Journal
COMPUTERS AND ELECTRONICS IN AGRICULTURE
Volume 117, Issue -, Pages 200-213Publisher
ELSEVIER SCI LTD
DOI: 10.1016/j.compag.2015.08.001
Keywords
Nitrogen status; Water status; N fertilization; Spectral bands; Principal Component Analysis; Partial Least Squares Regression
Funding
- Research Council of Norway (Program: Bionaer)
Ask authors/readers for more resources
The common method of estimating N demand in cereals by spectral measurements may be negatively affected by variation in other crop properties, in particular by crop water status. In this study, we tested whether it is possible to distinguish between N and water status in spring wheat, at the time of split fertilization (plant growth stage - BBCH 32), by means of spectral reflectance data obtained from a sensor with an effective spectral range of 400-950 nm (i.e. without utilizing the distinct short-wave infrared (SWIR) water bands, which requires a more costly sensor). In 2012 and 2013 we ran a spring wheat field experiment in SE Norway. The experiment comprised 36 treatment plots arranged in a split-plot design, with N fertilization on major plots (either 70 or 100 kg N ha(-1) applied at sowing) and water regime on sub-plots (either limited water supply, natural (rain-fed) water supply, or natural water supply + irrigation). Canopy reflectance was measured on all plots at BBCH 32, using a portable field spectroradiometer (tec5 AG, Germany). Immediately afterwards, aboveground wheat biomass was sampled from 0.25 m(2) quadrats on each plot, and analyzed for total N, fresh and dry matter. The spectral data were first preprocessed by logarithm linearization, 1st derivative filtering using the Savitzky-Golay method and mean normalization. Principal Component Analysis (PCA) performed at the treatment level (i.e. not utilizing ground truth measurements) showed that the first component in all datasets (2012, 2013 and combined) was related to the water regime. The second component in the single year datasets (or the third in the combined dataset) was related to N fertilizer. When combining the spectral information and the ground truth data by means of Powered Partial Least Squares Regression (PPLS), we were able to calibrate models (combined dataset) which fitted well with measured nitrogen at ratio of performance to deviation: RPD = 2.52 (R-2 = 0.86) and water concentration at RPD = 2.35 (R-2 = 0.84) in aboveground spring wheat biomass. The multivariate calibration method revealed several distinct spectral regions containing information related to either wheat plant N or water concentration. This method was superior to an index-based approach, with the best model performance of RPD = 2.26 (R-2 = 0.83) and RPD = 1.49 (R-2 = 0.68) for N and water concentration, respectively. (C) 2015 Elsevier B.V. All rights reserved.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available