Flow in complex karst aquifers is challenging to conceptualize, especially in poorly investigated areas, in semi-arid climates and under changing climate conditions. Based on more than three years of high resolution continuous monitoring, a semi-distributed lumped model (MikeShe 2016) was calibrated and validated for the Qachqouch karst spring, North of Beirut (Lebanon). Time series analyses and decomposition of spring hydrographs showed that the system has a high regulatory behaviour with a considerable storage capacity providing a stable flow depletion (minimum flow of 0.2 m3/s) during the entire dry season, with flow rates reaching more than 10 m3/s during the wet season, similarly to other karst aquifers in the region. Acquiring this detailed understanding of the hydrodynamic of the system allowed refining and validating the developed semi-distributed conceptual model of three linear reservoirs used to reproduce the combination of the contribution of the different flow components of the systems, cause of the high contrasts of flow rates in the spring discharge. The obtained satisfactory precision (Nash-Sutcliff index 0.7) is due to the use of sensitive parameters for the model calibration deducted from the performed time series analyses, such as the number of reservoirs and their time constant. Climate change conditions (+1 to +3°C warming, -10 to -30 % less precipitation annually, and increased intensity) were applied to an average climatic year to produce scenarios of expected spring flow responses. The comparison of future simulations to the baseline scenario showed that Qachqouch karst aquifer tends to be sensitive mostly to rainfall decrease, leading to more important recessions with flow rates decreased by 34% and duration dry periods approximately one month longer. Since the influence of snow on the spring flow rate revealed to be minimal, a warming effect on climate exacerbates to a relative lesser extent the spring conditions than loss in precipitation. Although the model shows that increasing daily rain intensity creates higher flood flow that could lead to longer recessions, thus slightly reducing the length of the low flow period in comparison with the baseline, the real impact of high intensity precipitation events is uncertain. This work shows that calibrating a semi-distributed lumped model using time series analyses seems to be an efficient approach to enhance the quality of simulations for complex karst aquifers. This is expected to be especially true for karst aquifers where subsurface characterization is very difficult, thus providing a more detailed model for a better long-term sustainable water management.