6/29/2010 Wan-Shu Wu1 GSI Tutorial 2010 Background and Observation Error Estimation and Tuning.
AN OBSERVATION ERROR MODEL FOR NORTH ......SCRS/2016/024 Collect. Vol. Sci. Pap. ICCAT, 73(4):...
Transcript of AN OBSERVATION ERROR MODEL FOR NORTH ......SCRS/2016/024 Collect. Vol. Sci. Pap. ICCAT, 73(4):...
SCRS/2016/024 Collect. Vol. Sci. Pap. ICCAT, 73(4): 1328-1338 (2017)
1328
AN OBSERVATION ERROR MODEL FOR NORTH ATLANTIC ALBACORE
Laurence Kell1, Haritz Arrizabalaga2, and Paul De Bruyn1,
Gorka Merino2, Iago Mosqueira3, Rishi Sharma4 and Jose-Maria Ortiz de Urbina5
SUMMARY
In this paper we describe the development of an Observation Error for North Atlantic albacore.
When conducting Management Strategy Evaluation an Operating Model is used to simulate
resource dynamics in simulation trials in order to evaluate the performance of a Management
Procedure, where the Management Procedure is the combination of pre-defined data, together
with an algorithm to which such data are input to provide a value for a management control
measure. To link the Operating Model and the Management Procedure it is necessary to
develop an Observation Error Model to generate fishery-dependent or fishery-independent
resource monitoring data. The Observation Error reflects the uncertainties between the actual
dynamics of the resource and the perceptions arising from observations and assumptions by
modelling the differences between the measured value of a resource index and the actual value
in the Operating Model. We show how an analysis of catch per unit effort data and hypotheses
about fisher behaviour can be used to parameterise an Observation Error Model.
RÉSUMÉ
Le développement d’une erreur d'observation concernant le germon de l’Atlantique Nord est
décrit dans le présent document. Pour réaliser une évaluation de la stratégie de gestion, un
modèle opérationnel est utilisé pour simuler les dynamiques de la ressource au moyen d'essais
de simulation visant à évaluer la performance d’une procédure de gestion, dans le cadre duquel
la procédure de gestion est une combinaison de données prédéfinies et d’un algorithme dans
lequel ces données sont saisies afin d’obtenir une valeur d’une mesure de contrôle de la
gestion. Pour relier le modèle opérationnel et la procédure de gestion, il est nécessaire
d’élaborer un modèle d’erreur d’observation afin de générer des données de suivi de la
ressource dépendantes ou indépendantes des pêcheries. L'erreur d'observation reflète
l’incertitude existant entre les dynamiques réelles de la ressource et les perceptions découlant
des observations et des postulats en modélisant les différences entre la valeur mesurée d’un
indice de la ressource et la valeur réelle dans le modèle opérationnel. Le présent document
montre la façon dont une analyse des données de prise par unité d'effort et les postulats sur le
comportement des pêcheurs peuvent être utilisés pour paramétriser un modèle d’erreur
d'observation.
RESUMEN
En este documento se describe el desarrollo de un error de observación para el atún blanco del
Atlántico norte. Al realizar la evaluación de estrategias de ordenación se utiliza un modelo
operativo para simular la dinámica del recurso en ensayos de simulación para evaluar el
rendimiento del procedimiento de ordenación, donde el procedimiento de ordenación es una
combinación de datos predefinidos, junto con un algoritmo en el que se introducen dichos datos
para proporcionar un valor para una medida de control de ordenación. Para vincular el
1 ICCAT Secretariat, C/Corazón de María, 8. 28002 Madrid, Spain; [email protected]; Phone: +34 914 165 600; Fax: +34 914 152 612. 2 AZTI -Tecnalia, Herrera Kaia Portualdea, 20110, Pasaia, Spain 3 European Commission Joint Research Centre (JRC), Institute for the Protection and Security of the Citizen (IPSC), Maritime Affairs
Unit, Via E. Fermi 2749, 21027, Ispra, VA, Italy 4 U.S. National Marine Fisheries Service, Southeast Fisheries Center, Sustainable Fisheries Division, 75 Virginia Beach Drive, Miami, FL,
33149-1099, USA 5 Instituto Español de Oceanografía (IEO- CO Málaga).
1329
modelo operativo y el procedimiento de ordenación es necesario desarrollar un modelo de
error de observación con miras a generar datos de seguimiento del recurso dependientes o
independientes de la pesquería. El error de observación refleja las incertidumbres entre la
dinámica real del recurso y la percepción procedente de las observaciones y supuestos
mediante la modelación de las diferencias entre el valor medido de un índice de recurso y el
valor real en el modelo operativo. Se muestra cómo pueden utilizarse un análisis de datos de
captura por unidad de esfuerzo y las hipótesis sobre la conducta de los pescadores para
parametrizar un modelo de error de observación.
KEYWORDS
Albacore, Observation Error Model, Operating Model,
Management Strategy Evaluation, Measurement Error, Uncertainty
Introduction
In Management Strategy Evaluation (MSE) an Operating Model (OM) is used to simulate resource dynamics in
simulation trials in order to evaluate the performance of a Management Procedure (MP). Where the MP is the
combination of pre-defined data, together with an algorithm to which such data are input to provide a value for a
management control measure. To link the OM and the MP it is necessary to develop an Observation Error Model
(OEM) to generate fishery-dependent or fishery-independent resource monitoring data. The OEM reflects the
uncertainties, between the actual dynamics of the resource and perceptions arising from observations and
assumptions by modelling the differences between the measured value of a resource index and the actual value in
the OM. In this paper we describe the development of an OEM for North Atlantic albacore for use in MSE.
The OM is based upon the Multifan-CL assessment conducted during the 2013 ICCAT North Atlantic Albacore
stock assessment (Kell et al. in press a). We first summarise the characteristics of the CPUE series to allow
CPUEs with different properties to me simulated then provide example simulations under a variety of
assumptions.
Operating Model
Ten scenarios were considered when fitting Multifan-CL, the scenarios are described and the results summarised
in Kell et al. (in press a).
When fitting the albacore stock assessments commercial catch per unit effort (CPUE) is used as a proxy for
relative abundance. The CPUE series (per fleets) used in the Multifan-CL assessment are summarised in Table
1.
It has long been recognised, however, that time series of CPUE may not accurately reflect trends in population
abundance (e.g. Beverton and Holt, 1993; Maunder et al., 2006; McKechnie et al., 2013; Polacheck, 2006).
Particularly since changes can occur in in the spatial distribution of populations and the allocation of effort in
response to management. In the latter case economic drivers can affect catch and effort independent of stock
abundance (e.g. Paloheimo and Dickie, 1964; Tidd et al., 2011). Furthermore Interactions between changes in
oceanographic conditions and exploitation can drive spatial and temporal dynamics (Fromentin et al., 2013). In
addition, technological developments can produce bias on the relation between stock abundance and fleet’s
CPUE. Therefore CPUE may not be proportional to stock abundance (Hilborn et al., 1992). For example CPUE
may be hyperstable where catches remain high as a population declines (Erisman et al., 2011) or hyperdepletion
may occur where catches decline faster than the population.
CPUE Series
First the selection patterns of the fleets and the error structure of the CPUEs are explored. Then, simulations are
conducted to evaluate the bias and uncertainties for different assumptions. The selection patterns for the 10
CPUE series are plotted in Figures 1 and 2. The residuals are plotted to check the error model and to look for
systematic patterns that may suggest that some assumptions are violated. In Figure 3 the residuals are plotted
(on the log scale) against year to check for systematic patterns over time. The residuals are then plotted against
the fitted values to check the variance function (Figure 4). Figure 5 plots residuals with 1 year lag to check for
auto-correlation, and quantile-quantile plots to check for error assumptions are provided in Figure 6.
1330
Simulations
Eight simulations were conducted corresponding to different assumptions about the potential error structure and
biases in the CPUE series, the index was based on biomass in all but one case, a lognormal error with a CV of
30%. The index is simulated using the Multifan-CL Base Case estimates of numbers-at-age (ICCAT, in press).
The simulations were i) unbiased, ii) hyperstability, iii) trend in catchability, iv) auto-correlation, v) a decrease in
variance with time, vi) an index based on juvenile age-classes, vii) index based on mature age-classes, and viii)
index based on numbers. The simulated CPUE series are shown in Figure 7, the ribbons show the 50th and 80th
percentiles. Two individual simulations are also shown (red and cyan), the blue line an unbiased index and the
black line the median of the simulated series. The unbiased index captures the stock trends although individual
CPUE series show a wide variation. The same random deviates were used for all simulated series and therefore
in all cases, but for auto-correlation, the peaks of the indices coincide. For a trend in catchability and
hyperstability the trend in the index underestimates the true decline. For a decrease in variance no bias was seen.
The scenario based on a juvenile index showed more variability, due to the effect of year class strength, using an
index based on the mature biomass alone or numbers provided results similar to the unbiased index with slightly
more variation.
Summary
The residual patterns in the fits of the CPUE series could be used to simulate alternative hypotheses about
processes that could affect catch and effort. The properties of the indices, e.g. ages selected, catchability,
availability or vulnerability could also be used simulated hypotheses about the processes to be implemented as
scenarios in an MSE. The impact of using alternative simulated CPUE series to plug into a Management
Procedure need to be evaluated through Management Strategy Evaluation (see paper Kell et al. in press b).
1331
References
Canty, A. J., Davison, A. C., Hinkley, D. V., and Ventura, V. (2006). Bootstrap diagnostics and remedies.
Canadian Journal of Statistics, 34(1):5–27.jack-knife,
R. Beverton and S. Holt. On the dynamics of exploited fish populations, volume 11. Springer, 1993.
B. E. Erisman, L. G. Allen, J. T. Claisse, D. J. Pondella, E. F. Miller, J. H. Murray, and C. Walters. The illusion
of plenty: hyperstability masks collapses in two recreational fisheries that target fish spawning aggregations.
Canadian Journal of Fisheries and Aquatic Sciences, 68(10):1705–1716, 2011.
J.-M. Fromentin, G. Reygondeau, S. Bonhommeau, and G. Beaugrand. Oceanographic changes and exploitation
drive the spatio-temporal dynamics of Atlantic bluefin tuna (Thunnus thynnus). Fisheries Oceanography,
2013.
R. Hilborn, C. J. Walters, et al. Quantitative fisheries stock assessment. Choice, dynamics and uncertainty.
Reviews in Fish Biology and Fisheries, 2(2):177–178, 1992.
ICCAT. (in press). Report for biennial period, 2014-15, Part II (2015), Appendix 10. Report of the Sub-
Committee on Statistics meeting. 15 pp. Document SCRS/2016/013.
L.T. Kell, H. Arrizabalaga, G. Merino, and P. De Bruyn. (in press a). Conditioning an operating model for North
Atlantic Albacore. Document SCRS/2016/023: 32 p.
L.T. Kell, H. Arrizabalaga, G. Merino, and P. De Bruyn. (in press b). Cross testing of biodyn an R package to
implement management procedures based on biomass dynamic models. SCRS/2016/026: 5 p.
M. N. Maunder, J. R. Sibert, A. Fonteneau, J. Hampton, P. Kleiber, and S. J. Harley. Interpreting catch per unit
effort data to assess the status of individual stocks and communities. ICES Journal of Marine Science:
Journal du Conseil, 63(8):1373–1385, 2006.
S. McKechnie, S. Hoyle, and S. Harley. Longline CPUE series that account for changes in the spatial extent of
fisheries. Technical report, WCPFC-SC9-2013/SA-IP-05, 2013.
J. Paloheimo and L. Dickie. Abundance and fishing success. Rapports et Proces-Verbaux des Runions du
Conseil International pour lExploration de la Mer, 155, 1964.
T. Polacheck. Tuna longline catch rates in the Indian Ocean: Did industrial fishing result in a 90% rapid decline
in the abundance of large predatory species? Marine Policy, 30(5):470–482, 2006.
A. N. Tidd, T. Hutton, L. Kell, and G. Padda. Exit and entry of fishing vessels: an evaluation of factors affecting
investment decisions in the North Sea English beam trawl fleet. ICES Journal of Marine Science: Journal du
Conseil, 68(5):961–971, 2011.
1332
Table 1. Catch per unit effort series used in the 2013 assessment.
Figure 1. A comparison of fleet selection patterns.
1333
Figure 2. Selection patterns by fleet.
1334
Figure 3. Plots of residuals against year to check for systematic patterns.
1335
Figure 4. Plots of residuals against fitted values to check variance function.
1336
Figure 5. Plots of residuals with 1 year lag to check for auto-correlation.
1337
Figure 6. Quantile-quantile plots to check for normal error.
1338
Figure 7. Simulated CPUE series.