Environmental DNA (eDNA) has increasingly been used to detect rare species (e.g., newly introduced nonindigenous species) in both terrestrial and aquatic ecosystems, often with distinct advantages over traditional methods. However, whether water eDNA signals can be used to inform invasion risks remains debatable owing to inherent uncertainties associated with the methods used and the varying conditions among study systems. Here, we sampled eDNA from canals of the central route of the South‐to‐North Water Diversion Project (hereafter SNWDP) in China to investigate eDNA distribution and efficacy to inform invasion risks in a unique lotic system. We first conducted a total of 16 monthly surveys in this system (two sites in the source reservoir and four sites in the main canal) to test if eDNA could be applied to detect an invasive, biofouling bivalve, the golden mussel Limnoperna fortunei. Second, we initiated a one‐time survey in a sub‐canal of the SNWDP using refined sampling (12 sites in ~22 km canal) and considered a few environmental predictors. We found that detection of target eDNA in the main canal was achieved up to 1100 km from the putative source population but was restricted to the warmer months (May–November). Detection probability exhibited a significant positive relationship with average daily minimum air temperature and with water temperature, consistent with the expected spawning season. eDNA concentration in the main canal generally fluctuated across months and sites and was generally higher in warmer months. Golden mussel eDNA concentration in the sub‐canal decreased significantly with distance from the source and with increasing water temperature and became almost undetectable at ~22 km distance. Given the enormity of the SNWDP, golden mussels may eventually expand their distribution in the main canal, with established “bridgehead” populations facilitating further spread. Our findings suggest an elevated invasion risk of golden mussels in the SNWDP in warm months, highlighting the critical period for spread and, possibly, management.