Comparison of Leptosira concentrations in two rivers within contrasting catchments in a leptospirosis hotspot.

Leptospirosis is a neglected disease that primarily impacts small islands in the South Pacific region. We compared pathogenic Leptospira concentration in rivers from two contrasting catchments in New Caledonia, a leptospirosis hot spot. The Tho and Foa catchments are located on the humid east coast and on the drier west coast, respectively. Similar concentrations of pathogenic Leptospira (pL) were found in both catchments although no extreme events were sampled in Foa and two cyclones were sampled in Tho. Moreover, significant pL concentrations were found in Foa during a 25 mm rainfall event as well as during the low flow sampling. A total of 402 measurements (226 in Tho and 176 in Foa) and 804 augmented data points were used to build a feedforward neural network model. This model explained 71% and 74% of the variance of the decimal logarithm of pL concentration data, when trained with rainfall or water level data, respectively. However, we were unable to extend a model trained on the Tho or Foa catchment to the other one. The upper envelope of the decimal logarithm of pL concentration, representing the maximum concentration expected over a few hours period was identified as the relevant signal for issuing warnings to the local population. Neural network models designed for the envelope were built and trained more quickly than augmented data models. It is concluded that envelope models based on rainfall data extracted from weather forecasts could be used to create local early warning systems for leptospirosis.