Biomonitoring of complex heterogeneous environments is highly challenging. Fish in deep water bodies occupy different habitats, therefore a combination of survey methods has traditionally been used. Environmental DNA (eDNA) metabarcoding is a novel monitoring tool that can overcome spatial heterogeneity in a highly sensitive and entirely non-invasive manner. However, taxon detection probability is dependent on real-time environmental variables. In this study, three reservoirs were sampled in two seasons using a spatiotemporally distributed sampling design covering major environmental gradients. In all sampling campaigns, 31 fish taxa were detected which exceeded expectations. Data reliability was confirmed by a tight positive correlation between individual taxon scores derived from gillnet sampling and eDNA site occupancy. Analyses confirmed anticipated trends, such as the highest number of taxa were observed in the largest water body, and more taxa were detected in inflows and littoral regions compared to open water. The most important factors for fish distribution were temperature, age and trophic status (expressed as total Chlorophyll a concentration) of water bodies. Taxon detection reflected ecological niches of individual species, e.g. warm water wels catfish (Silurus glanis) and cold water salmonids. This study provides further evidence that eDNA metabarcoding is suitable for ecological study in heterogeneous environments and may substitute conventional fish sampling techniques.