Estimating Productivity Measures in Guayule Using UAS Imagery and Sentinel-2 Satellite Data

Guayule (Parthenium argentatum Gray) is a perennial desert shrub currently under investigation as a viable commercial alternative to the Para rubber tree (Hevea brasiliensis), the traditional source of natural rubber. Previous studies on guayule have shown a close association between morphological traits or biomass and rubber content. We collected multispectral and RGB-derived Structure-from-motion (SfM) data using an unmanned aircraft system (UAS; drone) to determine if incorporating both high-resolution normalized difference vegetation index (NDVI; an indicator of plant health) and canopy height (CH) information could support model predictions of crop productivity. Ground-truth resource allocation in guayule was measured at four elevations (i.e., tiers) along the crop's vertical profile using both traditional biomass measurement techniques and a novel volumetric measurement technique. Multiple linear regression models estimating fresh weight (FW), dry weight (DW), fresh volume (FV), fresh-weight-density (FWD), and dry-weight-density (DWD) were developed and their performance compared. Of the crop productivity measures considered, a model predicting FWD (i.e., the fresh weight of plant material adjusted by its freshly harvested volume) and incorporating NDVI, CH, NDVI:CH interaction, and tier parameters reported the lowest mean absolute percentage error (MAPE) between field measurements and predictions, ranging from 9 to 13%. A reduced FWD model incorporating only NDVI and tier parameters was developed to explore the scalability of model predictions to medium spatial resolutions with Sentinel-2 satellite data. Across all UAS surveys and corresponding satellite imagery compared, MAPE between FWD model predictions for UAS and satellite data were below 3% irrespective of soil pixel influence.