Predicting alfalfa leaf area index by non-linear models and deep learning models.

Area of Science:

• Agricultural Science
• Plant Physiology
• Machine Learning

Background:

• Leaf Area Index (LAI) is vital for alfalfa growth and yield prediction.
• Traditional non-linear models struggle to integrate environmental factors for accurate LAI prediction.
• Environmental variables like temperature and soil moisture significantly influence alfalfa LAI.

Purpose of the Study:

• To evaluate classical non-linear models for alfalfa LAI prediction.
• To develop and assess a novel deep learning model for enhanced alfalfa LAI prediction.
• To overcome the limitations of non-linear models in incorporating environmental data.

Main Methods:

• Developed Logistic, Gompertz, and Richards non-linear models based on growth days.
• Proposed a time series prediction model using mutation point detection and an encoder-attention-decoder BiLSTM network (TMEAD-BiLSTM).
• Integrated environmental factors (temperature, soil moisture) into the TMEAD-BiLSTM model.

Main Results:

• The TMEAD-BiLSTM model achieved superior prediction accuracy (R² > 0.99).
• Classical non-linear models showed lower accuracy (R² > 0.78).
• The TMEAD-BiLSTM model effectively integrated environmental factors for improved predictions.

Conclusions:

• The TMEAD-BiLSTM model offers rapid and accurate alfalfa LAI predictions.
• This deep learning approach overcomes limitations of traditional models in integrating environmental data.
• Findings provide valuable insights for alfalfa growth monitoring and field management practices.