Improving ecological modeling: Integrating CNOP-P and adjoint assimilation in a coupled ecological model

• The CNOP-P and adjoint assimilation methods are integrated for the first time. • Integrated method pinpoints sensitive parameters, reducing computational needs. • The integrated method outperforms other sensitivity analysis methods. • This method markedly improves the model's predictive skill. Ecological modeling is an important methodology for studying the spatio-temporal evolution of marine ecosystem. Given the significant role of model parameters as a major source of uncertainty in ecological models, we propose a novel approach by combining the Conditional Nonlinear Optimal Perturbation related to Parameters (CNOP-P) method with the adjoint assimilation method to enhance predictive accuracy. CNOP-P denotes the parameter perturbation that leads to the greatest deviation of the model's development from the reference state. In comparison to other sensitivity analysis methods, this combined approach proves to be more efficient. Considering the nonlinearity of the model structure, the maximum development of the model does not consistently align with the extreme parameter values within the confidence interval. Minor parameter errors can lead to substantial model development, significantly impacting the precision of ecological models. Notably, traditional sensitivity analysis methods such as one-at-a-time (OAT) sensitivity analysis and global sensitivity analysis (GSA) methods fail to capture this characteristic. On the other hand, the GSA methods incur substantial computational costs and tends to overestimate the sensitivity of the most sensitive parameters while underestimating the sensitivity of less sensitive parameters. The combined approach of CNOP-P and adjoint assimilation enables the assimilation of satellite data and the simultaneous optimization of model parameters alongside the CNOP-P calculations. This integration substantially improves both efficiency and precision of the ecological model, thereby improving predictive skill.