Journal of Big Data Информационные системы
Graph-based learning for multi-horizon Martian atmospheric forecasting
Аннотация
Abstract Purpose Martian weather forecasting is important for future exploration, but atmospheric behaviour on Mars combines spatial, temporal, vertical, and dust-driven processes in ways that challenge current modelling and forecasting approaches. Existing machine learning studies often reduce this structure to local time series, which limits their ability to capture wider atmospheric dynamics. Methods This paper introduces a graph-based data engineering framework called MaGMA (Martian Graph-based Multi-horizon Atmospheric Forecasting) that transforms OpenMARS reanalysis fields into structured learning objects for Martian atmospheric forecasting. The framework represents local atmospheric patches as graph nodes and links them through neighbouring regions, successive time steps, longer temporal dependencies, and dynamically similar atmospheric states. It integrates recent atmospheric history, engineered physical descriptors, and vertical atmospheric information to support forecasting across multiple horizons. Results We evaluate MaGMA across five unseen Martian years, including regular years and a global dust storm year. In regular years, the model achieves overall coefficients of determination of approximately 0.73–0.85, indicating that it captures a substantial proportion of the variation in the target atmospheric variables. For dust-column forecasting, the model outperforms classical and deep temporal baselines in most year–horizon comparisons. In the global dust storm year, dust-column prediction remains strong at shorter horizons, with coefficients of determination above 0.8 for the first two horizons, but broader multivariate performance declines, showing that extreme regimes still challenge generalisation. Conclusion The study shows that graph-based data engineering can create reusable and diagnostically useful representations for planetary atmospheric forecasting. It also identifies two priorities for future work: improving learning under rare extreme regimes and making better use of vertical atmospheric structure.