Soil Reflectance Composites—Improved Thresholding and Performance Evaluation

Reflectance composites that capture bare soil pixels from multispectral image data are increasingly being analysed to model constituents such as organic carbon. These temporal used instead of single-date images account for the frequent vegetation cover soils and, thus, get broader spatial coverage pixels. Most compositing techniques require thresholds derived spectral indices Normalised Difference Vegetation Index (NDVI) and Burn Ratio 2 (NBR2) separate all other land types. However, threshold derivation is handled based on expert knowledge a specific area, statistical percentile definitions or in situ data. For operational processors, site-specific partly manual strategies not applicable. There need more generic solution derive large-scale processing without intervention. This study presents novel HIstogram SEparation Threshold (HISET) methodology deriving index testing them Sentinel-2 stack. The technique index-independent, data-driven can be evaluated quality score. We tested HISET building six reflectance (SRC) using NDVI, NBR2 new combining NDVI short-wave infrared (SWIR) band (PV+IR2). A comprehensive analysis performance accuracy resulting SRCs proves flexibility validity HISET. Disturbance effects confusion with non-photosynthetic-active (NPV) could reduced by choosing grassland crops input LC NBR2-based SRC spectra showed highest similarity LUCAS spectra, broadest least number valid observations per pixel. validated against database Integrated Administration Control System (IACS) European Commission. Validation results show PV+IR2-based outperform two indices, especially spectrally mixed areas soil, photosynthetic-active NPV. NDVI-based lowest confidence values (95%) bands. In future, shall different environmental conditions characteristics evaluate if findings this also valid.