Research on enhanced calibration of biological models using hybrid particle swarm optimization and gradient descent method

As environmental changes increasingly impact ecosystems, accurately predicting ecological shifts has become a crucial area of research in environmental science. Biological models are essential for predicting and analyzing these changes, and their effectiveness directly influences the success of ecological prediction and resource management efforts. However, many models face significant challenges, such as parameter sensitivity and susceptibility to local optima, which severely limit their predictive capabilities and broader applications. To tackle these challenges, this paper combines the Particle Swarm Optimization (PSO) algorithm with traditional optimization methods for effective calibration of biological models. We employ PSO to conduct a comprehensive global parameter search, paired with an improved gradient descent method to further refine the parameters and thresholds identified by PSO. This results in the development of an enhanced PSO-Gradient Descent hybrid model, specifically designed for reliably predicting ecological changes. Comparative experiments using environmental variable and species distribution data from a rainforest, collected from 2012 to 2023, demonstrate that the improved intelligent heuristic optimization model significantly enhances prediction accuracy, reducing the relative error rate from 5.12% to 2.45%. This improvement has important implications for better ecological management and conservation strategies in the future.