The potential of hyperspectral reflectance data could be explored to assess severity of any disease in a crop. Here is a case study of assessing severity of yellow rust disease (Biotroph Pucciniastriiformis) in winter wheat crop.
Case study
The hyperspectral remote sensing data was collected for winter wheat (Triticum aestivum L.) crop at different levels of disease infestation using field spectroradiometer over the spectral range of 350 to 2500nm. The partial least squares (PLS) and multiple linear (MLR) regression techniques were used to identify suitable bands and develop spectral models for assessing severity of yellow rust disease in winter wheat
Fig: Wheat crop in study area with different disease
severity scores of yellow rust
crop. The PLS model based on the full spectral range and n = 36, yielded a coefficient of determination (R2) of 0.96, a standard error of cross validation (SECV) of 12.74 and a root mean square error of cross validation (RMSECV) of 12.41. The validation analysis of this PLS model yielded r2 as 0.93 with a SEP (Standard Error of Prediction) of 7.80 and a RMSEP (Root Mean Square Error of prediction) of 7.46. The loading weights of latent variables from PLS model were used to identify sensitive wavelengths.
To assess their suitability multiple linear regression (MLR) model was applied on these wavelengths which resulted in a MLR model with three identified wavelength bands (428nm, 672nm and 1399nm). MLR model yielded acceptable results in the form of r2 as 0.89 for calibration and 0.90 for validation with SEP of 3.90 and RMSEP of 3.70.
The result showed that the developed model had a great potential for precise delineation and detection of yellow rust disease in winter wheat crop.
Salient findings-
The study successfully executes spectroscopy in VNIR and SWIR regions for yellow rust severity detection in winter wheat crop.
By means of PLS regression, ANOVA and MLR, a robust model was developed based on reflectance spectra and yellow rust disease severity scores of wheat crop.
The model developed through this study reflects strong correlation and low error.
Six latent variables in the form of principal components were derived through PLSR and three highly significant wavelengths (428nm, 672nm and 1399nm) were determined for yellow rust severity.
