[Retrieval of leaf water content of winter wheat from canopy hyperspectral data using partial least square...

Accurate estimation of leaf water content (LWC) from remote sensing can assist in determining vegetation physiological status, and further has important implications for drought monitoring and fire risk evaluation. This paper focuses on retrieving LWC from canopy spectra of winter wheat measured with ASD FieldSpec Pro. The experimental plots were treated with five levels of irrigation (0, 200, 300, 400 and 500 mm) in growing season, and each treatment had three replications. Canopy spectra and LWC were collected at three wheat growth stages (booting, flowering, and milking). The temporal variations of LWC, spectral reflectance, and their correlations were analyzed in detail. Partial least square regression embedded iterative feature-eliminating was designed and employed to obtain diagnostic bands and build prediction models for each stage. The results indicate that LWC decreases quickly along with the winter wheat growth. The mean values of LWC for the three stages are respectively 338.49%, 269.65%, and 230.90%. The spectral regions correlated strongly with LWC are 1 587-1 662 and 1 692-1 732 nm (booting), 617-687 and 1 447-1 467 nm (flowering), and 1 457-1 557 nm (milking). As far as the LWC prediction models are concerned, the optimum modes of spectral data are respectively logarithmic, 1st order derivative and plain reflectance. The diagnostic bands detected by PLS are from SWIR, NIR, and SWIR. Retrieval accuracy at the flowering stage is the highest (R2(cv) = 0.889) due to the enhancement of leaf water information at canopy scale via multiple scattering. At the booting and milking stage, accuracies are relatively lower (R2(cv) = 0.750, 0.696), because the retrieval of LWC is negatively affected by soil background and dry matter absorption respectively. This research demonstrated clearly that the spectral response and retrieval of LWC has distinct temporal characteristics, which should not be neglected when developing remote sensing product of crop water content in the future.