Ability of models with effective wavelengths to monitor nitrogen and phosphorus status of winter oilseed rape leaves using in situ canopy spectroscopy

• Published in: Field crops research Vol. 215; pp. 173 - 186
• Main Authors: Li, Lantao, Wang, Shanqin, Ren, Tao, Wei, Quanquan, Ming, Jin, Li, Jing, Li, Xiaokun, Cong, Rihuan, Lu, Jianwei
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.01.2018
• Subjects: First derivative reflectance; Hyperspectral remote sensing; Oilseed rape farming; Partial least square; Precision nutrient management; First derivative reflectance; Hyperspectral remote sensing; Oilseed rape farming; Partial least square; Precision nutrient management
• ISSN: 0378-4290; 1872-6852
• DOI: 10.1016/j.fcr.2017.10.018

•Four spectral techniques were proposed to monitor LNC/LPC in winter oilseed rape.•The first derivative was the most accurate technique for LNC/LPC estimation.•Effective wavelengths (EW) were selected using the variable importance in projection.•Field experiments and farmers’ fields were used to revalidate the performance of EW. Till date, studies using canopy hyperspectral data to monitor crop nutrient status have focused mainly on biomass, water and nitrogen (N) prediction, and only a few have attempted to monitor phosphorus (P). This study aimed to evaluate the potential of the canopy raw spectra (R) in combination with a partial least square (PLS) regression model for estimating the leaf N and P concentration (LNC and LPC), compared to the potential of other hyperspectral transformation techniques such as log-transformed spectra (Log(1/R)), the continuum removal (CR) method and first derivative reflectance (FDR) for winter oilseed rape. Field experiments were conducted over three consecutive growing seasons (2013–2016) at different sites (Wuxue, Wuhan and Shayang) in Hubei, China, using different N and P application rates, planting patterns, cultivars and ecological sites. Data from the conventionally managed fields of 25 farmers in 2015–2016 were also collected to test the transferability of the established optimal monitoring model for LNC and LPC prediction. Canopy hyperspectral reflectance data were acquired over a wavelength range from 400 to1300nm (the visible and near-infrared region, VNIR), and quantitative correlations between LNC and LPC and their spectra were determined. The results showed that the FDR-PLS model yielded the highest retrieval accuracy for LNC and LPC predictions. The coefficient of determination of the validation dataset (r2val) between the observations and predictions was 0.89 for LNC and 0.82 for LPC, with a relative percent deviation (RPDval) of 2.41 and 2.22, respectively. The variable importance in projection (VIP) values of the FDR-PLS model with full spectral range were applied to identify the effective wavelengths and to decrease the high dimensionality of the canopy hyperspectral reflectance dataset. Seven wavelengths centred at 445, 556, 657, 764, 985, 1082, and 1194nm and six wavelengths at 755, 832, 891, 999, 1196, and 1267nm were identified as effective wavelengths for predicting the LNC and LPC values. The newly-developed FDR-PLS models for LNC (r2val=0.85, RPDval=2.10) and LPC (r2val=0.78, RPDval=1.94) provided accurate estimations based on field experiment validations using the effective wavelengths. The validation in the farmers’ fields also indicated an excellent accuracy between the observed and predicted values for LNC (r2val=0.82, RPDval=2.09) and LPC (r2val=0.75, RPDval=2.01). The overall results demonstrated the applicability and feasibility of the FDR-PLS model for estimating the N and P status of winter oilseed rape using in situ canopy hyperspectral reflectance data.