Premium
Site–Year Characteristics Have a Critical Impact on Crop Sensor Calibrations for Nitrogen Recommendations
Author(s) -
Colaço A.F.,
Bramley R.G.V.
Publication year - 2019
Publication title -
agronomy journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.752
H-Index - 131
eISSN - 1435-0645
pISSN - 0002-1962
DOI - 10.2134/agronj2018.11.0726
Subject(s) - calibration , environmental science , crop , crop simulation model , univariate , yield (engineering) , agricultural engineering , vegetation (pathology) , sowing , nitrogen , remote sensing , statistics , agronomy , multivariate statistics , soil science , mathematics , geography , engineering , chemistry , medicine , materials science , pathology , metallurgy , biology , organic chemistry
Core Ideas Reflectance sensors are often calibrated to predict yield potential and crop nitrogen response. Site–year characteristics affect crop sensor calibrations. Univariate sensor models (e.g. those focused solely on normalized difference vegetation index) derived from multi site–year experimentation can have low accuracy. Soil moisture and crop sensor information should be combined for more holistic calibrations.Sensor‐based approaches to delivering nitrogen (N) fertilizer recommendations often rely on field experimentation to build calibrations between sensor readings and particular crop variables of interest. One important question is the extent to which calibrations derived from specific locations in specific seasons can be confidently extrapolated to other locations in future seasons. In this study, we examined a dataset covering 19 yr of wheat ( Triticum aestivum L.) yield potential and crop N response calibrations to help answer two questions. First, what drives the variability in sensor calibration between different years? Second, what are the consequences of seasonal variability for the accuracy of sensor predictions and N recommendations? Deep soil moisture information, measured between 50 and 90 d after sowing, helped to explain the variability in the R 2 and equation coefficients of yield potential and response index sensor calibrations between years. When data from multiple experimental years were combined to build generalized models, the R 2 was quite low (∼0.25 for all 19 yr of data). Consequently, the benefit of the N rates recommended by the sensor over a conventional application based on historical optimum N rates was insignificant. We suggest that, rather than increasing experimental years indiscriminately, the relevant seasonal covariates should be incorporated into the sensor predictions in order build holistic sensor models. The results also highlight the likelihood that relevant site‐specific information, such as soil moisture status, may also explain between‐site variation in sensor calibration and so promote an enhanced ability to extrapolate calibrations from one location to another.