BO-CNN-BiLSTM deep learning model integrating multisource remote sensing data for improving winter wheat yield...

Area of Science:

• Agricultural Science
• Remote Sensing
• Data Science

Background:

• Accurate winter wheat yield estimation is vital for agricultural policy and food security, especially amidst climate variability.
• Remote sensing and deep learning offer advanced tools for crop monitoring and yield prediction.
• Solar-induced chlorophyll fluorescence (SIF) shows promise for crop photosynthesis monitoring but requires further exploration for yield estimation.

Purpose of the Study:

• To develop and evaluate a deep learning model for accurate winter wheat yield estimation.
• To investigate the effectiveness of integrating SIF data with traditional remote sensing and climate data.
• To explore the capabilities of the developed model in predicting yield at different stages of crop growth.

Main Methods:

• A deep learning model, Bayesian Optimization-Convolutional Neural Network-Bidirectional Long Short-Term Memory (BO-CNN-BiLSTM or BCBL), was developed.
• The model integrated traditional remote sensing variables (TS), solar-induced chlorophyll fluorescence (SIF), and climate data.
• Bayesian Optimization (BOM) was used for hyperparameter tuning to optimize model performance.

Main Results:

• The BCBL model integrating TS, climate, and SIF data achieved high estimation accuracy (R²=0.81, RMSE=616.99 kg/ha, MRE=7.14%).
• The model accurately identified the critical winter wheat yield formation period (early March to early May).
• High yield estimation accuracy was achieved approximately 25 days before harvest, demonstrating model stability and generalization.

Conclusions:

• The BCBL model combined with SIF data provides reliable winter wheat yield estimates.
• This approach offers significant potential for agricultural policymaking and field management.
• The study highlights the value of SIF data in enhancing crop yield prediction models.