Simultaneous retrieval of sugarcane variables from Sentinel-2 data using Bayesian regularized neural network

Area of Science:

• Agricultural remote sensing
• Biophysical and biochemical vegetation analysis
• Machine learning applications in agriculture

Background:

• Precision agriculture relies on accurate quantification of vegetation variables.
• Leaf Area Index (LAI), leaf sheath moisture (LSM), leaf chlorophyll content (LCC), and leaf nitrogen concentration (LNC) are key sugarcane indicators.
• Traditional methods for retrieving these variables can be time-consuming and less accurate.

Purpose of the Study:

• To develop and evaluate a Bayesian Regularized Artificial Neural Network (BRANN) model for simultaneous retrieval of multiple sugarcane vegetation variables.
• To assess the performance of BRANN compared to conventional Artificial Neural Networks (ANNs).
• To map vegetation variables across sugarcane fields using Sentinel-2 imagery.

Main Methods:

• Utilizing Sentinel-2 spectral data as input for the ANN models.
• Employing Bayesian Regularized Artificial Neural Networks (BRANN) to mitigate overfitting and enhance generalizability.
• Implementing a simultaneous retrieval approach for LAI, LSM, LCC, and LNC.
• Comparing BRANN performance against ANNs trained with the Levenberg-Marquardt algorithm.

Main Results:

• Achieved high accuracy with RMSE values of 0.48 (m²/m²) for LAI, 2.36 (% wb) for LSM, 5.85 (microg/cm²) for LCC, and 0.23 (%) for LNC.
• Demonstrated that simultaneous retrievals outperformed individual retrievals.
• Confirmed the superiority of BRANN over conventional ANNs through statistical analysis.
• Generated maps showing reasonable spatial and temporal variations of the retrieved variables.

Conclusions:

• BRANN enables accurate and simultaneous quantification of multiple sugarcane vegetation variables from Sentinel-2 data.
• The simultaneous retrieval approach with BRANN is more efficient and accurate than individual retrievals and conventional ANNs.
• This methodology holds significant potential for advancing precision agriculture through improved vegetation monitoring.