Tuna is a crucial economic fish species in the ocean, and the research on predicting the location of tuna fishing grounds has always been a key issue in deep-sea fisheries. Tuna, being a highly active fish that moves vertically in the ocean, is extremely sensitive to changes in marine environmental factors. This study aimed to investigate the impact of environmental factors and water depth changes on the accuracy of tuna fishing ground prediction. To achieve this, convolutional neural networks (CNNs) were utilized to train datasets using four key environmental factors - concentration of chlorophyll-a (CHLa), seawater temperature (ST), concentration of dissolved molecular oxygen (Oxy), and seawater salinity (SS) - at various depths ranging from 0 to 300 meters. A binary classification fishing ground prediction model for high and low yield areas of albacore and yellowfin tuna in the Western and Central Pacific Ocean was constructed. The goal was to identify the optimal fishing ground prediction model trained under these conditions and to explore the potential impact of environmental factors and water depth on the prediction accuracy of the model. The data used in this study included information from the Western and Central Pacific Ocean albacore and yellowfin tuna fisheries from 2003 to 2016, along with corresponding environmental data, which were used as the training and validation sets. The fishing data and environmental data from 2017 were then utilized as the test set to develop fishing ground prediction models for the two fish species. The results indicate that using convolutional neural networks to construct fishing ground prediction models is indeed feasible, and spatio-temporal information plays a significant role in optimizing predicting models for tuna fishing ground. Among the models developed for the same fish species using the four marine environmental factors at different depths, the model based on CHLa at each depth was found to be the most effective, while the model constructed using Oxy at the same depth performed the worst. Furthermore, the albacore fishing ground prediction model demonstrated significantly higher efficacy compared to the yellowfin tuna fishing ground prediction model when using the same marine environmental factors. It was also observed that the accuracy of fishing ground prediction models for the same species varied depending on water depth, even when based on the same marine environmental factors. These findings highlight the importance of considering both environmental factors and water depth in predicting tuna fishing grounds accurately.