Improved estimation of carbon dioxide and methane using machine learning with satellite observations over the Arabian Peninsula

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: Jan. 4, 2025

Estimating spatiotemporal maps of greenhouse gases (GHGs) is important for understanding climate change and developing mitigation strategies. However, current methods face challenges, including the coarse resolution numerical models, gaps in satellite data, making it essential to improve estimation GHGs. This study aims develop an advanced technique produce high-fidelity (1 km) CO2 CH4 over Arabian Peninsula, a highly vulnerable region change. Using XGBoost, columnar carbon dioxide (XCO2) methane (XCH4) concentrations using data from OCO-2 Sentinel-5P (the target variables) were downscaled, with ancillary CarbonTracker, MODIS Terra, ERA-5 input variables). The model trained validated against these datasets, achieving high performance XCO2 (R2 = 0.98, RMSE 0.58 ppm) moderate accuracy XCH4 0.63, 13.26 ppb). Seasonal cycles long-term trends identified, higher observed summer, emission hotspots urban industrial areas. Comparisons EDGAR inventory highlighted significant contributions power, oil, transportation sectors GHG emissions. These results demonstrate value high-resolution local-scale monitoring, supporting targeted strategies sustainable policymaking region. Future work could integrate ground-based observations further enhance monitoring accuracy.

Language: Английский

Multi-Scale Temporal Integration for Enhanced Greenhouse Gas Forecasting: Advancing Climate Sustainability

Sustainability, Journal Year: 2025, Volume and Issue: 17(8), P. 3436 - 3436

Published: April 12, 2025

Greenhouse gases (GHGs) significantly shape global climate systems by driving temperature rises, disrupting weather patterns, and intensifying environmental imbalances, with direct consequences for human life, including rising sea levels, extreme weather, threats to food security. Accurate forecasting of GHG concentrations is crucial crafting effective policies, curbing carbon emissions, fostering sustainable development. However, current models often struggle capture multi-scale temporal patterns demand substantial computational resources, limiting their practicality. This study presents MST-GHF (Multi-Scale Temporal Gas Forecasting), an innovative framework that integrates daily monthly CO2 data through a multi-encoder architecture address these challenges. It leverages Input Attention encoder manage short-term fluctuations, Autoformer long-term trends, mechanism ensure stability across scales. Evaluated on fifty-year NOAA dataset from Mauna Loa, Barrow, American Samoa, Antarctica, surpasses 14 baseline models, achieving Test_R2 0.9627 Test_MAPE 1.47%, notable in forecasting. By providing precise predictions, empowers policymakers reliable targeted policies conducting scenario simulations enabling proactive adjustments emission reduction strategies enhancing sustainability aligning interventions goals. Its optimized efficiency, reducing resource demands compared Transformer-based further strengthens modeling, making it deployable resource-limited settings. Ultimately, serves as robust tool mitigate impacts advancing societal domains.

Language: Английский

Predicting Building Primary Energy Use Based on Machine Learning: Evidence from Portland

International Journal of Architectural Engineering Technology, Journal Year: 2024, Volume and Issue: 11, P. 124 - 139

Published: Dec. 28, 2024

Accurately predicting equivalent primary energy use (EPEU) in buildings is crucial for advancing energy-efficient design, optimizing operational strategies, and achieving sustainability goals the built environment. This study aims to develop reliable prediction models EPEU by leveraging a comprehensive high-quality dataset from Portland, USA. To achieve this, systematic machine learning framework adopted, encompassing feature selection, data preprocessing, model training, performance evaluation. Several state-of-the-art algorithms are applied, including Random Forest (RF), Gradient Boosting Decision Tree (GBDT), Back-Propagation Neural Networks (BP). These trained using key features such as building type, gross floor area, construction year, various characteristics that known significantly influence consumption patterns. The carefully cleaned normalized ensure generalizability minimize bias. Model assessed standard statistical metrics, coefficient of determination (R²), Mean Absolute Error (MAE), Root Squared (RMSE). Among tested models, ensemble methods—particularly RF GBDT—consistently outperform others terms accuracy, robustness, stability across different types. results this not only highlight potential tasks but also provide actionable insights architects, engineers, facility managers, policymakers. By identifying most influential variables employing effective predictive research supports data-driven decision-making processes aimed at improving performance.

Language: Английский