This dataset provides standardised, image-based estimates of percentage flower cover derived from the Grassworks-vegetation plots surveyed four times per year in parallel with pollinator assessments. Within 50 m × 5 m transects, four 1 m × 1 m subplots were photographed vertically using RGB cameras. Approximately 12,000 images were geometrically corrected, cropped to a standardised frame, and merged into mosaics representing vegetation status per site and survey period. Mosaics were assigned artificial spatial references for processing in a GIS environment. Flower cover was quantified through supervised colour classification using a Maximum Likelihood algorithm in ArcGIS Pro. Three flower colour classes were defined: white, yellow, and a combined blue–violet–red spectrum ("red-blue"), including transitional hues. Training data were compiled from randomly selected pixels of images across all sites and months, covering a wide range of colour shades, species, and light conditions. Classification outputs were visually checked and iteratively refined. An attribute table was created to record, for each image, whether flowers of the respective color category were present. Incorrectly classified flower coverings for a flower color not present in the image were set to zero. Percentage cover was calculated as the proportion of classified pixels per colour class; total flower cover equals the sum of the three classes. Validation with flower-free control images indicated low false-positive rates (~0.4% mean total error), with comparable uncertainty expected for images containing flowers. The dataset includes image ID, sub-transect ID, UTC timestamp (ISO 8601), visually assessed light conditions, and percentage cover values per colour class and in total. Data were collected at multiple grassland sites across central, northern, and southern Germany in 2022 and 2023, with site coordinates reported at two-decimal precision (blurred for privacy).