Biases in estimated vegetation indices from spectral below cloud reflectance measurements

Vegetation indices (VIs) derived from ground-based or unmanned aerial vehicle measurements use relative reflectance measurements obtained by flying over well-defined reflectance panels (RP). The RP overflights provide a form of transfer calibration to determine reflectances over the actual vegetated areas. It is assumed that environmental conditions, i.e., solar zenith angle (SZA) and cloud optical thickness (COT), remain constant between RP overflights. During typical 10-minute intervals between RP overflights, the COT varies especially during broken cloud conditions. Although days with low-level and mid-level clouds are avoided during vegetation remote sensing, even optically thin cirrus affects the radiation reaching the surface and therefore the reflectance measurements. The clouds change the absolute value of the incoming irradiance, but also the spectral signature by scattering radiation primarily in the visible–near-infrared wavelength range and absorbing radiation in the shortwave–infrared wavelength range. Consequently, a change in cloud cover between RP overflights distorts the measured reflectance. To systematically investigate the effects of COT changes on VI estimates between RP overflights, we present coupled radiative transfer simulations using the library for radiative transfer model (libRadtran) and the Soil Canopy Observation of Photosynthesis and Energy fluxes (SCOPE2.0) model. Simulations were performed for liquid water and ice clouds, and combinations of COT between the assumed state, i.e., during an RP overflight, and the true state, i.e., during the actual measurement. The COT was varied between 0 and 10, which is representative for cirrus. For comparability, the same range was used for the mid-level liquid water cloud.  In general, biases in estimated VI are most sensitive to COT, respond non-linear, and are further governed by the SZA. The simulations performed indicate that the normalized vegetation index (NDVI) is less sensitive to cloud effects than the enhanced vegetation index (EVI). For SZA less than 50°, a higher than assumed COT leads to an overestimation of the EVI, while for lower than assumed COT it leads to an underestimation of the EVI. For a more general assessment, the spectral effects on narrow-band ratios of the form (ρ(λ₁) - ρ(λ₂)) / (ρ(λ₁) + ρ(λ₂)), with ρ(λ) the spectral reflectance at wavelengths λ_1,2 ∈ [400,2400~nm], were investigated. The proposed presentation will outline the raised problems and present the results from the coupled simulations.