Spatial analysis of COVID-19 and the Russia–Ukraine war impacts on natural gas flows using statistical and machine learning models

The natural gas market has faced unprecedented disruptions due to the COVID-19 pandemic and the Russia–Ukraine war, both of which have significantly affected gas flow patterns. In this study, we present a spatial analysis to identify the nodes – points where gas is injected, withdrawn, or stored – and the pipelines connecting them mostly impacted by these events, using both statistical and machine learning models. We employ ARIMA, Long Short-Term Memory (LSTM), and a hybrid CNN-LSTM model to forecast hourly gas flows. Models are trained using lagged, temporal, and weather characteristics, with nodes grouped into clusters. For each cluster, we train and optimize the model’s hyperparameters on vertically concatenated time series data and compute prediction errors. To assess the impact of disruption, we construct a composite indicator based on forecasting errors and compare the values before and after each event. We find a positive correlation between the maximum daily percentage variation in gas flow and the change in the composite indicator—approximately 20% for COVID-19 and 17% for the Russia–Ukraine war—indicating the method’s effectiveness in detecting affected nodes. A spatial aggregation of impacted nodes is compared with official regional statistics on industrial gas consumption, revealing additional positive correlations. The results show that industrial and municipal nodes were most severely affected during the COVID-19 pandemic, probably due to lockdowns. For the Russia–Ukraine conflict, municipal outflows and inflows were most affected, possibly due to rising prices and increased imports. In general, while both events caused significant disturbances, the COVID-19 pandemic had a greater impact, probably because Russia continued to export energy during the war.