Abstract. In the field of rainfall-induced landslides on sloping covers, models for early warning predictions require an adequate trade-off between two aspects: prediction accuracy and promptness. When a cover's initial hydrological state is a determining factor in triggering landslides, taking evaporative losses into account (or not) could significantly affect both aspects. This study evaluates the performance of three physically based predictive models, converting precipitation and evaporation fluxes into 10 hydrological variables useful in assessing slope safety conditions. Two of the models incorporate evaporation, with one representing evaporation as both a boundary and internal phenomenon, and the other only a boundary phenomenon. The third model totally disregards evaporation. Model performances are assessed by analysing a well-documented case study involving a two-meter thick sloping volcanic cover. The large amount of monitoring data collected for the soil involved in the case study, reconstituted in a suitably equipped lysimeter, makes it possible to propose procedures for calibrating and validating the 15 parameters of the models. All predictions indicate a hydrological singularity at the landslide time (alarm). Comparison of the models' predictions also indicates that the greater the complexity and completeness of the model, the lower the number of predicted hydrological singularities when no landslides occur (false alarms).