Validation of UAV Estimates of Canopy Height and Boll Count with Manual Measurements for Two Cotton Cultivars

Site-specific validation of crop phenological estimates from Unmanned Aerial Vehicle (UAV) collected data is necessary before using that data in large applications and for crop modeling. In this study, both UAV-based and manual measurements were made in a cotton irrigation experiment at Halfway, TX in 2019. The data was collected for two cultivars (FM2011 and FM2484) from four replicated plots of six irrigation treatments. In each plot, five plants were manually monitored, and measurements of canopy height and boll count were taken at bi-weekly interval. Two UAVs, Phantom 4 Pro (with an RGB sensor) and DJI Matrice M200 (with a multispectral sensor) were used for collecting aerial data on every other visit when manual measurements were made. The UAV-collected data were processed to estimate canopy height and open boll count, and then compared with manual measurements. Canopy Height Models were developed from UAV-collected data using Agisoft Metashape software and the canopy height for each replication/plot was calculated by dividing the area covered by 5 monitored plants into three segments. The highest canopy height for each segment was determined and the average of those three measurements was considered as the canopy height for that replication/plot. For estimation of open boll count, number of open bolls in the area covered by 5 monitored plants were counted from a binary cotton boll classified image and the total count was divided by 5 to calculate the open boll count per plant. The UAV-derived estimates matched well with manual measurements as indicated by an average error of 4.4% and 9.5% for cultivars FM2011 and FM2484, respectively for canopy height, and an average error of 3.2% and -11.3% for cultivars FM2011 and FM2484, respectively for open boll count estimation. Efforts to estimate cotton lint yield from the open boll count and area data are in progress.