머신러닝 기반 남극은암치(Pleuragramma antarctica) NASC 예측 모델 개발
Antarctic Silverfish Pleuragramma antarctica Nautical Area Scattering Coefficient (NASC) Prediction Using a Machine Learning-Based Model
2
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
This study aimed to develop a machine learning-based prediction model for the nautical area scattering coefficient (NASC) of the Antarctic silverfish Pleuragramma antarcticum, a key species in the Southern Ocean. Acoustic survey data from the Ross Sea from 2018 to 2023 were integrated with environmental variables, including depth, temperature, salinity, survey period, survey area, and grid location, to construct Random Forest regression models. Separate models were trained on the adults and juveniles. For adults, continuous variables were standardized using z-scores. Meanwhile, juvenile models were standardized using raw values. Model training was performed using MATLAB TreeBagger with grid search optimization. The performance was evaluated by hold-out validation. The adult model achieved high accuracy (R²≈0.76, RMSE≈2.10), with depth, temperature, and salinity identified as the most influential predictors. The juvenile model showed lower explanatory power (R²≈0.38, RMSE≈2.54), often underestimating high NASC values. Adults are more strongly governed by physical conditions, whereas juveniles are influenced by additional biological or ecological factors. Random Forest models can effectively predict adult silverfish NASC using limited environmental inputs, supporting the improved interpretation of acoustic data and ecosystem-based management in polar environments.
Abstract
This study aimed to develop a machine learning-based prediction model for the nautical area scattering coefficient (NASC) of the Antarctic silverfish Pleuragramma antarcticum, a key species in the Southern Ocean. Acoustic survey data from the Ross Sea from 2018 to 2023 were integrated with environmental variables, including depth, temperature, salinity, survey period, survey area, and grid location, to construct Random Forest regression models. Separate models were trained on the adults and juveniles. For adults, continuous variables were standardized using z-scores. Meanwhile, juvenile models were standardized using raw values. Model training was performed using MATLAB TreeBagger with grid search optimization. The performance was evaluated by hold-out validation. The adult model achieved high accuracy (R²≈0.76, RMSE≈2.10), with depth, temperature, and salinity identified as the most influential predictors. The juvenile model showed lower explanatory power (R²≈0.38, RMSE≈2.54), often underestimating high NASC values. Adults are more strongly governed by physical conditions, whereas juveniles are influenced by additional biological or ecological factors. Random Forest models can effectively predict adult silverfish NASC using limited environmental inputs, supporting the improved interpretation of acoustic data and ecosystem-based management in polar environments.