Page 4 - Soldatini2014

Local and Global Climate Effects on Storm Petrel Demography
Table 2. Results of goodness of fit test for the multistate model.
Test component
Global test
Test 3G (test for transience)
Test M (test for trap-dependence)
doi:10.1371/journal.pone.0094526.t002
breeding. For an individual in state B, W2 represents the
probability that a breeder survives. After the first capture, pre-
breeders are not observed as they do not attend the colony;
skippers are not present at the colony; we then estimated breeders
detection as the detection parameter in the ‘‘breeder’’ state is the
product of the probability of not skipping breeding and of the
probability of detection given breeding.
Goodness-of-fit tests [43] were performed using U-CARE [44].
The results of the GOF test are summarised in Table 2.
We analysed 25 model combinations in which Y was modeled
as a constant or functions of age and time, either singly, additively,
or with interactions. The model assumes that there are two age
classes (juvenile and adult) and there is an age from which the
probability to breed ai does not change. Full recruitment age is the
age at which any individual has become a breeder, it is equivalent
to age ‘‘m’’ as defined by Gauthier and colleagues [45], which
considers age ‘‘m’’ as the age at which recruitment probability can
be considered constant for all subsequent ages. Juvenile survival
and breeder recapture probabilities were allowed to be constant or
to vary with time.
We distinguished the first from the following encounter
occasions, as the encounter history is conditional on being caught
in the first period and the following detection probabilities depend
on the state and the time occasions. We started running a model
assuming that there is no change after 10 years. As we obtained
that after age 7 there is no improvement in model fitting we
reduced the full breeding age to 7 (namely the 6th year of age)
(Table 3). For a full breeding probability we fixed Y relative to this
hypothetical age 1. Thus we tested which is the full recruitment
age (probability of becoming a breeder) by fixing to 1 the potential
age of full recruitment, specifying initial parameter values we
constrained beta values before running the model.
We tested the time effect on survival and the effect of time and
sex (the last as group effect) on recruitment only. We limited the
tests in this way because we primarily sexed adults and we have
information on chicks sex only after 2007.
Bearing in mind previous results, to assess climatic effects on
survival and recruitment, model selection began with the model
with age and time effects on survival and sex effect on recruitment
Table 3. Selection of the full breeding age (f.b.), only best models are presented.
Model Deviance NP AIC
f. b. year 7 (age 6) 3127.74 29 3185.74
f. b. year 5 (age 4) 3224.01 27 3278.01
f. b. year 6 (age 5) 3224.01 28 3280.01
f. b. year 8 (age 7) 3224.01 30 3284.01
f. b. year 9 (age 8) 3224.01 31 3286.01
Deviance (Deviance), number of parameters (NP) and AIC [47] are reported. Preferred model is in bold.
doi:10.1371/journal.pone.0094526.t003
PLOS ONE | www.plosone.org
X2 d.f. P
48.568 52 0.610
22.264 23 0.504
26.304 29 0.609
age. More in details, after testing the single effect of the individual
covariates on survival (W) and recruitment (Y) we tested the
additive effects of different individual covariates and time on
survival (W) and the additive effects of different individual
covariates, time and sex on recruitment (Y).
Models were fitted with the E-SURGE program [46] (for model
implementation in E-Surge see Material S1). Model selection was
based on the Akaike Information Criterion (AIC) [47,48].
Effects of winter climate on survival over time
Because data on environmental conditions are available only
from 2001 to 2010, we used a subset of birds ringed as chicks and
adults in the same period (2918), modelling potential effect of
covariates only on adults survival. We carried out goodness-of-fit
for the single state model using U-CARE [6]. Results of the GOF
test are summarized in Table 4. Due to the presence of transient
individuals, we used 2-age-class models and a variance inflation
factor (c-hat = 1.4). Following the single state, as simplification of
the scheme reported above, we implemented the model in
program E-SURGE [44] (for model implementation in E-Surge
see Material S1).
We analysed 24 model combinations in which W was modeled
as a constant or function of time, we then explored singly and
additively interaction with environmental (external) covariates
dependent models. The model assumes that there are two age
classes (young, first year of age, and adult, older than 1 year),
survival and breeder recapture probabilities were allowed to be
constant or to vary with time.
Model selection procedure: model selection was based on the
Akaike’s information criterion (AIC) [47], adjusted for over-
dispersion (Q). After identifying time-dependent parameters, we
investigated the effects of SST and CHL in describing the time
variation with an analysis of deviance (ANODEV) and calculated
the proportion of variance explained by the environmental
(external) covariates using the R2 statistic [35].
Ethics Statement
Field work was carried out under permission from the Marine
Protected Area. Permits included obtaining DNA samples for
DAIC AIC weight
0.00 1.000
92.27 0.000
94.27 0.000
98.27 0.000
100.27 0.000
4 April 2014 | Volume 9 | Issue 4 | e94526