20
21
22
17
18
19
12
13
14
15
16
10
11
7
8
9
4
5
6
1
2
3
Supporting Information
Supporting Methods
Oceanographic model
We used a buoyancy-, wind-, and tidally-driven Lagrangian particle dispersal model for the coastal region of North America from California to southeastern Alaska. This model uses depth-varying ocean velocities to simulate the movement of individual particles from a given starting location and depth. Ocean velocities were computed with a highresolution diagnostic finite element model of the northeast Pacific Ocean forced with seasonal climatologies of temperature, salinity, and wind stress (Foreman et al.
2008).
While the model was developed to provide seasonal estimates of dynamic ocean topography, and in this capacity was validated against satellite altimetry and coastal tide gauge measurements, it also produces three-dimensional velocity fields that can be used for transport and dispersion studies. In particular, depth-varying velocities are computed at the vertices of an unstructured triangular grid with 169,869 triangles and sides whose lengths vary from 100 m in some narrow coastal channels to 70 km in the deep ocean
(Foreman et al.
2008). We used the summer-averaged velocities to simulate the spawning season of P. miniata . These included a time-invariant background buoyancy- and wind-driven flow field and a tidal current that is comprised of time-varying flows from eight constituents (the four largest semi-diurnals and diurnals), comprising approximately 85% of the total tidal range.

42
43
44
45
39
40
41
34
35
36
37
38
29
30
31
32
33
23
24
25
26
27
28
We used these velocity fields with the drogue-tracking algorithm implemented in
DROG3D (Blanton 1995) to trace passive Lagrangian particles through the combined tidal and buoyancy-wind velocity fields with a 0.5 hour time step. This particle tracking simulation was deterministic (without a random diffusion step), but outcomes were variable depending on the half-hour at which the drogue was released because of differences in the tidal cycle, and we therefore obtained high variation among particles released from the same location at different release times. We held particles at 1m of depth from the surface, based on observation of other echinoderm larvae (Pennington &
Emlet 1986; Martin et al.
1997) and observations of negative-geotactic swimming behaviour in Pisaster ochraceus (Crawford & Jackson 2002) and Patiria miniata (JS personal observation). During particle tracking, some simulated larvae were grounded at the coastline at the point of release. We assumed that these larvae did not survive until the competency period and did not include them as viable recruits. This error indicates that the model lacked precision for coastally-entrained particles, and removing them likely lead to an underestimate of regional recruitment. However, it is not likely to affect estimates of cross-regional connectivity.
Seeding locations
We defined 10 coastal regions of 110 km stretches of coastline within the northern end of the P. miniata range (Fig.1). These were in sites where P. miniata had been observed, or where their habitat (shallow rocky substratum) is known to exist. Within each region, we established 40 seeding locations along the coastline each ~2km offshore to reduce the number of particles that became grounded along the coastline near the release sites. We

65
66
67
68
62
63
64
57
58
59
60
61
52
53
54
55
56
46
47
48
49
50
51 tracked simulated larvae from each location starting at 25 different times throughout the breeding season, producing 1000 tracks from each region. These regions included areas of the coastline which were not sampled genetically, but provide “stepping stones” for more realistic dispersal between the 5 sampled populations. We refer to geneticallysampled populations using bold capitals ( BA, WH, BB, HG, AK ) and to stepping stone populations using lower case italics ( cvi, scc, shg, kit, pru; see Table 1 for definitions).
Connectivity matrix
To estimate the number of simulated larvae recruited into each region, we defined an area bounded by the 50km buffer from the center of each region on the coast using
Geographic Information Systems (ESRI, ArcGIS 10) (Fig.1a). A larva was considered to settle within any such settlement region through which it passed during its competency period, which we defined as being 40-70 days, the time between which we expected larvae to be capable of settling and still viable (Basch 1996). We assumed that a constant fraction survived across this period. We chose not to impose an arbitrary daily mortality rate, but note that our estimates of connectivity are likely overestimates as a result. We calculated dispersal probability between pairs of settlement regions based on the number of arrivals out of the total number of larvae started from each region.
From preliminary work we determined that there was no evidence of direct connectivity between sites in California and those in British Columbia and Alaska. This is to be expected, as the predominant wind, and thus surface current direction, in summer is southward. In addition, we had lower confidence in dispersal projections from this region of the model due to lower grid cell resolution and fewer available salinity and

88
89
90
85
86
87
80
81
82
83
84
75
76
77
72
73
74
78
79
69
70
71 temperature observations. Therefore, we did not include California populations in the connectivity matrix, and focussed on hypotheses of genetic structure within the northern part of the P. miniata range only.
Genetic model
To predict the genetic structure arising from the estimated connectivity matrix, we ran a genetic simulation similar to that in Galindo et a l. (2006). We simulated 50 unlinked loci each with 2 alleles starting at a frequency of 0.5 in each population. All populations had
N pop
individuals, and in every generation, N larv
larvae were randomly sampled from parents, and dispersed to each of the other populations according to probabilities in the connectivity matrix. When new recruits arrived at each population, we randomly subsampled N pop
individuals in the new population to simulate post-settlement mortality, thus the population sizes remained stable after each generation. There were thus two episodes of genetic drift (random sampling of allele frequencies) in every generation: once when larvae were sampled from parental populations, and again when the population was reduced to its carrying capacity after new recruitment.
We measured overall and pairwise F
ST
as the standardized variance in allele frequency
(p):
F
ST
=var(p)/(mean(p)*(1-mean(p)))
We found that pairwise F
ST
values varied greatly between model runs (simulating different loci), with a mean stabilizing at ~50 model runs. We therefore used the mean
F
ST
from 50 model runs (or loci) to generate robust predictions for pairwise F
ST
.

107
108
109
110
111
101
102
103
104
105
106
97
98
99
100
91
92
93
94
95
96
We ran the model for sufficient time for genetic structure to become independent of starting conditions, but before many alleles became fixed within populations. Longer time periods did not strongly affect patterns of spatial differentiation, but increased variability among model runs as more populations became fixed for single alleles (see
Fig. S1). Values of N pop
and N larv
affected the rate at which populations reached driftmigration equilibrium but did not qualitatively or quantitatively affect F
ST
values (Fig.
S1). We present results from N pop
=200 and N larv
=100 here, for which pairwise F
ST became independent of starting conditions within ~500 generations (Fig. S1). We therefore used the used F
ST
at 500 generations as predicted population connectivity.
Displacement distance
We calculated the displacement distance for each simulated larva as the shortest absolute distance between starting locations and ending locations midway through the competency period (55 days). For comparison, we also calculated the predicted dispersal distance derived by Siegel et al.
(2003) for a larva with the same pelagic duration. This relationship was based on the regression between larval duration and empirically-derived dispersal distances from genetic isolation-by-distance analyses of 32 species from a range of locations (Siegel et al . 2003): displacement distance (km) = 1.33*(Pelagic larval duration, days) 1.30
.

112
113
114
115
116
Fig. S1. Pairwise F
ST between two example population pairs with time, under different model parameters for population size and number of larvae successfully dispersing from each. Comparisons between Bamfield ( BA ) and Alaska ( AK ) (black), and Bamfield and
Winter Harbour ( WH ) (green) are shown from 50 independent model runs with the same starting conditions. Thick lines show mean F
ST
values across model runs.
117
118

127
128
129
130
131
132
133
119
120
121
122
123
124
125
126
134
135
136
137
138
139
140
141
References for Electronic Supplementary Information:
Basch LV (1996) Effects of algal and larval densities on development and survival of asteroid larvae. Marine Biology 126 , 693-701.
Blanton BO (1995) DROG3D: User’s manual for 3-dimensional drogue tracking on a finite element grid with linear finite elements, Ocean Processes Numerical
Methods Laboratory, University of North Carolina at Chapel Hill.
Crawford BJ, Jackson D (2002) Effect of microgravity on the swimming behaviour of larvae of the starfish Pisaster ochraceus . Canadian Journal of Zoology/Revue
Canadienne De Zoologie 80 , 2218-2225.
Foreman MGG, Crawford WR, Cherniawsky JY, Galbraith J (2008) Dynamic ocean topography for the northeast Pacific and its continental margins. Geophysical
Research Letters 35 , L22606, doi:10.1029/2008GL035152.
Galindo HM, Olson DB, Palumbi SR (2006) Seascape genetics: A coupled oceanographic-genetic model predicts population structure of Caribbean corals.
Current Biology 16 , 1622-1626.
Martin D, Claret M, Pinedo S, Sarda R (1997) Vertical and spatial distribution of the near-shore littoral meroplankton off the Bay of Blanes (NW Mediterranean Sea).
Journal of Plankton Research 19 , 2079-2089.
Pennington JT, Emlet RB (1986) Ontogenetic and diel vertical migration of a planktonic echinoid larva, Dendraster excentricus (Eschscholtz) - occurrence, causes, and probably consequences. Journal of Experimental Marine Biology and Ecology
104 , 69-95.

142
143
144
Siegel D, Kinlan B, Gaylord B, Gaines S (2003) Lagrangian descriptions of marine larval dispersion. Marine Ecology Progress Series 260 , 83–96.