Comparison of three algorithms for estimating crop model parameters based on multi-source data: A case study using...

Area of Science:

• Agricultural Science
• Computational Biology
• Environmental Science

Background:

• Accurate crop phenological stage prediction is vital for effective agricultural management and understanding crop responses to environmental shifts.
• Soybean phenology modeling requires precise calibration of cultivar-specific parameters (CSPs) for reliable simulation.

Purpose of the Study:

• To compare the performance of Modified Non-dominated Sorting Genetic Algorithm (MNSGA-II), Generalized Likelihood Uncertainty Estimation (GLUE), and Differential Evolution (DE) in calibrating CSPs for the CROPGRO-Soybean phenological model.
• To evaluate the accuracy and stability of different algorithms in simulating soybean phenology using multi-source datasets.

Main Methods:

• Calibrated CROPGRO-Soybean phenological model parameters using MNSGA-II, GLUE, and DE algorithms.
• Utilized multi-site, multi-year, and multi-cultivar soybean datasets for calibration.
• Validated the calibrated model using independent experimental data and evaluated performance using RMSE, MAE, and R2 metrics.

Main Results:

• MNSGA-II, GLUE, and DE showed comparable simulation accuracy, with RMSEs of 4.28, 4.76, and 5.17 days, respectively.
• MNSGA-II demonstrated a slight advantage in calibration effectiveness.
• GLUE exhibited the highest stability across repeated calibration runs.

Conclusions:

• MNSGA-II is a suitable algorithm for crop model parameter estimation.
• The choice of algorithm for calibrating crop model parameters should align with specific project requirements.
• Results offer guidance for selecting appropriate algorithms for crop model parameter estimation.