If all you have is a hammer, everything looks like a nail: spectral trends as a measure of ecological change in the Arctic

The Arctic is warming four times faster than lower latitudes. Observations from spatially limited tundra field sites show increased vegetation growth, expansion of woody shrub cover, decreased snow cover and extreme permafrost thaw disturbances. Satellite imagery is fundamental to our understanding of these land-surface changes across the Arctic and, thereby, to our ability to predict feedbacks to global climate. Positive trends in the satellite-derived normalised difference vegetation index (NDVI) have been broadly observed and attributed to increased vegetation productivity, instigating a discourse of Arctic greening, while negative trends, or browning, are less common and broader in attribution. However, methodological issues such as spectral mixing within satellite pixels, the saturation of the NDVI, and the period of the analyses have emerged as a source of significant uncertainty in the detection and ecological interpretation of spectral greening and browning trends. In this thesis, I use high-resolution drone and satellite imagery to assess the effect of snow, fractional vegetation cover, and permafrost thaw slumps on our ability to detect and interpret spectral trends. Snow cover has decreased in extent and duration across much of the Arctic but is poorly accounted for in spectral trend analyses. By using high-resolution drone imagery from one Arctic and one sub-Arctic site, I found that fine-scale snow persistence within satellite pixels is associated with both reduced magnitude and delayed timing of annual peak NDVI, the base metric of spectral trend analyses. These findings indicate that unaccounted changes in fine-scale snow persistence may contribute to Arctic spectral greening and browning trends through either biotic responses of vegetation to snow cover or abiotic integration of snow within the estimated peak NDVI. Across the Arctic, changing snow persistence may drive both underestimation and overestimation of changes in vegetation productivity. Fractional vegetation cover corresponds with spectral mixing and the saturation of the NDVI. However, vegetation cover is difficult to calculate at the scale of satellite pixels, and the relationship between vegetation cover and spectral trends therefore remains unknown. I found that spectral Sentinel-2 data can predict vegetation cover at a high-latitude and low-latitude tundra site, and subsequently observed that predicted vegetation cover differed significantly between pixels with and without spectral trends. These results suggest that a pixel’s vegetation cover may affect our ability to detect spectral trends, due to spectral mixing within low vegetation cover pixels and saturation of the NDVI within high vegetation cover pixels. Spatial variation in spectral greening and browning across the Arctic may, in places, reflect underlying patterns in fractional vegetation cover more than the presence or absence of vegetation change. Permafrost disturbance events are an often-cited source of spectral browning, however, the effect of their timing and subsequent recovery on trend detection has received limited attention. I use a pan-Arctic dataset of retrogressive thaw slumps to examine the representation of permafrost disturbance events in spectral trends derived from Landsat imagery. I found that spectral browning occurred over less than half of the analysed thaw slumps (~49%) due to post-disturbance vegetation recovery and the time period of analysis. Ultimately, this may lead to an underestimation of permafrost disturbance-related change across the Arctic. Together, my thesis findings demonstrate that spectral trend analyses, although familiar, are a somewhat blunt tool for inferring Arctic vegetation change from satellite imagery. Attributing spectral trends to field observations of ecological change is complicated by a lack of methodological nuance, where unaccounted variation in snow, vegetation cover and dynamic permafrost thaw disturbances may obscure the detection or interpretation of trends. Overall, this thesis highlights three confounding effects on spectral trend analyses that should be considered to improve future assessments of Arctic land-surface change.