Abstract 

This paper studies a critical topic in environmental policy: how the 'minimum mapping threshold' of satellite-based monitoring data shapes our understanding of deforestation and the effectiveness of related policies. Leveraging an novel big-data framework, we analyzed ~180 billion pixels using Google Earth Engine and R, achieving unprecedented detail in deforestation analysis. By comparing high-resolution land-cover and land-use change data from MapBiomas with official government data from PRODES, we uncovered significant measurement discrepancies. Official data, which only captures deforestation patches larger than 6.25 hectares, underestimates annual deforestation by up to threefold. Additionally, we demonstrate how landholders strategically fragment their activities into smaller, undetectable patches to evade monitoring, highlighting the urgent need for more refined and comprehensive tracking systems. Our interdisciplinary approach integrates methodologies from agroecology, geography, and economics. Using the FragStats-based framework, we quantify patch-level metrics to reveal how forest fragmentation accelerates ecosystem degradation. Advanced causal inference techniques assess the impact of Brazil's Blacklisting policy, showing how variations in monitoring intensity influence landholder behavior. Policies based on official data are perceived to be nearly twice as effective as they are in reality, emphasizing the need for robust validation tools. This research bridges disciplines to advance land systems science, applied economics, and environmental policy, providing actionable insights to enhance policy design and implementation.

Paper selected for Data, Information, and the Environment Workshop with Journal of Environmental Economics and Management and CAGE, Warwick 2024