Fieldwork and laboratory analyses were conducted by a collaborative network covering the global range of two dominant and broadly distributed kelp species (S. latissima and L. hyperborea) (Fig. 1). Field decomposition rates of kelp detritus were quantified in concurrent, standardized litterbag experiments deployed in 12 regions throughout the northern hemisphere. Litterbag experiments are widely used to quantify decomposition rates in the field 66 by measuring the mass loss of plant material enclosed in mesh bags that allow water flow and microbial colonization while excluding large grazers and preventing biomass advection. In each region, three sites, approximately 0.5 to 10 km apart, were selected on sand or coarse sediment adjacent to rocky reefs in areas with low to moderate wave and current exposure (Supplementary Table 1). Litterbags were pre-assembled and shipped to all partners, ensuring identical treatments were deployed in all regions. We targeted overall patterns of kelp loss rather than attempting to distinguish between mesograzers (or detritivores) and microbial activity. Consequently, we did not vary mesh size of the litterbags as this can substantially alter light and water flow, which may affect kelp decomposition.
In each of the 12 regions divers haphazardly collected 24 adult blades with minimal to no epibionts of each targeted species. Six regions collected and deployed two species (S. latissima and L. hyperborea), and six regions deployed one species (S. latissima) (Supplementary Table 1). From each blade a 20-g piece of kelp tissue was sectioned ~15 cm from the base and at least 15 cm from the distal end and weighed to the nearest 0.01 g. This approach was chosen to maximize blade uniformity across regions as older distal tissue would be less uniform depending on age and fouling. Using newly formed basal tissue also minimized phenological or seasonal differences in detrital material from slight variation in timing of the trials across regions, which may influence the decomposition rates. Additional kelp samples (n = 8-10) were collected for each species for baseline C:N analyses.
A single kelp piece was loosely packed into each litterbag (plastic ~1 x 1-cm mesh bags) and placed into cages (four litterbags in each of the two cages for each species at each site), to allow access of smaller mesograzers. Cages were 20 cm by 20 cm by 40 cm and made of plastic 1 x 1 cm mesh (‘gutter guard’). Each cage was tethered with cable ties to a weight on the seafloor at ~8-m depth. In order to accurately quantify the impact of ocean climate on decomposition, we selected this cage size to exclude grazing by large herbivores in our experiments, which can drive localized increases in the turnover, size, and availability of kelp detritus in some areas 36 and could overwhelm measures of turnover in areas where they were locally abundant. All kelp pieces were kept damp after collection, stored in a dark cooler and deployed within 24 hours of collection.
Environmental variables known to influence decomposition were measured concurrently throughout the experiment at each site. Hourly light and temperature were measured by an Onset HOBO pendant temperature and light logger fixed to the top of a cage at each site. Only light records for the first two weeks of deployment were used to account for fouling of the sensor, which could shade and confound measurements over time. To estimate wave action, an Onset HOBO G logger was placed inside a mesh bag and added to a cage at each site to log hourly movement of the litterbags. We used the average product of logged acceleration along 3 axes (x, y and z, units of g3) of the period as a relative measure of movement of the litterbags.
Approximately 4-6 weeks into the experiment half the litterbags were collected (two from each cage). The remaining litterbags were collected after 12-18 weeks. Litterbags were lost at some sites (Supplementary Table 1). Samples were processed within 10 h of collection. All kelp fragments were removed from bags, patted dry, and weighed to the nearest 0.01 g. Weighed samples were rinsed in distilled water, oven dried at 60 °C for 48 h, and then shipped to the University of California (Davis, CA, USA) where they were analyzed for nitrogen and carbon tissue content.
We compared the obtained values of kelp decomposition to that of other marine detritus using data from litterbags or incubations obtained from the literature (Supplementary Table 4). For each type of organic material (seaweed, seagrass, mangrove, other particulate detritus (e.g., marine snow, zooplankton feces or debris), and dissolved organic material (DOM)) we calculated residence times (days to 50% loss). This metric enabled comparison between materials with different decay functions. We did not include refractory pools of DOM or below-ground decomposition.
Rates of kelp loss (average rate of biomass loss for each retrieval time at each site) as a function of environmental conditions and kelp tissue properties were analyzed by generalized linear mixed effects models. Sites were averaged because litterbags in the same cage were not independent replicates. We also calculated k values, using the equation y = e-kt, where y is the proportion of biomass remaining at a time point and t is the time elapsed since the beginning of the experiment (days), but linear rates of loss fit our dataset better. The lagged onset of decomposition in some of our study regions may explain why our linear decomposition rates, although similar to other regional decomposition experiments on kelp detritus 50,55, deviated from patterns of exponential decay shown for other types of organic material 67. Because we were examining kelp decomposition, any negative rates of loss (biomass increase or growth) were assigned a value of 0 in our model, as a growing kelp is undergoing little to no decomposition. This was important for sites in the subarctic, where kelp detritus continued to grow after deployment. Our predictor variables were obtained from logger data and stable isotope measures. The fixed effects were kelp species, average water temperature, range in water temperature, average light conditions, and relative water movement during the experimental period, as well as site nested within region as the random effects. We used two variables to capture temperature conditions, the average temperature over the deployment and the temperature range (the difference between the 10th and 90th percentiles) as temperature ranges varied markedly, from 0.6 to 18.6 °C. Average temperatures and peak temperatures (90th percentile) were highly correlated among sites (Pearson’s R = 0.96, p < 0.001), so peak temperatures were not included in our model. Temperature loggers were lost in the Gulf of Maine region, so temperatures were obtained from the closest meteorological weather buoy (19 km away).
We accounted for differences in starting kelp conditions using initial % carbon content in kelp tissue as fixed effects in the model. This variable was correlated with initial % nitrogen and C:N ratio (Pearson’s correlation tests, R > 0.7), so only initial % carbon was included in the model. Carbon content was modelled separately using a subset of the data, because these measures were not available for Gulf of Maine, Rhode I Sound, and the Gulf of St Lawrence. The main relationships between the other key variables (light, temperature, species) were similar in both models. To confirm the latitudinal gradient was statistically significant, we ran another model using the continuous variable of ‘latitude’ as a predictor of biomass loss instead of a categorical variable (region name) (Supplementary Information 1). We did not use ‘latitude’ in our final model because it was correlated with temperature and the environmental gradients underlying these latitudinal differences provided more interesting and operational information on spatial patterns of carbon turnover.
We tested for significant nitrogen enrichment of S. latissima and L. hyperborea using a 2-way ANOVA comparing %N at the start and end of the experiment among regions. Post-hoc comparisons were conducted for each region using Tukey’s tests.
All analyses were conducted in R (version 3.5.3). We used the glmer function from package lme4 to fit the generalized linear mixed-effects models. Decomposition models were fit with a gamma distribution and identity link function. We checked model residuals for violation of model assumptions and to investigate the suitability of the chosen distribution (i.e. deviance residuals vs. theoretical quantiles), dispersion and heteroscedasticity, using package DHARMa (Supplementary Information 1). To stabilize parameter estimation, we standardized mean light by dividing it by 100, so it matched the scale of the other predictor variables. We used likelihood ratio tests with single-term deletions to assess the importance of each fixed effect predictor in the models. Relationships between the most important predictor variables and decomposition rates were illustrated with package visreg, which shows the relationship between a single predictor and the model outcome while holding the other predictors constant 68.