While environmental DNA (eDNA) metabarcoding holds promise as a holistic approach to assess vegetation changes and community composition across diverse spatial and temporal scales, systematic investigations of its efficacy compared to conventional field surveys remain scarce in the literature. The present study explores the differences in plant diversity recovered from field surveys and captured with a multi-marker eDNA metabarcoding approach (two nrDNA ITS1 and ITS2, and two cpDNA rbcL and trnL) from river water samples. The eDNA metabarcoding approach retrieved 46 aquatic plants (hydrophytes and helophytes) and 245 terrestrial plants, compared to 24 and 127 species identified from field surveys. On average, eDNA samples collected immediately downstream of the survey sites recovered 43% and 39% of the aquatic and terrestrial species observed, respectively. Discrepancies were explained by differences in taxonomic resolution, the stochasticity of the retrieval of rare and elusive species, and the presence of reference sequences. We found a significant positive correlation between spatial and community distances at scales ranging from 2-9 km and identified turnover as the driving force of these differences. Metabarcoding demonstrated sensitivity to community changes and both approaches converge on a similar community structure. Interestingly, eDNA samples collected immediately upstream of the survey sites exhibited significant species overlap with the downstream samples (c. 100 m apart). Overall our results demonstrate that, with adequate sampling and a multi-marker metabarcoding approach, eDNA has the potential to approximate catchment gamma diversity while still being informative of the local flora.