Accurate and precise estimation of discards is a major objective of data collection programs throughout the world. Discard reduction is also a major topic of the new Common Fisheries Policy (CFP) and the future Data Collection Multi-Annual Programme (DC-MAP). Using data from the Portuguese on-board observer programme that samples two otter trawl fisheries in ICES Division IXa, we compare two different approaches for estimating the sampling effort required to attain "assessment grade" discard estimates: a model-based approach (exponential-decay models) and a probability-based approach (based on classic sampling theory). We show that both approaches attain comparable sample size estimates and that the sample size required to attain precision objectives
varies across species and across fisheries being likely influenced by discard motifs. We demonstrate that sampling levels at least two fold higher than the present sampling levels would be required to attain the precision levels set in the current Data Collection Framework (DCF). We discuss the implications of these results in light of the future ability of European onboard sampling programs to detect, e.g., progressive reductions in discard levels.
300003-World Science Day For Peace And Development.pptx
Portuguese Market and On-board Sampling Effort Review
1. Sampling
review
Jardim, E.,
Prista, N. &
Dias, M.
Introduction Portuguese Market and On-board
Data
Methods
Sampling Effort Review
Working document presented to PGCCDBS, 7-11 February 2011
Results
Conclusions
Jardim, E., Prista, N. & Dias, M.
February 5, 2011
2. Sampling
review
Jardim, E., Outline
Prista, N. &
Dias, M.
Introduction
Data
1 Introduction
Methods
Results
Conclusions 2 Data
3 Methods
4 Results
5 Conclusions
3. Sampling
review
Jardim, E., Introduction
Prista, N. &
Dias, M.
Introduction The implementation of the metier approach resulted in:
Data
a decrease in the precision of the length frequencies
Methods
estimates by species, due to the spread of sampling effort
Results
to new species and the reduction of trips sampled.
Conclusions
an increase in the number of strata to be sampled
on-board
The objective of this work is to optimize sampling effort by
computing the number of samples required to achieve the
precision levels defined by the DCF:
for length frequencies of the landings sampled at the
market
for total discards sampled on-board
4. Sampling
review
Jardim, E., Data
Prista, N. &
Dias, M.
Introduction
Data
Methods
On market:
Results number of individuals estimated by trip
Conclusions
data from 2009-2010,
by REGION, GEAR, SPECIES & QUARTER
On board:
weight discarded by trip
data from 2004-2010
by METIER (OTBDEF, OTBCRU) & QUARTER
Data is scarce and the breakdown by metier makes it
even scarcer, it was necessary to aggregate.
5. Sampling
review
Jardim, E., Methods
Prista, N. &
Dias, M.
Introduction
Data
Model CV = f (N) using exponential decay models (N
Methods
being number of samples)
Results
Conclusions
Compute N to achive 12.5% CV for market sampling or
20% for on-board
Compute 95% percentile of N as an indicator of a high
probability to achieve the objective and cover species with
more variability than average
Review the sampling plans
(Lots of technical details and statistical mambo-jambo to be
provided if requested)
6. Sampling
review
Jardim, E., Data aggregation for
Prista, N. &
Dias, M. market sampling
Introduction
Data
Methods Each pair used in model refers to
Results
the CV of the total number of individuals sampled
Conclusions the number of samples collected from which the CV above
was computed
Each pair was computed by GEAR (aggregation of
metiers), QUARTER, REGION & SPECIES
Each model was fit to distinct dimensions of the data
collapsing all other dimensions
for each REGION
for each GEAR
for each combination of REGION and GEAR
8. Sampling
review
Jardim, E., Models for on-board
Prista, N. &
Dias, M. sampling
Introduction
Data
Methods
Results
Conclusions
9. Sampling
review
Jardim, E., Preliminary conclusions
Prista, N. &
Dias, M. for market sampling
Introduction
Region Metier SampEff.2010 SampEff.2011
Data 1 North FPO MOL >=29 0 0 5 17
2 North GNS DEF 80-99 0 0 2 21
Methods 3 North GNS DEF 60-79 0 0 3 21
4 North GTR DEF >=100 0 0 8 12
Results
5 North LLS DEF 0 0 0 1 17
Conclusions 6 North OTB DEF 65-69 0 0 6 18
7 North PS SPF >=16 0 0 7 13
8 North TBB CRU >=20 0 0 1 17
9 Center FPO MOL >=29 0 0 5 17
10 Center GNS DEF 80-99 0 0 3 19
11 Center GTR DEF >=100 0 0 7 4
12 Center LLD LPF 0 0 0 2 2
13 Center LLS DEF 0 0 0 2 17
14 Center LLS DWS 0 0 0 2 17
15 Center OTB CRU >=70 0 0 6 4
16 Center OTB CRU 55-59 0 0
17 Center OTB DEF 65-69 0 0 1 4
18 Center PS SPF >=16 0 0 5 13
19 South FPO MOL >=29 0 0 5 17
20 South GNS DEF 80-99 0 0 2 21
21 South LLD LPF 0 0 0 1 1
22 South LLS DEF 0 0 0 1 17
23 South OTB CRU >=70 0 0 5 18
24 South OTB CRU 55-59 0 0
25 South OTB DEF 65-69 0 0 2 18
26 South PS SPF >=16 0 0 2 13
27 Total 84 338
10. Sampling
review
Jardim, E., Preliminary conclusions
Prista, N. &
Dias, M. for on-board sampling
Introduction
Data
Methods Model point estimate is 15 samples per quarter for both
Results
metiers
Conclusions
Sampling theory estimate is 18-20 samples per quarter
95 percentile is 48 samples per quarter
Increase sampling effort up to 192 trips per year for each
metier
The sampling effort is not applicable due to high costs
and lack of human resources. In 2011 on-board sampling
effort will be increased up to the maximum possible,
taking into account other metiers and resources available.
11. Sampling
review
Jardim, E., The End
Prista, N. &
Dias, M.
Introduction
Data
Methods
Results
Conclusions
12. Sampling
review
Jardim, E., Details: codes
Prista, N. &
Dias, M.
Introduction
Data
Methods
Results N, C, S = Norte, Centro Sul
Conclusions
OTB, PS, GTR, GNS, FPO, LLS = trawl, purse seine,
trammel nets, gill nets, traps, longliners
Models: exp, strexp, exp log, strexp log = exponential,
streched exponential, exponential with log errors, streched
exponential with log errors.
13. Sampling
review
Jardim, E., Details: Methods
Prista, N. &
Dias, M.
Models are fit to distinct data breakdowns, All, by
Introduction
REGION, by GEAR, by REGION & GEAR = 60 models
Data
Methods
(only market)
Results Models are fit to both metiers merged (only on-board)
Conclusions Fits are analysed by visual inspection of residuals, AIC,
likelihood, precision of parameters, precision of the
estimated number of samples to achieve objective.
Fits selected are averaged considering the inverse of the
residuals variance (only market)
Number of samples are allocated considering the highest
number for each combination of GEAR & REGION (only
market).
Number of samples are estimated by the best model as
well as with sampling theory (only on-board).
14. Sampling
review
Jardim, E., Details CV (µ) = CV (τ )
Prista, N. &
Dias, M.
Introduction
Data
Methods
Results
τ =C ∗µ
ˆ ˆ
Conclusions
var (ˆ) = C 2 ∗ var (ˆ)
τ µ
C 2 ∗ var (ˆ)
µ var (ˆ)
µ
CV (ˆ) =
τ = = CV (ˆ)
µ
C ∗µ ˆ µ
ˆ
15. Sampling
review
Jardim, E., Details τ & var (τ )
Prista, N. &
Dias, M.
Introduction Consider N the number of individuals, i = 1 . . . l to index
Data length classes and j = 1 . . . s to index sampled trips.
Methods
Results N= Ni
Conclusions i
Σ = var (N) = var (Ni ) + 2 ∗ cov (Ni , Nj )
i i j=i+1
Ni = Nij
j
Nij ∗w
j ( wj − Ni )2
var (Ni ) =
s ∗ (s − 1)
16. Sampling
review
Jardim, E., Details DPUE & var (DPUE )
Prista, N. &
Dias, M.
Introduction
Data
Methods Let i be the index of the number of hauls sampled in trip j
Results (i = 1, 2, .., nj , j = 1, 2, .., nt ), d be total weight discarded (in
Conclusions
kg) and h be the haul duration (in hours)
nj di,j
i=1 hi,j
DPUE j = nj
and nt
DPUE j
DPUE = i=1
nt
nt
j=1 (DPUEj −DPUE )2
VAR(DPUE ) = nt (nt −1)
17. Sampling
review
Jardim, E., Details residuals of
Prista, N. &
Dias, M. on-board model
Introduction
Data
Methods
Results
Conclusions