Maize yield estimation at different growth stage using weather variables by LASSO, elastic net and stepwise multiple...

Area of Science:

• Agricultural Science
• Data Science
• Statistical Modeling

Background:

• Accurate maize yield estimation is vital for food security and agricultural planning.
• Traditional methods often lack precision, necessitating advanced statistical approaches.
• Understanding the impact of weather parameters on crop development is key.

Purpose of the Study:

• To evaluate four statistical modeling approaches for maize yield estimation at various growth stages.
• To identify the most accurate model for predicting maize yield based on weather data.
• To determine the most influential weather parameters affecting maize yield.

Main Methods:

• Utilized Least Absolute Shrinkage and Selection Operator (LASSO), Elastic Net, Stepwise Multiple Linear Regression (SMLR), and PCA-SMLR.
• Developed models using historical maize yield data (1984-2021) and daily weather parameters.
• Validated models for the 2020 and 2021 kharif seasons at vegetative, flowering, and grain-filling stages.

Main Results:

• The Elastic Net model demonstrated the lowest Root Mean Square Error (RMSE) and normalized RMSE (nRMSE), indicating superior performance.
• Percentage deviations between estimated and observed yields ranged from 4.8-29.1% across growth stages.
• Temperature and bright sunshine hours were identified as the most significant predictors of maize yield.

Conclusions:

• The Elastic Net model is the most reliable for maize yield estimation at different growth stages.
• LASSO and SMLR also provide valuable maize yield predictions.
• Weather parameters, particularly temperature and sunshine, are critical drivers of maize yield.