Monitoring Atmospheric Anomalies from Space: A Topological Approach (MAASTA)

Carbon dioxide (CO₂) has experienced an alarming increase since the Industrial Revolution due to anthropogenic factors like fossil fuel combustion and deforestation. Before the Industrial Revolution, atmospheric CO₂ levels were relatively stable, around 280 parts per million (ppm). By the 21st century, human activities have increased CO₂ concentrations beyond 400 ppm, reaching levels not seen in at least 800,000 years. In recent years, satellite observations have enabled global monitoring of CO₂ and other atmospheric trace gases such as nitrogen dioxide (NO₂), carbon monoxide (CO) and methane (CH4), and indicators like solar induced fluorescence (SIF), which indicates photosynthetic activity. Anthropogenic emissions have a significant impact in the carbon cycle, and they are difficult to monitor accurately at large scale. Additionally, due to the ongoing climate change, biospheric CO₂ fluxes are also showing altered patterns and behaviours. This project aims at developing a data-driven method to detect spatial and temporal anomalies in satellite-based CO₂ observations. This will have as starting point a rigorous mathematical approach based on topological data analysis. Topological data analysis (TDA) is a technique in data science that has its origins in algebraic topology, an abstract area of mathematics. It focuses on the study of shapes and properties of space invariant under continuous transformations. TDA can detect high level features in data that are often overlooked by traditional methods, and is robust against outliers and noise. At the same time, it is an intuitive and interpretable tool that has a solid theoretical foundation. We will use TDA to find spatio-temporal anomalies in CO₂ data, and will combine these with other datasets (NO₂, CO, CH4, SIF) to distinguish between types of anomalies (e.g., anthropogenic from biogenic). For this, unsupervised machine learning methods such as hierarchical clustering and DBSCAN (Density-Based Spatial Clustering of Applications with Noise) will be explored. These techniques are known for their capability to discern patterns and clusters within large datasets without prior labelling. With this novel approach, we aim at building a comprehensive dataset of anomalies characterized by distinct features, thereby enhancing our understanding of the various factors influencing CO₂ concentrations and providing tools for more effective and near-real time monitoring and mitigation strategies.