Using field-derived hyperspectral reflectance measurement to identify the essential wavelengths for predicting nitrogen uptake of rice at panicle initiation

B.W. Dunn, Remy Dehaan, Leigh Schmidtke, T.S. Dunn, R. Meder

Research output: Contribution to journalArticle

3 Citations (Scopus)

Abstract

Rice growers are reluctant to physically sample their crops, with many relying on visual estimates of crop nitrogen status to determine mid-season nitrogen topdressing requirements. There is increased interest in using remote sensing to determine how accurately mid-season rice crop nitrogen status (nitrogen concentration and content in shoots) can be predicted. Field-derived hyperspectral reflectance measurements were collected over three seasons to evaluate the optimal wavelengths for the prediction of in-crop nitrogen at panicle initiation. Rice plots were planted and differing nitrogen rates applied to establish a range of crop nitrogen uptake levels at the panicle initiation stage of crop development. Hyperspectral canopy reflectance (350'2500 nm) data were collected from each plot at the panicle initiation growth stage and plant samples were collected at each scanning location. Calibration models were developed for plant nitrogen concentration, dry matter and nitrogen uptake using partial least squares regression and the influential wavelengths used in the nitrogen uptake prediction model identified. When the nitrogen uptake model, which included samples collected in three crop seasons, was used to predict nitrogen uptake in an independent set of samples, r2 was 0.84 and the root-mean-square error of prediction (RMSEP) was 16.5 kg N ha'1. The important wavelengths utilised for predicting nitrogen uptake were determined from the regression coefficients and a new calibration created by multiple linear regression (MLR) using only four wavelengths (738 nm, 1362 nm, 1835 nm and 1859 nm). The MLR calibration using the four wavelengths had a prediction accuracy of r2 = 0.82 and RMSEP = 18.4 kg N ha'1, which shows potential for commercial applications. These findings will encourage the development of a remote sensing method for predicting nitrogen uptake for rice crops using the identified wavelengths in an inexpensive multiband instrument.
Original languageEnglish
Pages (from-to)473-483
Number of pages11
JournalJournal of Near Infrared Spectroscopy
Volume24
Issue number5
DOIs
Publication statusPublished - Oct 2016

Fingerprint

Reflectometers
Nitrogen
Wavelength
Crops
Calibration
Linear regression
Mean square error
Remote sensing

Cite this

@article{32d00591f0d34f11ac7bb74aef358162,
title = "Using field-derived hyperspectral reflectance measurement to identify the essential wavelengths for predicting nitrogen uptake of rice at panicle initiation",
abstract = "Rice growers are reluctant to physically sample their crops, with many relying on visual estimates of crop nitrogen status to determine mid-season nitrogen topdressing requirements. There is increased interest in using remote sensing to determine how accurately mid-season rice crop nitrogen status (nitrogen concentration and content in shoots) can be predicted. Field-derived hyperspectral reflectance measurements were collected over three seasons to evaluate the optimal wavelengths for the prediction of in-crop nitrogen at panicle initiation. Rice plots were planted and differing nitrogen rates applied to establish a range of crop nitrogen uptake levels at the panicle initiation stage of crop development. Hyperspectral canopy reflectance (350'2500 nm) data were collected from each plot at the panicle initiation growth stage and plant samples were collected at each scanning location. Calibration models were developed for plant nitrogen concentration, dry matter and nitrogen uptake using partial least squares regression and the influential wavelengths used in the nitrogen uptake prediction model identified. When the nitrogen uptake model, which included samples collected in three crop seasons, was used to predict nitrogen uptake in an independent set of samples, r2 was 0.84 and the root-mean-square error of prediction (RMSEP) was 16.5 kg N ha'1. The important wavelengths utilised for predicting nitrogen uptake were determined from the regression coefficients and a new calibration created by multiple linear regression (MLR) using only four wavelengths (738 nm, 1362 nm, 1835 nm and 1859 nm). The MLR calibration using the four wavelengths had a prediction accuracy of r2 = 0.82 and RMSEP = 18.4 kg N ha'1, which shows potential for commercial applications. These findings will encourage the development of a remote sensing method for predicting nitrogen uptake for rice crops using the identified wavelengths in an inexpensive multiband instrument.",
keywords = "Rice, nitrogen, panicle initiation, remote sensing",
author = "B.W. Dunn and Remy Dehaan and Leigh Schmidtke and T.S. Dunn and R. Meder",
note = "Imported on 12 Apr 2017 - DigiTool details were: month (773h) = October; Journal title (773t) = Journal of Near Infrared Spectroscopy. ISSNs: 0967-0335;",
year = "2016",
month = "10",
doi = "10.1255/jnirs.1246",
language = "English",
volume = "24",
pages = "473--483",
journal = "Journal of Near Infrared Spectroscopy",
issn = "0967-0335",
publisher = "I M Publications",
number = "5",

}

TY - JOUR

T1 - Using field-derived hyperspectral reflectance measurement to identify the essential wavelengths for predicting nitrogen uptake of rice at panicle initiation

AU - Dunn, B.W.

AU - Dehaan, Remy

AU - Schmidtke, Leigh

AU - Dunn, T.S.

AU - Meder, R.

N1 - Imported on 12 Apr 2017 - DigiTool details were: month (773h) = October; Journal title (773t) = Journal of Near Infrared Spectroscopy. ISSNs: 0967-0335;

PY - 2016/10

Y1 - 2016/10

N2 - Rice growers are reluctant to physically sample their crops, with many relying on visual estimates of crop nitrogen status to determine mid-season nitrogen topdressing requirements. There is increased interest in using remote sensing to determine how accurately mid-season rice crop nitrogen status (nitrogen concentration and content in shoots) can be predicted. Field-derived hyperspectral reflectance measurements were collected over three seasons to evaluate the optimal wavelengths for the prediction of in-crop nitrogen at panicle initiation. Rice plots were planted and differing nitrogen rates applied to establish a range of crop nitrogen uptake levels at the panicle initiation stage of crop development. Hyperspectral canopy reflectance (350'2500 nm) data were collected from each plot at the panicle initiation growth stage and plant samples were collected at each scanning location. Calibration models were developed for plant nitrogen concentration, dry matter and nitrogen uptake using partial least squares regression and the influential wavelengths used in the nitrogen uptake prediction model identified. When the nitrogen uptake model, which included samples collected in three crop seasons, was used to predict nitrogen uptake in an independent set of samples, r2 was 0.84 and the root-mean-square error of prediction (RMSEP) was 16.5 kg N ha'1. The important wavelengths utilised for predicting nitrogen uptake were determined from the regression coefficients and a new calibration created by multiple linear regression (MLR) using only four wavelengths (738 nm, 1362 nm, 1835 nm and 1859 nm). The MLR calibration using the four wavelengths had a prediction accuracy of r2 = 0.82 and RMSEP = 18.4 kg N ha'1, which shows potential for commercial applications. These findings will encourage the development of a remote sensing method for predicting nitrogen uptake for rice crops using the identified wavelengths in an inexpensive multiband instrument.

AB - Rice growers are reluctant to physically sample their crops, with many relying on visual estimates of crop nitrogen status to determine mid-season nitrogen topdressing requirements. There is increased interest in using remote sensing to determine how accurately mid-season rice crop nitrogen status (nitrogen concentration and content in shoots) can be predicted. Field-derived hyperspectral reflectance measurements were collected over three seasons to evaluate the optimal wavelengths for the prediction of in-crop nitrogen at panicle initiation. Rice plots were planted and differing nitrogen rates applied to establish a range of crop nitrogen uptake levels at the panicle initiation stage of crop development. Hyperspectral canopy reflectance (350'2500 nm) data were collected from each plot at the panicle initiation growth stage and plant samples were collected at each scanning location. Calibration models were developed for plant nitrogen concentration, dry matter and nitrogen uptake using partial least squares regression and the influential wavelengths used in the nitrogen uptake prediction model identified. When the nitrogen uptake model, which included samples collected in three crop seasons, was used to predict nitrogen uptake in an independent set of samples, r2 was 0.84 and the root-mean-square error of prediction (RMSEP) was 16.5 kg N ha'1. The important wavelengths utilised for predicting nitrogen uptake were determined from the regression coefficients and a new calibration created by multiple linear regression (MLR) using only four wavelengths (738 nm, 1362 nm, 1835 nm and 1859 nm). The MLR calibration using the four wavelengths had a prediction accuracy of r2 = 0.82 and RMSEP = 18.4 kg N ha'1, which shows potential for commercial applications. These findings will encourage the development of a remote sensing method for predicting nitrogen uptake for rice crops using the identified wavelengths in an inexpensive multiband instrument.

KW - Rice, nitrogen, panicle initiation, remote sensing

U2 - 10.1255/jnirs.1246

DO - 10.1255/jnirs.1246

M3 - Article

VL - 24

SP - 473

EP - 483

JO - Journal of Near Infrared Spectroscopy

JF - Journal of Near Infrared Spectroscopy

SN - 0967-0335

IS - 5

ER -