Leptospirosis is a water-related zoonotic disease with complex transmission pathways, including direct transmission from infected animals and indirect transmission through contaminated soil and water. Identifying key areas to implement targeted infection prevention and control strategies is challenging, as a range of risk factors across different scales can drive human infection. We aimed to develop an epidemiological modelling approach to predict key transmission pathways driving leptospirosis infection, including risks ranging from household level factors to the movement of pathogens across watersheds. We combined a causal Bayesian network with a novel hydrological pathogen transport model to predict leptospirosis across Fiji and found key infection hotspots adjacent to rivers and within degraded watersheds; a dynamic overlooked by previous epidemiological models. We used a wide range of data for model parameterisation (e.g., expert elicitation, epidemiological surveys, and environmental data) and found that predictive validity improved when expert input on risk factors was used to guide model parameterisation, improving R2 for predicted versus observed seroprevalence from 0.71 to 0.91. Our One Health modelling approach can be used to support the design and evaluation of environment-based disease prevention strategies at national scales.