Abstract
Leptospirosis is a zoonosis caused by Leptospira bacteria present in the urine of mammals. Leptospira is able to survive in soils and can be resuspended during rain events. Here, we analyzed the pathogenic Leptospira concentration as a function of hydrological variables in a leptospirosis hot spot. A total of 226 samples were collected at the outlet of a 3 km2 watershed degraded by ungulate mammals (deer and feral pigs) and rats which are reservoirs for leptospirosis. Water samples collected at the beginning of a rain event following a dry period contained high concentrations of pathogenic Leptospira. The concentration was generally correlated with the water level and the suspended matter concentration (SMC) during the main flood event. A secondary peak of pathogenic Leptospira was sometimes detected after the main flood and in slightly turbid waters. Lastly, the pathogenic Leptospira concentration was extremely high at the end of a wet season. The pathogenic Leptospira concentrations could not be explained by a linear combination of hydrological variables (e.g. the rainfall, water level, SMC and soil moisture). However, nonlinear machine learning models of rainfall data only provided a fair fit to the observations and explained 75 % of the variance in the log10-transformed pathogenic Leptospira concentration. A comparison of identical machine learning models for the water level, SMC and pathogenic Leptospira concentration showed that the residual error in the Leptospira concentration was due to not only the small dataset but also the intrinsic characteristics of the signal. Our results support the hypothesis whereby pathogenic Leptospira survive at different depths in soils and superficial river sediments (depending on their water saturation) and are transferred to surface water during erosion. These results might help to refine leptospirosis warnings given to the local population. Future research should be focused on larger watersheds in more densely populated areas.