Naslov
Use of Field Spectroscopy for Assessment of Nitrogen Use Efficiency in Winter Wheat

Autori
Šestak, Ivana

Vrsta, podvrsta i kategorija rada
Ocjenski radovi, doktorska disertacija

Fakultet
Agronomski fakultet

Mjesto
Zagreb

Datum
07.12

Godina
2011

Stranica
159

Mentor
Mesić, Milan

Ključne riječi
winter wheat; nitrogen fertilization; nitrogen use efficiency; leaf reflectance; vegetation index; neural networks; linear modeling; classification

Sažetak
The objective of doctoral study was to evaluate the ability of field VNIR spectroscopy to estimate in-season N status and harvest variables of winter wheat under field conditions, and to determine the effects of N fertilization on spectral reflectance and agronomic characteristics. Agronomic data included leaf total N content, CCI, grain total N content, yield and NUE. Research was conducted on experimental field in Western Pannonian subregion of Croatia. To simulate different crop growth scenarios, field measurements with reflectance (350-1050 nm) were acquired from winter wheat flag leaves grown under different mineral N fertilization treatments ranging from 0 to 300 kg N ha-1, during stem extension (F8) and heading stage (F10.5) of growing period 2008 with cultivar “Fiesta” and 2010 with cultivar “Lucija”. Linear statistical models (SLR, MLR, PLSR), non-linear pattern analysis (ANN) and classification analysis (DA, CLA, ANN) were generated to estimate crop biophysical variables and to discriminate between nitrogen fertilization categories, based on the 1st derivative of reflectance in form of PCs and VIs. N fertilization and high level of greenness during F8 reduced visible and increased NIR spectra. These changes were related to lower levels of chlorophyll and earlier senescence in the N-limited plots and F10.5 stage. The greatest spectral differences between N fertilization treatments and the highest correlations with the winter wheat variables were found in the visible and the red edge region which contributed most to the PC development. High and robust correlations between VIs calculated from reflectance that was acquired at F8-“Lucija”, and all crop variables among each other proved to choose yield [r (NDVI : yield) = 0.91] as input to prediction models development. The absence of treatment x cultivar/year interaction for leaf TN content measured at F8 which indicates possibility for spectral sensing of winter wheat in stem extension stage regardless of cultivar differences. MLR identified a 7PC leaf reflectance model that explained 83 % of the variability in winter wheat yield, and accounted for a large variance in leaf and grain biochemical concentration (R2 ~ 0.8, P < 0.05) (F8, cultivar „Lucija“). Forecasting NUE from harvest data estimated by ground-based remote sensing was feasible (r for observed vs. predicted NUE was 0.81, p < 0.05). This relationship is the most encouraging result due to the high predictive ability of the models based on F8 spectral data to accurately estimate winter wheat harvest variables. ANN models were the most efficient in capturing the complex link between yield and leaf reflectance spectra (train and test dataset with r = 0.95 and r = 0.92, RMSEC = 2.57 dt ha-1 and RMSEP = 4.41 dt ha-1, respectively) compared to corresponding SLR-VIs, MLR and PLSR models. Performance of the 8 factor PLSR model indicated the highest consistency due to the small difference between RMSEC (4.10 dt ha-1) and RMSEP (4.61 dt ha-1) besides high prediction ability (validation R2 = 0.84), and showed that it is possible to predict grain yield using hyperspectral field spectroscopy data. Still, NDVI and RVI reached a very strong relationship with yield due to the cross-validation results: R2 = 0.80 and R2 = 0.73, respectively. Classification analysis indicated that hyperspectral measurements are best at detecting where N is limited rather than where it is in excess during the fast vegetation development (F8) which is of great interest in precision agriculture by optimizing N top-dressing, identifying crop stress patterns and aid in yield forecasting. Irrespective of pre- analysis type of classification approaches, extreme groups were completely separable [I (Control, N0) and III (N250, N250 + amendments, N300)], and the small proportion of variability in intermediate N non-limited treatments [II (N100, N150, N200)] was probably result of some additional factors during the growing season. The results obtained in this doctoral thesis confirm the high information potential and feasibility of field spectroscopy for estimation of winter wheat conditions during development and harvest because they are scalable and applicable in high range of N stress and non-N-limited environments. Key spectral features and algorithms should help to support non-destructive and real-time monitoring of N status in wheat production by using hyperspectral remote sensing. Further efforts should be taken to increase the amount, complexity and representation of samples so that the derived model can be applied under varied conditions.

Izvorni jezik
Engleski

Znanstvena područja
Poljoprivreda (agronomija)

POVEZANOST RADA