Successfully reported this slideshow.
Upcoming SlideShare
×

Bayesian networks for environmental management including fisheries The BayFish suite of models

1,252 views

Published on

Presented at the Basin Focal Project Fisheries Workshop at Cali, Colombia, Feb 2008

Published in: Education
• Full Name
Comment goes here.

Are you sure you want to Yes No
• Be the first to comment

• Be the first to like this

Bayesian networks for environmental management including fisheries The BayFish suite of models

1. 1. people science environment partners Bayesian networks for environmental management including fisheries The BayFish suite of models Eric BARAN Martin VAN BRAKEL
2. 2. Bayesian models Networks of variables Example: medical diagnosis of a baby with a probability of interaction between variables • Each box is a variable with 2 or 3 states • Choice of variables and weight of each variable is based on hard data OR on expert knowledge • Once the network is set up and each variable parametrized, the software computes probabilities resulting from multiple interactions
3. 3. Probabilities based on a time series analysis COMBINATION OF PROBABILITIES Probabilities based on mapping
4. 4. Scenario analysis 1 Scenario analysis 2
5. 5. Bayesian Networks • BNs define a system as a network of variables linked by probabilistic interactions • For each variable a small number of states is defined • Variables can be driving (parent) or driven (child) Network variables Attached probability tables Justification (driving variables) (based on data or on knowledge) Natural fish stock Number of fishing boats Natural fish stock There are 74% chances that the fish >100,000 tons > 500 units <100000 tons 300 to 500 units stock is superior to 100,000 tons >100,000 tons 74 < 300 units There are 26% chances that the fish <100000 tons 26 stock is inferior to 100,000 tons There are 25% chances that the number of Number of fishing boats fishing boats is superior to 500 units > 500 units 25 There are 50% chances that the number of Fish catch 300 to 500 units 50 fishing boats is between 300 and 500 units > 50,000 tons There are 25% chances that the number of < 300 units 25 25,000 - 50,000 tons < 25,000tons fishing boats is inferior to 300 units
6. 6. Driven Variables Driving variables and probability Justification Computer calculation table of the driven variables (detailed example for one combination) (based on Bayes formula) of the probability of having a certain catch given all the Natural fish stock Number of fishing boats previous given probabilities >100 ,000 tons > 500 units <1 00 000 tons 300 to 500 units < 300 units If the stock is >100,000 tons If there are 300 to 500 boats Stock Boats Catch Then > 50,000 tons 25,000 to 50,000 tons 25,000tons < Fish catch - there are 60% chances that >100,000 tons> 500 units 90 10 0 >100,000 tons300 to 500 units 60 30 10 the catch is superior to 50,000 tons > 50,000 tons >100,000 tons< 300 units 20 20 60 - there are 30% chances that 25,000 - 50,000 tons <100000 tons > 500 units 10 70 20 25,000 tons < catch < 50,000 tons < 25,000tons <100000 tons 300 to 500 units 10 30 60 - there are 10% chances that <100000 tons < 300 units 0 10 90 the catch is inferior to 25,000 tons
7. 7. Building a Bayesian network 1. Network development • Identify the major variables of the system studied • Delete variables for which no information is available • Arrange remaining variables in a meaningful way 2. Variables specification • Define the relevant states of each variable • Specify the weight of each variable 3. Parameterization (“elicitation of probabilities”) • Define the probability of each state of each driving variable • For each driven variable, define the weight of each driving variable.
8. 8. Thus Bayesian models • Quantify trends resulting from multiple influences • Allow integration of quantitative and qualitative information (e.g. databases and local knowledge) • Are made for scenario analysis • Allow analyzing trade-offs • Act as decision support tools
9. 9. BayFish Bac Lieu
10. 10. Bac Lieu: inland coastal system characterized by: - a saline intrusion; - a series of sluice gates; - conflicting land uses (basically rice vs. shrimp farming) 1 Shrimp farming area 2 Shrimp+rice area 3 Rice farming area 3 2 1 Cau Sap r oa d Vinh My stal Coa Thanh Tri Ho Phong Pho Sinh Sea Lang Tram
11. 11. Objectives 1. Help optimize operation of sluice gates 2. Assist decision making about water management options 3. Inform stakeholders about SLUICE GATE production trade-offs 4. Involve stakeholders in the PHYSICAL WATER SOURCES management process WATER AND SOIL CHEMICAL PROPERTIES RICE FISH CRAB SHRIMP TOTAL PRODUCTION
12. 12. Management options, outcomes and trade-offs Economic return Food security Ecosystem health SLUICE GATE SLUICE GATE SLUICE GATE PHYSICAL WATER SOURCES PHYSICAL WATER SOURCES PHYSICAL WATER SOURCES WATER AND SOIL CHEMICAL PROPERTIES WATER AND SOIL CHEMICAL PROPERTIES WATER AND SOIL CHEMICAL PROPERTIES TOTAL PRODUCTION TOTAL PRODUCTION TOTAL PRODUCTION
13. 13. Gate operation Sea vs. Mekong water flows Saline vs. fresh water Pollution (acid soils, farming, shrimp factories) IF freshwater IF no pollution IF saline water RICE FISH/CRAB SHRIMP Low income Medium income High income Bad environment Good environment Bad environment Food security Food security No food security TRADE OFFS