Assessment of the role of mangroves for Periophthalmus modestus applying machine learning and remote sensing: a case study in a large estuary from Vietnam

Early stages of fish are easily sensitive to any alteration of environments, thus understanding their dispersions in a dynamic system like estuaries are important in protection and conservation of fish diversity and fishery resources. Especially, it is more significant for a mangrove-associated mudskipper, Periophthalmus modestus, which are distributed in mudflats along the Northwest Pacific, and now are listed as near threatened due to climate change and human activities. In the present study, a hybrid model, Extreme gradient boosting (XGBoost)-Artificial neural networks (ANN), was applied to forecast the distribution of P. modestus larvae and juveniles collected from a large estuary in northern Vietnam, which are driven by temperature and mangrove changes. Present results demonstrate the usefulness and applicability of ANN-XGBoost model in ecological studies, with an excellent estimation accuracy. Furthermore, employing Generative adversarial networks (GANs) model, this study exhibits a decrease in mangrove areas due to human activities between 2010 and 2023. This change with a rise in temperatre during this period would have impacted on P. modestus larvae and juveniles, which tend to be distributed in mangroves and avoid human-affected areas. Thus, it is concluded that changes in P. modestus’ environment like mangroves have a significant influence on their distribution and survival. Applying a novel model in ecological research, this work further indicates the importance of mangrove forests for aquatic organisms, especially mudskippers. This research will allow scientists and biological managers to make more precise forecasts regarding the spread of P. modestus, while also helping to the protection of this mudskipper and other species. Protecting and developing mangrove forests are the first and crucial action to supply a suitable habitat for any fish species. The models employed in this work will be helpful for other relevant studies when obtaining a highly accurate performance.