Hyperspectral Reflectance Data Processing Through Cluster and Principal Component Analysis for Estimating Irrigation and Yield-Related Indicators

Abstract

Management of agricultural practices such as irrigation by using remotely sensed data requires background data obtained from field experiments carried out under controlled conditions. In this study, spectral and agronomic data from field trials consisting of six different irrigation treatments were used to derive new spectral indicators for estimating growth level and water use status of dwarf green beans. Spectral reflectance (Ref) values were smoothed and first-order derivative spectra (ρ) were calculated. Linear regression and multivariate analysis (cluster and principal component analysis) were done between agronomic indicators and both the smoothed spectral reflectance (R) and ρ of each individual wavelength between 650 and 1100. nm. Based on those calculations, the most appropriate wavelengths were selected for each agronomic indicator and new combinations were calculated by using rationing, differencing, normalized differencing and multiple regression. The ratio between ρ measured at 950 or 960. nm and 1020. nm wavelengths provided estimates in an error band of 2.47. bar for Leaf Water Potential (LWP) and 3.18% for Leaf Water Content (LWC). An equation based on ρ740 and ρ980 was developed to estimate Leaf Relative Water Content (LRWC). In the same manner, the ρ at 820 and 970. nm provided a good estimate of crop water use and the ρ values at 770 and 960. nm were critical for the calculation of Leaf Area Index (LAI) and dry biomass. It was also determined that the ratio of R930 to R670 can be applied to yield estimation. © 2011 Elsevier B.V.