When foraging for clams, sea otters cause physical disturbance to marine sediments. Some of their preferred prey (e.g. butter and steamer clams) live intertidally, alongside the seagrass beds that fringe the Alaskan shorelines. It was recorded during the 2017 field season that the number of pits dug outside and at the edge of seagrass beds often exceeded the number within the same bed. Seagrass biomass is actively removed with such bioturbation and it was questioned whether increased digging activity at the bed’s edge reduces seagrass extent in the intertidal.

This was investigated observationally and experimentally: [Observational] The elevation of the upper edges (diffuse and continuous) was measured across a sea otter gradient, the hypothesis being that the upper extent of seagrass beds will be lower in elevation in areas with higher sea otter presence compared to seagrass beds minimally influenced by sea otters. The upper and lower extent of the Fucus band was also measured to assess potential differences in elevation due to exposure. In addition to elevation data, the width of the diffuse seagrass bed was measured, as well as seagrass density of the diffuse edge, continuous edge, and inside the bed. [Experimental] Cages and control plots (n = 3; ~ 1.05 x 1.60 m) were installed along the edge of a seagrass bed at 10 sites located in regions with elevated sea otter presence. Half of the plots consisted of existing vegetation and the other half bare sediment above the bed. These plots will help determine if the upper edge of the seagrass bed is capable of extending higher into the intertidal if protected from sea otter digging activity. The control plots will also be used to estimate the rate of pits dug and the area disturbed.


For the Analysis, we need the following information:



An introduction to the sites and site data

We measured elevations at 26 sites across POW, and installed cages at 10 sites from Naukati to Klawock.
























Great. Now that we know where we are working, let’s compile a comprehensive dataframe of site characteristics. The following shows all of the site-level variables that we have and that could be considered when investigating relationships with the upper extent of seagrass. All of these variables were collected/calculated in the same way as they were for the 2017 season.

After a flurry of code…

##  [1] "site"                  "longitude"            
##  [3] "latitude"              "date_grass_MM.DD.YY"  
##  [5] "date_otts1_MM.DD.YY"   "date_otts2_MM.DD.YY"  
##  [7] "so_region1"            "so_region2"           
##  [9] "beach_slope"           "site_polygon_area_km2"
## [11] "so_index"              "so_duration"          
## [13] "pop_dens_surv_km2"     "dens_otter1"          
## [15] "dens_otter2"           "dens_otter_avg"       
## [17] "prop_otter_shells"     "sed1_qual"            
## [19] "sed2_qual"             "sed1_edge"            
## [21] "sed1_inside"           "sed1_outside"         
## [23] "sed2_edge"             "sed2_inside"          
## [25] "sed2_outside"          "sed1_site_avg"        
## [27] "sed2_site_avg"         "sedall_site_avg"      
## [29] "sed1_abovegrass"       "sed2_abovegrass"      
## [31] "pits_edge"             "pits_inside"          
## [33] "pits_outside"          "pits_site_sum"        
## [35] "pits_abovegrass"



The sea otter index:



The index on the left uses identical input values as the index from last year, where the boat-based sea otter counts were included as two seperate input variables. The index on the right uses the means of these two sea otter surveys per site. Franz Mueter recommended that the two surveys be included, seperately. Interpretting the graphs with such recommendation in mind, I do find the larger jump between “S Fish Egg Island” and “Garcia Cove” appropriate after working in the collective sites for the past 1-2 years. As such, I will proceed with the index on the left but wanted to present both options to the group, here.






Determining elevations to assess upper extent of seagrass

…and inclusion of seagrass density



First, here is a schematic of what elevation data we collected in the field (along a 50-m transect):



The true elevation of the seagrass and Fucus measurements were derived from knowing the elevation of the waterline at a specific time. For determining the elevation of the waterline, it is tempting to rely tidal elevation provided by a nearby tidal station – this is fine for some sites due to their close proximity, but for others, the true waterline elevation will diverge from such measurements due to spatiotemporal influences on water movement. I corrected for this by reaching out to someone connected to the US Coast Guard and NOAA branches responsible for assessing the accuracy of marine benthic maps used for navigation. They re-surveyed a good chunk of POW in 2008, and their work required applying spatiotemporal corrections for tidal amplitude in subregions across POW (example shown below). Lucky us!




Great. Below is a list of elevation data available for analysis.

Note: The “edge_diffuse_elev_corrct” and “edge_continuous_elev_corrct” data reflect seagrass edge elevations that were corrected using Fucus elevations, potentially correcting for varation in exposure. In future analysis, I am interested in calcuating an exposure variable, based on shoreline aspect and fetch potential.

##  [1] "site"                         "fucus_upper_elev_mllw_cm"    
##  [3] "fucus_lower_elev_mllw_cm"     "fucus_mean_elev_mllw_cm"     
##  [5] "fucus_band_height_cm"         "edge_diffuse_elev_mllw_cm"   
##  [7] "edge_continuous_elev_mllw_cm" "edge_diffuse_elev_corrct"    
##  [9] "edge_continuous_elev_corrct"  "diffuse_band_distance_cm"    
## [11] "diffuse_band_height_cm"       "diffuse_band_area_m2"


These data columns will be added to the site dataframe, as will the seagrass density data below. We have a new master dataframe!

## [1] "site"                         "shoot_density_inside_m2"     
## [3] "shoot_density_continuous_m2"  "shoot_density_diffuse_m2"    
## [5] "flower_density_inside_m2"     "flower_density_continuous_m2"
## [7] "flower_density_diffuse_m2"






SO THEN, let’s do that analysis thing


Question: Do sea otters reduce the upper extent of seagrass beds via bioturbation?

To address this, we can look at the following dependent variables:


First, are the data normally distributed? In short, not really (other than the Fucus elevations). For now, I will show the results using both un-transformed and transformed data, where appropriate, but we should talk about which transformations we want to choose for analysis. I have included histogram plots for all dependent variables on this page: Follow this link





Looking at Fucus

I’m not expecting for this to be a reported result in any manuscript, but if we choose to consider Fucus as an indicator of exposure (like people have with barnacles), we could use these data to correct our seagrass elevation data for exposure. An alternative is to generate an exposure value based on maximum fetch potential, but that will take a bit more time/fiddling to determine.

Below, we see that the mean (of upper and lower Fucus elevations) suggests that (1) the range of elevations are similar and (2) there is a slight, non-significant increase from low to high. We do, however, see a larger range in the Fucus elevations in the low sea otter region.

NOTE: The “low” and “high” sea otter regions were determined using the sea otter index, where all sites that are above the obvious break (i.e. between “S Fish Egg Island” and “Garcia Cove”) are considered as “high” sites.






Looking at Zostera elevations



Let’s start with comparing values in “low” and “high” sea otter categories

NOTE: The “low” and “high” sea otter regions were determined using the sea otter index, where all sites that are above the obvious break (i.e. between “S Fish Egg Island” and “Garcia Cove”) are considered as “high” sites.


The two plots below suggest that the mean elevation of both the diffuse and continuous edges are lower in areas with higher sea otters than in areas with lower otters. Neither, however, are significant when a t-test is applied (see table)…but the latter is pretty close for an ecological question that probably has a lot of noise.



A table to support the values and t-test for the above plots:

SO Region   | Diffuse edge (cm, MLLW)  | Continuous edge (cm, MLLW) | p-value
------------|--------------------------|----------------------------|----------
Low         | +10.32                   | -2.35                      | 0.124
High        | -10.45                   | -27.62                     | 0.073
Difference  | 20.77                    | 25.27                      | n/a



That said, it is also worth mentioning that some some sites were just weird, without clear explanation. For example, the seagrass in a few sites grew very high in the intertidal (up to the Fucus band) with almost no diffuse growth and these sites were noted as inconsistant with the majority of sites found on POW. These sites include Big Tree Bay, Big Clam Bay, and both Fish Egg sites. Out of interest, these sites all had sands that were more “sugar-like”, a shallow beach slope, and with Phyllospadix growing in them (but so did Blanquizal Bay and Goat Mouth Inlet). This is what I mean by noise. I think that with slightly higher replication, this noise would have been reduced. “Ecosystems are complex, AF.” With the followinbg tables, I’m not advocating for removing any sites, I’m exploring potential explanations for noise in the dataframe.

The same table but with all noticebly high-growing seagrass sites removed:

    SO Region   | Diffuse edge (cm, MLLW)   | Continuous edge (cm, MLLW)  | p-value
------------|-------------------------------|-----------------------------|----------
Low         | -1.04                         | -14.22                      | 0.137
High        | -14.21                        | -36.06                      | 0.037
Difference  | 13.17                         | 21.84                       | n/a


The same table but with all Phyllospadix sites removed:

SO Region   | Diffuse edge (cm, MLLW)   | Continuous edge (cm, MLLW)  | p-value
------------|---------------------------|-----------------------------|----------
Low         | +2.85                     | -11.16                      | 0.059
High        | -18.89                    | -36.67                      | 0.023
Difference  |  21.74                    | 25.51                       | n/a






Let’s move into a sea otter gradient.

  1. Looking at the upper edges against the SOI: are those intermediate disturbance curves? It looks like it…regardless, the plots below show that the elevation of the diffuse band scales very well with the elevation of the continuous band. This could allow us to move forward with the mean of the two edges to simply wording in the results/discussion.






  1. Looking at area of the diffuse band: There appears to be no difference in the mean widths of the diffuse growth band, compared as binary categories (low and high sea otters) or against the sea otter gradient. There is a hint of intermediate disturbance in the latter, but the p-value is high with a low r^2 value.






How about direct interactions with specific independent variables?

  1. Sea otter densities derived from boat surveys: There are no apparent relationships when directly comparing the mean seagrass edge elevation and the mean sea otter densities within a 2 nm swimming radius. Note how the sea otter index values fall across the spread of the scatterplots.


  1. Tinker population density model: This component of the sea otter index appears to be an important component in the model.


  1. Number of sea otter pits: In the left plot, are the untransformed data fit with a polynomial curve – I haven’t figured out how to plot an extinction curve (or get the model statistics), yet. The plot on the right show the same data but with the dependent variable (“mean edge elevation”) log-transformed. The stats are similar enough.


  1. Sea otter cracked shells: The relationship with the log-transformed data (right) is weaker.


  1. Sediments: The sites that had “sugar sand”, or fine + semi-compacted sand throughout much of the site seemed to have a markedly higher seagrass edge (mean = +51.4 cm MLLW). The edge elevations at muddy and sandy sites (mud: -14.7; sand: -19.0 cm MLLW) appear to be similar across sites.




Can we estimate area seagrass gained and/or lost due to intermediate and high sea otter occupation?

We can do a “back-of-the-envelope” calculation by determining the mean difference in elevation of seagrass edges between low and high sea otter regions, and then applying the slope to estimate the width of the band that sea otters might have bioturbated away. We just need to decide if we believe the patterns and then whether this “back-of-the-envelope” calculation can be streamlined in some way.

  1. Intermediate disturbance: Taking the mean beach slope and the mean difference in elevation of the edges at low and mid sea otter index values, we can estimate that intermediate bioturbation may increase the upper extent of the bed by a mean of 5.34 m^2 seagrass per m shoreline of appropriate habitat. Why is this? If this is a real effect (I’m not totally convinced ATM), then my best guess for the driver of this pattern is an interaction with clams…where in the absence of sea otters, larger clams and seagrass rhizomes clash and prevent seagrass from growing higher in the intertidal. This competition may be relieved by intermediate foraging along the edge and outside the bed, thus resulting in an increase in edge elevation. The reason that I’m not yet totally convinced by this apparent effect is because of variation in site attributes that seem to align clump in the intermediate sites (e.g. sugar sand and “odd sites”).

  2. Persistant disturbance: For sites with values in the highest third of the sea otter index (estimated here as an appropriate breaking point for when bioturbation forcing exceeds intermediate), the data suggest a potential reduction of 8.84 m^2 seagrass per m shoreline, when transitioning from intermediate to persistant sea otterness. Grouping the 8 lowest and 8 highest index values and “low” and “high” otter sites suggests that high sea otter sites have 3.32 m^2 of seagrass less per m shoreline compared to sites with minimal sea otter influence. The lower mean edge elevation, here, is likely from direct removal during high disturbance and not from competition with clams as hypothesized above.


P.S. We notice that the slopes appear slightly different across the low and high zones. Using a t-test, the two groups are not significantly different (p = 0.407). Additionally, there is no compelling correlation between the slope of a beach and the mean elevation of the seagrass edge…I am not concerned about variation in slope for this study.






Looking at Zostera densities


This section will expand when we get seagrass biomass data, but see below. Although there are no significant relationships with seagrass shoot density and the sea otter index, there is a relatively strong positive relationship between the elevation of the continuous edge and the density of seagrass (first plot). It helps explain the almost-identifiable parabolic curve in the third plot (so index vs continuous edge density)…since a similar shape is seen with the elevation data and density positively scales with elevation. This is inconsequential for THIS study, but is a fun thing to consider for morphopetrics and 3D-structure of the bed for other studies.







Looking at experimental cage data


I’m still working on this. After revisiting all cages at the very end of the summer, I learned the following:


(Left) Experimental cage at Garcia Cove. (Right) Control plot with a sizable excavation in the unvegetated side, Guktu Cove.



Light was also measured within and outside of a cage replica in the air, below 1 m of seawater, and below 5 m of seawater. There is a detectable difference, where seagrass growing within cages should receive slightly less light, on average, compared to control plots. I doubt that this will be strong enough to inhibit expansion of the upper edge, but it could explain any differences in rates of productivity or speed of expansion (both of which we are not quantifying).