Imprint of Climate Change on Pan-Arctic Marine Vegetation

Abstract

The Arctic climate is changing rapidly. The warming and resultant longer open water periods suggest a potential for expansion of marine vegetation along the vast Arctic coastline. We compiled and reviewed the scattered time series on Arctic marine vegetation and explored trends for macroalgae and eelgrass (Zostera marina). We identified a total of 38 sites, distributed between Arctic coastal regions in Alaska, Canada, Greenland, Iceland, Norway/Svalbard, and Russia, having time series extending into the 21st Century. The majority of these exhibited increase in abundance, productivity or species richness, and/or expansion of geographical distribution limits, several time series showed no significant trend. Only four time series displayed a negative trend, largely due to urchin grazing or increased turbidity. Overall, the observations support with medium confidence (i.e., 5–8 in 10 chance of being correct, adopting the IPCC confidence scale) the prediction that macrophytes are expanding in the Arctic. Species distribution modeling was challenged by limited observations and lack of information on substrate, but suggested a current (2000–2017) potential pan-Arctic macroalgal distribution area of 820.000 km2 (145.000 km2 intertidal, 675.000 km2 subtidal), representing an increase of about 30% for subtidal- and 6% for intertidal macroalgae since 1940–1950, and associated polar migration rates averaging 18–23 km decade–1. Adjusting the potential macroalgal distribution area by the fraction of shores represented by cliffs halves the estimate (412,634 km2). Warming and reduced sea ice cover along the Arctic coastlines are expected to stimulate further expansion of marine vegetation from boreal latitudes. The changes likely affect the functioning of coastal Arctic ecosystems because of the vegetation’s roles as habitat, and for carbon and nutrient cycling and storage. We encourage a pan-Arctic science- and management agenda to incorporate marine vegetation into a coherent understanding of Arctic changes by quantifying distribution and status beyond the scattered studies now available to develop sustainable management strategies for these important ecosystems.

Biologically meaningful distribution models highlight the benefits of the Paris Agreement for demersal fishing targets in the North Atlantic Ocean

“…The distribution of species was modelled by ensembling (Araújo & New, 2007) the response of boosted regression trees (BRT) and adaptive boosting (AdaBoost), two machine learning algorithms with high predictive performances in SDMs (Assis et al., 2018; Krause‐Jensen et al., 2020) that automatically fit complex interactions between predictor variables and nonlinear relationships, while reducing overfitting by optimal parametrization and monotonicity responses (Elith et al., 2008; Hofner et al., 2011). The algorithms fitted occurrence records per species (presences and pseudo‐absences) against the predictor variables.…”

“…A cross‐validation (CV) framework with 10‐fold latitudinal bands was implemented to find the optimal hyperparameter combination to reduce overfitting and improve the potential for transferability of the model (Assis, Araújo, et al., 2018; Krause‐Jensen et al., 2020; Vignali et al., 2020). The “grid search” method was used to test all possible hyperparameter combinations of the number of trees (50–1,000, step 50), tree complexity (1–6) and learning rate (.01, .005 and .001) for BRT, and the number of interactions (50–250, step 50), shrinkage (.25–1, step .25) and degrees of freedom (1–12) for AdaBoost (Fragkopoulou et al., 2021; Krause‐Jensen et al., 2020). Models trained the different hyperparameters interactively by fitting occurrence records with one withheld latitudinal band at a time, where performance was tested with the area under the curve (AUC) of the receiver operating characteristic curve (Fragkopoulou et al., 2021; Vignali et al., 2020).…”

“…Models trained the different hyperparameters interactively by fitting occurrence records with one withheld latitudinal band at a time, where performance was tested with the area under the curve (AUC) of the receiver operating characteristic curve (Fragkopoulou et al., 2021; Vignali et al., 2020). The optimal combination of hyperparameters was determined as the one producing models with the highest AUC in CV (Fragkopoulou et al., 2021; Krause‐Jensen et al., 2020; Vignali et al., 2020). Overfitting was controlled further by forcing predictors to fit monotonic responses based on the biological effect (positive or negative) expected on the probability of occurrence of species (Hofner et al., 2011).…”

“…The depth range, predictor choice and monotonic responses of species were set based on the literature (Supporting Information Appendix S1). The cross‐validation framework also allowed estimation of the performance and potential transferability of the models in independent data, by reporting the average AUC obtained in CV when modelling with the optimal hyperparameters (Assis, Araújo, et al., 2018; Boavida et al., 2016; Fragkopoulou et al., 2021; Krause‐Jensen et al., 2020; Wenger & Olden, 2012).…”

Summer Heatwave Impacts on the European Kelp Saccharina latissima Across Its Latitudinal Distribution Gradient

“…In Europe, species distribution models project a northward expansion of different kelp species (Müller et al, 2009;Assis et al, 2018). Newly exposed hard substrates in Arctic regions are already colonized by kelps (Krause-Jensen et al, 2020), whereas drastic loss of kelps is occurring at the southern distribution boundary (e.g., Voerman et al, 2013;Filbee-Dexter et al, 2020).…”