Relating EOF with literature knowledge
Most of the available knowledge available for our species focuses on
spawning habitats (Table 2; Figure 1). By crossing the spatio-temporal
patterns in the EOF with this knowledge it is possible to identify EOF
principal components that spatially match with the spawning grounds and
verify if the temporal dynamics of the associated loading factor match
with the expected seasonality of spawning. It is then possible to
investigate the intra-annual variability and the environmental drivers
of spawning.
Investigating inter-annual variability
Reproduction is known to face intra-annual variability and is partly
driven by the temperature (Fincham et al., 2013; Huret et al., 2018).
Specifically, for sole, hake and sea bass, some studies have
investigated the relationship between reproduction timing and SST and
have evidenced an optimal range of temperature for reproduction
(references are in Table 2).
We hypothesized that the peak of the loading factor associated with the
spawning season represents the peak of the reproduction season, and we
investigate the inter-annual variability of the peak.
We check if the spawning peak identified from EOF matches with the
period of temperature optimal range (e.g. see Figure 4). SST data
were extracted from the Marine Copernicus platform
(https://marine.copernicus.eu/).
Results
Dimension filtrations and extracting average spatial
patterns: Pre-analysis of the EOF
For sole , we select the six first EOF dimensions that capture
50% of the variance (Figure 3). For hake , we filter the two
first dimensions that capture 30% of variance. For seabass , we
select the first dimension only. It captures 30% of the variance. Other
dimensions are not considered in this analysis as they are considered as
noise.
The averaged spatial distribution (denoted\(\overset{\overline{}}{S(x,\cdot)}\) in EOF equations) reveal
specific average patterns for each species (Figure S1). EOF results
presented in figure 4 have to be analyzed relatively with their average
spatial pattern. For sole , average distributions are relatively
coastal with high biomass offshore the Gironde Estuary (2°W - 45°N). Forhake , average spatial distribution is more offshore and
corresponds to the slope area. For sea bass, the mean pattern is very
coastal. Biomass is high along the Vendée coast (2°W-46°N to 3°W-47°N)
with a hotspot near Belle Île (3°W - 47°N), and along the Landes coast
(1.5°W-44°N to 1.5°W-45.5°N).
Identifying the essential habitats and associated
seasons
Taking sole as illustration, we perform a clustering analysis on the
loading factors (time steps) and eigen vectors (locations). This allows
to identify several areas and seasons that characterize sole
spatio-temporal dynamics (Figure 5, center and right). Six spatial
clusters and three seasons can be identified (Figure 5 - Note that the
clustering trees are available in Figure S3 and S4): (i) an area
constituted by clusters 1-3 that is mainly correlated to winter months
(November to February – Figure 5, left); (ii) a coastal area
constituted by clusters 5 and 6 that mainly correlates to summer, autumn
and early winter months (July to November, Figure 5, left); (iii) and an
area constituted by cluster 4 that mainly corresponds to the average
distribution for spring and early summer (March to June, Figure 5,
left).
A deeper insight on the temporal dynamics of each cluster is given in
the appendix (Figure S5). The clustering was also performed on the other
species and are presented in the supplementary material (all figures
after Figure S6).
Crossing the available knowledge with EOF to infer
spawning phenology
All species present a strong seasonal pattern (Figure 4).
For sole , a periodic signal is revealed in the loading factors.
Dimension 1 and 2 highlight high biomass in offshore areas in winter
(December to April) and relatively coastal distribution in summer. EOF1
mainly captures the coastal and offshore seasonal migrations without
highlighting spawning areas per se. Whereas in EOF2, offshore areas
correspond to reproduction grounds highlighted in Figure 1. Also, PC2
maximums fall within the period where SST are favorable for
reproduction. Then, for sole, dimension 2 seems to be the best
descriptor for reproduction phenology: orange areas in EOF2 are spawning
areas and PC2 maximums are spawning peaks. For hake , similar
seasonal patterns can be evidenced in dimension 1 and 2. There are (1)
shelf areas that are occupied during summer and (2) offshore areas on
the edge of the shelf that are occupied during winter which coincides
with spawning grounds from Figure 1. The maximum of PC1 falls within the
period when SST is favorable for reproduction. EOF2 represents offshore
and coastal seasonal migrations with less emphasis on reproduction.
Hence, we consider EOF1 as the dimension that best corresponds to
reproduction and we retain this dimension and the maximum values of PC1
to investigate the phenology of reproduction of hake.
For sea bass , the EOF 1 captures the variability off the
central shelf of the Bay of Biscay i.e. off the Gironde estuary.
The corresponding time amplitude showed a very strong seasonal pattern
with high positive peaks occurring in January/February. These peaks
match the period where SST is favorable for reproduction and the
spawning areas from Figure 1.
Inter-annual variability of reproduction and relationship
with SST
Our results also highlight inter-annual variability in reproduction
phenology for the three species (Figure 6).
For sole , the months of reproduction identified through PC2
falls between January and March. In 2012/2013, reproduction seems to be
a bit earlier; Maximum of PC2 is in December and falls outside the
period where temperature is favorable for reproduction. When looking at
the PC time series for 2012/2013 (Figure 4, PC2 for sole), the PC is
flatter than for the other years and reproduction could also occur later
on (there is another peak in March).
For hake , months of reproduction are a bit earlier and fall
between December and February.
For sole and hake , both the reproduction period and
the time range where temperature is favorable look to be relatively
stable.
For sea bass , reproduction months emphasize more variability.
The maximum of the PC1 time series falls between February and November
specifically at the beginning of the time series. By contrast, the
period where SST is favorable for reproduction is steady. This suggests
that other covariates than temperature may strongly affect reproduction
timing.
Discussion
‘VMS x logook’ data opens new gates to realize ecological analysis at a
much finer spatio-temporal resolution than ever before (Azevedo and
Silva, 2020; Gerritsen and Lordan, 2011; Murray et al., 2013), but still
only few applications have evidenced this potential through concrete
analysis on large time periods with massive amount of ‘VMS x logbook’
data.
In this paper, we combined an existing spatio-temporal model with a
dimension-reduction approach (EOF) to investigate the phenology of three
species of the Bay of Biscay (sole, hake and sea bass) based on ‘VMS x
logbook’ data.
Here, ‘VMS x logbook’ data give access to monthly distributions.
Combined with our modeling framework, it provides a way to analyze
spawning phenology at a much finer temporal scale as other data sources
(e.g., scientific survey) that would give access to quarterly
distribution at best. Such modeling approach synthesizing inter and
intra annual variability in spatial variation is a major result that
should support a broader access to VMS data for science (Hintzen et al.,
2012).