCpLP1iHvvPMO0tPTkZqa2mmP4HalDm1hamoqpkyZ4vdiN9Dt3DNnziAiIgJutxthYWEoLCzE2rVrgZb5xowZI5wA+D++nPjSarV45JFHWr2L63A4cPjwYcyYMUOYJu9jAwD79+9HeHg4wsLChLvI/GOJ/dHJkyexbt064QeOadOmYd++fYiLi1N8JLW6uhrx8fGYMGEC0EeTcL6vIP9ITXudOnUK9957LziOa1Mb49vu9OnTO5T4ybndbvzoRz9CTEwM1qxZI1mmUn/JAwcOICwszOcuI982+F8qp02bhr179yI+Ph5ms7nV8t5O3Jb5ZL+wsBCfffaZ37YsPlZiSueaAwcOCOcaMf6cxo8FSL4jjov8Ark98W1NoPUQ0h5K/SUPHDgAjUbj0y5v5XvlVubtTfjvH3G/wIMHDwrxkn9XyfsQiuvy/Q2VvlvNZjOGDRuGCRMmCAn43LlzcfTo0V6ZgLe2r+25LnC73ZgxYwYGDRqEDz74AMHBwfIqnUqpv+TBgwfxve99L+C219fX47PPPsNTTz3VroS3J3UooQSAGTNmwGQyKV5UR0dHw+l0Yv369UDLRcbvf/97oOXu5b59+/DEE0+AMSb59V4+H/FlsVhw6NAhyf9NJpPkRRm5ubkoKSkR/g8A69evlyQ6AIRHi/l4u91u7Nq1C5MnTxY+qOHh4SgvLxfmEffB7A/Wrl0r+XGjsrISOp0O58+fF+7AyONttVpx5swZ4f9K/ZsgOnbiJL8tZT0tULuIiYnxiYdYSUkJPv30U+HFOvyy+H7TaOkrKW5j69evR1xcnKTt3gr+cdp3333X52TOfyY2bNgA1vIiGvFnIi8vDyUlJWCMYc2aNT5tIzMzE3V1dUhLS2u1XIw/3p2VNHcWpbbM9zEVx4K3YcMG4ViJ94Pf3/Xr10vimp2djbNnzyqe00aNGiVM64/4cz0fX3FcOI5Dfn6+EH++D5C43e7evRvZ2dk+5x05q9UqWY/VapWsh5CO4tvlX/7yF+HlL+J2WVBQAIPBAK/XC41Gg/DwcKxYsUJ4NHHz5s0YO3Zsq/3HbmXe3iQtLQ0qlconXllZWQgLC8PMmTNhMBjg8Xj8xpavy3+3vvPOO8KduS1btuChhx6CVqvFyZMnUVNTI3zu+X6KgZKcnsLHZePGjcK+fvTRR8jKykJ4eDhmzZolxEWOP5/xiVlhYSGam5tRUVHR5ckkRJ+BjRs3wuPxSI6TeNvld4Y3btyIoUOHYvz48bfF3UkgwLgM8iL5sCHi4SeUhjzgx98DwCIiIthvf/tbptPpWE1NDUtISBDK+D9+XrPCuIm9ZSgFeUy6kl5h3E9/8RMP38AfC/E0fixJpeEWxMdJviymcDwCDZnR2boz3m1VWVkpiYFSvOXHjh/nUx5LiMZyDVTWG8iHr+CH2xC3C/FwGHw85PXkQ4go1REPvcGPUdhZsZCvS2md/HiLfBk/TIZ435SGU6isrFQcBoMnLlfajuXLl/ssVx737pabmyuJBT+GqFIs+GOltB+MMWY0GiXLysnJYV6vV3FsUL6sJ/X0sCH8+JDyuHg8HuZqGVNSr9cL22g0GiVjr/F1eXl5eT5jm2VkZLAvvvjC73p6UncNG2I0Gllubq7i/hqNRqZWq9myZcsUy8UsFgsbPHgwy8jIUBwSo7Vyfl3+yjtLdw4bwkT7xbetl156iTU3NzOXy8UmTZrEcnJyhONssVhYVFSUpH3KhwYRt3Fxndbm7WqdMWwIY4yZTCYWFBQkiVdTUxNztYzBmZOTIxw/f3V5FouFRUdHC+UTJ06UDCkijmVERIRkWI6O6ophQ1jLvgYHBwvb++Mf/5jdvHlTGCYkJyeH3bx5U9hn8bluyZIlrLm5WbGMjws/PMjMmTMl7VVe3hEmk4mFhIQobvvkyZOFbec5HA6WkpLCli5d6vNZ9bd9TqczYFln8jdsCMfEPy+LcBwn+eW5s5SUlODkyZOSx1hzc3Nx7733+n3JT2/RVTEhyijevYdKpYLH4+l1v1z2dRT3nqNWq9HU1HT7/DrcxwQFBeHGjRtd9riX2+3G008/jb1792LixInYtm2b8DZSh8OB8ePHS+7OJyQk4JNPPsGQIUNES/lOfn4+Pv30Uxw8eBAnTpzAk08+iS1btghvcS4oKMChQ4cUy+vr6zF+/HgMHToUW7duxYcffgiv14uioqIuaXvBwcH49ttvO62fF/lOaGio8OKb/uyOO+5AQ0MDQkND5UWkj7jzzjvhdDoxYMAAyfTOP1u1gbg/mdFohMlkQkZGhrwaIYQQQkinmzt3LubOnQuDweCTuGm1WuFxQMYYLBYL3G63MNyZmKNlXN6NGzciNjYWU6ZMQU5ODrZu3QrGGBwt4wGKy/V6PbZu3Qqv14s333wTY8eOxZ49ezBw4EAUFRVhwYIFPttECCG9WbefsRYsWIDNmzcL/TM2bdqEK1eutOuNgIQQQgghHbV69Wph3N9b5XQ6sX//fqH/qt1ul/QBlpfbbDaMGjUKly9fxunTpzFv3jxKIAkht7UeOYOJh72gN7gSQgghpLey2WwIDw9XHABdq9WirKwMixYtQkpKCuLi4hAVFYXi4mKoVCq/5UVFRcJQLdXV1QgKCgLHcdDpdGhoaJCsgxBCerseSSgJIYQQQno7h8OBoqIizJ49GzExMfJiAMD48eOh0+lw9uxZxTG4A5Xzg8/fvHkTLpcLQUFBmDdvnuIbKwkhpLeihJIQQgghRMbtdiMvLw9jx47126/R4XAgLS0Nr776KjweD1wuFy5fvoz8/Hx4vV6/5QUFBfB4PIqDz//tb3/zO9YrIYT0Rr5nR0IIIYSQfox/C2x0dDRWr16tmEyiZZD4uLg4jB8/HhzHSZJCu90esJx/wc/ly5d9XvZDCCG3E+UzJCGEEEJIPzV37lx4vV6sWLHCJ5nMz8+HwWCA1+sFWgZPP3PmjJAU1tTUIC4uDmFhYQHLtVotBg4ciBUrVgjL2rJlCx5++GFotVphfYQQ0tt1+ziUtzOKSfeiePceNB5iz6C49xwah7JndfU4lPn5+Vi7dq3kOyYjIwPbtm1DbW0tsrOz8fXXX0vmycjIwJ/+9Cfk5eVBq9UKdy7ly4qIiMDu3buRnp4ujEO5Zs0axfITJ05I1pWRkYGtW7di4MCBknV3FhqHsmvQOJTfoXEo+z5/41BSQtkOFJPuRfHuPSix6RkU955DCWXP6uqEsr+ihLJrUEL5HUoo+z5/CSV9UxJCCCGEEEII6RBKKAkhhBBCCCGEdAgllIQQQgghhBBCOiRgH0pCCCGEEEIIIQQArl+/7tOHMmBC6aeo3+J+I02y2dsUn65E8e496OUwPUP1GxUY/t3uvW97wYGOQXdQ/1YNL/tuKAcA8PzKAxVHD/V0F3opT9cIDg5G8xvNkufTbr51E8HqYHE10k6h/x2Km56bwv9vvHUDIer+94IeeilP30cv5elklNwQQroTJZM9h5JJ0ldRMtn5+msySfo3+obsAEomux/FnPRnlEz2HEomSV/V9MsmSiY7GSWTpL+ib8l2osSm+1HMSX9GyWTPoWSS9FVNv2xCkIrGouxMlEyS/oz6ULYDxaR7Ubx7D+pD2TMo7j1HrVajqakJKhUllD2B+lB2jeDgYHz77bcICqJksjOFhobC5XIhJKR/J5TUh7Lvoz6UhBBCCCEi+fn5qKysvKUfL00mE4KCgrB8+XJ4vf9+kRMAOBwOjBgxAjqdDg0NDZIyQgjpKyihJIQQQki/U1ZWhnXr1qG8vBwul0teDLfbjezsbKhUKnAcB47jkJiYiEuXLgGiZHHatGnweDx4/fXXkZSUJJSXlZXhvvvuw/r16zFlyhTcdddd2LFjxy0lr4QQ0htRQkkIIYSQfsVqtWLjxo344x//2Opjzdu3b4fX6wVjDOfOncOQIUMAAFqtFjU1NUKZxWLBP//5T5jNZlgsFqxatQpVVVVIT09HcXExDAYD3n33XTQ2NspXQQght7XAZ1FCCCGEkD6moqIC06dPR3R0dJf0UbbZbGhqapIsf8SIETh//rzi3VBCCLmdUUJJCCGEkH7DZDLBZrNh9uzZrd6dBICnnnoKKpUKkZGROH78uN9HVm02G8LDw5GWloaamhrExcUhLCxMKI+NjUVERIRkHkII6QtaP5MSQgghhPQBbrcbFRUVmDNnjiTZU6LRaPDRRx8Jj7S++eabePDBB7Fz506fpNLhcKCoqAizZ89GTExMm+968v0wxf00xX9KL/ohhJDehhJKQgghhPQLq1atQkxMDKZOndrmpI9XVFQEvV6PrVu3ShJKt9uNvLw8jB07FgsWLGjTXU+evB+m/O+1115r1/IIIaQn0FmKEEIIIf3CyZMnsW7dOqjVanAch6lTp+Ljjz9GQkJCwMdZ/XG73XjmmWcQHR2N1atXC8mfUn/JAwcOICwsrNU7o4QQcruhhJIQQggh/cLq1asldwONRiOysrJQW1uL+++/HwUFBTAYDIqPmVqtVlRVVeGZZ54R7m7OnTsXHo8HK1askNxJTEtLA8dx2LBhA7xeL9xuN3bv3o3s7GxKKAkhfQ4llIQQQgjp99xuN2w2G06dOgW0JJBRUVFCf8bU1FQsXLhQeFyWTzD37duH73//+0I9nU6HO++8E2VlZVi0aBHUajXCwsIQFRWFoqIieoSVENLntPmsdujQIVgsFvlkCYvFgueeew5ut1teBKPRCI7jUFJSIi8KWOZwOJCYmIjIyMhW198XGY1G5ObmyicLMVMqay8+xpmZmYrHDqL1BarTH+Tm5sJoNMon+7BYLIiMjOxwvNq6np5mNBqhUqmQmZkZ8FX4FosFgwYNQm5urt9Hytq6rM7Ar8tgMChuj8PhQFJSUru2JS8vL+Axa628t+KPXXti4U+guJpMJqhUKuj1esVj0l+ZTCao1Wro9XrFu2ZiDocDw4cPR2Zmpt+xDk0mE/Ly8nyW1Z719CVTpkzBnj17EB4eDo1Ggz179mDt2rVQqVQYPXo0Ll++7LdPo1I5Ywz79u3DwIEDMWXKFDQ3NwvT+Udt+xuHw4Hk5GTodDo0NDTIi31UVVUhODgYOp0O33zzjbwYVVVVyM/Ph8fjkRe1Ou/toL6+HikpKW3ah7bUraqqQkhICDIyMiR1rFYrtFotdDodrl27JpmnN6qvr8fIkSNveXv5eCxdulSxDaGlTkFBAZqamuRFHSLe9qtXr8qL2y3Q9gUq61LMD3mRXq9nBoNBMk3M5XIxnU7HAEjq8dMTEhJYTU0N0+l0TKfTMZfLFbDMbrezhIQEBoBVVlayyspKBoAlJCQwu90uWXd3kcekK4njyceEMSbERa/XC/8OFBOz2cwiIiIYAMkfv0yDwcAiIiKY2WxmBoNBiDePX4e4fqB20Jm6M95twcdHfDyU6PV64Zjw7bayslJyTPk/pWPX1vV0J47jmNfrFf5vt9tZYmIi0+l0rLGx0W+7cLlcLDMzkyUkJDCbzcb0ej2LiIhg1dXVQh1/yzKbzSwyMlKx7TY2NkrW01b8uuTHwGazCXVKSkpYZGQkq66uZgaDgXEcx3bu3ClZjlxJSQnjOM7vtsnL+bgE2g6mEPfulpubK2xXZWWlEAv5NrXlWPmLK39M9Ho9s9lsLDExUTEW3U2lUjGPxyOf3G3sdjtLSkpiOTk5zGazsaSkJJaQkMAuXbokr8qYKL7Hjx8X2tuOHTuEY+VyuVhWVhbjOI5lZGSwhoYGxjqwnu6iVqtZc3OzfDK5RUFBQaypqUk+ucuUlpayQYMGsWPHjrGSkhKmVqvZ9u3bfc4hjDHmcDjYiBEjWEZGBrt27RorLS1lBoNB+By6XC6WnZ0ttOFr1661ed6uFhISwm7cuCGf3G5lZWVs8ODB7NixY6y0tJSp1Wq2bds2xf3g6x49elSxrsPhYMnJyUJMysrKmMFgYM3NzaygoIAlJSWxCxcuMKPRyIKDg9nWrVsV19MeoaGh7F//+pd88i2T72tQUJDi9rpcLvbYY48xlUolfA/Fx8ez8+fPM7vdzlJSUhjHcT5l4vMkP//EiRPZ1atXJcvviPLychYVFcWOHDkibPuWLVt8tp0xxqxWK9NqtZJtBMAmTpzInE4nc7lcbPLkycL2OZ1OYd5AZZ1pwIAB7Pr16/LJzO8VO9qZUJrNZjZmzBi2YMECyYUwP11+0dxamV6v97mg5i/I9Xq9pG53kcekK+n1elZZWckMBoNPHMTMZjMbMWIEM5vN8iJFfAz5C3b5vPL16fX6fhHv1vBtddWqVQGPh91uZ2PGjJHElI8hH3txwi7X1vV0N3li09Z2UVlZ6ZMc5ObmSs4lubm5TK/XK15giPFJWKDzUHvxiRCfMJrNZpacnCxJeEtKSvwmiqxlngceeEA4ZvJ6fPnKlSt9EsrWElV53LuT3W5nDzzwgCQWHTlWXq/Xb1wzMzMV4yWv2xN6OqGUM5vNLCUlhR0/flxepFjGx5dPHPPy8tjOnTuZwWCQTJezWCzCslo7zl2JEsqu0Z0JpcViYSNHjmSff/650JZKS0tZZmYm++abb+TVWX5+PtPr9X6Pe35+PtuxY4fQtsUJZWvzdrXOSCgtFgv7wQ9+wI4dOybEq6ysjGVlZUn2ta11CwoKmF6v9zneDoeDjR07VjKvv7rt1RUJpcViYT/84Q/Z0aNHffZVnvDxCeH27dtbPX9bLBam1WolienMmTPZ9u3bWUlJieLy28tqtQrbzq+jvLycZWVltSnh45PEZcuWsaamJp/tEy8jUFln8pdQtvmR19bs378fI0eOxI9//GN89dVXOHPmjFBWW1uL6upqSX2eUpnD4cDhw4dRWFgIjUYjTNdoNCgsLMThw4fhcDgk8/Q1a9euxdSpU+WTb9nBgwdRV1eHF198ETabDc3NzYiOjhbKk5OTUVdXB5fLBYfDgdOnT6OwsFCyjP6ovLwczz33nCRW/jidTuzfvx8Q9cm599575dUUtWc9PcXhcOCLL75oc7uIj4+XvIRi1KhR2LVrl6SNzZ8/v9VX+PNt94UXXpAXdRqlz4TS2xrFKioq8OyzzyI6OlpxH/hyrVarWN6b8W2ZMdautvzJJ58Ix4rjONjtdp+4is815NbY7XY0NTUFjO/q1asxbdq0264NktsXfz7l+6Gi5Xx64cIFn0ey6+vr8cUXX2Du3Ll++5h++OGHeOKJJ3zacFvmvR3wn+OoqChhmjhe3/3O/u+6fGyV6vLfrXPmzFF8zPrq1as4cOCA8MIou92Oe+65xye2vUF74nKr3n//fTz55JOd1o6UPgPDhw/HxYsX4XK5Wt32Q4cOoba2Fs8++yzUarWwfUrHNFBZd+iUiLndbuzatQszZszA8OHDMXToUOGCOjU1FVOmTMG0adN8+pP5K6uvr0dQUBBiY2OFurzY2FgEBQWhvr5eXtTvWCwWZGVlYfbs2UhNTZUXK9q0aRMeeeQRaLVafPnllz4X+7GxsYiIiABajgMAVFdXCy8bkB/D/sBoNMJms+Hll1+WF/nQarWoqKjAwoULkZycjLi4OMTGxmLBggVCHf6iTt4vuD3r6Un19fVgjPm0C3+JQaCkgW9jZrNZGNjb37I2b96McePGQavVyos6zGazYeDAgUhPTwcA1NTUIC4uzu9nQo4/ZrNnz5YXAW0o5y+OIiMjYTab5cU9SqvVory8HK+//jpSUlIQFxcHrVaL4uLiVi86Nm3aJDlWSueamJgYn7haLBZMmjQJs2bNavM5rT+wWCzIzs7GzJkzkZaWJi9uc3xbY7VaMWnSJGE9rR1nQgKpqanBsGHDEB4eLkzjz6fytsV/r5jNZgQHB4NrebmRv36BYrcyb2/Cf/+Eh4cL8eE/x/J4iWOrVPfy5ctAy3drSEgIOI4T+lBGR0ejtLQUb731FkaOHIm4uDhERkaiqKiox5KRQFrbVyVPP/001Go1IiIicOzYMcXEzW63IywsrEvPdWfOnMHdd9+NsLAwYR2Brinktm7digcffBAxMTFdto2dpVMSyjNnzqCxsRHp6enCXcRdu3YJicfatWthNpthtVoRFhYmefmOUpnSXQLyb/wLX9LS0uB0OiWxDsRiseDIkSNtvrOEljvIn376KRhjwkX+q6++Kq/WZ7ndbpSXl/vcLQ9kwoQJ0Ol0qKmpgdPpFKZrNBrs3bsXLY+aY9GiRUhLS4PRaOzQenpSXV2dpF1wHIc5c+bIq2HChAlISEhAbGwsuJbkc+HChZI6fBvzeDySZYm/APi2q3Qnk3/RC798+V+gF+8UFhbiZz/7WYeSVLfbjYqKCsybN09xGIBA5RqNBh9//LHQFt544w3cf//9qKyslNTraf7aciCBjpU//It/0tLScOXKFezevbtN57S+zmKxYPDgwUhLS8M//vEP4c5+Z7NarRg8eDBSU1O7dD2EBMJ/r9y4cQMulwvBwcGYN2+e3xeniN3KvH1VXV0dDh8+jG+//RYulwshISGYN28empub8eijjyIjIwPnzp1r87n9dqDRaLBr1y54PB4wxvDLX/4S48aNw/bt2yUvG6uvr8eCBQswc+ZMDBkypM3fVfzLdfgfwOV/S5cuRXNzs3y2Djlx4gSOHDmCV199tVcm+nKdklDu378fn3/+uZBBT5s2Dfv27cPBgweFOqmpqbhy5QoqKyuxePFiyZ0Zednly5fpLmQAfLwYY7Db7airq8N7770nr+ajvLwcDz30ULt++U9ISMCSJUuAfvbIMe+9995DbGxsmx8/djgcuO+++1BYWCgkWzabTfFtvAsWLIBer8emTZvavZ6eFh8fj8WLFwMt7WL+/Pk4fPgw7Ha7pJ48cWKMwWAwIDY2VkicExISsHjxYnAcJ1mWuI1VVFTgwQcfVGy7Wq0WZ8+elaxD/Ldw4UKfLwu32w29Xo9x48a16Y6bkpUrVyImJsbvMWutXKy4uBh6vR6bN29WTH57gsPhwOjRozFv3jx4vV64Wh5RzsvLC7iNgY6VP3wiw5/Tzp8/j/feey/gevqD1NRUfP3110JcLl68iJUrV3Z6XEaPHt0t6yEkkPj4ePzud7+DSqWCRqPBvHnzcOTIEdjt9lbb4q3M21fxMVGr1UJMjh49CovFgvT0dPz0pz9FU1MTXC4Xrly5goKCgk5LhnqLwsJC5ObmYvv27cKPC263G/n5+RgzZky778pGR0fj9OnTkrFsxX8///nPERQUJJ+tQ9555x2kp6cjNTW10x7B7Uq3vIUOhwO///3v5ZOBlsee5NLT0zFw4EDYbDZ5kVDGGENzc7Pw2KzY/v37MXToUAwfPlxe1C9ptVo88sgjOHnypLxIgu+XOmPGDGGavI8NWuIbHh6OsLAw4Q5xf07sT548iXXr1gm/PvE/lsTFxSkOY1NdXY34+HhMmDABaEcS3t719CS+ryD/SE17nTp1Cvfeey84jmtTG+Pb7vTp0zuU+Mm53W786Ec/QkxMDNasWSNZplJ/yQMHDiAsLMznLiN/zPhfKqdNm4a9e/ciPj4eZrO51fLeTtyW+WS/sLAQn332md+2LD5WYkrnmgMHDgjnGjH+nMaPBUi+I46L/AK5PfFtTaD1ENIeSv0lDxw4AI1G49Mub+V75Vbm7U347x9xv8CDBw8K8ZJ/V8n7EIrr8v0Nlb5bzWYzhg0bhgkTJggJ+Ny5c3H06NFemYC3tq/tuS5wu92YMWMGBg0ahA8++ADBwcHyKp1Kqb/kwYMH8b3vfS/gttfX1+Ozzz7DU0891a6EtyfdckLJv1CHb4T8X2VlJQ4fPoz/+q//klwQV1dXo6GhAbGxsXj//fcVy9LT0/HKK6/43Mk0Go1YuHDhbfNYYFewWCw4dOiQ5P8mk0nyoozc3FyfMT3Xr18vSXTQksDzZRD1hZ08ebLwQQ0PD0d5ebkwj7gPZn+wdu1an3at0+lw/vx54Q6MPN5Wq1XyUiql/k0QHbsZM2a0aT29RaB2ERMT4xMPsZKSEnz66afCi3X4ZVVUVAgn282bN0va2Pr16xEXFydpu7eCf5z23Xff9TmZ85+JDRs2gLW8iEb8mcjLy0NJSQkYY1izZo3PMcvMzERdXR3S0tJaLRfj20JnJc2dRakt831MxbHgbdiwQThW4v3g93f9+vWSuGZnZ+Ps2bOK57RRo0YJ0/oj/lzPx1ccF47jkJ+fL8Sf7wMkbre7d+9Gdna2z3lHzmq1StZjtVol6yGko/h2+Ze//EV4+Yu4XRYUFMBgMMDr9UKj0SA8PBwrVqwQHk3cvHkzxo4d22r/sVuZtzdJS0uDSqXyiVdWVhbCwsIwc+ZMGAwGeDwev7Hl6/Lfre+8845wZ27Lli146KGHoNVqcfLkSdTU1Aife76fYqAkp6fwcdm4caOwrx999BGysrIQHh6OWbNmCXGR489nfGJWWFiI5uZmVFRUdHkyCdFnYOPGjfB4PJLjJN52+Z3hjRs3YujQoRg/fvxtcXcSCDAug7xIr9cz8Zgo/NiF/oYP4Mcv/O1vf+szDiI/bIJZYYxE8ZAK8nXKy7ubPCZdSWnfdTodq6mpYQkt43Pyf+L4Kw2twh8LpeEW+HESlZbFFI5Rdw5l0Z3xbqvKykpJDJTiLT92/GdFHkvIxmwVk6+np8mHr+CH2xC3C/FwGHw85PXkQ4go1REPvcGPUegvTu0lX5fSOvnxFvkyfpgM8b4pDadQWVmpOAwGT1yutB3Lly/3Wa487t0tNzdXEgt+DFGlWPDHSmk/GGPMaDRKlpWTk8O8Xq/i2KB8WU/q6WFD+PEh5XHxeDzM1TKmpF6vF7bRaDRKxl7j6/Ly8vJ8xjbLyMhgX3zxhd/19CQaNqRrdOewIaylXarVaqFtvfTSS6y5uZm5XC42adIklpOTIxxni8XCoqKiJO1TPjSIuI2L67Q2b1frjGFDGGPMZDKxoKAgSbyampqYq2UMzpycHOH4+avLs1gsLDo6WiifOHGiZEgRcSwjIiIkw3J0VFcMG8Ja9jU4OFjY3h//+Mfs5s2bwjAhOTk57ObNm8I+i891S5YsYc3NzYplfFz44UFmzpwpaa/y8o4wmUwsJCREcdsnT54sbDvP4XCwlJQUtnTpUp/Pqr/tczqdAcs6k79hQzgm/nlZhOM4yS/PhGLS3SjevYdKpYLH4+l1v1z2dRT3nqNWq9HU1HT7/DrcxwQFBeHGjRtd/riXyWTCpk2b8OGHH/oca5PJhCeffBKLFy/GggULfMrFrFYrsrOzcc8992Dr1q0YOHBgu8r5dY0fP16xvLMEBwfj22+/7bR+XuQ7oaGhwotv+rM77rgDDQ0NCA0NlReRPuLOO++E0+nEgAEDJNP9nx0JIYQQQvogt9uNSZMmYdq0afjqq68kbxV2OBwYMWIEpk6dCo/Hg9dffx3Dhw/HpUuXJMvg5efnY/r06aiurkZxcTEiIiKwY8cO4QfRgoICv+X8ukpLS+F0OjF16lS8//77kjdSEkJIb0cJJSGEEEL6lblz52Lu3LkwGAw+dx61Wq3Qv4wxBovFArfbDbPZ7PPUjMPhwBdffIGNGzciNjYWU6ZMQU5ODrZu3SokjKdPn5aU6/V6bN26FV6vF2+++SbGjh2LPXv2YODAgSgqKmr1bighhPQ2dMYihBBCSL+yevVqTJs2rVMeJ3c6ndi/f7/wQiS73S55qZS83GazYdSoUbh8+TJOnz6NefPmUQJJCLmt0RmMEEIIIcQPm82G8PBw4Y2NYlqtFmVlZVi0aBFSUlIQFxeHqKgoFBcXQ6VS+S0vKioSxv6srq5GUFAQOI6DTqdDQ0ODZB2EENLbUUJJCCGEEKLA4XCgqKgIs2fPRkxMjLwYADB+/HjodDqcPXsWTqdTXhywvK6uDocPH8bNmzfhcrkQFBSEefPmKQ6BQAghvRUllIQQQgghMm63G3l5eRg7dqzffo0OhwNpaWl49dVX4fF44HK5cPnyZeTn58Pr9fotLygogMfjQXx8PH73u99BrVZDo9Fg3rx5+Nvf/gaHwyFfFSGE9Fq+Z0dCCCGEkH7M7Xbj6aefRnR0NFavXq2YTAKA2WxGXFwcxo8fD47jJEmh3W4PWM6/4Ofy5cs+L/shhJDbifIZkhBCCCGkn5o7dy68Xi9WrFjhk0zm5+fDYDAIQ3tYrVacOXNGSApramoQFxeHsLCwgOVarRYDBw7EihUrhGVt2bIFDz/8MLRarbA+Qgjp7Tjm52cxGlTeF8Wke1G8ew+VSgWPx+PzQgrStSjuPUetVqOpqcknmSDdIygoCDdu3IBarZYXdYr8/HysXbtW8h2TkZGBbdu2oba2FtnZ2fj6668l82RkZOBPf/oT8vLyoNVqhTuX8mVFRERg9+7dSE9PB8dxKCgowJo1axTLT5w4IVlXRkYGtm7dioEDB0rW3VmCg4Px7bffIigoSF5EbkFoaChcLhdCQkLkRf3KHXfcgYaGBoSGhsqLSB9x5513wul0YsCAAZLplFC2A8Wke1G8ew9KbHoGxb3nUELZs7o6oeyvKKHsGpRQfocSyr7PX0JJ35SEEEIIIYQQQjqEEkpCCCGEEEIIIR1CCSUhhBBCCCGEkA4J2IeSEEIIIYQQQggBgOvXr/v0oQyYUPop6re43/w7yWZvU2y6mjjeoJj3KHo5TM9Q/UYFhu/avfdtLzhQ/LuL+rdqeNl3Qzl4fuWBiqMHeroTvZSnawQHB6P5jWbh+bSbb91EsDpYXo20U+h/h+Km5yYA4MZbNxCi7p8v56GX8vR99FKeTkSJTfejmJP+jJLJnkPJJOmrmn7ZRMlkJ+vPySTp3+hbsp0osel+FHPSn1Ey2XMomSR9VdMvmxCkoqFDOhMlk6Q/o29K0qtRMkkIIYR0HkomCSGdjRLK9vi1fALpUhRv0t/9BmjpQkm6G8W+X8jPz0dlZWXAd0Y4HA6MGDECmZmZaGhokBfDZDIhKCgIer0eXu93/W5JP/NfADzyiYT0H5RQEkIIIaTfKSsrw7p161BeXg6XyyUvBlrq3HfffVi/fj0ef/xx3HXXXdixYwcYY0KiuXHjRly4cAFHjhzB8OHDcenSJVgsFkRFRYHjOMmfTqdTTEoJIeR2RgklIYQQQvoVq9WKjRs34o9//CNUKuVLIavVipUrV2LXrl1IT09HcXExDAYD3n33XTQ2NkKr1aKmpgbr1q1DbGws/vrXvyI0NBT19fUYPXo0Ll++DMYYGGNwuVyYNGkSJk+ejLCwMPmqCCHktqZ8FiWEEEII6aMqKiowffp0REdH+x0OyWazobm5WbjTCAAjRozA+fPn/d7R9OeTTz5BbW0tnn/+eb8JLCGE3K7orEYIIYSQfsNkMsFms2H27NkBk7uamhrExcVJ7ijGxsYiIiJCUg8tdzOzs7ORn5+PtLQ0nyR1y5YtePjhh6HVaiXTCSGkL/B/JiWEEEII6UPcbjcqKiowZ86cTnn01Gq1IioqCqmpqfj666+xe/dun7uXVqsVR44cwdy5c30SWL4fpkql8ulvyXEcli9fTi/6IYT0epRQEkIIIaRfWLVqFWJiYjB16lSfu4gdIe4rabfb8dVXX2HlypWSJHDFihUYM2YMUlNTfdbJ98P0er1Cf0vx32uvveaThBJCSG9DZylCCCGE9AsnT57EunXroFarwXEcpk6dio8//hgJCQk4fvy4ZPgQpf6SBw4cQFhYmOLdTa1Wi3HjxuHvf/+7MM3hcODw4cN45plnKDEkhPRZdHYjhBBCSL+wevVqyd1Ao9GIrKws1NbW4v7770dBQQEMBgO8Xq/QF3LDhg3wer1wu93YvXs3srOzERYWBqvVikOHDglJqNVqRVVVFe655x5hfX/5y18wbNgwjB8/3ufuJCGE9BWUUBJCCCGk33O73bDZbDh16hTQcsexrKwMixYtglqtRlhYGKKiolBcXAyVSoWoqCjMmjVLuNuZmpqK7OxsLFiwACqVCg6HA3/4wx8wadIkxTuahBDSV1BC2csZjUbk5ubKJwtyc3NhNBrlk9vNaDSC4zhkZmbC7XZLyiwWCyIjIwNuB+kcfKyVjkNvZDQaoVKpkJmZ6fMiCjGLxYJBgwa1Wk8Jv47c3FzJ42i3gl+mwWBQXGZr5WKt1b2Vfe8N2rP9fF29Xq8YC5PJ5DdWfJm/efsrk8kEtVoNvV5/yy9n4ZelFH+Hw4Hhw4cjMzMTjY2NkrK+bMqUKdizZw/Cw8Oh0WiwZ88erF27Vng8dcqUKWhubhbuaK5bt04oU+r/qFTe3/tBOhwOJCcnQ6fToaGhQV7caaqqqhAcHIxly5bd8melJ9XX1yMlJQU6nQ7ffPONvNhHVVUV8vPz0dzcLC8CWikPVNbb1NfXY+TIkdDpdLh27Zq8WNGsWbOwc+dOn/bA73dTU5NkOlqeNIiJiWnXeloj3varV6/Ki9ulqqoKoaGhWLJkieJx48szMjJueV3twvyQF+n1egZA8mcwGPyWR0REMLPZLFnG7U4ek67kcrmYTqdjAJhOp2Mul0tehRkMhoDlTLYc/i8hIYHZ7XbGGGN2u50lJCQIyzAYDMJx5efl6+v1+m49rt0Zb39ai59coPqByljLZ4j/f2VlJQPAKisr5avoERzHMa/XK/zfbrezxMREptPpWGNjo6TdyPH7ZbPZWGVlJeM4ju3cuZOxlnhlZmb6xMRmswnr0Ov1zGazscTERKGso/hlKq2vLeXtWRZjjOXm5irue6D9FpPHvbv52375NvH7w9fNzc1lkZGRrLq6mrFWYtUVx7kzqFQq5vF45JO7jd1uZ0lJSSwnJ4fZbDaWlJTEEhIS2KVLl+RVmcvlYllZWYzjOEl8+br8suTx58tLSkpYZGQkO378OCspKWEcx7EdO3b4HOfupFarWXNzs3wyuUVBQUGsqalJPrnLlJaWskGDBrFjx46xkpISplar2fbt233alsvlYtnZ2ZI2HB8fzy5evMi8Xi+zWCwsKipK0oYBsIyMDHb69Gk2YsQIv/N2h5CQEHbjxg355HYrKytjgwcPZseOHWOlpaVMrVazbdu2KZ6L+JipVCo2ceJEdu3atTaXy8uuXr0qmbejQkND2b/+9S/55FvGx+Xo0aOstLSUBQUFsa1btyrGhVdWVsaCgoIk++dyudhjjz3md79nzpzJkpKS2Pnz55nJZGLBwcGtrqc15eXlLCoqih05ckTY9i1btigu02q1Mq1WK2nLANjEiRPZ//t//4+lpKT4tPPz588zr9fLHA4HS0lJYRkZGczpdLKysjJmMBg6/fM+YMAAdv36dflk5veKHQoJpfiC0WAwCBe/SuV9kTwmXUmv17PKykpmMBgUE0az2czGjBnDVq1apVjO45MYf4mJXq9ner1ePpkxxlhlZaXkGLNuPs7dGW9/WoufXKD6gcrsdjsbM2aMJFkPdGy6mzyxaeu22e129sADDwiJBWtJUvR6PfN6vUIiohQTObPZzJKTkyXLulVms5lFRkYKCa5ca+Vi8rqB9r2xsZFlZma2ulx53LtToO2Xb5PRaGSJiYk+ybTBYPCpy2Sxkpd3xXHuiJ5OKOXMZjNLSUlhx48flxcJCaVSPJVYLBY2aNAgtmPHDsly+XlLSkpYZmYma2hokM/abSih7BrdmVBaLBY2cuRI9vnnnwttq7S0lGVmZrJvvvlGUpdPcHbs2NGmzx1ff9myZT71+eRz+/btPmVdpTMSSovFwn7wgx+wY8eOCfEqKytjWVlZPskiY4wVFBSwHTt2sJKSEsU6gcrlZfLEqqO6IqG0WCzshz/8ITt69KhPXPxtt8ViYWPHjmV//OMf2aRJk4R6M2fO9LvfDoeDPfzww5L1zJw5k+n1enbz5k2hXntYrVZh2/m2WF5ezrKyspjT6ZRX9+FyudjkyZPZsmXLfD63fPLJJ7y3uq1t5S+h7PAzGMnJyfJJpBOtXbsWU6dOlU8WlJeX47nnnkN0dLS8qM0cDgdOnz6NwsJCeZEgPj5e0vfj3nvvxa5du26LxzFvN06nE/v37wdEfXnuvfdeebUe53A48MUXXwRsN2JOpxMHDhwAZPvVH15QwR9TxlivPqb+tGf76Vxxe7LZbGhqakJ0dLTwmUxOTkZdXV2rjzgTEojNZkNzczOioqKEtjVixAhcuHDhlh+rPnToEGpra/H888/3me8Su92OpqYmREVFCdPE8ZI/pv7BBx/giSee8Ps4daDyQGW9TXvjAgDvvvsunn76acTExEimv//++wH3++rVqzh48KDwEi6bzYZ77rnHb/3WKH0Ghg8fjosXL8Llciluuxjfzp999lmo1Wp5sYC/Lnv11VcD1utKHYqQw+HA/Pnz8corr0Cr1cqLSRczGo2w2Wx4+eWX5UV+TZs2DRzHITIyEhaLBWh5phsAqqurhUGU5X336KLiO0rxCyRQfaUyrVaLiooKLFy4EMnJyYiLi0NsbCwWLFggmbc3qK+vB2PMp90otROtVovy8nK89tprwn7FxMSguLhYUk8cE7PZLClDS9+8SZMm4Sc/+QnS0tLkxR1ms9kwcOBApKeny4uANpSLyevy+/76668jJSVFsu/8F8sTTzwRcL97ktL2a7VayfaLtedcYbfbhViJl8Uf51mzZiE1NVUyT39msViQnZ2NmTNnBmz/Tz75JFQqFSIjI1FdXS0vFthsNoSHhyM9PR01NTU+PwbExsYiMjJSMg8h7VVTU4Nhw4YhPDxcmBYbG4uIiAjFcwgAPPXUU1Cr1YiMjMTnn3/u94J78+bNGDt2LGJiYnyWZbfbER4eLryl93ZRU1ODuLg4hIeHC9sdExMTMF79gbgdtSUuVVVV+OqrrzBz5sx2JVfR0dEoKSnBW2+9hZEjRyI+Ph6RkZEoLCxs13LEzpw5g7vvvhthYWHCtvKfgbbYunUrHnzwQcV2zp/H09LS8I9//AOMMVgsFtxxxx3gOK7b+1C2K6FcuHAhOI5DTEwMKioqfC52+XKO45CYmAiHwyEpJ7fO7XajvLwchYWF0Gg08mIfGo0Ge/fuRcvjzVi0aBHS0tKEF/nU1tbi008/BWNMuBh89dVXAQATJkxAfHy80JA5jsPChQsly+/rWoufXKD6gcrQEm+dToeamho4nU75onuVuro6SbvhOA5z5syRVwNa2S+NRoOPP/5YEpP09HRUVlYCope8pKWl4cqVK9i1a5dP0uJwOJCUlCS0Ufmf0gtI+PkKCwvxs5/9TPGHsdbKxfi6P/3pTyV15fvOb4d8v9944w3cf//9wn73FvLt92f8+PGIj49HbGysEHf59wOP/0FSHCv5cd69ezfd2WyJy+DBg4ULBqX2j5b2tGfPHuHlMG+88QbGjBmDnTt3+rR9cVuV/3pPSE/RaDTYvXu30IbffPNNjB07Fjt27PB5oYrVasWRI0cwd+5cnztH9fX1KCoqwk9+8hPhfET6D7fbjXfeeQevvPIKBg4cKC9u1aOPPgqdTodz587hypUrPudP/uU6KpXK51qD4zgsXbpU8UU5HXHixAkcOXJE8a5jfX09FixYgIKCAqGd19XV4fDhw/jnP/8Jl8uFO+64A4WFhYovHuoK7Uoo+Qszs9mMvLw8n4tqvpwxhnPnzrV6EUba77333kNsbGzAx2EDWbBgAfR6PTZt2gQASEhIwJIlS4CWE3phYSEOHz4Mh8PhkwAxxmAwGBAbG9umZLYvksevNYHqi8scDgfuu+8+FBYWgrUkaTabrde+WTc+Ph6LFy8GWtrN/PnzcfjwYdjtdkk9h8OB0aNHY/78+cJ+2e125OXl+Zyo0RKT3NxcbN68GYwxpKamCr+82e12nD9/HitXrpTMo9VqcfbsWUk7Ff/xP3SJud1u6PV6jBs3TvGOW2vlYuK6CxYsEOry+z5v3jxh3x0Oh+K+FxcXQ6/XC/vdG4i33+v1Btx+eYLMGENJSYnPucJfrJSO83vvveeznv4mNTUVX3/9tRCXixcvYuXKla3Gpbi4GLm5udiyZYukrtvtRm5uLh5++OFW2zUhPamoqAh6vR7btm3zae8rVqzAmDFjkJqaKmnDbrcbeXl5ePDBB4WhXUj/8v777yMqKgqPP/54u49/fX09xowZg5dffhlNTU1wuVxwOp2YOXOmkJRFR0fj9OnTkjc7i/9+/vOfIygoSL7oDnnnnXeQnp6O1NRUyb643W4UFBRgzJgxKCoqEtYXHx+P//7v/0ZQUBA0Gg3mzJmDY8eOwWaz+XyGukL7ot0iNTUVU6ZMUbxIJl3r5MmTWLdunfBryLRp07Bv3z7ExcX5PFrZGr7/Jf/oa1ucPHnSbx8q0nHV1dWIj4/HhAkTAIXkvjfh+1pdvnxZXuSjuroacXFxGD9+PCBLPtu7X1qtFo888ghOnjx5SydHt9uNH/3oR4iJicGaNWt8LqpbKxcLVFfpmM6fPx+fffZZu/e9J4i3n+M4oU22dfvl54pAsRLjjzM/FiD5jjguHWn/brcbTz/9NLRaLVavXi1coCj1l9y/fz/CwsJo7ERyS5T6Sx44cAAajabDbau+vh6fffYZnn76aZ9kcvr06Rg8eDBWr17tc0fndjBixAicP39e0i/w4MGDQrz8nTP7OqX+kv7i8ve//x1//vOfERISAo7jhGGBkpKScOzYsYDnTrPZjLvvvhvjx4+HSqWCRqPB3LlzcfToUdjt9oDz+qPUX/LgwYP43ve+57PtYnw75x8B57ndbjz77LOIjIzEBx98gODgYAAQ+mhevny5Q9vZGTqUUALAjBkzYDKZ2p3EkFuzdu1aya8hlZWV0Ol0OH/+vNDnKDc3FyUlJfJZgZZHqEwmE2bMmIGwsDCEh4ejvLxcKN+0aRMeeeQRxbvLJSUlOHz4MF588UV5Ub8hjh+vrfGWk5dZrVacOXNGKP/yyy99+jb1BoHaTUxMjE882rNffEymT58Oq9WKQ4cO+ZTd6gt95syZA8YY3n33XcXlBCrPy8tDSUmJcMIOVBcK+873kZHvu3i/lZbTU+Tb/+WXXwrbL4+FWGlpKQ4fPowXXnhB2B9/sbJYLIrHedSoUcK0/oiPCx9fcVw4jkN+fr7f+PN1n3nmGSHWc+fOhdfrxTvvvCP5tZvvZ7Zhwwawlpcv7d69G9nZ2T7tlJD24NvWX/7yF+ElJ+K2VVBQAIPB4PNIK1rOPVVVVT6J44YNGyQX/bx58+bB4/FgxYoVt2UyiZZ4qVQqn3hlZWUhLCwMM2fOhMFggMfjkc/ap/Fx2bhxoxCXjz76CFlZWQgPD8esWbOEuLz//vuScWNNJhMmTZqEs2fP4oEHHmj1+/XkyZOoqakRzqtK/Tfbg/8MbNy4ER6PR3JMxdsuf0x248aNGDp0qE875x9hLS8vF5JJtPxgHR4ejnfffVdoH3z/y2579Fv+2leevEg+XAQ/BAI/TV7eF8lj0pX0CuN+Kg0PUllZKZnOHxd+SAez2cwiIiIkyxEfJ3m5eFnyMvkQIl2tO+PtjzwG8vi1J96BypjCMe/OMT9bIx++gh/2gd9WfjxKfhgQ8ZAiSvvFDwkhX444JkpjFyoNWdEeSusTb3+gcrvdLuyb1+sNWLexsZGxlqEzxGX8vivNu3z5cp99k8e9u+Xm5krGvOK3X3yclWIhH0dSXi6O1ZdffulznHNycnp0v1kvGDZEaezInJwc5vF4hGFC9Ho983g8zGw2s0GDBknqitsTP0yIPP4ZGRmsoaGBGY1GplKpfNbTk2jYkK7RncOGsJYhhdRqtdC2XnrpJdbc3MxcLhebNGkSy8nJYc3NzYrjTC5dulTSDh0OBxsxYoTPUCFK8/LtWz6URlfpjGFDGGPMZDKxoKAgSbyampqEYVJycnKE41dQUCD53KJlvEJ+nwOV+yvzNwxHW3XFsCGsJS7BwcHCtv74xz9mN2/eZK6WcSVzcnIUh8swmUw+w4aI26N8v+Xld911Fzty5MgtfR+ZTCYWEhKiuO2TJ0/22XZ+PMmlS5dKPquBxqh0Op3MYrFIyvnpnc3fsCEcU/p5EwDHcYq/fPZnFJPuRfHuPVQqFTweT/f8ykUEFPeeo1ar0dTU1O5+OKRzBAUF4caNG11+t8lkMmHTpk348MMPfY61yWTCk08+iRdffBFr1qzxKRezWq3Izs7GPffcg61bt/q8EMThcGD8+PEYOnSoTzk/b3Z2dpc/rhkcHIxvv/220/p5ke+EhobC5XIhJCREXtSv3HHHHWhoaEBoaKi8iPQRd955J5xOJwYMGCCZ7v/sSAghhBDSB7ndbkyaNAnTpk3DV199BbforcIOhwMjRozAxo0bceHCBRw5cgTDhw/HpUuXJMvg5efnY/r06aiurkZxcTEiIiKwY8cO4QfRsrIy3HfffVi/fj2mTJmCu+66Czt27IDL5UJ2drYwr1qtRnR0NI4fP04/phJCbiuUUBJCCCGkX5k7dy7mzp0Lg8Hgc+dRq9WipqYG69atQ2xsLP76178iJCRE8YUXjpYBxTdu3IjY2FhMmTIFOTk52Lp1KxhjsFqtWLVqFaqqqpCeno7i4mIYDAa8++67qKysRG1tLQ4ePIghQ4Zg9erVmDZtGg4ePOizHkII6c0ooSSEEEJIv8Inb53xOLnT6cT+/fuFlxrZ7XbhpVI2mw1NTU3C27Ehepun2+32eUnXqFGj8NFHHymON0oIIb0VJZSEEEIIIQr4/o0FBQU+4x6i5W5mWVkZFi1ahJSUFMTFxSEqKkoYB1Hpzc6xsbGIiIgAAJw/f56SR0LIbY8SSkIIIYQQEavViqioKKSmpuLrr7/G7t27/SZ+48ePh06nw9mzZ+F0OuXFfj366KOIj4/H3XffDa5lbOni4mLFYTQIIaQ3o4SSEEIIIURk9OjRQp9Ju92Or776CitXrvRJ9hwOB9LS0vDqq6/C4/HA5XLh8uXLyM/P96krp9FosGfPHni9XrCWcfNKS0sxdOhQaDQaeXVCCOm1KKEkhBBCCPFDq9Vi3Lhx+Pvf/y4vgtlsRlxcHMaPHw+O46DRaDBv3jz87W9/g91uF/pLiu9uHjhwAGFhYZLHYHmnTp3CPffc4/NoLSGE9GaUUBJCCCGEtLBarTh06JDwplWr1Yqqqircc889QMswIQaDQbgDabVacebMGaG+uN9kWloaOI7Dhg0b4PV64Xa7sXv3bmRnZ/sklGVlZfj000/x/PPP+7x5lhBCejOO+Xk3NQ0q74ti0r0o3r2HSqWCx+OhX827GcW956jVajQ1NdGFfQ8JCgrCjRs3oFar5UWdIj8/H2vXrpV8x2RkZGDbtm345z//ifHjx+PMmTNCWU5ODtasWYPr16/jmWeeQXR0NFavXg2VSuWzrIiICOzevRvp6engOA4mkwlPPvkkPB6PsKzVq1fj1KlTyM7Oxtdffw0AiI+Px6FDhzBkyBBhvZ0tODgY3377LYKCguRF5BaEhobC5XIhJCREXtSv3HHHHWhoaEBoaKi8iPQRd955J5xOJwYMGCCZTgllO1BMuhfFu/egxKZnUNx7DiWUPaurE8r+ihLKrkEJ5Xcooez7/CWU9E1JCCGEEEIIIaRDKKEkhBBCCCGEENIhlFASQgghhBBCCOmQgH0oCSGEEEIIIYQQALh+/bpPH8qACaWfon6L+400yWZvU3y6EsW796CXw/QM1W9UYPh3u/e+7QUHOgbdQf1bNbzs3wPTe37lgYqjh3q6C72Up2sEBwej+Y1myfNpN9+6iWB1sLgaaafQ/w7FTc9N4f833rqBEHX/e0EPvZSn76OX8nQySm4IId2JksmeQ8kk6asomex8/TWZJP0bfUO2E3ubUTLZzSjmpD/zvu0Fe5tRMtkDPL/ygL3NKJkkfU7TL5vA3maUTHaiG2/dAHubUTJJ+iX6liS9GiWShBBCSOdp+mUTglQ0DiUhpPNQH8p2oJh0L4p370F9KHsGxb3nqNVqNDU1QaWi3117AvWh7BrBwcH49ttvERRECWVnCg0NhcvlQkhI/747SX0o+z7qQ0kIIYQQAsBkMkGtVkOv18Pr/ffLl5RYrVZERUW1Wjc/Px+VlZX0QyghpN+hhJIQQggh/YLD4cCIESOwceNGXLx4EUeOHMHw4cNx6dIleVW43W5kZ2fjmWeeQXV1NdRqNaKionD8+HGfpLGsrAzr1q1DeXk5GhsbhSSU4zjJn06nQ0NDg2ReQgi53VFCSQghhJB+QavVoqamBuvWrUNsbCz++te/IiQkBJcvX/ZJEj/55BPU1tbik08+wZAhQ7B69WpMnToVBw8elNS1Wq3YuHEj/vCHPwiPSI8ePVpYJmMMLpcLkyZNwuTJkxEWFiZaCyGE3P4ooSSEEEIIURAfHy9JAO+991589NFHcLlcwrSKigpMnz4dWq3Wb39nPjl9/vnnqV8uIaTPobMaIYQQQvodq9WK7OxsFBQUIDU1VTEZrKurkySPciaTCTabDT/5yU8CJopbtmzBww8/DK1WKy8ihJDbnv+zHyGEEEJIH8P3b0xNTcXXX3+N3bt3w+12y6th/PjxSEhIwN133y30gSwuLhZezON2u7FixQrMmTMH4eHh8tkFVqsVR44cwdy5c32STr5Pp0ql8ulvyXEcli9fHvBFQIQQ0htQQkkIIYSQfkPcv9Fut+Orr77Ce++955O4aTQafPTRR/B6vUJfyNLSUgwdOhRhYWFYtWoVtFotpk6dqnh3k7dixQqMGTNG8S4o36dTvA7x32uvveaThBJCSG9DZylCCCGE9EtarRbjxo3D3//+d3mRopMnT+Kee+4R/r1u3Tqo1WpwHIepU6fi448/RmJiovAmWIfDgcOHD+OZZ56hxJAQ0mfR2Y0QQggh/YLVasWhQ4eEt7RarVZUVVUJSWJ+fj4MBoPP3Uq0DA1y+PBh4cU6q1evltxZNBqNyMrKwrlz53D//feD4zj85S9/wbBhwzB+/Hifu5OEENJXUEJJCCGEkH4hKioKs2bNEvospqamIjs7GwsWLMD169dhs9lw6tQpQNTXku/P+O677+LAgQMYMmSIfLGKHA4H/vCHP2DSpEk0VAghpE9rU0JpNBqRm5srmeZ2u/Hcc8/BYrH4TM/MzATHcSgpKZGUkfZTir1Ybm4ujEajfLIio9EIjuOQmZkpeQGBw+FAYmKiz3Sx1raD+PIXb56/Y9fafL2J0WiESqVCZmZmwDchtkVry8rLy1OMV0fw6zIYDD5jz7WlXIm/7eOXpdfrfZZlsVgwaNAg5Obm+pT1Fvw2+jsuYm2pazKZkJeX57O/JpPJb5z6M5PJBLVaDb1er3jXrD0cDgeGDx+OzMxMNDY2yov7Bb7Porif4rp166BSqaDRaLBnzx6sXbsWKpXKZyzJc+fOBUwmp0yZgj179mDgwIGAaF39vR+kw+FAcnIydDodGhoa5MVEpr6+HikpKdDpdPjmm2/kxRKt1bVardBqtcjJyUFzc3Oby3qj+vp6jBw5EjqdDteuXZMXK5o1axZ27twpOXdWVVUhJCQEL730EpqamiT1u4p4269evSovbreqqioUFBT4bH9VVRVCQ0O7dd8EzA9xkdlsZhEREayyslKYZjAYmE6nYy6XizHGmMvlYjqdjgFgOp2O1dTUsISEBAZAmE+v1zMAkj+DwSAss6vo9Xqm1+vlk9stQLg6nTyefJzFDAZDwHKe3W5nCQkJQj2DwSDE3WAwsIiICGY2m4XliY9zW7ajq3RnvP0R7z//l5CQwOx2u7yqIFC8eUrHri3z9RSO45jX6xX+b7fbWWJiItPpdKyxsdHvtprNZhYZGenzuefna+uyDAYD4zhOMl9H8OuSH0+bzdamcn9KSkp8to9fll6vZzabjSUmJgrLcrlcLDMzU/h/bm4ui4iIYNXV1ZLlyuPe3XJzc4VtrKysZBzHsZ07dypuU2t1+X32d/yV4tSTVCoV83g88sndxm63s6SkJJaTk8NsNhtLSkpiCQkJ7NKlS/KqzOVysaysLMZxnKTdiuuWlJSwyMhIdvz4caG97tixgzU2NrY6b09Qq9WsublZPpncoqCgINbU1CSf3GVKS0vZoEGD2LFjx1hJSQlTq9Vs+/btPucQl8vFsrOzJe0wPj6eXbx4kXm9XmaxWFhUVJTk3AyAZWRksEuXLrFJkyb5nbc7hISEsBs3bsgnt1tZWRkbPHgwO3bsGCstLWVqtZpt27ZN8VzE1z169KhPXT6eiYmJ7MKFCyw/P58NGjSIHT16lDU2NkrKCgoKhLJbjVdoaCj717/+JZ98y+T7GhQUxLZu3aoYF15ZWRkLCgpiEydOZFevXmV2u52lpKSwl156iV28eJGlpKSwhIQEdv78eaGNabVaSTsCIMzfUeXl5SwqKoodOXJE2PYtW7YobrvVavW7DU6nk7lcLjZ58mSmUqmEaYwx5nA4FPetrq7ulo+p3IABA9j169flk5nfK3bILuYNBoNwIc1f+IoTD71e32rCodfrJReKBllS2lVux4RSr9ezyspKvzEym81szJgxbNWqVYrlYv7232w2sxEjRjCz2SxMk6+vte3oSt0Zb3/4hFLc1lvjL948f8eutfl6kjyx6ei28kmF+DyQm5vL9Hq935Oe2WxmDzzwgBCvW0ko5fiEd+fOnfIixtpQztqxfWazmSUnJ7Pq6mpWWVnJEhMTJUlTbm4uMxgMkjjI496d7HY7e+CBByRJrr9j1Za6ubm5bOfOncxgMLDMzMw2xakn9XRCKWc2m1lKSgo7fvy4vEhIKP0l+xaLRZiXLy8pKWGZmZnMZrMFnLenUELZNbozobRYLGzkyJHs888/F9pWaWkpy8zMZN98842kLp8A7dixo02fO77+smXLWENDQ7vm7QqdkVBaLBb2gx/8gB07dkyIV1lZGcvKymLXrl1rV12TycRGjBjBLly4IJQXFBSw5cuXs507d/otu9XPXFcklBaLhf3whz+UJLz8vvpL9CwWCxs7diz74x//yCZNmqRYT2m5Yi6Xiz322GNs2bJlHY6L1WoV1sG3zfLycpaVlSUkg4HwCeSyZctYU1MTmzlzJtu+fTsrKSkJuAx+vUeOHOn0z4S/hLLNz2C8+OKLAID169fjF7/4BR555BFMnToVaHmc4fDhwygsLIRGo5HN2Xb8o5d8f4XIyEjJI7Wtlefm5gplmZmZqK+vR2ZmJtatW4d169bdVo/hrl27VoivkvLycjz33HOIjo6WF0k4HA6cPn0ahYWF8iLYbDY0NzdLlpGcnCwZyLm17SBSgeLNUzp2bZmvt3A4HPjiiy86tK0HDx5EbW0tXnjhBUC03/Pnz/f7woqKigo8++yzrbb1niLePn/7oCQ+Pl7Sr+ree+/Frl27etVjzk6nE/v37wdjDG63GzabDffee6+8GtCGumvWrMG0adPaFSPSOWw2G5qamiRtVH6uJ6Sz8dcYfD9UABgxYgQuXLhwy49cHzp0CLW1tcILkvoCu92OpqYmREVFCdPE8RJ3BbDb7UJs/dWNi4tDeHi4EPtRo0Zhz549aGxsDFjW27QnLrx3330XTz/9NGJiYuRFbca3sWeffbbDbUzpMzB8+HBcvHgRLpdLcdvFxNugVqvx/vvv48knn4RarZZX7XFtjpBWq0VFRQUWLlwIk8kkuZisrq4GAKSnp4vmCMxoNGLhwoVCEup2u5GTk4NXXnlF6K+waNEivPbaa3C73a2WWywWnD59Gna7HYwx7N27F9HR0di7dy/0er3QL2fBggXyTbntGI1G2Gw2vPzyy/IiH/X19UDLMRIn2263G19++aXPRW1sbCwiIiJESyAAhAth+Y8YcoHijQDHrrX5epP6+nowxny2tS0Xpps3b8YjjzwCrVYLiPbbbDYLL8kQL8toNOLSpUuYPXu2ZDmdxWazYeDAgX7PXa2V88ezte2zWCyYNGkSZs2ahdTUVADo9RfzWq0W5eXleP3115GSkoK4uDhotVoUFxf7JIX890Nb6gaiFCfyXVyys7Mxc+ZMpKWlyYsFTz75JFQqFSIjI4XvZQB+z/WRkZHC/+XztnahQ0hrampqMGzYMISHhwvT+GsMf+eFp556Cmq1GpGRkfj888/9tsPNmzdj7NixkoShrfP+/+3df0wT5x8H8PeVitug+NcEymCUOLcYzWgN0YlZEQbEfbc/hJjMbICA/8wfjB9LlmxZsv0zxobKr2SRmekWoxCHZgOkTIYIdTFLpC3qNsUBkaznzz8GjcuU9r5/jHtyd71rKfJTPq+k//S55/rc567XPvfcPZ+F6vr1634dPaPRqBovaWy1lh0ZGdHscA0PD2uWLTRT2VapM2fOYHR0FMXFxZodL6fTia1btyIvLw/r169XXc+pU6ewceNGGI1G1fKpGBwcREJCAgwGA1tHKP+zT506hQ0bNoTUBpfLJdu26XaGQxXSp4hBWLFiRcDRAnFSEfElnaji/fffB8dxePPNN4HJ0RqPx4PBwUGMjY2xkVAASE9Px+joKAYHB4OWA8DQ0JDsR/RJ5PF4UFNTE9Jo8NDQEPr6+iAIAvsDu2fPHuViREVkZCS6urpkFzEsFovq5CsirXgH23da9Rai4eFhWVs5jsPevXuVi8k4HA5cvHjRbzRS3G6v1ytb1/j4OGpra1FaWhpwhkSe57Fq1SrZOUf60ppYh+d5lJaW4t1332Ud3FDKPR4PamtrUVJSotk+cZIai8WCe/fuwWazwePxIC0tDSaTCXFxcaydC/FiV1paGjIyMnD9+nXcv39fWSxjtVqnvKySVpyWOofDgZUrV8JiseDu3bvo6OhQvQghTiYjprD48MMPkZKSgh9//FH12JfSqtva2hq0LiEzJTIyEjabjR2HH330EV555RX88MMPfpNROZ1OXLx4Efv27WOTKU217lJhtVphMplgMpnYxdqysjJMTEzAarUiKSlJtWyx83g8qK+vx+7du9nkWFLiREQWiwW3bt1CZ2en6qiseIzt2bMHer2evS9OriPGTfmqqqqasTi6XC7WBq2OsZTL5YLRaITZbAbP85rbNltC6lDW1NSwg/T48ePs/bi4OOj1ejba8MYbb7A/mhkZGZI1gP25EwQBbrcbw8PD6OnpwV9//eX3JyQmJoYdEMHKzWYzzp49i507d4JTdGKfJIcOHUJcXFxIt6EmJSXh888/ByZP2qWlpbDb7eB5XrkoCaKiogL5+fk4efKksojRivcnn3wScN9p1VuI+8lkMqGyshKYbOt7770Hu90Ot9utXJSpra3Fhg0b/EaekpKSUFlZCY7jZOv69NNPYTQa8b///U+2vFJsbCxu3LjBzivKl3gRS8rj8SA/Px+pqamqo2jBygGgsbERRqNRc39i8rx09+5ddr4bGRnBoUOHEBERgbNnz8rauX//fsTFxalebJgPPM8jOTkZJSUl8Pl8GB8fB8/zqjO0hrKsGq04TaXuk8xsNuPOnTssLjdv3kRjY2PQuJSXl6OgoAAtLS1Bl1V6nLqEzJSysjLk5+fj9OnTfsdhXV0dUlJSYDabVc/NgeouFREREbDZbPB6vew35uDBg0hISEBMTIxmmdbF0cXi8OHDiI6Oxuuvv646MpecnAye5+Hz+cDzPHiex9dffw2v1ytbrqGhAevXr/c7xmJiYvD777/L8s9KXx988IGsA/o46uvrWRvUtkXp5Zdfhtvt9tu2mergBhO8hZPa2tpgt9vR0NCA0tJSVFZWslv/YmJiMDExge7ubmW1gAwGA0wmE65du6YsAiavBASaYlpZbjabce/ePfT396OioiLgrYmL1cDAAHselJsc6f3555+RmJiour3iSLLY2Zd66aWX/G676+7uRlRU1KI/qcyXQPF2uVya+07sNKrVW2jE57Bu376tLNLETz5nvX37dr+TMzS2W4yXeCVQjJfJZEJ/f79y8SnzeDzYtm0bjEYjjh496veHJFi5SPwuStvX1dWl2b7Y2Fhs3ryZ5bhTGhgY0Hw+cT5cunQJJpMJaWlprLNfWlqKCxcu+F3k6O/vn/KywQSL01IljUuof5LVnpfs7u6GwWCgcz2ZNWrPS547dw6RkZHTPu5u3bqFCxcuICcnR/PcvFitXr3a7zbVnp4eFi/p9qo9Q6i1rOjy5ctYu3ZtyGXzLZRtvXLlCo4dO4bw8HBwHMdS+axatQq//vqr7NwZExOD1NRUXL16VTaSLR5j27Ztm9LIYCBqz0v29PQgIiLCr+1SYhvE27hDJW7bb7/9FvLvxXRNqUMp3qq3e/duxMbGIi0tDcnJyaipqQEUz1eGMulNT08PnE4n0tPT2TNK0pHPmpoaxMfH44UXXghaLiUducTkZBdPim+//VZ2NaS1tRUZGRkYGRlhIz8FBQVsPxgMBkRFRbF9BQAnT57E5s2b2ciPGFOPx4OOjg5s3bp1wYySLDQOhwPt7e3Yvn07e2+q8VaOSEn33auvvqpZT+12y/kUaBuNRqMsHqLjx48jMTERVqtV9r64rtraWnbSE5+z/Omnn1TjNTw8HPA5smD27t0LQRDQ0NCgejIPVL5z507s378fgiDg6NGjfu177bXXWPscDgd6e3tZXfHYWbdunWydAHDgwAHY7Xbs2LHD7zPnk9PpZI8UYPJZvMTERBgMBlksgi0bSChxWkrEuIjxlcaF4zgUFhbK4q+s297ejtzcXHAcB4vFAo7jcOLECQiTkybZbDZkZ2f77R+n0ymrS8h0icddU1MTfD6f33FXVFSE6upq1dtSnU4nzpw549dxPHHiBBISEmC1WjVHbbTqLnQWiwU6nc4vXpmZmTAYDCguLkZ1dTW8Xq9mbMVllWpqatDX16c6wUygsoVAjEtzczPb1s7OTmRmZiIqKgq7du1icTl8+DAmJibY73J7ezuysrJw48YNhIeHo6+vjx1v4nGyZs0a2XY3NTUhPj4+4DE2VeJ+am5uhtfrle0naduVo4jNzc0htcHlcqG3t5dtm8vlYts2Z98B5bSvImmRNGWISC03pfieNHeKtF6+Sh7KQPW1PlOtvLW1VbZeaUoDaT1lfrtQSGMy29RipZa2o7W1Vfa+mOZCa/uV6wkUNyGEdsyGuYy3FmXsoDiGQo23lHLfTbXefFCmrxDTaUjbOjY2xtKCSOMh5hnU+u5prUtJjJda2VQpP0v5mYHK3W432za1KcZbW1tl6TDUclrm5eUJPp/P73O08i4q4z7XCgoKZPmwxFyZ0v0sTQuitqzWusS4Xrt2TTNO82m+04aIeSiVcRFzzGVmZgr5+fmC1+sV+vv7hWeffVa27JdffimLYVtbm6DT6fzW5XA4gtadD3OVNqStrU0oKChQ3ddtbW1CWFgYi1Uw4vLp6el+aTECcTgcwsqVK4W8vLxZ3+a5TBsiSGIiHlvvvPOOMDExIYyPjwtZWVlsm9XyTFZVVcnizvO8sHr1auGLL76QvT+VurNtJtKGCIIgtLe3C3q9XhavR48esTQpeXl5bP9pLStMxiQmJoaVmUwmliYkUNnjmo20IcLkti5btoy1+e233xYePnzIUnvk5eUJDx8+VFYT2tvbWdoQMQ+l9HdIXI9IzOdYVVU1Y9/F9vZ2ITw83O8zxZQgyrZL26D8rhYXF8u+T5jMUXn16tWg2zZTtNKGcILa5U0AHMepXvlcyigmc4vivXDodDp4vd65u9JFAIr7vAoLC8OjR4+mdHWYzDy9Xo9///13Wrd7TYXH40FOTg66urqwZcsWnD59ms1GyvM8rFYrNm7ciMrKSmzZsgVerxfnz5/Hc889p1wVWz4hIQEtLS345ptv4PP5UFZWhoGBAWRnZ+POnTuyOunp6fjuu+9QVFSEP//8Ez09Pfj444/R1taGjo4OzZknH9eyZcvwzz//zNhzXuQ/y5cvx/j4OMLDw5VFS8pTTz2Fv//+G8uXL1cWkSfEM888g/v37+Ppp5+WvU8dyhBQTOYWxXvhoI7N/KC4zx/qUM6v2e5QFhYWIicnB4ODg+jo6EBLS4ssvYWU0+nEjh07cOzYMXYLm1RhYSF8Ph+OHDkS9HjxeDzIzc1FZmYmXnzxRZSXl+PcuXOso1pUVIQ1a9agvLw86LqmgzqUs4M6lP+hDuWTT6tDOfNnK0IIIYSQBezIkSMsv/Dj4Hkef/zxB0pKSqa0rvPnz2NoaAhvvfUWdDqdX37QdevWobOzUzU9DCGELFTUoSSEEEIIUeF0OpGdnY2ioiK/FAIAcPv2bQiCgEuXLkGv14PjOGRkZGjOUN/S0oJNmzaxydaUs+8SQshiRB1KQgghhBAJp9OJ6OholgvUZrNpdvyGh4dht9vx6NEjjI+PQ6/Xs5ysUmKy9H379kGn07EE8wkJCeAm00mVl5f71SOEkIWOOpSEEEIIIRLJycls9NHtdmN0dBSNjY2qnT2TyYTPPvsMOp0OkZGRKCkpwS+//AK32y1brq6uDikpKWykMzIyEp2dnbIk6QcOHEB8fDyl7iKELCrUoSSEEEII0RAbG4vU1FRcuXJFWYTo6GhwHMc6n1p4nofdbkdubm7AyXYWcoJ5QgjRon1WI4QQQghZYpxOJ3p7e1kHUUyAvnbtWmByVtfq6mr4fD4YDAZERUWhrq6OLf/9999j06ZNMBqNbJ1NTU14/vnnYbVaNTuLBw8eRF9fH5uwhxBCFgs6YxFCCCFkSSksLIROp0N5eTm6urqwYsUKZGRkYGxsDNHR0di1axd0Oh04joPZbEZ2djYqKirw4MEDuN1uXL58GQAQGRmJqqoq2Gw2hIWFgeM43Lx5E3V1daxTyPM8vvrqK2RlZclmdBWf0xSfn6yvr5elECGEkMWC8lCGgGIytyjeCwflQ5wfFPf5Q3ko59ds56FcqigP5eygPJT/oTyUTz7KQ0kIIYQQQgghZEZRh5IQQgghhBBCyLRQh5IQQgghhBBCyLQEfIaSEEIIIYQQQggBgAcPHvg9Q6nZoSSEEEIIIYQQQgKhW14JIYQQQgghhEwLdSgJIYQQQgghhEwLdSgJIYQQQgghhEzL/wGrkOhBOx8UWQAAAABJRU5ErkJggg==\\\" width=\\\"609\\\" height=\\\"225\\\"\\u003e\\u003c/em\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThe impact of spatial aggregation on model performance is examined in Section \\u003cspan refid=\\\"Sec14\\\" class=\\\"InternalRef\\\"\\u003e3.5\\u003c/span\\u003e and summarized in Table\\u0026nbsp;\\u003cspan refid=\\\"Tab2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003e.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec13\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.4 SHAP-Based Insights into Pollution Drivers\\u003c/h2\\u003e \\u003cp\\u003eBeyond predictive accuracy, interpretability is essential for policy relevance. SHAP analysis applied to the XGBoost model (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig6\\\" class=\\\"InternalRef\\\"\\u003e6\\u003c/span\\u003e) reveals that short-term persistence is the dominant driver of PM₂.₅ forecasts in Tashkent, with PM₂.₅_lag_1h exerting the strongest influence on model output. High lagged concentrations consistently produce large positive SHAP values, directly amplifying predicted pollution levels.\\u003c/p\\u003e \\u003cp\\u003eMeteorological variables, particularly humidity, exhibit secondary but interpretable effects. Elevated humidity tends to increase PM₂.₅ predictions, consistent with enhanced secondary aerosol formation and reduced dispersion under moist conditions. Temperature effects are generally negative, aligning with improved atmospheric mixing during warmer periods. Temporal indicators (hour, month, season) cluster near zero contribution, indicating that their influence is largely mediated through persistence and meteorology.\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003eSHAP values quantify the contribution of each feature to PM₂.₅ predictions. High values of lag-1 PM₂.₅ exert strong positive influence on forecasts, while meteorological variables such as humidity and temperature show secondary but interpretable effects. Temporal indicators cluster near zero, indicating limited independent contribution beyond persistence.\\u003c/p\\u003e \\u003cp\\u003eThese SHAP-based findings provide actionable insight into dominant pollution drivers, moving the analysis beyond black-box prediction. For policymakers, this suggests that short-term PM₂.₅ exceedances are more effectively anticipated through persistence-aware monitoring rather than reliance on meteorological forecasts alone.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec14\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.5 Single-Station versus Aggregated Multi-Station Forecasting\\u003c/h2\\u003e \\u003cp\\u003eBuilding on the single-station results (Table\\u0026nbsp;\\u003cspan refid=\\\"Tab1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e), this section evaluates model performance after aggregating data across multiple monitoring stations. As summarized in Table\\u0026nbsp;\\u003cspan refid=\\\"Tab2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003e, spatial aggregation substantially improves predictive accuracy and stability across all models, reflected by markedly higher R\\u0026sup2; values at the city scale. This improvement arises from the smoothing of localized noise and short-lived extreme events, which enhances model generalization while reducing sensitivity to station-specific PM₂.₅ peaks. Regularized linear models exhibit the highest stability under aggregation, while ensemble and deep-learning models retain strong performance with slightly increased variance.\\u003c/p\\u003e \\u003cp\\u003e \\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab2\\\" border=\\\"1\\\"\\u003e \\u003ccaption language=\\\"En\\\"\\u003e \\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 2\\u003c/div\\u003e \\u003cdiv class=\\\"CaptionContent\\\"\\u003e \\u003cp\\u003ePredictive performance of machine-learning models for hourly PM₂.₅ forecasting using aggregated multi-station data across Tashkent. Results are reported as mean values and standard deviations across five-fold time-series cross-validation.\\u003c/p\\u003e \\u003c/div\\u003e \\u003c/caption\\u003e \\u003ccolgroup cols=\\\"7\\\"\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c3\\\" colnum=\\\"3\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c4\\\" colnum=\\\"4\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c5\\\" colnum=\\\"5\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c6\\\" colnum=\\\"6\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c7\\\" colnum=\\\"7\\\"\\u003e\\u003c/div\\u003e \\u003cthead\\u003e \\u003ctr\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eModel\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eRMSE_mean\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eRMSE_std\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eMAE_mean\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003eMAE_std\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003eR\\u0026sup2;_mean\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003eR\\u0026sup2;_std\\u003c/p\\u003e \\u003c/th\\u003e \\u003c/tr\\u003e \\u003c/thead\\u003e \\u003ctbody\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eLinear\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e7.863\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e2.892\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e3.906\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e1.568\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.950\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.025\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eRidge\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e7.862\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e2.892\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e3.907\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e1.569\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.950\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.025\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eLASSO\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e7.863\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e2.892\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e3.906\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e1.568\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.950\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.025\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eRF\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e9.256\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e3.902\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e4.409\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e1.765\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.934\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.027\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eXGB\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e11.129\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e5.395\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e4.945\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e2.333\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.915\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.026\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eLSTM\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e9.396\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.000\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e4.364\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.000\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.931\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.000\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003c/tbody\\u003e \\u003c/colgroup\\u003e \\u003c/table\\u003e\\u003c/div\\u003e \\u003c/p\\u003e \\u003cp\\u003eThe zero variance observed for the LSTM model reflects deterministic convergence under fixed initialization and limited training epochs rather than superior robustness. XGBoost performance improves markedly after spatial aggregation, with R\\u0026sup2; increasing from approximately 0.58 in the single-station configuration to about 0.92 for the aggregated multi-station dataset. This gain reflects the smoothing of localized noise and extreme events through spatial averaging, which enhances model generalization at the city scale while reducing sensitivity to short-lived local peaks.\\u003c/p\\u003e \\u003cp\\u003eFrom a policy perspective, however, aggregation improves spatial representativeness and robustness, making it more suitable for city-scale early-warning systems. Ensemble models remain the most resilient under aggregation, confirming their applicability for regional forecasting rather than site-specific peak reconstruction. This trade-off between local accuracy and urban-scale robustness has been documented in prior ESPR studies employing multi-site or aggregated air-quality data.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec15\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.6 Implications for Multi-Horizon Forecasting and Policy\\u003c/h2\\u003e \\u003cp\\u003eMulti-horizon evaluation shows that near-term forecasts (1\\u0026ndash;6 h) retain meaningful skill even under data scarcity, supporting their use in operational alert systems. Forecast skill deteriorates rapidly at longer horizons, particularly for aggregated data, underscoring the limitations of data-driven models in the absence of emissions inventories or chemical-transport coupling.\\u003c/p\\u003e \\u003cp\\u003eNevertheless, the ability to generate reliable short-term forecasts using only routinely observed variables represents a data-efficient pathway for developing regions. For Uzbekistan, where monitoring networks are expanding but still incomplete, such models provide a pragmatic alternative to resource-intensive deterministic approaches. By emphasizing interpretability and robustness over complexity, the proposed framework aligns closely with the applied, policy-oriented scope of Environmental Science and Pollution Research.\\u003c/p\\u003e \\u003c/div\\u003e\"},{\"header\":\"4. Conclusions\",\"content\":\"\\u003cp\\u003eThis study demonstrates that effective PM₂.₅ forecasting in data-scarce urban environments is achievable through carefully selected, interpretable machine-learning models. Under the fragmented monitoring conditions characteristic of Uzbekistan, short-term persistence dominates predictive skill, constraining both linear and nonlinear approaches.\\u003c/p\\u003e \\u003cp\\u003eXGBoost emerges as the most reliable high-performance model, outperforming LSTM in accuracy and stability under limited data availability. SHAP-based interpretation provides transparent insight into dominant pollution drivers, enhancing policy relevance. Spatial aggregation improves robustness at the city scale while reducing peak predictability, positioning the framework for early-warning rather than extreme-event reconstruction.\\u003c/p\\u003e \\u003cp\\u003eCollectively, these findings establish a realistic, data-efficient forecasting paradigm tailored to emerging monitoring networks, contributing directly to the Environmental Science and Pollution Research literature on applied, policy-relevant air-quality modeling.\\u003c/p\\u003e\"},{\"header\":\"Declarations\",\"content\":\"\\u003cp\\u003e\\u003cstrong\\u003eAcknowledgments\\u003c/strong\\u003e\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003eThis work was supported by the Uzbekistan-China International Science and Technology Innovation Cooperation fund under grant IL-8724053120-R11.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eConflict of Interest\\u0026nbsp;\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThere are no conflicts of interest to declare.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eDeclaration of Generative AI and AI-assisted technologies in the writing process\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eDuring the preparation of this work, the author(s) used Grammarly to improve the readability and language of the article. After using this tool, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eAuthor\\u003c/strong\\u003e\\u003cstrong\\u003es’\\u003c/strong\\u003e\\u003cstrong\\u003e\\u0026nbsp;Contribution\\u003c/strong\\u003e\\u003cstrong\\u003es:\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eMoulay Rachid Babaa (MRB)\\u003c/strong\\u003e conceived the study, designed the methodology, curated and analyzed the data, developed the machine-learning models, performed the statistical analysis, interpreted the results, and wrote the original draft of the manuscript. \\u003cstrong\\u003eOrabek Arabayev (OA)\\u003c/strong\\u003e contributed to model verification, validation of results, and Python code review and optimization. Both authors reviewed and approved the final manuscript.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eData Availability Statement\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThe air-quality and meteorological data underlying this study were obtained exclusively from the \\u003cstrong\\u003eOpen Data Tashkent initiative\\u003c/strong\\u003e and are publicly accessible through the corresponding municipal open-data portal. All data preprocessing steps, feature definitions, and model inputs are fully described in the main text and the Supplementary Information. The computational procedures used for data preprocessing, model calibration, validation, and prediction were implemented in Python and are available at https://github.com/0221eng/Early-Warning-Air-Quality-Management-\\u003c/p\\u003e\"},{\"header\":\"References\",\"content\":\"\\u003col\\u003e\\n\\u003cli\\u003eTursumbayeva M et al. (2023) Cities of Central Asia: new hotspots of air pollution in the world. Atmospheric Environment 309:119901. https://doi.org/10.1016/j.atmosenv.2023.119901\\u003c/li\\u003e\\n\\u003cli\\u003eWorld Bank (2022) The global health cost of PM\\u003csub\\u003e2.5\\u003c/sub\\u003e air pollution: a case for action beyond 2021. International Development in Focus. World Bank, Washington, DC\\u003c/li\\u003e\\n\\u003cli\\u003eWorld Bank (2024) Air quality assessment for Tashkent and the roadmap for air quality management improvement in Uzbekistan. World Bank, Washington, DC\\u003c/li\\u003e\\n\\u003cli\\u003eLin Y et al. (2025) Health effects and economic losses due to PM\\u003csub\\u003e2.5\\u003c/sub\\u003e pollution from oil consumption by mobile sources in China. Frontiers in Environmental Science 13:1554150. https://doi.org/10.3389/fenvs.2025.1554150\\u003c/li\\u003e\\n\\u003cli\\u003eGarcia A et al. (2023) Toxicological effects of fine particulate matter (PM\\u003csub\\u003e2.5\\u003c/sub\\u003e): health risks and associated systemic injuries\\u0026mdash;systematic review. Water, Air, \\u0026amp; Soil Pollution 234(6):346. https://doi.org/10.1007/s11270-023-06278-9\\u003c/li\\u003e\\n\\u003cli\\u003eSangkham S et al. (2024) An update on adverse health effects from exposure to PM\\u003csub\\u003e2.5\\u003c/sub\\u003e. Environmental Advances 18:100603. https://doi.org/10.1016/j.envadv.2024.100603\\u003c/li\\u003e\\n\\u003cli\\u003eAljafen BN et al. (2025) Effect of environmental pollutants particulate matter (PM\\u003csub\\u003e2.5\\u003c/sub\\u003e, PM10), nitrogen dioxide (NO\\u003csub\\u003e2\\u003c/sub\\u003e), sulfur dioxide (SO\\u003csub\\u003e2\\u003c/sub\\u003e), carbon monoxide (CO) and ground-level ozone (O\\u003csub\\u003e3\\u003c/sub\\u003e) on epilepsy. BMC Neurology 25:133. https://doi.org/10.1186/s12883-025-04142-3\\u003c/li\\u003e\\n\\u003cli\\u003eWu J, Xu C, Wang Q, Cheng W (2016) Potential sources and formations of the PM\\u003csub\\u003e2.5\\u003c/sub\\u003e pollution in urban Hangzhou. Atmosphere 7(8):100. https://doi.org/10.3390/atmos7080100\\u003c/li\\u003e\\n\\u003cli\\u003eTessum MW et al. (2022) Sources of ambient PM\\u003csub\\u003e2.5\\u003c/sub\\u003e exposure in 96 global cities. Atmospheric Environment 286:119234. https://doi.org/10.1016/j.atmosenv.2022.119234\\u003c/li\\u003e\\n\\u003cli\\u003eChen Z et al. (2020) Influence of meteorological conditions on PM\\u003csub\\u003e2.5\\u003c/sub\\u003e concentrations across China: a review of methodology and mechanism. Environment International 139:105558. https://doi.org/10.1016/j.envint.2020.105558\\u003c/li\\u003e\\n\\u003cli\\u003eYang H, Peng Q, Zhou J, et al. (2020) The unidirectional causality influence of factors on PM\\u003csub\\u003e2.5 \\u003c/sub\\u003ein Shenyang city of China. Scientific Reports 10:8403. https://doi.org/10.1038/s41598-020-65391-5\\u003c/li\\u003e\\n\\u003cli\\u003eYan AP et al. (2025) A roadmap to implementing machine learning in healthcare: from concept to practice. Frontiers in Digital Health 7:1462751. https://doi.org/10.3389/fdgth.2025.1462751\\u003c/li\\u003e\\n\\u003cli\\u003eErsozlu Z, Taheri S, Koch I (2024) A review of machine learning methods used for educational data. Education and Information Technologies 29:22125\\u0026ndash;22145. https://doi.org/10.1007/s10639-024-12704-0\\u003c/li\\u003e\\n\\u003cli\\u003eTorres AI, Ferreira J, Pedemonte M (2025) Machine learning and process systems engineering for sustainable chemical processes: a short review. Current Opinion in Green and Sustainable Chemistry 51:100982. https://doi.org/10.1016/j.cogsc.2024.100982\\u003c/li\\u003e\\n\\u003cli\\u003eEichenseer P, Hans L, Winkler H (2025) A data-driven machine learning model for forecasting delivery positions in logistics for workforce planning. Supply Chain Analytics 9:100099. https://doi.org/10.1016/j.sca.2024.100099\\u003c/li\\u003e\\n\\u003cli\\u003eAlotaibi E, Nassif N (2024) Artificial intelligence in environmental monitoring: in-depth analysis. Discover Artificial Intelligence 4:84. https://doi.org/10.1007/s44163-024-00198-1\\u003c/li\\u003e\\n\\u003cli\\u003eKhanam MM, Uddin MK, Kazi JU (2025) Advances in machine learning for the detection and characterization of microplastics in the environment. Frontiers in Environmental Science 13:1573579. https://doi.org/10.3389/fenvs.2025.1573579\\u003c/li\\u003e\\n\\u003cli\\u003eZhang M et al. (2024) Predicting the impacts of urban development on urban thermal environment using machine learning algorithms in Nanjing, China. Journal of Environmental Management 356:120560. https://doi.org/10.1016/j.jenvman.2024.120560\\u003c/li\\u003e\\n\\u003cli\\u003eMohan S, Kumar B, Nejadhashemi AP (2025) Integration of machine learning and remote sensing for water quality monitoring and prediction: a review. Sustainability 17:998. https://doi.org/10.3390/su17030998\\u003c/li\\u003e\\n\\u003cli\\u003eTang D, Zhan Y, Yang F (2024) A review of machine learning for modeling air quality: overlooked but important issues. Atmospheric Research 300:107261. https://doi.org/10.1016/j.atmosres.2024.107261\\u003c/li\\u003e\\n\\u003cli\\u003eZhou S et al. (2024) Deep-learning architecture for PM\\u003csub\\u003e2.5\\u003c/sub\\u003e concentration prediction: a review. Environmental Science and Ecotechnology 21:100400. https://doi.org/10.1016/j.ese.2024.100400\\u003c/li\\u003e\\n\\u003cli\\u003eWu C et al. (2025) Time-series data-driven PM\\u003csub\\u003e2.5\\u003c/sub\\u003e forecasting: from theoretical framework to empirical analysis. Atmosphere 16:292. https://doi.org/10.3390/atmos16030292\\u003c/li\\u003e\\n\\u003cli\\u003eDoreswamy N, S HK, Km Y, Gad I (2020) Forecasting air pollution particulate matter (PM\\u003csub\\u003e2.5\\u003c/sub\\u003e) using machine learning regression models. Procedia Computer Science 171:2057\\u0026ndash;2066. https://doi.org/10.1016/j.procs.2020.04.221\\u003c/li\\u003e\\n\\u003cli\\u003eMohammadi F et al. (2024) Prediction of atmospheric PM\\u003csub\\u003e2.5\\u003c/sub\\u003e level by machine learning techniques in Isfahan, Iran. Scientific Reports 14:52617. https://doi.org/10.1038/s41598-024-52617-z\\u003c/li\\u003e\\n\\u003cli\\u003eMahmud S et al. (2022) Implementing machine learning algorithms to predict particulate matter (PM\\u003csub\\u003e2.5\\u003c/sub\\u003e): a case study in the Paso del Norte region. Atmosphere 13(12):2100. https://doi.org/10.3390/atmos13122100\\u003c/li\\u003e\\n\\u003cli\\u003eKumar R, Uppaluri VSR (2025) Identification of best machine learning model for the real-time vehicular data-based prediction of PM\\u003csub\\u003e2.5\\u003c/sub\\u003e and PM\\u003csub\\u003e10\\u003c/sub\\u003e. Atmospheric Pollution Research 16(9):102575. https://doi.org/10.1016/j.apr.2025.102575\\u003c/li\\u003e\\n\\u003cli\\u003eBai W, Li F (2023) PM\\u003csub\\u003e2.5\\u003c/sub\\u003e concentration prediction using deep learning in Internet of Things air monitoring systems. Environmental Engineering Research 28(1):210456. https://doi.org/10.4491/eer.2021.456\\u003c/li\\u003e\\n\\u003cli\\u003eSingh S, Srivastava A (2025) Machine learning approach to PM2.5 forecasting and health risk assessment during stubble burning period in Delhi. Aerosol Science and Technology 1-20. https://doi.org/10.1080/02786826.2025.2513503\\u003c/li\\u003e\\n\\u003cli\\u003eZaini N, Ean LW, Ahmed AN, Abdul Malek M, Chow MF (2022) PM\\u003csub\\u003e2.5 \\u003c/sub\\u003eforecasting for an urban area based on deep learning and decomposition method. Scientific Reports 12:17565. https://doi.org/10.1038/s41598-022-21769-1\\u003c/li\\u003e\\n\\u003cli\\u003eShah I, Mubassir P, Ali S, Albalawi O (2024) A functional autoregressive approach for modeling and forecasting short-term air temperature. Frontiers in Environmental Science 12:1411237. https://doi.org/10.3389/fenvs.2024.1411237\\u003c/li\\u003e\\n\\u003cli\\u003eCatalano S, De Felice M, Crisci A (2016) Improving the prediction of air pollution peak episodes: the role of high-resolution meteorological data in air quality modeling. Environmental Science and Pollution Research 23(7):6287\\u0026ndash;6299. https://doi.org/10.1007/s11356-015-5391-x\\u003c/li\\u003e\\n\\u003cli\\u003ePetrić I, Božić D, Žunić D (2024) Improving prediction accuracy for hourly concentrations of pollutants using advanced statistical models. Atmospheric Environment 289:119345. https://doi.org/10.1016/j.atmosenv.2024.119345\\u003c/li\\u003e\\n\\u003cli\\u003eEspinosa A, Ibarra-Berastegi G, Sarriegi JM (2021) A time series forecasting based multi-criteria methodology for pollutant forecasting. Environmental Modelling \\u0026amp; Software 135:104888. https://doi.org/10.1016/j.envsoft.2020.104888\\u003c/li\\u003e\\n\\u003cli\\u003eKholmatjanov BM et al. (2020) Analysis of temperature change in Uzbekistan and the regional atmospheric circulation of Middle Asia during 1961-2016. Climate 8:101. https://doi.org/10.3390/cli8090101\\u003c/li\\u003e\\n\\n\\u003c/ol\\u003e\"}],\"fulltextSource\":\"\",\"fullText\":\"\",\"funders\":[],\"hasAdminPriorityOnWorkflow\":false,\"hasManuscriptDocX\":true,\"hasOptedInToPreprint\":true,\"hasPassedJournalQc\":\"\",\"hasAnyPriority\":true,\"hideJournal\":true,\"highlight\":\"\",\"institution\":\"\",\"isAcceptedByJournal\":false,\"isAuthorSuppliedPdf\":false,\"isDeskRejected\":\"\",\"isHiddenFromSearch\":false,\"isInQc\":false,\"isInWorkflow\":false,\"isPdf\":false,\"isPdfUpToDate\":true,\"isWithdrawnOrRetracted\":false,\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"researchsquare\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":true,\"externalIdentity\":\"\",\"sideBox\":\"\",\"snPcode\":\"\",\"submissionUrl\":\"/submission\",\"title\":\"Research Square\",\"twitterHandle\":\"researchsquare\",\"acdcEnabled\":true,\"dfaEnabled\":false,\"editorialSystem\":\"\",\"reportingPortfolio\":\"\",\"inReviewEnabled\":false,\"inReviewRevisionsEnabled\":true},\"keywords\":\"PM₂.₅ forecasting, machine learning, data-scarce environments, XGBoost, LSTM, feature interpretability, multi-horizon prediction, air-quality early warning\",\"lastPublishedDoi\":\"10.21203/rs.3.rs-8387439/v1\",\"lastPublishedDoiUrl\":\"https://doi.org/10.21203/rs.3.rs-8387439/v1\",\"license\":{\"name\":\"CC BY 4.0\",\"url\":\"https://creativecommons.org/licenses/by/4.0/\"},\"manuscriptAbstract\":\"\\u003cp\\u003eFine particulate matter (PM₂.₅) poses a major environmental and public health risk in Central Asia, yet predictive air-quality modeling remains limited due to fragmented monitoring networks and data scarcity. This study presents an interpretable multi-horizon machine-learning framework for PM₂.₅ forecasting in Tashkent, Uzbekistan, representing the first such analysis for the country. Six models, linear regression, ridge regression, LASSO, random forest, XGBoost, and long short-term memory (LSTM), were developed and evaluated under realistic data-limited conditions using hourly air-quality and meteorological observations. Forecasts were generated for three operational horizons (1 h, 24 h, and 168 h). Results show that short-term PM₂.₅ persistence dominates predictive skill, with XGBoost achieving the highest accuracy and stability, outperforming LSTM under fragmented datasets. Feature-selection and SHAP analyses provide transparent insight into dominant pollution drivers, enhancing policy relevance. Spatial aggregation across monitoring stations improves robustness for city-scale early-warning applications, albeit with reduced peak sensitivity. The proposed framework offers a data-efficient and interpretable pathway for operational air-quality management in emerging monitoring contexts.\\u003c/p\\u003e\",\"manuscriptTitle\":\"Interpretable Multi-Horizon Machine Learning Framework for PM₂.₅ Forecasting in Tashkent: Toward Early-Warning Air Quality Management\",\"msid\":\"\",\"msnumber\":\"\",\"nonDraftVersions\":[{\"code\":1,\"date\":\"2025-12-18 11:54:51\",\"doi\":\"10.21203/rs.3.rs-8387439/v1\",\"editorialEvents\":[{\"type\":\"communityComments\",\"content\":0}],\"status\":\"published\",\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"researchsquare\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":true,\"externalIdentity\":\"\",\"sideBox\":\"\",\"snPcode\":\"\",\"submissionUrl\":\"/submission\",\"title\":\"Research Square\",\"twitterHandle\":\"researchsquare\",\"acdcEnabled\":true,\"dfaEnabled\":false,\"editorialSystem\":\"\",\"reportingPortfolio\":\"\",\"inReviewEnabled\":false,\"inReviewRevisionsEnabled\":true}}],\"origin\":\"\",\"ownerIdentity\":\"17d1728d-7d97-412f-b5c3-f532a5952d2a\",\"owner\":[],\"postedDate\":\"December 18th, 2025\",\"published\":true,\"recentEditorialEvents\":[],\"rejectedJournal\":[],\"revision\":\"\",\"amendment\":\"\",\"status\":\"posted\",\"subjectAreas\":[{\"id\":59884401,\"name\":\"Environmental Engineering\"},{\"id\":59884402,\"name\":\"Artificial Intelligence and Machine Learning\"}],\"tags\":[],\"updatedAt\":\"2025-12-18T11:54:51+00:00\",\"versionOfRecord\":[],\"versionCreatedAt\":\"2025-12-18 11:54:51\",\"video\":\"\",\"vorDoi\":\"\",\"vorDoiUrl\":\"\",\"workflowStages\":[]},\"version\":\"v1\",\"identity\":\"rs-8387439\",\"journalConfig\":\"researchsquare\"},\"__N_SSP\":true},\"page\":\"/article/[identity]/[[...version]]\",\"query\":{\"redirect\":\"/article/rs-8387439\",\"identity\":\"rs-8387439\",\"version\":[\"v1\"]},\"buildId\":\"8U1c8b4HqxoKbykW_rLl7\",\"isFallback\":false,\"isExperimentalCompile\":false,\"dynamicIds\":[84888],\"gssp\":true,\"scriptLoader\":[]}","source_license":"CC-BY-4.0","license_restricted":false}