Predicting occurrences of landslides or debris flows at a specific slope is critical for early warning. Instead of the groundwater height, discharge of a spring at a slope adjacent to a primary school was observed and simulated by a functional model, which includes long‐ and short‐term precipitation terms. The long‐term precipitation term was determined by fitting the observed spring discharge, whereas the short‐term precipitation term was determined so that the model yields significantly large spring discharges during recent disasters in the district. The comparison between the observed and calculated discharge showed 0.8 of the Nash‐Sutcliffe efficiency factor. The simulation of the spring discharge from 1976 to 2020 revealed that the calculated discharges during the recent disasters were within the greatest 0.1% of the entire period. These results suggest the robustness of our simple functional model and less failure of the disaster prediction.