Machine Learning Assessment of Freshwater Fish Habitat Under Climate Warming

Abstract

Introduction: The habitats of freshwater fishes are being impacted more frequently due to climate warming in terms of changes in temperature, oxygen, flow, precipitation, and other environmental parameters. The conventional methods of habitat assessment have certain limitations in terms of depicting the interaction among these parameters which is generally nonlinear in nature. This research aims to develop an ML-based method for habitat assessment of freshwater fishes.

Materials & Methods: The environmental and fish habitats’ attributes underwent missing values handling, standardization, and feature selection processes. Four machine learning algorithms, which included Random Forest (RF), Support Vector Machine (SVM), XGBoost, and Artificial Neural Network (ANN), were constructed and evaluated based on their accuracy, precision, recall, F1 score, and ROC-AUC metrics. The important features of the models were determined by performing feature importance analysis. The top-performing algorithm was used to assess the changes in habitats’ suitability under baseline and warming conditions.

Results: The XGBoost model attained the best performance in terms of prediction with the highest accuracy of 87.8%, precision of 87.4%, recall of 87.1%, F1-score of 87.2%, and ROC-AUC of 0.92. The Random Forest model attained 84.6% accuracy, while ANN and SVM models attained 83.7% and 81.9%, respectively. Water temperature and dissolved oxygen were found to be significant predictors of habitat suitability. Climate warming analysis revealed that there was decreased suitability where there was increased thermal stress.

Conclusion: Results show how machine learning can be applied to assess freshwater fish habitats under varying climatic conditions reliably. XGBoost was found to perform better in prediction and captured the nonlinear relationship between the environmental conditions and suitability of habitats. The machine learning approach may be used to identify sensitive habitats as well as help in freshwater fish ecosystem management and monitoring