Short-Term Prediction of Hilsa (Tenualosa ilisha) Catch in the Northern Bay of Bengal Using Advanced Machine Learning Algorithms

Hilsa is a vital transboundary fishery resource in the Bay of Bengal (BoB), holding commercial, ecological, and cultural importance. This study aims to develop a short-term prediction of Hilsa catch in the northern BoB using a machine learning (ML) model. The prediction technique was developed considering the georeferenced Hilsa catch per unit effort (CPUE) as a function of environmental variables like surface salinity, sea surface temperature (SST), surface current speed, and direction. We employed two advanced ML algorithms, viz., random forest (RF) and 5. (XGBoost) to examine their efficacy in the short-term prediction of Hilsa for the northern BoB and compared the model performance with a baseline information obtained through multiple linear regression (MLR). Our analysis showed significant improvement in the prediction accuracy using advanced ML techniques where XGBoost again outperformed RF. The root mean square error (RMSE) values between observed and predicted CPUE for RF and XGBoost models were 5.72 and 5.63 kg/h, respectively. The correlation coefficient (r) between the observed and predicted catch were 0.90 and 0.93 for RF and XGBoost, respectively. SHapley Additive exPlanations (SHAP) analysis revealed the highest influence (58.38%) of surface current speed on the Hilsa CPUE. We generated the spatial prediction maps of Hilsa CPUE using the best performing (XGBoost) model with 85% prediction efficiency. This study showed the potential of the XGBoost model in developing a short-term prediction for Hilsa in the northern BoB, towards developing Hilsa fishery advisory for sustainable management of these fishery resources.