West Nile virus (WNV) persists in an enzootic cycle between birds and mosquitoes. Human infections are incidental and usually sub-clinical with some cases of neuroinvasive disease. Mitigation relies on surveillance to guide decision making, including RT-qPCR testing pools of mosquitoes. Cycle threshold (Ct) values from these pools---semi-quantitative proxies for viral load---are binarised to estimate WNV prevalence in mosquitoes as a proxy for human risk. We showed that Ct value variation in Colorado and Nebraska (2022-2024) cannot be explained by laboratory factors, suggesting a role for biological and epidemiological mechanisms. We developed a multiscale model linking pooled Ct values to mosquito viral kinetics, bird-to-mosquito transmission, and seasonal force of infection. A novel method estimating WNV prevalence using pooled Ct values outperformed existing approaches at high prevalence. We demonstrate the importance of treating environmental viral load samples as quantitative, as binarising masks key information for arbovirus surveillance and risk mitigation.