computational challenges in assessment of marine resources
DESCRIPTION
Computational challenges in assessment of marine resources. Hans Julius skaug Talk given at Inst. of Informatics, UiB Nov. 28, 2002. Outline. Mathematical ecology Management of living resources Computation and statistics Models Numerical integration in high dimensions - PowerPoint PPT PresentationTRANSCRIPT
![Page 1: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/1.jpg)
Computational challenges in assessment of marine resources
Hans Julius skaugTalk given at Inst. of Informatics, UiB
Nov. 28, 2002
![Page 2: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/2.jpg)
Outline
• Mathematical ecology– Management of living resources
• Computation and statistics– Models– Numerical integration in high dimensions– Automatic differentiation
• Examples of ’assessment problems’– Harp seals
![Page 3: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/3.jpg)
Institute of Marine Research
• Units– Bergen, Tromsø, Flødevigen, Austevoll, Matre
• 500 employees – 135 researches
• Center for marine resources– Fish (cod, herring, capelin, …)– Whales and seals
![Page 4: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/4.jpg)
Management of living resources• ‘Assessment’ = Estimation of
– Current stock size– Biological parameters: mortality rates, birth rates, …
• Prediction of future stock size– Given today’s status– Under a given management regime (quota)
• Principles– Maximize harvest– Minimize risk of extinction
• Uncertainty– Important for determining risk
![Page 5: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/5.jpg)
The complication
• Acoustic and trawl surveys do NOT give absolute estimates of stock size
I = q · U
where
I = survey index
U = stock size
![Page 6: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/6.jpg)
Population dynamics ut = stock size at time t
• Exponential growth
ut = ut-1 (1 + 1)
• Density regulation
ut = ut-1 [1 + 1(1- ut-1/ 2)]
![Page 7: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/7.jpg)
0 50 100 150 200
02
00
40
06
00
80
01
00
0
Population growth
Time (t)
Po
pu
latio
n s
ize
(u
)
2
1 = 0.05
2 = 500
![Page 8: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/8.jpg)
Population dynamics
• Exponential growth
ut = ut-1 (1 + 1)• Density regulation
ut = ut-1 [1 + 1(1- ut-1/ 2)] • Harvest
ut = (ut-1 - ct-1) [1 + 1(1- ut-1/ 2)]
• Stochastic growth rate
ut = (ut-1 - ct-1) [1+1(1- ut-1/ 2) + N(0, 3)t ]
![Page 9: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/9.jpg)
Additonal complexity
• Age structure
• Spatial structure
• Interactions with other species
• Effect of environment
![Page 10: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/10.jpg)
Models in statistics
• Parameters structural parameter, dim() 20– u state parameter, dim(u) 1000
• Data– y ‘measurements’
• Probability distributions– p(y | u, ) probability of data given u and – p(u | ) probability of u given – p() ’prior’ on
![Page 11: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/11.jpg)
State-space model
• State equation p(u | )
ut = (ut-1 - ct-1) [1+1(1- ut-1/ 2) + N(0, 3)t ]
• Observation equation p(y | u, )
yt = 4 ut + N(0, 5)t
![Page 12: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/12.jpg)
Models in statistics
• Joint probability distribution
p(y,u,) = p(y|u,) p(u|) p()
• Marginal distribution
p(y,) = p(y,u,) du
![Page 13: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/13.jpg)
Statistical computation
• Major challenge (today)
p(y,) = p(y,u,) du
• Maximum likelihood estimation
= argmax p(y,)
![Page 14: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/14.jpg)
Algorithms
• Simple Monte Carlo integration
• Importance sampling
• Markov Chain Monte Carlo– Metropolis-Hastings algorithm– Gibbs sampler
• EM-algorithm
• Kalman filter (nonlinear)
![Page 15: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/15.jpg)
Laplace approximation
p(y,) = p(y,u,) du
L |detH | 1/2py, u,
u arg maxu
py, u, ,
H 2
u2 logpy, u, |u u
![Page 16: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/16.jpg)
Pierre-Simon Laplace (1749-1827)
• Bayes formula
p(u|y) = p(y|u)p(u) / p(y)
where
p(y) = p(y|u)p(u) du• Replaces integration with
maximization• Exact for the Gaussian case
![Page 17: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/17.jpg)
Maximum likelihood estimation
• Estimate by maximizing L()
• Gradient L by automatic differentiation
![Page 18: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/18.jpg)
AD (Automatic Differentiation)• Given that we have a code for L• AD gives you L
– at machine precision– ’no’ extra programming– efficiently (reverse mode AD)
cost(L) 4cost(L)
• AD people– Bischof, Griewank, Steihaug, ….
• Software: ADIFOR, ADOL-C, ADMB …
![Page 19: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/19.jpg)
Laplace approximation again
where
L |detH | 1/2py, u,
u arg maxu
py, u, ,
H 2
u2 logpy, u, |u u
![Page 20: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/20.jpg)
Research questions
• AD versus analytical formulae– Which is the most efficient?
• Higher order derivatives (3th order)– Less efficient than the gradient
• Sparseness structure
![Page 21: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/21.jpg)
Constrained optimization formulation
Maximize jointly w.r.t. u and
under the constraint
L , u |detHu, | 1/2py, u,
u
logpy, u, 0
![Page 22: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/22.jpg)
BeMatA project • NFR- funded (2002-2005)
• Collaboration between:– Inst. of Marine Research– Inst. of informatics, UiB– University of Oslo– Norwegian Computing Center, Oslo
• Joins expertise in– Statistical modeling– Computer science / Optimization
![Page 23: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/23.jpg)
Fisheries assessments
• Age determination of fishes invented around 1900– like counting growth layers in a tree
• ’Fundamental law’
ut,a = (ut-1,a-1 - ct-1,a-1)( 1 - )– t is an index of year– a is an index of year is mortality rate
![Page 24: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/24.jpg)
Recruitment mechanism
• Number of spawners
Ut = a ut,a
• Number of fish ’born’
ut,0 = R( Ut, )
• Strong relationship: seals
• Weak relationship: herring
Barents Sea cod 1946-1998
0
200000
400000
600000
800000
1000000
1200000
1400000
1600000
1800000
2000000
0 200000 400000 600000 800000 1000000 1200000 1400000
U
R
![Page 25: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/25.jpg)
Case study: Harp seals
• Barents Sea stock– Currently 2 million individuals
• Data– Historical catch records– Only pups can be counted – Age samples on whelping grounds
• Questions– Pre-exploitation stock size– Lowest level attained by the stock
![Page 26: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/26.jpg)
1880 1920 1960 2000
a: 1+ population
Year
Nu
mb
er
of
1+
an
ima
ls (
in m
illio
ns)
01
23
45
0.2
0.5
1880 1920 1960 2000
b: Pups
Year
Nu
mb
er
of p
up
s (i
n m
illio
ns)
00
.10
.20
.30
.40
.5
xxx
x
xxxx
0.2
0.5
![Page 27: Computational challenges in assessment of marine resources](https://reader035.vdocuments.site/reader035/viewer/2022062519/56815022550346895dbe07b9/html5/thumbnails/27.jpg)
Annotated referencesBeMatA project
Title: Latent variable models handled with optimization aided by automatic differentiation; application to marine resource
http://bemata.imr.no/
State-space models in fisheries
Gudmundsson, G. (1994). Time-series analysis of catch-at-age observations, Applied Statistics, 43, p. 117-126.
AD
Griewank, A. (2000). Evaluating Derivatives: Principles and Techniques of Algorithmic Differentiation, SIAM, Philadelphia.
AD and statistics
Skaug, H.J. (2002). Automatic differentiation to facilitate maximum likelihood estimation in nonlinear random effects models. Journal of Computational and Graphical Statistics. 11 p. 458-470.
Laplace approximation in statistics
Tierney, L. and Kadane, J.B. (1986). Accurate approximations for posterior moments and marginal distributions. Journal of the American Statistical Association 81 p. 82-86