Neural network models for the prediction of Indian mackerel catch using environmental variables in Visakhapatnam fishing harbour

Abstract

Artificial Neural Network (ANN) models were developed here to forecast daily catches of Indian Mackerel [Rastrelliger kanagurta (Cuvier, 1816)] at Visakhapatnam Fishing Harbour (VFH), located along the Bay of Bengal (BOB), Andhra Pradesh, India. The study utilised daily mackerel catch data alongside two Satellite-derived variables, Chlorophyll-a (CHL-a) and Sea Surface Temperature (SST), as well as two meteorological variables, Air Temperature (AT) and Wind Speed (WS), from January to September 2024. Fourteen Neural Network Models (NNMs) were developed to predict September 2024 (30 days) mackerel catch With Ban (WB, 7 models) and WithOut Ban (WOB, 7 models). The predicted catches were then compared with the actual catches of September 2024. Among the WB models MAC_ALL_WB (4-4-1) model performed better and MAC_SAT_WOB (2-4-1) model performed better among the 7 WOB models, achieving a low Mean Squared Error (MSE) of 0.012 and 0.009, respectively. Satellite-derived data emerged as the most influential variable for predicting mackerel catch, and it was concluded that predictions with WOB models were more accurate than those with WB models.