Transcript FishBase goes FishBayes R, JAGS and Bayesian Statistics

FishBase goes FishBayes
Summarizing all available information
for all fishes
Rainer Froese
14th FishBase Symposium
2nd September 2013, Thessaloniki, Greece
A Human Dream: Encyclopedias
• „… to collect knowledge distributed around
the globe…to set forth … to the men with
whom we live, and transmit it to those who
will come after us… so that the work of
preceding centuries will not become useless
…and so that our offspring, becoming better
instructed, will at the same time become
more virtuous and happy...” Diderot (1751)
A New Challenge:
Knowledge Explosion
• In most fields, the number of annually
published studies far exceeds the ability of the
specialists to absorb them or even know
about them
• „Best available knowledge“ summarizes in a
scientifically correct manner all available
knowledge, including related knowledge, with
indication of uncertainty
A Case in Point:
FishBase has compiled thousands of studies on
growth, maturity, reproduction, diet, etc
• How can the information be summarized?
• How can new studies be informed?
• How can best estimates for species without
studies be derived?
Heavy Computing to the Rescue
• Assemble all relevant facts, with probability
distributions
• Establish their correlations, with probability
distributions
• Select suitable models to explain data and
predict key parameters
• Let the computer test all possible combinations
and select those with highest overall probability
Example I
MSY from Catch Data
Catch-MSY in a Nutshell
•
•
•
•
Get reliable catch data
Select a population growth model
Try all possible parameter combinations
Select those that, given the catches, do
neither crash the stock nor overshoot carrying
capacity of the ecosystem
Excellent Match with Estimates from full Stock Assessments
Available from FishBase, but runs for several minutes
Challenge: General Scientific Method
to Summarize Widely Different Information
Bayesian Inference in a Nutshell
• Prior: express existing knowledge (textbook,
common sense, logic, best guess, previous
studies) with a central value (such as a mean)
and a distribution around it (such as a normal
distribution and a standard deviation).
• Likelihood function: analyze new data, get the
mean and distribution
• Posterior: Combine prior and likelihood into a
new, intermediate mean and distribution
Example II
Length Weight Relationships (in press)
Length Weight Relationships
LWR Across All Studies
LWR for Many Studies
LWR for One Study Only
LWR Priors
FishBase Online
FishBase Online
Estimating LWR for ALL Fishes
• Best LWRs are obtained from studies for the
species, including relatives with similar body
shapes if needed (< 5 LWR)
Rank
Based on LWR estimates…
Species
1
for this Species
410
2
for this Species & Genus-body shape
700
3
for this Species and Subfamily-body shape
1,419
4
for relatives in Genus-body shape
1,222
5
for relatives in Subfamily-body shape
18,410
6
for other species with this body shape
9,711
Self-Learning Database
• When Daniel and Rainer first discussed about
FishBase, they envisioned an artificial
intelligence system
• Some years (decades) later, we are getting
there:
the addition of LWRs for 16 species improved LWR
quality for over 400 species
Next Steps
• Repeat exercise with growth estimates
(ongoing)
• Repeat exercise with mortality and maturity
• Estimate intrinisc rate of population increase
(the holy grail in biology)
Thank You
Questions?