Drivers of spectro-temporal variability in multi-spectral soil reflectance measured from space: A continental-scale analysis

Abstract

In Europe, bare soil is typically exposed only during short time windows and optical sensing by satellite constellations, such as Sentinel-2 and Landsat, therefore commonly relies on aggregates of soil reflectance spectra over extended periods of time. Depending on the length of the time series, substantial spectral variability can be observed in the soil spectra. To select appropriate aggregation strategies, it is critical to understand the mechanisms that drive temporal variability in soil reflectance. In this study, we quantify the variance induced into soil reflectance time series by two properties that modify the spectral response: surface soil moisture and the illumination geometry. We analyze seven-year time series of surface reflectance (Level-2A corrected) satellite imagery of Landsat 8 and Sentinel-2 at 27,819 LUCAS sites distributed across the European continent. The sites are correlated with independent datasets from Copernicus and ECMWF. A rigorous statistical framework is applied to estimate the global sensitivity for both properties, using pooled standardized effect sizes ( β ¯ ∗ ) and its isolated contribution to spectral variance, using pooled partial eta squared ( η ¯ P 2 ) for each spectral band. Furthermore, we explicitly separate illumination effects attributable to atmospheric influences (e.g., haze, dust, light-scattering) from those arising from ground-level surface–light interactions (e.g., shadows cast on rough surfaces at shallow elevation angles). Results show that the dominant driver of temporal variability in measured soil reflectance is the interaction between the rough bare soil surface and changes in solar incidence angle ( η ¯ P 2 ≈ 0.2 ) throughout the entire spectral domain. In contrast, surface soil moisture plays a comparatively minor role at all bands ( η P 2 ¯ ≈ 0.06 ). The large volume of data further enables a stratified analysis by soil texture, revealing that soils with different textural properties respond with varying strength to both drivers. For example, the reflectance of clay-rich soils is influenced more strongly by changes in the illumination geometry. Further, soils with a high content of sand tend to exhibit a weaker response to soil moisture than finer-textured soils with a greater water-holding capacity. The presented variance decomposition enables both, the implementation of targeted corrective measures, such as soil-specific BRDF corrections, and the exploitation of variability patterns caused by different soil aggregates themselves as proxies for soil property modeling.