Evaluating Mesquite Distribution Using Unpiloted Aerial Vehicles and Satellite Imagery

Encroachment of woody plant species on rangelands is of critical concern for rangeland health and function. Recent advances in unpiloted aerial vehicles (UAVs) technology have opened new opportunities for natural resource personnel to quantify features within the landscape. Our goal was to combine information derived from UAV and Sentinel-2 imagery to quantify the presence and spatial distribution of mesquite (Prosopis glandulosa). The specific objectives were to 1) evaluate the accuracy of UAVs to estimate mesquite height and 2) classify mesquite using a combination of height metrics and spectral information by combining UAV and Sentinel-2 remote sensing data. We conducted our study in three different ecoregions in Texas. We collected on-site tree height field measurements and created two UAV-derived height outputs at 50 m and 100 m above ground level (AGL). We then used a random forest classification with data derived from the UAV to inform Sentinel-2 imagery of mesquite locations to assess its presence at the landscape level. Linear regression analysis showed that 50-m AGL mesquite height estimates from UAV explained 95% of the variability. Variability explained from 100 m AGL height estimates was slightly lower at 92%. Accuracy assessments from using UAV training data to predict mesquite presence using Sentinel-2 imagery yielded overall accuracy of > 80% for all sites, user accuracies up to 86%, and producer accuracies up to 92%. UAV-derived heights of mesquite were reliable. UAV imagery can be used as training data for Sentinel-2 satellite imagery to assess mesquite presence on Texas rangelands. Being able to combine the high spatial resolution of the UAV imagery with the high temporal resolution of the Sentinel-2 satellite imagery could improve our ability to use UAVs to monitor rangelands.