1. Modeling fishermen behavior in the Santa Barbara Channel Islands geog.Mike & econ.John www.ucsb.edu

2. Images: William B. Folsom, NMFS, http://www.photolib.noaa.gov/fish/ Commercially harvested sea urchins ready for offload at the Ventura marina. Inspecting sea urchin innards.

3. Fisherman behavior Adaptability –Fishing is fraught with physical and financial risk and is undertaken in a constantly changing environment. –Successful fishermen exhibit an ability to change their behavior with varying conditions and information. –Learning and communication are critical components of fishing activity.

4. Fisherman behavior Adaptability –How do fishermen learn about their environment? How does this affect their efficiency and success? –How are risk behaviors and decision paradigms set? –Do these factors change significantly over seasons, over years, or as catch is (or is not) accumulated?

5. Research Heterogeneous effort distribution –Satisficer-optimizer continuum Learning & memory –Bayesian updating Communication –Information exchange matrix

6. Heterogeneous effort distribution

7. Effort distribution model Random fishing probabilities (~U [0,1]) applied to each fisherman in fleet –Probably closer to an exponential or highly skewed gamma distribution Variable weather Temporal closures

8. Effort distribution model Utility Function to decide most attractive patch where: F = “attractiveness” weight applied to patch D = distance of patch from home port

9. Effort distribution model We let the utility function be probabilities and treat as a cdf –Sort probabilities in ascending order –Keep track of corresponding patch numbers

10. Effort distribution model If a fisherman decides to go fishing we draw a (constrained) random number and find the corresponding value (inverse CDF) in the range 1:Npatches (# of patches)

11. Effort distribution model Determine whether or not each fisherman in the fleet goes fishing –decision paradigm –weather, temporal closures, gear maintenance Determine where each fisherman goes fishing –patch attractiveness –information & learning (eventually) Determine "attractiveness" of each patch for next day –count number of "hits" to each patch & reduce abundance –attractiveness should increase at first as a patch is exploited and then decrease as fish are removed Loop over the year Loop over multiple years

12. Results PatchID Initial Patch Attractiveness Average Hits 0-----5574 10.30332505 20.37466452 30.118090 40.119030 50.744982329 60.186520 70.169720 80.0252140 90.57544266 100.413540

13. Results

14. FishermanID Decision Paradigm Average Fishing Effort 10.2079152 20.45583112 30.85126213 40.77714197 50.75392185 60.52262119 70.57483148 80.77253192 90.9271233 100.66157163 110.1214428 120.2354958 130.73946184 FishermanID Decision Paradigm Average Fishing Effort 140.90988222 150.78506199 160.47394117 170.3754197 180.50925114 190.04331910 200.64754162 210.81629198 220.48307121 230.068818 240.99486248 250.69553167 Gung-Ho! (optimizer)

15. Results FishermanID Decision Paradigm Average Fishing Effort 10.2079152 20.45583112 30.85126213 40.77714197 50.75392185 60.52262119 70.57483148 80.77253192 90.9271233 100.66157163 110.1214428 120.2354958 130.73946184 FishermanID Decision Paradigm Average Fishing Effort 140.90988222 150.78506199 160.47394117 170.3754197 180.50925114 190.04331910 200.64754162 210.81629198 220.48307121 230.068818 240.99486248 250.69553167 Gun shy! (satisficer)

16. Fish block/simulation comparison 2004 DFG urchin block dataUrchin fishing simulation

17. Learning, memory, and communication

18. Learning  Information received from an individual’s daily fishing effort Communication  Information received from the rest of the fleet Bayesian updating: –Modify an individual’s belief about the environment if they go fishing –Influence their decision the following day based on the beliefs of the entire fleet

19. Learning & memory Bayesian updating (DeGroot, 1970)  Good signals (S a > α 0 ) increase expected abundance at site a  Noisy signals (large σ 2 s ) are given less weight α = abundance (α 0 = mean) S a = signal, which is normally distributed A fisherman updates beliefs about abundance after acquiring signal S a from visiting site a. These updated beliefs follow a normal distribution.

20. Communication Information exchange matrix –Allen and McGlade, 1987 Matrix of “how well” and with whom information is shared between each member of fleet No sharing Perfect sharing Imperfect sharing Ignore information Developing sharing “weights” –Mean of guy that’s gone 200 times vs. mean of guy that’s only gone once –Variance of signal has an effect on “confidence in the signal”

21. Learning, memory, and communication Information exchange model output (red = fishing, blue = no fishing)

22. Fish block/simulation comparison Information exchange model output (red = fishing, blue = no fishing) 2004 DFG urchin block dataUrchin fishing simulation

23. Image: Wm. B. Dewey, www.islandpackers.com Questions?Suggestions? THANKS… Dave Siegel, Chris Costello, Kostas Goulias, Kristine Barsky, Chris Miller, Pete Halmay