Simplifying the calibration of ecological models by using the parameter estimation tool (PEST): The Curonian Lagoon case

In this study, we implemented an automated calibration procedure for an ecological model of the Curonian Lagoon, supported by a comprehensive two-year field observation dataset. Data from the second-year were used for model calibration, while first-year observations served for the validation of the model's performance in simulating nutrient dynamics. Calibration is essential for improving the accuracy and reliability of process-based ecological models. However, subjective and time-consuming manual (trial-and-error) calibration methods cannot ensure optimal parameter match.

To address this, we automated the calibration of a newly developed ecological model to improve the simulation of nutrient dynamics as ammonia, nitrate, and phosphate in the estuarine system (Curonian Lagoon). Calibration was carried out using Parameter Estimation (PEST) and PEST++ tools, focusing on three aforementioned limiting nutrient forms. We applied the method of Morris for global sensitivity analysis to determine the key parameters influencing model behavior. As biogeochemical models are highly nonlinear and multimodal, global methods are often assumed to provide a better fit. However, we challenged this assumption by initiating the inverse problem at different locations in the parameter space using a robust variant of a gradient-based method, which ultimately resulted in a better fit than global methods.

We tested four different optimization algorithms available in the PEST and PEST++ suites. The results demonstrated that PEST significantly improved model calibration performance followed the nutrient dynamics more effectively than more complex biogeochemical models for the Curonian Lagoon, and outperformed manual calibration methods. Furthermore, we employed an ensemble-based method within the PEST++ suite for parameter estimation and uncertainty quantification, significantly reducing the computational burden of these analyses.