Evaluating Trophic State of Lakes in Cold and Arid Zones: An Integrated Approach Based on Neural Networks and Genetic Algorithms

To investigate the nutritional status and spatiotemporal variations of Ulansuhai Lake—a typical lake in a cold and arid region—monitoring data from January and June 2019 were utilized to develop three evaluation models, a BP neural network model, a genetic algorithm-based Shepard similarity model, and a logistic curve model, for simulating and analyzing the eutrophication level of the lake. The results revealed that according to the BP neural network model, the lake was predominantly severely eutrophic during both ice-covered and ice-free periods, accounting for 80% and 55% of the nutritional state evaluations, respectively; the Shepard similarity model indicated that the lake was mainly moderately eutrophic during the ice-covered period and eutrophic during the ice-free period, representing 55% and 70% of the assessments, respectively; and the logistic curve model suggested that the lake was primarily moderately eutrophic in both periods, comprising 55% and 75% of the evaluations for the ice-covered and ice-free phases, respectively. Furthermore, during the ice-free period, all three models consistently demonstrated that eutrophication was more severe in the central–northern part of the lake and decreased gradiently from north to south, whereas no significant spatial trend in eutrophication was observed during the ice-covered period. These findings underscore the importance of adopting season-specific and spatially targeted eutrophication control strategies to improve water quality management and guide policy-making in cold and arid lacustrine environments.