Workshop on

Prediction for marine resources

Date: November, 19 2004

Place: Institute of Statistical Mathematics, Auditorium

Program

9:30-10:25
1. Byeong U. Park (Department of Statistics, Seoul National University)
Nonparametric regression techniques: A broad perspective

10:30-11:25
2. Tsukasa Hokimoto (Hokkaido University)
Predictability on spatiotemporal distribution of chlorophyll in the Sea of Japan

Lunch break................................ 11:25-13:00

13:00-13:55
3.Kazuhiko Hiramatsu (National Research Institute of Far Seas Fisheries)
Development of fisheries stock management using the operating model

14:00-15:05
4. Cleridy Lennert-Cody (Inter-American Tropical Tuna Commission)
An application of random forests to identify unusual observations in fisheries data

Tea break ..................................... 15:05-15:25

15:25-16:15
5. Masanori Kawakita (Graduate University for Advanced Studies)
Shark bycatch prediction by AdaBoost

16:20-17:10
6. Mihoko Minami (Institute of Statistical Mathematics and Graduate University for Advanced Studies)
Analysis of Shark bycatch counts by zero-inflated GAM models

Abstract

1. Byeong U. Park (Department of Statistics, Seoul National University)

Nonparametric regression techniques: A broad perspective

In this presentation, several nonparametric regression techniques are introduced. Thoseinclude local polynomial fitting, penalized least squares (spline smoothing), nearest neighbor methods, and wavelet smoothing. Also included are several techniques of fitting additive regression models, such as marginal integration, ordinary and smooth backfitting methods. Some extensions to local quasi-likelihood regression are also discussed.

2. Tsukasa Hokimoto (Hokkaido University)

Predictability on spatiotemporal distribution of chlorophyll in the Sea of Japan

In order to predict the change in fish distribution,the prediction of phytoplankton distribution in the sea has been an important problem. In this talk, we would like to present the methodology for predicting the change on chlorophyll-a concentration of the phytoplankton in the Sea of Japan, which is based on a spatiotemporal statistical model with spatially local structure.

3.Kazuhiko Hiramatsu (National Research Institute of Far Seas Fisheries)

Development of fisheries stock management using the operating model

Managing a fisheries stock is a difficult task.  There is large uncertainty of stock status, biology, and fisheries and conducting experiments in the marine environment is impossible.  Recently management procedures are investigated using the simulation model (operating model), which represents real system behavior. The operating model approach allows a laboratory testing of management procedures and developing the procedure which is robust to uncertainty.  Recent development of management procedures using the operating model approach is reviewed in this presentation.

4. Cleridy Lennert-Cody (Inter-American Tropical Tuna Commission)

An application of random forests to identify unusual observations in fisheries data

Fisheries observers of the Inter-American Tropical Tuna Commission collect data on catches and bycatches aboard purse-seine vessels of the international fishery for tunas in the eastern Pacific Ocean. A brief introduction to the fishery and the data will be presented, with emphasis on the some of the statistical challenges these data present, followed by an application of random forests to these data. A random forest algorithm is being developed to classify purse-seine sets according to their location, tuna catch compositions and environmental conditions. The algorithm is developed on data from the last two decades, using over 45 predictors. Applications of this algorithm will be discussed.

5. Masanori Kawakita (Graduate University for Advanced Studies)

Shark bycatch prediction by AdaBoost

 We studied the shark bycatch issue associated with tuna-fisheries in the eastern Pacific Ocean with AdaBoost. The shark bycatch issue can be regarded as a two-group classification problem whether one or more sharks are caught or not.  AdaBoost, which is a new statistical method for two-group classification, is applied to prediction of shark bycatch occurrence.  We show that AdaBoost has some favorable properties for this problem.

6. Mihoko Minami (Institute of Statistical Mathematics and Graduate University for Advanced Studies)

Analysis of Shark bycatch counts by zero-inflated GAM models

 Shark bycatch counts are characterized by a large number of zero observations. We believe that the large proportion of zeros in the data arise because sharks are not always associated with tunas and they are not caught in such cases. We apply zero-inflated Poisson regression model and zero-inflated Negative Binomial model with thin-plate regression splines so that bivariate smooth function of longitude and latitude can be include in the model.