A decision tree for nitrogen application based on a low cost radiometry

作者：Rodriguez Moreno F^*; Llera Cid F

来源：Precision Agriculture, 2012, 13(6): 646-660.

DOI：10.1007/s11119-012-9272-7

• Cereals
• Classification algorithms
• Leaf reflectance
• Principal component analysis
• Spectral vegetation indices

Fertilizer recommendations based on radiometry require studies to calibrate the relationships to scenario conditions, otherwise the effectiveness may be reduced. The objective of this study was to develop a decision tree to detect nitrogen deficiency with efficiency comparable to the analysis of the full spectral signature, with simplicity similar to a spectral index and valid over a wide range of development conditions and phenological stages. An agronomic trial with a dual-purpose triticale (X Triticosecale Wittmack) was used in this study having different planting densities, number of grazing events (regeneration from defoliation) and nitrogen fertilization. At different phenological stages, the spectral signatures of leaves were recorded with an ASD-FieldSpec3 spectroradiometer and the nitrogen concentrations were determined by the Kjeldahl method. Agronomic factors that affect the N concentration were identified using ANOVA; subsequently PCA was carried out on the set of spectral signatures representative of the groups formed according to nitrogen concentration. Linear regression was used to evaluate the relationship between the principal components and plant nitrogen concentration. Wavelengths with greater significance were used to construct a decision tree. The resulting decision tree defined for nitrogen using the Jenks Natural Breaks method had a success rate of 68.3%. The best spectral index had a R-2 = 0.31 while the estimate using the full spectral signature reached a R-2 = 0.68. Although further testing is needed, this work shows the approach was able to successfully categorize nitrogen deficiency.

• 出版日期2012-12