1. Data assimilation in APECOSM-E Sibylle Dueri, Olivier Maury

2. Industrial fisheries start in 1983, rapid increase
Tropical tuna exploitation in the Indian Ocean
Catch
tons/year
SKJ  
YFT  
BET  
Industrial fisheries start in 1983, rapid increase
Targeted tuna species
Skipjack (SKJ)
Yellowfin (YFT)
Bigeye (BET)
From IOTC – Report 2009

3. Skipjack tuna fishery: spatial distribution
Main fishing gears: purse seine, bait boat, gillnet and long line
From IOTC – Report 2009

4. Skipjack fishery: available data
Fishing gears
Purse seiners (French, Spanish and other)
Maldivian Baitboat
Data
Monthly catch 1x1 degree grid
Monthly size frequency 5x5 degree grid
Monthly fishing effort 1x1 degree grid
From IOTC – Report 2009

5. APECOSM-E
Dynamic, deterministic spatially explicit and size structured model
Single species of top predator (i.e. skipjack tuna)
Devoted to parameter estimation
State variable  Tuna biomass f(x,y,z,t,V)
Partial differential equation, discretized and solved numerically
48 parameters: 19 related to fisheries, 8 energetic (DEB), 21 ecological parameters

6. Discretisation
1º x 1º horizontal grid of the Indian Ocean (4230 cellules)
20 vertical layers, 10 m interval in the first 150m and max depth of 500m
83 size classes, from 1mm to 1m
1 day time step

7. APECOSM-E – Model Structure

8. Fishing mortality
4 fleets : French, Spanish and other (Seychelles, Ile Maurice, NEI) purse seiners and Maldivian Bait boat
Monthly value of spatial effort ek is applied on simulated biomass
Fishing power pk multiplies by an exponential function representing the increase in fishing efficiency due to technological development
Fish size and fishing depth selectivity functions

9. Observed catches vs distribution of simulated biomass
Mars 1993
Total biomass
Vertical transect of biomass distribution along the
Equator
Exploitable biomass + observed catch
Observed catch
Vertical transect along the Equator

10. Data assimilation 3 Steps
Identifiability of the model parameters (Hessian Matrix)
Parameter estimation
Sensitivity analysis of non-estimated parameters
Cost function combines
–log likelihood of catches JCAPT(K)
–log likelihood of size frequencies JFDT(K)
–log likelihood of parameters JPAR(K)
Minimization algorithm requires the derivation of the code  automatic differentiation tool (TAPENADE-Inria)
Derives the exact gradient of the cost function with respect to model parameters

11. Data assimilation: convergence
Data assimilation over 10 year period:
Optimization of fishery related parameters: fishing power, size and depth selectivity, …
Iterations

12. Overall catch: simulated vs observed
Temporal window used for optimisation

13. Skipjack tuna fisheries: fishing strategy
Peaks of catches occurring in association with Fish Aggregation Devices (FADs) are difficult to represent
Associative behavior of fishes
Increase the catchability?
Ecological trap? (Marsac et al. 2000, Hallier and Gaetner 2008)

14. Free-school catch

15. Size frequencies
Bimodal distribution

16. Size frequencies: temporal dynamics
mean

17. Sensitivity analysis
Based on the evaluation of the gradient of the cost function
The gradient represents the variation of the cost function for a change in parameters
Local sensitivity, changes in time
3 x 6 year periods
Ecological parameters
DEB Param.