Assessing 29 years of global land cover dynamics from satellite Earth Observation by Céline Lamarche

Louvain-La-Neuve

May 27, 2024

16h

Louvain-la-Neuve

Ocean room B002

Land use and land cover change contribute significantly to anthropogenic CO2 emissions and biodiversity loss. However, current inventory-based statistics miss year-to-year land changes, preventing a comprehensive global understanding.

Earth observation by satellite provides valuable information for mapping global annual changes at the pixel level. Yet, consistency is crucial to capture change signals among variability from natural surface fluctuations or evolving satellite mission capabilities. In this thesis, we co-develop and evaluate the very first global annual land cover change time series at 300 m from 1992 to 2020, maintaining consistency across space, time and satellite missions. This dataset, currently used for Intergovernmental Panel on Climate Change (IPCC) climate modeling and Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES) analysis, enables us to quantify and analyze the global dynamics of land cover change over the last 30 years.

This research also contributes to enhancing the qualification of land cover datasets. While validation guidelines are well-established for individual product assessment, we introduced a stratified random sampling method for product benchmarking, tailored to areas prone to discrepancies between products, to highlight the satellite product strengths and weaknesses. To improve climate and land surface, we enabled the conversion of land cover categories to plant functional types (PFT) composition per pixel. This was made possible thanks to the development of a globally applicable framework to seamlessly integrate multiple high-resolution datasets that might otherwise not be compatible.

Finally, we investigate the impact of uncertainty in Earth observation surface reflectance measurement on the categorical land cover classification process. Building on a Monte Carlo simulation, we quantify errors and propose a classification ensemble approach to effectively mitigate them.

This work contributes to more informed land accounting in the context of rapid anthropogenic change and climate evolution, supporting the assessment of the Food and Agriculture Organization (FAO) and Organization for Economic Co-operation and Development (OECD) environmental-economic accounting systems.