Environmental DNA (eDNA) samples capture mixtures of DNA from a wide range of organisms and often yield metabarcoding datasets that include taxa beyond the intended scope of the primer set used. The presence of this non-target DNA (termed "molecular bycatch") can make it difficult to interpret trends in the primary targets. Here, we present an example in which including non-target detections alters the ecological interpretation of a metabarcoding time series and highlight a quantitative approach that can account for molecular bycatch in a biologically meaningful way. We collected high-resolution temporal samples from an estuary (Hood Canal, Washington, USA), using a surface-stationed autonomous sampler over a 48-hour period. We metabarcoded samples with a primer set that targets fish, but also amplifies (non-target) marine mammals with lower efficiency. We then transformed sequence read proportions into absolute DNA concentrations using a mock community–based bias correction and single-species concentrations via droplet digital PCR (ddPCR). We found that non-target mammals were substantially under-represented in the raw metabarcoding data relative to their biased-corrected proportions. Cross-validation of this quantitative metabarcoding approach showed that predicted DNA concentrations closely matched independent ddPCR measurements for total 12S DNA and two single-species concentrations (Atlantic bottlenose dolphin and Pacific herring) in both mock communities and environmental samples. After estimating absolute species concentrations for all species, we revealed species-specific patterns that were not apparent in the observed read counts alone. Our results underscore the importance of correcting for species-specific amplification bias when evaluating community composition from metabarcoding data, particularly in the context of non-target detections.