Assessment of Sorghum Perennial Grass Root Biomass Using a Prototype 1.9 GHz Ground Penetrating Radar

Assessing and managing soil carbon storage is critical for addressing numerous issues in agriculture and carbon sequestration. Plant genotyping to enhance root biomass and depth provides for new opportunities for restoring degraded soils and improving soil quality. Specifically, accurate assessment of below-ground conditions are required to assess architecture, biomass quantity and agricultural system capacity. Ground penetrating radar (GPR) potentially provides such assessment capabilities, although numerous preprocessing and analysis issues must be overcome. Therefore, the objective of this research was to develop new GPR information-extraction solutions for quantitative prediction of biomass from rhizomes and fibrous roots. Sorghum plants were cultivated in troughs, harvested and measured. GPR data were collected using a multi-channel GPR system to assess biomass at three depth ranges. We examined a range of GPR preprocessing parameters to evaluate the sensitivity of surface removal, gain correction, band-pass filtering and migration on our ability to characterize root biomass variation. We utilized a frequency analysis approach to determine the relationship between extracted GPR information and measured biomass for all plots and depth ranges. Our preliminary results indicate statistically significant predictive capabilities using standard preprocessing approaches based upon human interpretation, although statistical results were not consistent with depth range. Sensitivity analysis revealed that band-pass filtering and surface removal significantly affect the magnitude of correlations found. Statistical results were highly variable, such that automation is required to obtain an optimal amount of explained variation. Our optimized results reveal that we can account for more than 90 percent of root biomass conditions for rhizomes and fibrous roots at all depths given our new optimized preprocessing and analysis procedures. Our results illustrate promising root detection and biomass prediction capabilities for agricultural applications.