Ruben Remelgado , Christopher Conrad, Carsten Meyer

Satellite remote sensing is vital for research, monitoring, and policy addressing environmental-change and sustainability issues from climate and ecosystem changes to food and water security. Here, Landsat satellite data play a crucial role, thanks to their unique global, long-term, and high-resolution coverage. Yet, spatial and temporal data gaps in the Landsat archive may propagate into derived remote-sensing products and thereby threaten the validity of downstream applications, especially when data users have limited training in remote sensing. To improve awareness of these issues, we here demonstrate that global, historical Landsat data provide a spatially and temporally highly uneven, frequently interrupted, and seasonally incomplete coverage. Using a causal-discovery framework, we moreover show that these inconsistencies are inherited in several state-of-the-art, global time-series products, causing pervasive biases in perceptions of changes in tropical moist forests, arable lands, and seasonal water resources (significant biassing effects detected in 84.6-93.6% of our country-specific tests, depending on land-change facet). These biases can impair reliable analyses and monitoring of diverse environmental changes and human development issues targeted by international policy frameworks including the Kunming-Montréal Global Biodiversity Framework, the Paris Agreement, and the Sustainable Development Goals. We discuss avenues towards better uncertainty reporting and bias control in satellite-based monitoring and related applications, highlighting needed contributions from both product developers and users.