Biodiversity monitoring is crucial for ecosystem conservation, yet field data collection is limited by costs, time, and extent. Remote sensing represents a convenient approach providing frequent, near-real-time information over wide areas. According to the Spectral Variation Hypothesis (SVH), spectral diversity (SD) is an effective proxy of environmental heterogeneity, which ultimately relates to plant diversity. So far, studies testing the relationship between SD and biodiversity have reported contradictory findings, calling for a thorough investigation of the key factors (e.g., metrics applied, ecosystem type) and the conditions under which such a relationship holds true. This study investigates the applicability of the SVH for plant diversity monitoring at the landscape scale by comparing the performance of three different types of SD metrics. Species richness and functional diversity were calculated for more than 2000 cells forming a grid covering the Czech Republic. Within each cell, we quantified SD using a Landsat-8 'greenest pixel' composite by applying: i) the standard deviation of NDVI, ii) Rao's Q entropy index, and iii) richness of 'spectral communities'. Habitat type (i.e., land cover) was included in the models describing the relationship between SD and ground biodiversity. Both species richness and functional diversity show positive and significant relationships with each SD metric tested. However, SD alone accounts for a small fraction of the deviance explained by the models. Furthermore, the strength of the relationship depends significantly on habitat type and is highest in natural transitional areas. Our results underline that, despite the stability in the significance of the link between SD and plant diversity at this scale, the applicability of SD for biodiversity monitoring is context-dependent and the factors mediating such a relationship must be carefully considered to avoid drawing misleading conclusions.