Summary of Abstract Submission

Abstract Submission No. IO50-09-0063PresentationPoster

PREDICTION OF INDIAN MACKEREL (RASTRELLIGER KANAGURTA) DISTRIBUTION IN THE ARABIAN SEA USING GAM (GENERALIZED ADDITIVE MODEL)

H. U. Solanki*1, Dhyey Bhatpuria2, Mini Raman1, Prakash Chauhan1, A Tiburtius3, Premchand3

1 Space Applications Centre, ISRO, Ahmedabad, India
2 Space Applications Centre, ISRO, Ahmedabad,, India
3 Fishery Survey of India, Mumbai, India

ABSTRACT :

Study aims to illustrate how remotely sensed oceanic variables and fishing operations data can be used to predict suitable habitat of fishery resources in GIS environment. Analyses of remotely-sensed oceanic variables and fisheries data in GIS environment have facilitated to understand fundamental relationships between fishery resources and their oceanic environment. GIS applications and statistical models provides platform for integrating diverse forms of data to provide scientifically hidden information for marine resource management. Time series satellite derived sea surface height (SSHa), sea surface temperature (SST), chlorophyll concentrations (CC), photosynthetically active radiation (PAR), and synchronous fishery data for year 1998-2010 fishing seasons to predict fishery species Indian Mackerel. Database model was developed to create satellite and fishery data base. Datasets were segregated randomly to create training and validation data with a ratio of 80:20. Catch was normalized into Catch Per Unit Effort (CPUE) with unit as kg/h. Generalized Additive Model (GAM) was performed on training data and then tested on the validation data. Suitable ranges of SST, CC, SSHA and PAR, for different fishery species distribution were derived using GAM. These variables were integrated to predict spatial distributions of fishery resources and validated with hind cast data set. Results indicate a good match between predicted and actual catch. Monthly probability map of predicted habitat areas coincides with the high catch of the particular month. Paper discusses methodological investigations for suitable ranges of different variable, integration of variables to generate predicted distribution maps of fishery resources and results of hindcast validation.