ELECTRONIC SUPPLEMENTARY MATERIAL
Affinity for natal environments by dispersers impacts reproduction and
explains geographic structure of a highly mobile bird
Robert J. Fletcher, Jr., Ellen P. Robertson, Rebecca Wilcox, Brian E. Reichert, James D. Austin,
and Wiley M. Kitchens
Additional Information on Modularity and Geographic Structure
Recently, emergent spatial structure in the movements of snail kites has been observed, which
was revealed through spatial modularity analysis [1]. Spatial modularity occurs where habitat
patches (or local populations) are tightly connected to other patches through movement of
individuals or their alleles but only weakly connected to the remaining patches in the landscape
[2]. In doing so, spatial modularity provides a formal description of the functional aggregation of
local populations, identifying a potentially critical scale for ecological and evolutionary
dynamics (e.g., ‘subpopulations’ or relevant ‘management units’). This approach is useful
because it honors complex dynamics that may arise, such as highly directional movement [3],
spatial variation in the resolution of critical scales (i.e., non-stationarity [4]), and effects of scale
beyond geographic distance alone [5, 6].
Formally, this approach is based on an extension of the Newman-Girvan algorithm from
statistical physics [7]. This approach defines modularity, Q, as:
Q
1
 ( Aij  Pij ) (C i , C j )
2m ij
where m is the total number of observed movements, A and P are square matrices where Aij
represents the number of observed movements from i to j and Pij is an expected value, and δ(Ci,
Cj) is an indicator matrix that is equal to 1 if i and j are members of the same module and zero
otherwise. We used a common expected value for directed networks of Pij = wi-outwj-in/w [8],
where wi-out is the emigration rate for patch i (i.e., number of dispersal events from habitat i) and
wj-in is the immigration rate for patch j (i.e., the number of dispersal events to breeding habitat j).
Consequently, this approach identifies spatial structure in dispersal beyond that expected based
on patch-specific emigration and immigration rates. We used a simulated annealing algorithm to
maximize the modularity function by iteratively searching for δ(Ci, Cj) that maximizes Q, on the
basis of the searching rule described in Guimera and Amaral [9]. This algorithm has been shown
to be very effective in identifying underlying structure on networks [9].
Previously, we found evidence of strong modular structure from mark-resight data, with
two modules occurring: one in the northern portion of their geographic range and a second,
larger module in the rest of their range [1]. That analysis was based on within-breeding season
movement data from 2005-2009. We have re-analyzed modularity at the annual time step across
the time frame considered in this work (1997-2013), finding nearly identical structure [10]. We
used the modules identified from the latter analysis here because it is more comparable to the
dataset used here to understand natal habitat preferences.
References
[1] Fletcher, R.J., Jr., Revell, A., Reichert, B.E., Kitchens, W.M., Dixon, J.D. & Austin, J.D.
2013 Network modularity reveals critical scales for connectivity in ecology and evolution.
Nature Communications 4, 2572.
[2] Fortuna, M.A., Albaladejo, R.G., Fernandez, L., Aparicio, A. & Bascompte, J. 2009
Networks of spatial genetic variation across species. Proc. Natl. Acad. Sci. USA 106, 1904419049. (doi:10.1073/pnas.0907704106).
[3] Fletcher, R.J., Jr., Acevedo, M.A., Reichert, B.E., Pias, K.E. & Kitchens, W.M. 2011 Social
network models predict movement and connectivity in ecological landscapes. Proc. Natl. Acad.
Sci. USA 108, 19282-19287.
[4] Fortin, M.J. & Dale, M. 2005 Spatial analysis: a guide for ecologists. Cambridge.
[5] Galpern, P., Manseau, M. & Wilson, P. 2012 Grains of connectivity: analysis at multiple
spatial scales in landscape genetics. Mol. Ecol. 21, 3996-4009. (doi:10.1111/j.1365294X.2012.05677.x).
[6] Expert, P., Evans, T.S., Blondel, V.D. & Lambiotte, R. 2011 Uncovering space-independent
communities in spatial networks. Proc. Natl. Acad. Sci. USA 108, 7663-7668.
(doi:10.1073/pnas.1018962108).
[7] Newman, M.E.J. 2006 Modularity and community structure in networks. Proc. Natl. Acad.
Sci. USA 103, 8577-8582. (doi:10.1073/pnas.0601602103).
[8] Leicht, E.A. & Newman, M.E.J. 2008 Community structure in directed networks. Physical
Review Letters 100. (doi:118703
10.1103/PhysRevLett.100.118703).
[9] Guimera, R. & Amaral, L.A.N. 2005 Functional cartography of complex metabolic networks.
Nature 433, 895-900. (doi:10.1038/nature03288).
[10] Reichert, B.E. 2014 Spatial structure in demography and movements of the endangered snail
kite: revealing multi-scale patterns and their implications for conservation [Dissertation].
Gainesville, FL, University of Florida.
Figure S1. Geographic variation in wetland types across the snail kite’s range. Lacustrine and
palustrine habitats taken from the National Wetlands Inventory, a standardized mapping system
for the United States administered by the U.S. Fish and Wildlife Service. We considered two
variables: the proportion of nest locations that occurred in palustrine and lacustrine habitats and
the proportion of lacustrine and palustrine habitat within 2 km surrounding each nest. Pie charts
show average proportion of lacustrine and palustrine habitats for each of the 19 wetlands where
we had >5 nests.
Figure S2. Silhouette plot from the cluster analysis (Partitioning around Medoids analysis; 2
cluster solution chosen because it had the greatest mean silhouette widths) on lacustrine and
palustrine habitats taken from the National Wetlands Inventory, a standardized mapping system
for the United States administered by the U.S. Fish and Wildlife Service. Observations with large
silhouette widths, si, (near 1) are well clustered, small si (near 0) suggest that observations lie
between two clusters, and negative si are likely placed an incorrect cluster. We considered two
variables: the proportion of nest locations that occurred in palustrine and lacustrine habitats and
the proportion of lacustrine and palustrine habitat within 2 km surrounding each nest. Y-axis
shows bars for each of 19 wetland sites considered, and the x-axis shows the silhouette width,
which describes how well a wetland fits within the cluster that it is grouped in contrast to other
clusters (the higher the silhouette values, the better the fit). Numbers on the right show average
silhouette values for the two groups. Site acronyms: GW: Grassy Waters; 2B: Water
Conservation Area 2B; Harns: Harns Marsh; Lox: Loxahatchee Wildlife Refuge; 3B: Water
Conservation Area 3B; 3A: Water Conservation Area 3A; ENP: Everglades National Park;
LakeJ: Lake Jackson; SJM: Saint John’s Marsh; BICY: Big Cypress National Preserve; STA1:
Stormwater Treatment Area 1; STA5: Stormwater Treatment Area 5; LakeR: Lake Runnymeade;
TOHO: Lake Tohopekeliga; ETOHO: Lake East Tohopekeliga ; Istok: Lake Istokpoga; OKEE:
Lake Okeechobee; KISS: Lake Kissimee; LakeH: Lake Hatchinea. We note that removing sites
that were less well classified provided qualitatively similar results in all analyses.
12
(a) Second-year birds
140
120
Frequency
10
100
8
80
6
60
4
40
2
20
0
0
0
120
50
100
150
200 250
300
0
350
(c) Birds born in lacustrine wetlands
100
Frequency
(b) All birds
25
50
100
150
200
250
300
350
(d) Birds born in palustrine wetlands
20
80
15
60
10
40
5
20
0
0
0
50
100
150
200 250
distance (km)
300
350
0
50
100
150
200
250
300
350
distance (km)
Figure S3. Observed dispersal distances from natal site origin to nesting site location for (a) 2nd
year birds (natal dispersal), (b) all birds, (c) all birds born in lacustrine wetlands, and (d) all birds
born in palustrine wetlands.
(a)
(b)
Figure S4. Spline correlograms for (a) 2nd year birds (natal dispersal), and (b) all birds.
Correlograms based on the residuals of the best models that explained dispersal (Tables S1, S2).
Confidence envelopes based on bootstrapping (n = 1000 samples).
Table S1. Model selection results for conditional logit models comparing wetland selection for
nesting by dispersing juvenile snail kites.
Model
K
LL
AICc
AICc
AICc
weight
Natal type similarity + distance
2
-89.0
181.9
0.00
0.76
Natal type similarity
1
-91.1
184.3
2.36
0.23
Natal nest similarity + distance
2
-95.5
195.1
13.16
0.00
Current wetland type
1
-96.6
195.2
13.28
0.00
Natal area similarity + distance
2
-95.6
195.3
13.36
0.00
Distance
1
-96.9
195.9
13.95
0.00
Natal nest similarity
1
-103.2
208.3
26.37
0.00
Natal area similarity
1
-106.6
215.3
33.35
0.00
Current wetland area
1
-110.1
222.2
40.29
0.00
Notes: Natal type similarity: categorical similarity of breeding site to natal site (palustrine versus
lacustrine wetland); Natal nest similarity: similarity of breeding site to natal site based on nesting
substrates used by kites (bray-curtis similarity); Natal area similarity: similarity of breeding site area to
natal site area (Euclidean distance); Current wetland type: type of wetland of current breeding site;
Distance = distance between natal site and breeding site (log-scale); Current wetland area: log(hectares)
of current breeding site; LL = log-likelihood; AICc = Akaike’s Information Criterion, corrected for
sample size; AICc = AICc for model i -minAICc.
Table S2. Model selection results from conditional logit models testing hypotheses for variation
in natal habitat preferences.
Model
K
LL
AICc
AICc
AICc
weight
Natal type similarity × age + natal type similarity ×
drought + distance
4 -1235.2
2478.4
0.00
0.79
Natal type similarity × age + distance
3 -1237.5
2481.0
2.61
0.21
Natal type similarity × drought + distance
3 -1247.7
2501.5
23.09
0.00
Natal type similarity + distance
2 -1250.4
2504.8
26.40
0.00
Distance
1 -1263.3
2528.6
50.22
0.00
Current wetland type
1 -1303.6
2609.1 130.74
0.00
Natal type similarity
1 -1309.5
2621.0 142.56
0.00
Notes: Natal type similarity: categorical similarity of breeding site to natal site (palustrine versus
lacustrine wetland); Age = age of individual; drought: categorical index of annual drought severity (from
Palmer’s index < -2); Distance = distance between natal site and breeding site (log-scale); Current
wetland type: type of wetland of current breeding site (palustrine versus lacustrine wetland); birth year =
year individual was born; current year = year of current breeding attempt; AICc = Akaike’s Information
Criterion, corrected for sample size; AICc = AICc for model i -minAICc. For models with interaction
terms, main effects are not shown. Age and drought effects did not vary within strata (paired use-available
data), so additive effects could not be considered.
Table S3. Model selection results from mixed effects models testing for variation in (a) nest
success (daily survival rates) and (b) the number of young fledged/successful nest.
Model
K
LL
AICc
AICc
AICc
weight
(a) Nest success
Natal type similarity + age
6
-561.0
1134.1
0.00
Natal type similarity
5
-562.8
1135.7
1.69
Natal type similarity × age
7
-560.9
1136.0
1.93
Current wetland type
5
-563.1
1136.3
2.22
Natal type similarity + distance
6
-562.5
1137.2
3.13
Age
5
-563.7
1137.4
3.35
Null
4
-564.8
1137.6
3.53
Natal type similarity × age + distance
8
-560.8
1137.9
3.81
Distance
5
-564.7
1139.6
5.49
(b) Number of young fledged/successful nest
Null
3
-161.7
329.7
0.00
Current wetland type
4
-161.6
331.6
1.92
Age
4
-161.7
331.7
2.03
Natal type similarity
4
-161.7
331.8
2.12
Distance
4
-161.7
331.8
2.13
Natal type similarity + age
5
-161.7
333.9
4.18
Natal type similarity + distance
5
-161.7
334.0
4.29
Natal type similarity × age
6
-161.4
335.5
5.79
Natal type similarity × age + distance
7
-161.4
337.7
8.02
Notes: Natal type similarity, or categorical similarity of breeding site to natal site (palustrine
0.34
0.15
0.13
0.11
0.07
0.06
0.06
0.05
0.02
0.36
0.14
0.13
0.13
0.13
0.05
0.04
0.02
0.01
versus lacustrine wetland); Age = age of individual; Distance = distance between natal site and breeding
site (log-scale); Current wetland type: type of wetland of current breeding site; LL = log-likelihood; AICc
= Akaike’s Information Criterion, corrected for sample size; AICc = AICc for model i -minAICc. For
models with interaction terms, main effects are not shown. For all models, individual and site were
considered as random effects. Models that included drought effects (see Table S1) did not converge. For
the analysis of nest success, date was included in all models (January 1 = 1, December 31 = 365).
Table S4. Model selection results from mixed effects models testing for factors predicting spatial
structure (modularity).
Model
K
LL
AICc
AICc
AICc weight
Wetland type similarity + distance
4
-73.04 154.31
0.00
1.00
Wetland type similarity
3
-84.36 174.86
20.54
0.00
Wetland type similarity + area
4
-84.04 176.33
22.02
0.00
Distance
3
-91.58 189.31
35.00
0.00
Distance + area
4
-91.01 190.25
35.94
0.00
Null
2 -118.29 240.65
86.34
0.00
Area
3 -118.28 242.71
88.39
0.00
Notes: Wetland type similarity: categorical similarity of wetland type between pairs of wetlands
(receiving site to sending site; palustrine versus lacustrine wetland); Distance = distance between natal
site and breeding site (log-scale); Area = similarity of habitat areas between pairs of wetlands (taken from
Euclidean dissimilarity matrix); Null = intercept-only model for fixed effects; AICc = Akaike’s
Information Criterion, corrected for sample size; AICc = AICc for model i -minAICc.
Table S5. Model results for best model explaining wetland breeding selection in (a) one-year old
birds and (b) all birds (from Tables S1, S2). For (b), we report robust SE that accounted for
repeated measures from individual kites over time for all inferences.
Model
 (SE)
Odds-ratio
exp()
95% CI
Z
P-value
(a) Juveniles
Similarity
Distance
2.04 (0.53)
-0.29 (0.20)
7.76
0.74
2.73-22.08
0.51-1.10
3.84
-1.47
0.00012
0.1404
(b) All birds
Similarity
Distance
Similarity  age
Similarity  drought
1.43 (0.29)
-0.51 (0.09)
-0.10 (0.03)
-1.81 (1.32)
4.18
0.60
0.90
0.16
2.34-7.46
0.50-0.72
0.85-0.97
0.01-2.16
4.84
-5.64
-2.94
-1.38
<0.0001
<0.0001
0.0033
0.169