Advanced Remote sensing Techniques for greenhouse gas EMISsion modeling In African lakes

The lacustrine emissions to the atmosphere of greenhouse gases (GHGs), mainly CO₂ and CH₄, are significant components of global GHG budgets. However, available estimates remain very uncertain due to sparse data coverage, in particular in tropical lakes. With over 1.4 million lakes (>0.1 km²) worldwide, the limited availability of sites-specific lacustrine CO₂ and CH₄ emission data presents a significant challenge for accurate global and regional assessments. African lakes, which comprise 66% of the tropical lake surface are, are critical contributors to these emissions, yet remain along the least studied. ARTEMISIA addresses this gap and aims to develop, after a proof of concept for a selection of African lakes, the first pan-African, Earth Observation (EO)-based framework for quantifying lake CO₂ and CH₄ emissions. Existing methods relying primarily on sparse in situ data and empirical models, cannot capture the spatial and temporal variability of these dynamic systems. Only through the integration of EO data we can generate robust, spatially explicit emission estimates. To achieve this, ARTEMISIA introduces an innovative combination of satellite EO, in situ measurements, and machine learning to estimate CO₂ and CH₄ emissions from lakes across the African continent. For the first time, a remote sensing-based approach will be at the continental scale will be attempted, delivering spatially explicit CO₂ and CH₄ emission estimates for lakes across Africa. ARTEMISIA will apply advanced atmospheric correction and quality control techniques and will enhance EO-based water quality retrieval using machine learning. These improvements will reduce uncertainties in key water quality products retrieved from EO data—namely chlorophyll-a (Chl-a), colored dissolved organic matter (CDOM), and total suspended matter (TSM)—which serve as critical inputs or proxies to the CO₂ and CH₄ emission models. Remote sensed lake surface temperature (LST) will also be integrated as an additional model input. Machine learning, specifically random forest algorithms or other methodologies, will be used to predict dissolved CO₂ and CH₄ concentrations based on EO-derived parameters and lake morphology (i.e. lake surface area and depth). These models will be trained on a large dataset of in situ GHG and Chl-a/CDOM/TSM concentrations collected from 146 African lakes, encompassing a wide range of sizes, depths, and productivity levels. Key predictors—Chl-a, CDOM, TSM, LST, lake surface area and depth—are selected based on their conceptual and empirical relevance to biogeochemical processes driving CO₂ and CH₄ dynamics in surface waters. Emission fluxes will be computed using these concentration estimates combined with gas transfer velocities derived from ECMWF ERA-5 wind speed products.
The result will be a consistent and scalable methodology for calculating monthly and annual CO₂ and CH₄ emissions from individual lakes, enabling aggregation at the national level. These outputs will be made available via a web interface and are designed for use in National Inventory Reports (NIRs) under the United Nations Framework Convention on Climate Change (UNFCCC). Beyond emission estimates, ARTEMISIA also seeks to understand emission drivers—internally (e.g. lake trophic state) and externally (e.g. erosion, land use, landslides). EO data and physically based models (e.g. SWAT+, LANDPLANER) will be used to simulate sediment and nutrient transport, linking catchment processes to lake biogeochemistry. Sentinel-1/-2, InSAR, and vegetation indices will support the detection of geomorphic disturbances, essential for understanding carbon loading dynamics. This approach will be demonstrated through two detailed case studies: Lake Tanganyika (including the Ruzizi River), and Lake Victoria. These case studies will go beyond lake-average assessments which are crude approximations in the case of large lakes with significant optical variability, focusing instead on high resolution, pixel-scale analyses to explore spatial and temporal variability in GHG emissions. For Lake Victoria, the study will also incorporate flux tower measurements and fish-kill event records to investigate the links between greenhouse gas fluxes, ecological responses, and remote sensing indicators. The ARTEMISIA framework will serve as a benchmark for other tropical regions with limited data, such as South America and Southeast Asia, where inland water GHG emissions remain poorly constrained. By integrating EO, in situ data, and machine learning, ARTEMISIA aims to deliver:
• After a proof of concept for a selection of African lakes, the first-ever pan-African CO₂ and CH₄ emission dataset from lakes >0.1 km², with monthly and annual temporal resolution and uncertainty estimates;
• Actionable tools and open-access datasets, including web-based dashboards and geospatial layers for national and regional reporting;
• New methodological advances in EO-based water quality retrieval and machine learning-driven GHG modeling, with peer-reviewed publications and operational tools.