Accuracy of NDVI-derived corn yield predictions is impacted by time of sensing

Abstract

Normalized difference vegetation index (NDVI) data derived from cameras mounted on unmanned aerial vehicles (UAVs) or from tractor-mounted sensors have the potential to aid in quick and accurate decision making for fertility management of corn (Zea mays L.). However, timing of sensing can impact the accuracy of yield predictions and hence nutrient need assessment. In this study the accuracy of yield predictions derived from NDVI data for corn silage and grain was evaluated using (a) an active sensor (GreenSeeker; handheld unit), and (b) a passive sensor (UAV-mounted camera). Two types of replicated trials were conducted in central New York: (a) N rate studies with N applied at V6 at rates ranging from 0 kg N ha−1 to 336 kg N ha−1 (2016, 2017, and 2018); and (b) timing of N application studies, with N applied at V4, V6, V8 or V10 (2017 and 2018). For each of the trials conducted in 2017 and 2018, the crop was sensed and NDVI was determined weekly from V4 through V12. An additional flight was done with the UAV-mounted camera at R2. In addition, the 2016 N rate study and timing of sidedress study in 2018 were sensed at V10 (single day) from 7 am through 5 pm on 2-hour intervals. Results showed that crop growth stage and timing of N application affected the accuracy of yield predictions. Scans within 2 weeks of sidedressing did not result in accurate yield predictions, independent of sensor type. Best yield estimates were obtained with a UAV-mounted camera late in the season (R2 growth stage). For in-season N management, a low-rate (<30 kg N ha−1) starter N application is recommended. Sensing must be delayed until just after initial N deficiencies become visible. Sidedressing at V10 still resulted in a yield increase beyond yields with starter N only, but if such late applications are done, accurate estimates of yield based on NDVI can only be obtained with sensing a second time after V12. Due to the reaction of corn plants to water stress, timing of sensing within a day showed the best yield predictions between 9 and 11 am and at 5 pm, independent of sensor used. For most accurate yield predictions, we recommend scanning between V12 and R2, avoiding early morning (dew) and mid-day hours (1–3 pm) if water stress in the crop is evident.