Abstract:
To explore the quantitative relationship between the natural reproduction of typical producing semi-buoyant eggs fish in the upper reaches of the Yangtze River and environmental characteristics, eight machine learning models, including decision tree, random forest, extra trees, gradient boosting classifier, Adaboost, XGBoost, LightGBM, and Catboost, were introduced to construct spawning prediction models for the target fish species of Coreius heterodon, Rhinogobio ventralis, and Rhinogobio cylindricus. The performance of each model was compared, and interpretable machine learning techniques were integrated to investigate the correlation between the target fish species and environmental characteristic indicators.The results showed that:for the spawning prediction of the three target fish species, seven ensemble models achieved high prediction accuracy, with the Catboost model performing the best overall. The most critical environmental factors influencing the spawning of three target fish species were daily average water temperature and daily average discharge, while short-term variations in hydrology, water temperature, and water environment had little impact in the evaluation of each model. Within a certain range of conditions, the spawning activity rate of Coreius heterodon and Rhinogobio cylindricus was positively correlated with both water temperature and discharge, indicating a preference for spawning under high-temperature and high-discharge conditions. In contrast, the spawning activity rate of Rhinogobio ventralis exhibited a unimodal relationship with water temperature and a negative correlation with discharge, and its optimal water temperature and discharge ranges for spawning were significantly lower than those of the other two species. Water temperature and discharge exerted a strong interactive effect on the spawning of all three target fish species, and suitable and mutually synergistic hydrological and water temperature conditions were identified as the key habitat characteristics promoting their spawning activities. Keywords: producing semi-buoyant eggs fish; ecological flow; ecological operation; ecology-hydrology response; machine learning model; upper reaches of the Yangtze River 〖FL