Comparing the ability of UAVs and high-resolution satellite data to resolve post-fire patterns of permafrost thaw and vegetation in Siberian larch forests

Abstract:

Vegetation change and permafrost thaw in response to amplified Arctic climate warming are likely to act as important global climate feedbacks over the next century through impacts on carbon cycling and surface energy dynamics. Wildfire can accelerate vegetation change and permafrost thaw, especially in northeastern Siberia where fire-adapted larch forests underlain by ice-rich yedoma permafrost dominate large portions of the continuous permafrost zone . Vegetation and surficial changes related to permafrost thaw can alter the spectral characteristics of the landscape. However, this typically occurs at spatial scales smaller than the pixel resolution of long-term satellite data. The objective of this study is to characterize patterns of vegetation cover and surface water ponding associated with permafrost degradation using ultra-high-resolution data collected using Uncrewed Aerial Vehicles (UAVs), and to determine whether these patterns can be detected using commercially available high-resolution satellite data.

Our study was located in upland larch forests near the town of Cherskiy in the northeast Sakha Republic, Russia. At two sites situated across the perimeter of a fire that occurred two decades ago, we collected UAV imagery using RGB and multispectral sensors in July 2019 and compared this to a Planet satellite image collected during the same week. Maps of NDVI generated using both Planet and UAV datasets reflected lower woody vegetation abundance in the burned areas relative to unburned areas. Areas of higher NDVI were typically associated with shrubs or trees, and heterogeneity in vegetation occurred at scales smaller than 3 m, and so could be detected using UAV but not satellite data. Surface water ponding associated with subsidence and permafrost thaw was best detected in burned areas with UAV RGB data. Surface water extent was underestimated with UAV multispectral data, due to emergent vegetation cover, and potentially shallow submerged vegetation. Most surface water ponding occurred at spatial scales too small to be explicitly detected with satellite data. Our results illustrate the necessity of RGB and multispectral UAV data for inferring the ecological mechanisms associated with spectral variability in high resolution satellite data in Arctic ecosystems.