Development of a Spectral Index for the Detection of Yellow-Flowering Vegetation

Abstract

Floral phenology as a special indicator of climate change and vegetation dynamics is drawing more attention. The long-term observations of flowering events collected at scattered ground sites have accumulated valuable priority on the understanding of floral phenology, but with insufficient investigation on the spatio-temporal dynamics at regional scale, which is mainly induced by the lack of effective ways to capture the pixel-based flower events from remote sensing images. The existing yellowness indices are constructed for rape (Brassica napus L.) with less suppression to the bright background and dark green vegetation, and further with inadequate consideration on physiological characteristics and the temporal spectral signature of investigated vegetation. In this paper, we examined rape and several other representative vegetation types to determine spectral features of yellow-flower period within the growing season, then selected the visible and near-infrared bands to construct a Novel Yellowness Index (NYI) with an enhancement on the physiological mechanism of plants. The proposed NYI were discussed on the variation of mathematical properties with representative instances, cross-compared with three typical yellowness indices—Ratio Yellowness Index (RYI), Normalized Difference Yellowness Index (NDYI), and Ashourloo Canola Index (ACI) —over various yellow-flowering vegetation species at multiple scales, and validated with ground observations of three available PhenoCam network stations and field phenological observations at Görlitz, Sachsen, and Germany. In addition, we applied NYI to detect the rape field using Sentinel-2 image at Görlitz with typical rape area as a case study. Results show that the proposed NYI exhibits the potential to capture yellow-flowering events with increased sensitivity to the variation of flower density, and reduction of noise introduced by bright background or dark green vegetation of multiple vegetation species at different scales. As the flower density increases from 33% to 78%, the relative differences of NYI captured can reach up to 74%, compared with other three indices which have the relative differences no more than 57%. The cross-comparison indicates NYI performs better with higher consistent with PhenoCam observation and Deutscher Wetterdienst phenological station than other yellowness indices in capturing the variation of yellow flower density. The case study of NYI application in the identification of rape field exhibits good accuracy with the overall accuracy up to 97.5%, the Kappa coefficient of 0.94, and F score of 0.96. Consequently, the satellite-derived yellowness index will be a potential means to investigate the flowering dynamics and planting range of yellow-flowering vegetation such as rape.