Research

Publications
Title: Remote sensing modeling of environmental influences on lake fish resources by machine learning: A practice in the largest freshwater lake of China
First author: Chen, Tan; Song, Chunqiao; Fan, Chenyu; Gao, Xin; Liu, Kai; Li, Zhen; Cheng, Jian; Zhan, Pengfei
Journal: FRONTIERS IN ENVIRONMENTAL SCIENCE
Years: 2022
Volume / issue: /
DOI: 10.3389/fenvs.2022.944319
Abstract: Climate change and human interference pose a significant threat to fishery habitats and fish biodiversity, leading to changes in fishery resources. However, the impact of environmental change on lake fishery resources has been largely blurred in assessments due to the complicated variables of the lake environment. Here, taking the largest freshwater lake (Poyang Lake) in China as a study case, we first proposed a conceptual model and simulated the effect of environmental variables on fish catches based on remote sensing techniques and machine learning algorithms. We found that the hydrometeorological conditions of fishery habitats are critical controlling factors affecting the fish catches in Poyang Lake through a long time series of simulations. Among the involved hydrometeorological variables, the temperature, precipitation, and water level are strongly correlated with the fish catches in the simulation experiments. Furthermore, we tested other experiments and found that the integration with water quality variables (correlation coefficient (R) increased by 11%, and root mean square error (RMSE) decreased by 2,600 tons) and water ecological variables (R increased by 17%, and RMSE decreased by 3,200 tons) can further improve the accuracy of fish catch simulation. The results also showed that fish catches of aquatic species in Poyang Lake are more susceptible to water ecological variables than water quality refers to the model performance improvements by different input variable selections. In addition, a multi-dimension variable combination involving hydrometeorological conditions, water quality, and water ecological variables derived from remote sensing can maximally optimize the model performance of fish catch simulation (R increased by 21%, and RMSE decreased by 4,300 tons). The approach developed in this study can save the labor and financial costs for large-area investigation and the assessment of lake fishery resources compared to conventional methods. It is expected to demonstrate an efficient way for public authorities, stakeholders, and decision-makers to guide fishery conservation and management strategies.