Evaluation framework for the generation of continental bare surface reflectance composites

Soils play a pivotal role in supporting ecosystems, human health, food security, and climate regulation. Since several years, temporal composites of bare soil reflectances derived from multispectral satellite data are used as input for soil property modeling. Due to the importance of these model inputs, the quality of the surface reflectance composites (SRC) is essential. The quality depends on the precise selection of pixels that are free of green and dry vegetation, cloud contamination and other atmospheric disturbances. However, there is a lack of suitable concepts and tools to evaluate the impact of the diverse processing parameters for the generation of SRC, especially for large areas such as continents. This study presents a novel approach to evaluate the process of computing bare SRCs across large geographical areas. It can estimate the theoretical limit achievable with defined processing parameters (spectral indices, thresholds, specific filtering, etc.) and it is also suitable to compare the performance of different SRC concepts from the literature. The performance is derived from the angular spectral distance between reference spectra derived from the LUCAS survey and the SRC spectra. It is demonstrated that a linear combination of two spectral indices complemented with a regional threshold dataset keep the complexity of threshold data sets low while performing well across Europe. The results also show that regionalization is as crucial as the choice of the index itself. The additional outlier removal focusing on clouds and haze marginally improved the SRC at the continental scale but can be very effective for areas with more frequent clouds. The proposed method offers two main advantages. First, it allows for parameter customization tailored to the region of interest, or, at minimum, to areas well represented by the reference data. Second, it facilitates the systematic evaluation of successive adaptations in the SRC generation process, eliminating the labor-intensive and error-prone task of visually comparing images to assess improvements in the SRC final product. The final bare surface reflectance composite for Europe and adjacent regions provids a robust foundation for future large-scale soil and bare surface monitoring. • Developed an automated evaluation method for bare surface reflectance composites. • Enhanced filtering of cloud and haze above bare surfaces using time series data. • Quantified the performance of various spectral indices for bare surface detection. • Compared several published bare surface reflectance composite techniques for Europe. • Demonstrated the superiority of regional thresholds over global ones.