Date: January 12
& 13, 2006
Place: Institute of Statistical Mathematics, the meeting room
This workshop is supported by The Institute of Statistical Mathematics,
Project Research (Developing predictive models for bycatch in tuna fisheries)
Program
“Applications of Tweedie distribution to the CPUE standardization”
“A spatiotemporal response model for chlorophyll-a distributions based on
some oceanographic factors”
“Population dynamics modeling for fisheries bycatch”
“Modelling shark bycatch: The zero-inflated negative binomial regression model with
smoothing”
“Species associations in purse-seine catch-bycatch in the eastern Pacific
Ocean”
“Methodological Issues in North Pacific Blue Shark Stock Assessment”
Abstract
“Applications of Tweedie
distribution to the CPUE standardization”
I focus on so-called zero-catch problem of CPUE (catch
per unit effort) standardization. Because the traditional CPUE model with
log-normal error structure can not be applied in this case, three methods have
been often used as follow:
1) Ad hoc method to add the small constant value to all
response variables.
2) Catch model with Poisson or negative binomial error
structure.
3) Delta-type two-step model. (After estimating the ratio
of zero-catch using logit or probit
model, typical model such as CPUE-LogNormal or
Catch-Poisson is applied to CPUE without zero-data.)
In this symposium, we mainly carried out the CPUE
standardization of by-catch data including many observations with zero-catch
caught by Japanese training vessels (e.g. bigeye
thresher shark or silky shark in the north Pacific Ocean) using Tweedie distribution model. Tweedie
distribution is a kind of extension of compound Poisson model derived from the
stochastic process and this model has been applied to the rainfall forecast in
the field of weather derivatives in Japan.
I performed CPUE standardization of actual by-catch data
using catch model with negative binomial error through SAS package (Version
9.1.3) and Tweedie model through R software (Version
2.0.1). I also compared both results based on n-fold cross-validation. MSE
(mean squared error) and Pearson's correlation coefficient was calculated based
on the observed CPUE and the corresponding predicted CPUE obtained from the
Catch Negative-Binomial model and/or Tweedie
distribution. In this Tweedie model of R software,
the power-parameter of variance function is initially computed using the grid
search, after that the parameters of regression coefficient are estimated based
on the framework of quasi-likelihood method.
I also introduce the results of comparison among Tweedie model, Catch Negative-Binomial model and/or CPUE-LogNormal model judging from the statistical performances,
MSE and Pearson's correlation coefficient. In order to check the reliability of
Tweedie model, simulation study assuming the
statistical distributions may be necessary and useful.
“A spatiotemporal
response model for chlorophyll-a distributions based on some oceanographic
factors”
This study aims
to develop a new spatiotemporal statistical model to evaluate spatiotemporal chlorophyll a (chl-a) distributions over the Sea
of Japan, derived from the satellite remote sensing data. Considering factors
affected to the chl-a distributions, we employed
satellite derived sea surface temperature (SST) and photosynthesis active
radiation (PAR). Because preliminary spatial analysis, chl-a exhibit anisotropy and SST and PAR exhibit almost isotropy
in the south-north and east-west directions. As a result of time series
analysis of spatial correlations, chl-a and PAR showed autocorrelation which indicate seasonal
cycles and SST showed stationary series, respectively. Based on the preliminary
analysis, we have conducted numerical prediction experiments with changing
temporal lag. We present results showing the ability of the proposed
spatiotemporal model to predict one-month ahead
changes in chl-a distribution. We evaluated the
prediction accuracy numerically, as well as the predicted distributions in each
season. The spatiotemporal prediction of remotely sensed data demonstrated in
this study offers a powerful and innovative way by which to determine the high
productivity areas that could be translated into potential fishing areas. The
spatiotemporal statistical model provides a new way to derive detailed
information on spatial and temporal dynamics of primary productivity in the
“Population dynamics modeling for fisheries bycatch”
I present a
population dynamics model of a species for which fisheries bycatch is a source
of mortality. The bycatch data, which provide useful information about the
population dynamics, are integrated with other data sources in the model.
“Modeling shark bycatch: The
zero-inflated negative binomial regression model with smoothing”
In contrast to catch data on target species, catch
data on non-target species may be characterized by many zero-valued
observations, and depending on the species, may also include large values when
aggregations of animals are incidentally killed. For modeling shark bycatch
counts, we introduce zero-inflated negative binomial model with
smoothing.
“Biological
characteristics and CPUE trend of silky shark caught by Japanese tuna longline in the Pacific
Ocean”
The silky shark
Carcharinus falciformis is
one of the major pelagic species, reaching 3.3 m in total length, and widely
distributed from tropical to subtropical waters. Tuna longline
and purse seine fisheries catch this species incidentally year-round in the Pacific Ocean. Silky shark was third dominant species
following blue shark and oceanic white-tip shark caught by the tuna longline fisheries in the tropical sea areas. The
international concern about the conservation and management of sharks has been
getting larger in recent years. It is necessary to accumulate the information
on biology, ecology and stock status of sharks for the proper conservation and
management. But there is not enough information about the silky shark.
Therefore, the biological characteristics of silky shark were briefly reviewed
and stock status in the Pacific Ocean was examined by the CPUE trend in this study.
Catch data of
silky shark obtained by Japanese research and training vessels with tuna longline from tropical to subtropical areas in the Pacific Ocean were used for the analysis. CPUE (catch
rate: number of catch/1000 hooks) of silky shark was large in the tropical
area. Standardized CPUE calculated with Generalized Linear Model (GLM) assuming
negative binominal distribution during 1992-2003 was fluctuated but neither
significant increasing nor decreasing trend was observed. Comparing CPUE from
above data with those from old ones during 1967-1971 in the tropical water,
there was not much difference between the two series of CPUE. Therefore, it is
suggested that tuna longline did not give a
significant impact on silky shark stock in the Pacific Ocean.
“Species associations in purse-seine catch-bycatch in the eastern Pacific
Ocean”
A preliminary
analysis of species associations in purse-seine catch data using cluster
analysis and nonmetric multidimensional scaling is
presented. Relationships between the main species associations identified in
these analyses and fishery and environmental factors are explored using the
random forest technique for classification. Results are compared for different
years.
“Methodological
Issues in North Pacific Blue Shark Stock Assessment”
In a recent stock assessment of blue shark (Prionace
glauca) in the North Pacific several special methods
were applied to overcome data limitations.
A key challenge facing the assessment was to derive species-specific
data from logbook records which contained only an aggregated category of
sharks. This situation was
complicated by instances of under- or non-reporting of by-caught sharks. Various methods for handling such data
were assessed and a reporting rate filter was applied. Another important methodological issue
involved assembling a total catch series from the various fleets catching blue
sharks in the North Pacific. This
issue was addressed by using standardized blue shark catch per unit effort from
Japan's longline fleets, and apportioned effort for other fleets,
to predict catches for the other fleets.
In order to determine whether catch figures were realistic, recent
estimates from the shark fin trade, of which blue shark is a
major component, were adjusted and compared to the catch series. The steps and assumptions involved in
these methods will be discussed in view of their potential application to
assessments of other species and areas.