Sensor Based N Rates Prescription Models for Maize: Sensitivity to Response Index Estimation and Other Parameters.

Several sensor based nitrogen prescription models (SBNPM) have been developed recently. The relative sensitivity to response index and other parameters as components of these models has not been fully explored. Physical or virtual reference was used to estimate response index (RI) and user customized parameters, such us potential yield or nitrogen use efficiency (NUE) were modified (expected local values±20%) to estimate its impact on the estimation of N rate. We used the SBNPM developed in Oklahoma State University (SBNRC). Sensor data were obtained from 14 maize fertilization experiments (N rates ranged 0 to 280 kg N ha^-1) carried-out from 2004 to 2008 in Paraná, (Argentina, 31º50’ S; 60 º31’ W). Response index was estimated based on sensor data (NDVI) from i) physical reference plot, and ii) percentiles of frequency distribution of NDVI on target plots (N0 and N70). Our results showed that RI estimations using virtual references were higher or similar than those using references plots. The best fit was recorded in fertilized target plots (N70) than in unfertilized target plots (N0). There were negligible differences in the fit of these relationship using average NDVI calculated from different percentiles of the target plots, i.e. from P(70) to P(95). Nitrogen rate prescribed using a virtual reference based on P(70), was however, more similar to the reference plots than those prescribed using higher percentiles, which exhibited higher NDVI than the reference plots. In fact, there were not significant differences (P<0.01) between NDVI from references plots and NDVI from P(70). As a consequence, all nitrogen rate prescribed using NDVI from percentile was over-estimated (>17% in P(70) in relation to the N rate estimated from the NDVI of reference plots). Not only the RI, but also potential yield and NUE affect the N rate prescriptions. The sensitivity of the SBRNC to the changes in potential yield was different according to the level of potential yield, i.e. ± 20% of potential yield leads to changes in 47 and -6% in prescribed N rate, respectively. In turn, changes of ± 20% in NUE leads to -17 to 25% in prescribed N rate, irrespective of potential yield. Our results supported previous reports that suggest the usefulness of virtual reference estimated on target plots, which could simplify the use of the SBNRC. Also, we showed the important role of the other user customized parameters of the SBNRC, such us potential yield and NUE, on the estimation of N rate.