Uncertainty and spatial analysis in wheat yield prediction based on robust inclusive multiple models.

Area of Science:

• Agricultural Science
• Data Science
• Environmental Science

Background:

• Accurate wheat yield prediction is crucial for agricultural management and food security.
• Existing modeling approaches face challenges in complexity and robustness.
• Decision-makers require reliable tools for supply chain management.

Purpose of the Study:

• To develop and evaluate an advanced ensemble model for predicting wheat yield.
• To optimize multilayer perceptron (MLP) model parameters using various algorithms.
• To introduce a novel hybrid gamma test for input selection and uncertainty analysis.

Main Methods:

• Developed an inclusive multiple MLP model (IMM) integrating optimized MLP models.
• Employed optimization algorithms: Particle Swarm Optimization (PSO), Honey Badger Algorithms (HBA), Sine-Cosine Algorithms (SCA), and Shark Algorithms (SA).
• Utilized a hybrid gamma test (HBA-GT) for input selection and Generalized Likelihood Uncertainty Estimation (GLUE) for uncertainty analysis.

Main Results:

• The IMM model significantly reduced Mean Absolute Error (MAE) compared to individual MLP models (up to 61% reduction).
• Key meteorological parameters identified: mean air temperature, wind speed, relative humidity, evapotranspiration, and precipitation.
• Uncertainty analysis showed lower uncertainty from input data compared to model parameters, with GLUE confirming prediction stability.

Conclusions:

• The developed IMM framework offers a robust and accurate method for wheat yield prediction.
• The study highlights the importance of meteorological factors and ensemble modeling for reliable agricultural forecasting.
• The approach provides valuable insights for rainfed agriculture and food policy analysis.