Making communications land - Are they received and understood as intended? we...
Bayesian networks for environmental management including fisheriesThe BayFish suite of models
1. people science environment partners
Bayesian networks
for environmental management
including fisheries
The BayFish suite of models
Eric BARAN
Martin VAN BRAKEL
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. Probabilities based on a
time series analysis
COMBINATION OF PROBABILITIES
Probabilities based on
mapping
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. 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. 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. 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
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. 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. 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. 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
14. Sluice gate operation
Baseline
All open
Sluice gates scenarios LT open
LT HP open
All closed
Mekong inflow Marine inflow Rainfall
Above mean 39.1 Above mean 39.1 Above 89mm 43.0
Below mean 60.9 Below mean 60.9 Below 89mm 57.0
Water quantity 3 ± 16 -53 ± 97 114 ± 93
Water salinity
Water acidity Above 10 40.6 Soil acidity
Water pollution
Between 4-10 28.8 Severe 20.6
Water quality Acceptable 93.1
Unacceptable 6.88 Below 4 30.6
Important 60.5
Medium 39.5
Negligible 39.5
13 ± 12 No 39.9
FWQuantity for rice WQual. for rice WQual. for aquaculture
High 40.3 Good 30.0 WQual. for est. fish WQual. for FW fish Good 54.6
Low 59.7 Bad 70.0 Good 52.7 Good 23.7 Bad 45.4
Bad 47.3 Bad 76.3
Estuarine fish Freshwater fish Fish Aquaculture
Production Good 43.2 Good 26.8 Good 48.2
Bad 56.8 Bad 73.2 Bad 51.8
Wild fish
Good 41.4
Bad 58.6
Rice production Fish production Crab Production Shrimp production
Good 48.4 Good 43.7 Good 54.6 Good 55.6
Bad 51.6 Bad 56.3 Bad 45.4 Bad 44.4
TOTAL INCOME FOOD SECURITY ENVIRONMENT
Global outcomes Good 53.4 Good 48.0 Good 46.9
Bad 46.6 Bad 52.0 Bad 53.1
15. Outcomes of BayFish – Bac Lieu
All Gates Closed
• Salinity very low, problems with acidity
• Aquaculture production very low
• Rice production increases a little (baseline +15%)
• Household income decreases (not balanced by better food security)
• Environmental conditions deteriorate
Lang Tram and Ho Phong sluice gates open
• High marine inflow
• High aquaculture production (baseline +80%)
• Decreased rice production (baseline –10%)
• Household income increases (baseline +50%)
• Food security and Environmental health held at baseline levels
etc
16. Details about BayFish – Bac Lieu:
Principles and structure
Google: “Developing a consultative Bayesian model”
18. Objectives
• To identify relationships between river hydrology, floodplain
habitats and fish production
• To predict the impact of environmental modification on fish
production
• To raise awareness among stakeholders about variables to be
encompassed in the management process
20. TS rainfall TS runoff
Tonle Sap Flooded
Mekong inflow water level Floodplain O2 vegetation
HYDROLOGICAL
SCENARIOS
Overland flow O2 for O2 for O2 for
TS migrants Mekong migrants residents
Built
Flood beginning Flood duration
Flood
Structures
MIGRATIONS HABITAT for
level
residents
Floodplain
refuges
HABITAT for
MIGRATIONS
TS migrants
of residents
Flood for fishes MIGRATION of
TS migrants HABITAT for
Mekong migrants
MIGRATIONS of
Mekong migrants
HYDROLOGY HABITAT
STOCK of
residents
STOCK of
# Khmer MS fishers
TS migrants
Pressure from
LS fishery MS gear efficiency
STOCK of # migrant MS fishers
Mekong migrants
# MS fishers
CATCH of
PRESSURE
residents Stock on residents
Pressure from # Viet./Cham MS fishers
MS fishery Activity of
SS fishers
CATCH of
# Viet./Cham SS fishers
TS migrants PRESSURE
on TS migrants Pressure from # SS fishers
SS fishery
# Khmer SS fishers
CATCH of Gear size
Mekong migrants
PRESSURE of SS fishers
on Mekong migrants
Catch FISHERIES
TOTAL
FISH CATCH
21. TS Runoff
Over Mean 42.5
Under mean 57.5
2.5e+004 ± 1.4e+004
Mekong flow
Over mean 47.6
Under mean 52.4
3.1e+004 ± 1.5e+004 Water level Kampong L
...
>9m 52.7
Overland flow <9m 47.3 Bank structures
Over Mean 42.9 7.9 ± 3.2
Many 5.00
Under mean 57.1 Few 95.0
7.3e+003 ± 4.7e+003
Floodplain oxygen Floodplain vegetation
> 4 mg/l 29.8 Grass 56.3
Flood beginning Flood duration Floodplain flood level 2<mg/l <4 27.6 Shrub 41.8
Before 20 June 30 > 13 weeks 30 > 9m 57.2 2 mg/l 42.6 Forest 1.94
Betwen 55 5< weeks <13 60 < 9m 42.8
After 20 Aug 15 < 5 weeks 10
O2 for Black fish O2 for White fish
Acceptable 58.4 Acceptable 67.9
HYDROLOGY Impossible 41.6 Impossible 32.1 HABITAT
WUP-JICA JICA
model GIS
WUP-JICA MRC WUP-FIN
database Consultation databases model Bibliography
22. HYDROLOGY HABITAT
TS rainfall TS runoff
Above 1000 45.5 Above 30000 53.7
Below 1000 54.5 Below 30000 46.3
1.2e+003 ± 8.7e+002 2.8e+004 ± 1.4e+004
TS water level Floodplain O2 Flooded vegetation
Mekong inflow Above 10m 29.6 Above 4mgl 16.6 Grass 54.6
HYDRO. SCENARIOS Above 34300 60.0 From 8 to 10 47.8 From 2 to 4 ... 23.5 Shrub 43.4
Baseline 100 Below 34300 40.0 Below 8m 22.6 Below 2mgl 59.9 Forest 1.98
High 0 3.4e+004 ± 1.6e+004 8.6 ± 2.6
Dam 0
Overland flow O2 for TS migrants O2 for Mekong migrants O2 for residents
Above 6400 60.0 Acceptable 28.4 Acceptable 16.6 Acceptable 40.1
Below 6400 40.0 Impossible 71.6 Impossible 83.4 Impossible 59.9
Flood beginning Flood duration
Flood level
Before mid J ... 40.0 More 11 we ... 26.2
Mid July to ... 40.0 Around 8 we ... 66.1 High 51.3 MIGRATIONS Built Structures
HABITAT for residents
Low 48.7 Blocking 3.64
After mid Au ... 20.0 Less 6 weeks 7.70 Open 96.4 Good 26.3
Bad 73.7
MIGRATIONS of residents Floodplain refuges
Flood for fishes Free 65.0 Perennial 73.2
HABITAT for TS migrants
Good 60.6 Blocked 35.0 Temporary 26.8
Good 18.5
Bad 39.4 Bad 81.5
MIGRATION of TS migra ...
Free 70.2
Blocked 29.8
HABITAT for Mekong mi ...
MIGR. of Mekong migra ... Good 10.7
Free 52.1 Bad 89.3
STOCK of residents Blocked 47.9
Abundant 48.5
Scarce 51.5
STOCK of TS migrants
Abundant 49.8
Scarce 50.2 Stock MS gear efficiency
Increasing 75.0 # Khmer MS fishers
STOCK of Mekong migr ... Stable 25.0 Increasing 50.0
Abundant 43.9 Pressure from LS fishery Stable 50.0
PRESSURE on residents
Scarce 56.1 Nil 41.0
Increasing 67.5 Blockage 59.0 # MS fishers # migrant MS fishers
CATCH of residents Stable 32.5
Increasing 60.0 Increasing 50.0
High 58.0 Stable 40.0 Stable 50.0
Low 42.0
PRESSURE on TS migra ... # Viet./Cham MS fishers
Increasing 69.6 Pressure from MS fishery Increasing 75.0
CATCH of TS migrants Stable 30.4 Increasing 64.5 Stable 25.0
High 59.7 Stable 35.5 Activity of SS fishers
Low 40.3 More Fishing 63.4
Catch More Farming 36.6
# Viet./Cham SS fishers
PRESSURE on Mekong ... Increasing 75.0
CATCH of Mekong migra ... Increasing 69.6 # SS fishers
Pressure from SS fishery Stable 25.0
High 56.7 Stable 30.4 Increasing 98.7
Increasing 73.4
Low 43.3 Stable 1.30
Stable 26.6 # Khmer SS fishers
Increasing 100
Gear size of SS fishers Stable 0
Increasing 25.0
TOTAL FISH CATCH Stable 75.0
High 57.0
FISHERIES
Low 43.0
23. Production of the dai fishery (x1000 tonnes)
Probability of a high stock in BayFish (%)
16
60
14
50
12
40
10
30 8
6
20
4
10
2
0 0
1995 1996 1997 1998 1999 2000 2001 2002 2003
Mekong migrant fish TS resident fish Dai fishery catch
24. Details about BayFish – Tonle Sap:
Principles and structure
http://www.mssanz.org.au/modsim05/papers/baran.pdf
Comprehensive report
http://www.ifredi.org/BS_project.asp
25. BayFish SWOT
STRENGTHS
• The only kind of decision support tool that can integrate quantitative
as well as qualitative information
• Can overcome the paucity of statistics and biological information in
zones poorly studied
• A way for a diversity of stakeholders to meet and talk about
management decisions
• Tool intuitive (no programming language needed) and user-friendly.
Open sources, models of very small size.
Software freely accessible on Internet to read models
(www.norsys.com)
26. BayFish SWOT
WEAKNESSES
• Not a dynamic model (snapshot of a system instead)
• Probabilities subjective -> good stakeholders selection is essential
• The tool must be simple enough to be acceptable -> difficult balance
between simplification and realism
OPPORTUNITIES
• Build models and partnerships between fish- and
agriculture- related institutions
• Address trade-offs in water uses and management