In tropical countries such as Sri Lanka, where leptospirosis—a deadly disease with a high mortality rate—is endemic, prediction is required for public health planning and resource allocation. Routinely collected meteorological data may offer an effective means of making such predictions. This study included monthly leptospirosis and meteorological data from January 2007 to April 2019 from Sri Lanka. Factor analysis was first used with rainfall data to classify districts into meteorological zones. We used a seasonal autoregressive integrated moving average (SARIMA) model for univariate predictions and an autoregressive distributed lag (ARDL) model for multivariable analysis of leptospirosis with monthly average rainfall, temperature, relative humidity (RH), solar radiation (SR), and the number of rainy days/month (RD). Districts were classified into wet (WZ) and dry (DZ) zones, and highlands (HL) based on the factor analysis of rainfall data. The WZ had the highest leptospirosis incidence; there was no difference in the incidence between the DZ and HL. Leptospirosis was fluctuated positively with rainfall, RH and RD, whereas temperature and SR were fluctuated negatively. The best-fitted SARIMA models in the three zones were different from each other. Despite its known association, rainfall was positively significant in the WZ only at lag 5 (P = 0.03) but was negatively associated at lag 2 and 3 (P = 0.04). RD was positively associated for all three zones. Temperature was positively associated at lag 0 for the WZ and HL (P < 0.009) and was negatively associated at lag 1 for the WZ (P = 0.01). There was no association with RH in contrast to previous studies. Based on altitude and rainfall data, meteorological variables could effectively predict the incidence of leptospirosis with different models for different climatic zones. These predictive models could be effectively used in public health planning purposes.