Biomass Assessment of Agricultural Crops Using Multi-temporal Dual-Polarimetric TerraSAR-X Data

Area of Science:

• Agricultural remote sensing
• Biomass estimation
• Synthetic Aperture Radar (SAR) applications

Background:

• Accurate crop biomass assessment is crucial for agricultural management and yield prediction.
• Remote sensing techniques, particularly SAR, offer potential for non-invasive biomass monitoring.
• Dual-polarimetric SAR data provides rich information for characterizing vegetation structure.

Purpose of the Study:

• To investigate the capability of multi-temporal dual-polarimetric TerraSAR-X data for estimating winter wheat, barley, and canola biomass.
• To compare the performance of multiple stepwise regression, the Water Cloud Model (WCM), and Random Forest (RF) machine learning for biomass retrieval.
• To evaluate the influence of different SAR polarization combinations on biomass estimation accuracy.

Main Methods:

• Extraction of radar backscattering coefficients (sigma nought) from dual-polarimetric HH and VV TerraSAR-X data.
• Application of multiple stepwise regression using combined polarizations (e.g., HH/VV, HH + VV, HH × VV).
• Utilized the semi-empirical Water Cloud Model (WCM) and Random Forest (RF) machine learning algorithms.
• Employed a split sampling approach (70% training, 30% testing) for model validation.

Main Results:

• Multiple stepwise regression achieved high accuracy (R² > 0.7) for dry biomass estimation without soil moisture data.
• Random Forest (RF) significantly outperformed WCM for fresh biomass estimation (R² > 0.68 vs. R² < 0.35).
• For dry biomass estimation, RF and WCM yielded comparable results.

Conclusions:

• Dual-polarimetric SAR data is a viable tool for estimating agricultural crop biomass.
• The Random Forest machine learning approach demonstrates superior performance for fresh biomass estimation compared to WCM.
• Stepwise regression provides a robust method for dry biomass retrieval, independent of soil moisture information.