PAWI: The first high-resolution, multi-class, Pan-Arctic Wetland Inventory

Abstract

Arctic wetlands are key regulators of global methane (CH₄) emissions, yet uncertainty in wetland spatial extent and class composition limits the accuracy of CH₄ budget models. Current mapping products lack circumpolar coverage and consistent thematic classification standards, constraining the ability to model ecosystem responses to Arctic warming. Here, we present the first high-resolution Pan-Arctic Wetlands Inventory (PAWI), produced at 10 m resolution using multi-sensor satellite imagery (e.g., Sentinel-1, Sentinel-2, and ALOS PALSAR-2), ArcticDEM topography, and environmental and hydrological datasets, using a machine learning Random Forest classifier. Wetlands are classified into bog, fen, swamp, marsh, and water. The final product achieves an overall accuracy of 89% (Kappa = 0.86) and estimates that 20% of the Arctic landmass is wetland. The PAWI provides a consistent, ecologically relevant baseline for improving CH₄ flux modeling, assessing climate vulnerability, and supporting conservation planning. By integrating advances in remote sensing, machine learning, and multi-national data harmonization, this work addresses a critical gap in Arctic wetland mapping and establishes a transferable framework for large-scale ecosystem classification in remote regions.