Early warning signal for river‐borne diseases with almost no data

Abstract Effective management of emerging river‐borne diseases requires early prediction of pathogen spatial distributions. However, data on pathogen locations are notoriously rare in the beginning of disease outbreaks and insufficient to feed existing predictive models. We extended hidden Markov models (HMM) to exploit detailed spatially structured data on environmental covariates that are readily available for many aquatic systems, augmented by arbitrarily sparse spatial data on actual pathogen occurrence, to predict pathogen distribution over large geographical scales. The extended HMM predicted the spread of whirling disease in the Oldman River, Canada, with 0.7 area under the receiver operating characteristic curve (AUC) in the absence of any disease test result, provided that the status of a single pixel can be estimated correctly. The AUC increased with the number of used test results, until it reached 0.9 when 100 test results were used. The model has the potential to provide early warning signals for emerging diseases prior to, or at the early stages of, their emergence in a river.