Simulated AVIRIS-3 detects species-specific patterns of vegetation succession in retrogressive thaw slumps

Retrogressive Thaw Slumps (RTS) are rapidly increasing in size and number across the Arctic. These mass-wasting features transform permafrost landscapes, influencing localized patterns of vegetation cover and associated carbon storage. Although hyperspectral imaging can detect fine-scale patterns of plant species and plant functional types (PFTs), due to its limited availability, and coverage, knowledge of the spatiotemporal patterns of vegetation distribution and change in such permafrost disturbed terrain spanning large spatial domains remains limited. Here, we evaluate the potential of hyperspectral image simulation for regional-scale mapping of fractional vegetation cover (fCover) in disturbed RTS terrain. We acquire intensive coincidental ground (i.e., vegetation surveys, hyperspectral, and LiDAR data), airborne (i.e., multispectral 5 cm Red Edge-M and Airborne visible/infrared imaging spectrometer; AVIRIS-3), and space-borne (i.e., Sentinel-2 Multi-Spectral Imager; MSI) datasets from five adjacent RTSs of different ages and vegetation composition in the Noatak National Preserve, Alaska. Using these datasets, we simulated AVIRIS-3 using reflectance from Sentinel-2 MSI with a spectral reconstruction method known as the Universal Pattern Decomposition Method (UPDM), to determine the impact of synchronous and asynchronously collected ground and airborne observations on fCover retrievals from seven dominant vegetation types among AVIRIS-3, simulated AVIRIS-3, and Sentinel-2. UPDM assumes that each pixel in an image can be decomposed into linear combination of different components present in it. The performance of the fCover models varied across sensors, with the simulated products outperforming Sentinel-2 resulting in an overall 16% lower RMSE. The models effectively captured vegetation patterns across RTS age classes, with older slumps dominated by shrubs and younger slumps by mosses and forbs. Seasonal fCover variability reflected phenological differences among PFTs, with older, shrub-dominated slumps showing greater change than younger, sparsely vegetated slumps, while interannual variability was < 1%, based on Analysis of Covariance (ANCOVA) test. Analysis of temporal variability revealed that by restricting regional simulation periods to Julian days 185–225 (i.e., within peak growing season) would markedly minimize the variability in modeled fCover by 0.79 ± 5.31%. These findings highlight the potential of spectral reconstruction methods to enhance our understanding of vegetation successional dynamics within rapidly expanding RTSs in regions, when hyperspectral datasets are limited.