Lahars impose significant secondary hazards on highly populated volcanic islands by remobilizing volcanic ash deposits. Karthala, on Grande Comore Island, is a relatively young and poorly eroded basaltic shield volcano with sporadic occurrence of ash-forming phreatic eruptions. In 2005, two mildly explosive episodes emplaced tephra blankets on the summit area. Heavy precipitation subsequently triggered the repetitive occurrence of small-volume secondary lahars up to 2012. These lahars damaged roads and hundreds of houses, affecting thousands of inhabitants at the foot of Karthala volcano, but were poorly documented at the time of their occurrence. Their future hazard remains unclear as well. This study aims at gaining insights into the main characteristics of these lahars, as well as testing and comparing the effectiveness of two numerical tools to simulate the extent of these small-volume lahars. To understand the physical characteristics of the lahars, we first documented the spatial extent and characteristics of the debris deposits at the foot of Karthala volcano and in the ravines that guided the flows. Our observations suggest that the debris were emplaced by small-scale (volumes ≤ 105 m3), rain-triggered and predominantly low sediment concentration lahars. The spatial extent of the deposits served to calibrate and compare numerical lahar simulations from the widely used volume-limited LaharZ model with simulations from Q-LavHA, a probabilistic flow model originally developed for lava flows. Q-LavHA mitigates some limitations of LaharZ, such as its ability to simulate flow bifurcations and the transition from constrained to unconstrained flow but comparison demonstrates that Q-LavHA typically yielded lower simulation accuracies compared to LaharZ simulations. The obtained accuracy values represent a rather poor performance for both models compared to existing studies on larger-volume lahars on stratovolcanoes, and are inferred to result mostly from difficulties in delineating lahar flow paths on the smooth, poorly eroded topography of the volcanic edifice. We therefore also evaluated the potential to increase simulation accuracy by updating a 10 m resolution Digital Elevation Model (DEM) with channel topography measurements. By using such updated DEMs, the correctly delineated area improved for both models. This approach, however, did not prevent simulations to sometimes miss the hazard-prone position of settlements which were actually affected by the hazard in the past. Our study shows the limitations of numerical volcanic flow simulation strategies on young and poorly eroded volcanic edifices, such as active basaltic shields. The results indicate that accurate topographic representations and detailed documentation of spatial extent of the impacted area and lahar deposit thickness are needed to produce accurate lahar simulations, as well as the further adaptation of existing numerical simulation tools to better suit these particular environmental settings.