Smart Agricultural Technology (Aug 2023)
Measuring soybean iron deficiency chlorosis progression and yield prediction with unmanned aerial vehicle
Abstract
Iron deficiency chlorosis (IDC), a symptom of reduction in chlorophyll and stunted growth, causes a great yield loss in soybean every year in the Midwest, USA and the most efficient method to manage IDC is to plant tolerant cultivars. The assessment of cultivars' tolerance is traditionally performed by visually rating the IDC symptoms based on leaves discoloration twice during the growing season. However, the visual rating method is time-consuming, subjective, not suitable at large scales, labor-intensive, and unaffordable for frequent observation. Therefore, in this study, we used an unmanned aerial vehicle (UAV) as a tool to monitor the soybean cultivars more frequently and more efficiently through image processing approach of the whole field. Images were taken with a DJI Phantom 4 and orthomosaicked in Agisoft Photoscan. A 40-cultivar soybean experimental plots (3000 m2; Image 1) at 5 locations in North Dakota, USA (Amenia, Colfax, Leonard (2), and Hunter) for 2 years (2016 and 2017) were used in the study. The orthomosaicked images were processed in MATLAB to calculate the dark green color index (DGCI), which is a good indicator of chlorophyll in soybean leaves. The grayscale DGCI images were then processed in ArcGIS to extract the average DGCI and canopy size (CS) for each plot for each flight. The area under the curve (AUC) was calculated for DGCI, CS, and CS × DGCI product (CDP) to aggregate the values of all flights within each year. The correlation of AUC of CDP and the yield was more consistent among both years and was the better predictor of yield (R2=0.74 and R2=0.79). The latest growth stage (more representative of yield) values of both years were combined to build models for yield prediction and the CDP produced the lowest error (11.72%). Future studies should look into IDC progress measurement involving more cultivars, geographical locations, frequent imaging, as well as methods applied to regular soybean production sites to evaluate various image-based parameters and their interaction for yield predictions.
