Proximal sensor-based algorithm for variable rate nitrogen application in maize in northeast U.S.A.

Ground-based proximal sensing can aid in within-field, mid-season, and on-the-go nitrogen (N) management of maize (Zea mays L.) but algorithms require regional calibration. Our objective was to develop algorithms for predicting N needs of maize in the northeastern US using the Normalized Difference Vegetation Index (NDVI). We established six replicated trials at research farms with up to six N rates (0, 56, 112, 168, 224, and 336 kg N ha−1) applied pre-plant, and eleven trials with five N rates (0, 56, 112, 168, 224 and 336 kg N ha−1 sidedressed at V7) and an N-rich treatment (168 kg ha−1 applied pre-plant and 168 kg N ha−1 sidedressed at V7) conducted on commercial farms. Sensor and yield data from the zero-N and N-rich plots of the on-farm trials were added to the data from the six pre-plant trials to determine a (1) yield prediction model, developed using NDVI obtained at V7, and (2) a Response Index (RIharvest), calculated as the yield in the highest N fertilized treatment divided by the yield in the low-N treatment. In addition, a virtual RI derived from NDVI values taken at V7 (RINDVI) was calculated. The most economic rate of sidedress N (MERN) was determined for the on-farm trials. The regression between RIalgo and virtual RINDVI resulted in two equations, as the correlation was heavily impacted by weather in the growing season; for years with normal precipitation, the response to N post-sensing was greater (RIalgo = 3.8 ∗ RINDVI − 2.38) than in years with severe drought (RIalgo = 1.45 ∗ RINDVI − 0.3). The N prescriptions (yield predictions and RIs combined) correlated well with the MERN (r2 = 0.82) across sites. We conclude that crop sensing is a promising technology for on-the-go N applications for maize in the Northeast but further studies are needed to determine RIs under a range of weather conditions and soils.