Estimating seasonal habitat suitability for migratory species in the Bohai Sea and Yellow Sea: A case study of Tanaka's snailfish (Liparis tanakae)

Acquiring a comprehensive and accurate understanding of habitat preference is essential for species conservation and fishery management, especially for mobile species that migrate seasonally. Presence and absence data from field surveys are recommended when available due to their high reliability. Using field survey data, we investigated seasonal habitat suitability requirements for Tanaka's snailfish (Liparis tanakae) in the Bohai Sea and Yellow Sea (BSYS) via a machine-learning method, random forests (RFs). Five environmental and biologically relevant variables (bottom temperature, bottom salinity, current velocity, depth and distance to shore) were used to build the ecological niches between the presence/absence data and suitable habitat. In addition, the degree to which false absence data might impact model performance was evaluated. Our results indicated that RFs provided accurate predictions, with seasonal habitat suitability maps of L. tanakae differing substantially. Bottom temperature and salinity were identified as important factors influencing the distribution of L. tanakae. False absence data were found to have negative effects on model performance and the decrease in evaluation metrics was usually significant (P<0.05) after 30% or more errors were added to the absence data. Through identifying highly suitable areas within its geographic range, our study provides a baseline for L. tanakae that can be further applied in ecosystem modelling and fishery management in the BSYS.

Automated summary

Acquiring a comprehensive and accurate understanding of habitat preference is crucial for species conservation and fishery management. This study used random forests to investigate the seasonal habitat suitability requirements for Tanaka's snailfish, and found that bottom temperature and salinity were important factors influencing its distribution. Additionally, the study evaluated the impact of false absence data on model performance.