French et al.: Strong relationship between catch of Hippoglossus hippoglossus and availability of habitat for juveniles 
1 13 
Availability of suitable habitat is the cornerstone of 
several ecological theories. In particular, the theory of 
density-dependent habitat selection describes how ani¬ 
mals occupy the most suitable habitat first and expand 
to more marginal habitat when competition for re¬ 
sources in the prime habitat reduces the fitness benefit 
of that area (Fretwell and Lucas, 1970; MacCall, 1990). 
In practice, this concept has been documented in mul¬ 
tiple marine systems (Swain and Wade, 1993; Marshall 
and Frank, 1995; Rangeley and Kramer, 1998; Shack- 
ell et al., 2005). There is also extensive evidence that 
range size and abundance of a species are correlated, 
and that availability of suitable habitat is strongly and 
positively correlated with total abundance (Gaston and 
Blackburn, 1996; Brosse et al., 1999; Holbrook et al., 
2000; VanDerWal et al., 2009). 
Any life history stage is subject to habitat limitation 
and to the nursery-size hypothesis that stems from the 
idea that recruitment and adult fish population den¬ 
sity can be constrained by the availability of nursery 
habitat because of density-dependent mortality during 
juvenile life stages (lies and Sinclair, 1982; Rijnsdorp 
et al., 1992; Gibson, 1994; Beverton, 1995). The idea 
has resurged in the results of recent studies that have 
shown a solid relationship between availability of nurs¬ 
ery habitat and the recruitment of adults (Sundblad et 
al., 2014; Wilson et al., 2016). It is theorized that be¬ 
cause younger fish are likely to have a narrower range 
of suitable habitat, regardless of how well-defined their 
nursery grounds are, juvenile habitat falls within a 
restricted domain of the adult range (Gibson, 1994; 
Beverton, 1995). To explore this notion, Sundblad et al. 
(2014) mapped nursery habitat distributions for preda¬ 
tory fish between mainland Finland and Sweden’s ar¬ 
chipelago region of the Baltic Sea, and found that they 
can be quantified and used to help estimate potential 
adult production. Similarly Wilson et al. (2016) added 
support to the nursery size hypothesis; they found a re¬ 
lationship between increased recruitment and the pres¬ 
ence of juvenile flatfish and concluded that accurate 
predictions of flatfish nursery locations can be useful 
for population management. 
We use these principles to support evidence-based 
management by quantifying suitable habitat for juve¬ 
nile Atlantic halibut and relating it to adult landings 
in directed fisheries. In this article, we expand upon 
the work of Shackell et al. (2016) who hypothesized 
that there was ample suitable habitat available in U.S. 
waters to support a larger juvenile population. We es¬ 
timate the distribution of suitable juvenile habitat 
in each NAFO division, within and outside Canada’s 
Exclusive Economic Zone (EEZ). We then express the 
amount of suitable habitat (SH) per NAFO division 
as a relative value (SH-based shares: the proportion 
of the total available suitable habitat that falls within 
each NAFO division), and show that these values are 
related to the proportional shares of commercial fish¬ 
ery landings (adult) among NAFO divisions allocated 
on the basis of total landings of Atlantic halibut (abun¬ 
dance-based shares [landings]). We propose that this 
relationship can be considered a baseline for expected 
juvenile and fishery production per NAFO division. We 
further propose that suitable habitat can be used as a 
proxy for production in the absence of more detailed 
ecological information. In areas where the relationship 
between expected juvenile habitat and fishery pro¬ 
ductivity substantially stray from a 1:1 relationship, 
further research is required to understand the mecha¬ 
nisms controlling the population, and to support the 
spatial management of this fishery. 
Material and methods 
Modeling species distribution with a correlative approach 
Species distribution models are designed to predict the 
limits of geographic range and habitat suitability for a 
selected species, by using functions that relate physio¬ 
logical (mechanistic) or distributional (correlative) data 
to areas of unknown occupancy (Kearney and Porter, 
2009; VanDerWal et al., 2009). Maximum entropy is a 
correlative approach that can be used for modeling spe¬ 
cies distribution because it describes the relationship 
between survey (coordinate) and environmental data 
(raster layers) (Elith et al., 2011). It is commonly used 
to study the distributions of invasive species, shifts in 
distribution related to climate change, and the spatial 
diversity of species (Phillips and Dudik, 2008; Elith et 
al., 2011; Fitzpatrick et al., 2013). 
As it pertains to the model, entropy is a measure 
of the uncertainty in a data set (Cover and Thomas, 
1991); a uniform distribution would represent com¬ 
plete uncertainty. A maximum entropy model uses 
multimodal logistic regression with information from 
observational data and maximizing entropy based on 
environmental constraints (Jaynes, 1957; Elith et al., 
2011). The logistic output from this model is a predic¬ 
tion layer that maps the relative habitat suitability of 
all locations on the basis of prevailing conditions where 
the species has been observed. From habitat suitabil¬ 
ity, we can infer the probability of species presence 
(VanDerWal et al., 2009). The output scales from 0 to 
1, where 0.5 represents the relative habitat suitability 
and probability of presence, where environmental con¬ 
ditions are ‘typical for presence’ (Phillips et al., 2006; 
Phillips and Dudik, 2008; Elith et al., 2011; details in 
Suppl. Material). VanDerWal et al. (2009) found that 
for many species, a positive relationship exists between 
habitat suitability and local abundance, and that these 
models provide useful insights into spatial patterns of 
abundance. 
We chose to model with maximum entropy to take 
advantage of its ability to accurately predict in unsam¬ 
pled locations, the several parameterizations available 
to account for biases in the data, and the continuous 
and common scale of model outputs that enable direct 
comparison among models (Phillips et al., 2006). Addi¬ 
tionally, unlike Shackell et al.’s (2016) use of a general¬ 
ized additive model for a similar analysis, our method 
