2023 IEEE ECICE 最佳論文獎 – Machine Learning-Based High-Resolution Estimation of Global Catch Distribution of Taiwan’s Distant Water Fisheries

Taiwan’s overseas fishing fleet is well-known worldwide for its size and operation efficiency. Due to technological advancement, Taiwan has an excellent coastal and offshore mapping system reaching 0.01 and 0.001 degrees in locating fishing grounds and plotting catch distribution. However, for overseas fisheries, traditional methods for mapping catch distribution use only the trajectory of the fishing vessel along with eLogbooks to evenly distribute the catch along the way on a coarse resolution geographic information system (GIS) of 1.0 degrees due to the size of the globe limiting the computation scale\cite{OFDC_summary}. This research presents an approach to map catch distributions of Taiwan’s overseas fishing fleet globally using artificial intelligence and high-resolution geographic information system (GIS) technology of 0.5 and 0.25 degrees. We integrate vessel monitoring system (VMS) data and eLogbooks records with artificial intelligence in recognizing fishing activities, improving the accuracy of pinpointing fishing grounds.

Our approach first combines the VMS data and eLogbook catch records of Taiwan fishing vessels to recognize the species of catch of each fishing vessel. After obtaining the species for each fishing vessel, the system employs machine learning to identify fishing and non-fishing to assign weighted catches based on the vessel operating state derived from a machine learning algorithm. Finally, the result is analyzed globally on a high-resolution scale of 0.25 degrees.

Preliminary findings show that the system can map the catch volumes of different fish species across regions around the globe and differentiate between fishing and non-fishing areas while the fishing vessel travels. This effectively removes catch distributions in areas where vessels are not fishing, providing a more accurate pinpoint of fishing grounds and enhancing the result of resource assessment. In addition, a distinctive feature to visualize quantitatively the distinct catch and species in a high-resolution domain is also made possible.