To improve the predicted performance of south Pacific Thunnus alalunga fisheries, in this paper a fishery prediction method based on the extreme learning machine (ELM) is proposed using the spatial and temporal information, the number of fish caught, the number of longline catches and three important environmental factors data: sea surface temperature, sea surface height and chlorophyll-a  concentration of Thunnus alalunga fishery in the South Pacific from 2000 to 2015. This method firstly introduces an one-hot liked coding algorithm, for effective digital coding of characteristics of several kinds of fisheries data, and then presents a Thunnus alalunga fishery forecasting model based on ELM, to learn the characteristics of various discretization encoded data and to obtain the stability prediction model parameters adaptively, for effective prediction of tuna fishing ground. Finally, the experimental results show that when the data of Thunnus alalunga in the South Pacific in 2015 are used as the test validation, the prediction accuracy of the model is 84.07% and the F1 Score index is 80.9%, which improves the performance of the fishery prediction compared with the conventional methods. This study provides a new idea for the research of fishery prediction.