Taiwan is a small, populous island. Because little flat land remains available for building, a trend in hillside development has recently been observed. During the storm season, especially during typhoons, heavy rainfall is common; however, the volume of rainfall is difficult to predict. Heavy precipitation causes flash floods as surface and groundwater levels overwhelm drainage systems. Thus people living on, or close to, hills are at risk of experiencing flooding with little warning.We investigated two models to discuss their differences and impact of rainfall in upland locations. The tank model describes and simulates the hydrology of an area, and the kinematic wave model simulates rainfall at upland locations. We used two natural rainfall events that occurred in the Tai-Chung basin as case studies for both models. We applied five statistical indexes, namely root mean squared error (RMSE), coefficient of efficiency (CE), percentage error of peak discharge (EQp), error in time to peak (ETp), and percentage error of total volume (EV) to evaluate and interpret hydraulic data and our simulation results. We determined both models to be applicable; the results showed the relative error in peak flow simulation for the upland tank model to be less than 10%. The total runoff forecasts for large-scale typhoon rainfall exhibited large errors in both models. Less severe rainfall events produced lower percentage errors in both models.The simulated hydrograph was compared with the measured hydrograph, and little difference was observed in the performance of the models. However, the simulation results of the kinematic wave model were slightly more favourable, possibly because the more complex wave model included more parameters representing physiographic factors, such as the river flow parameters P and K, equivalent roughness values of N, the average slope values, and the infiltration values, thus increasing the accuracy of the simulation. The upland tank model did not require complex physiographic factor parameters; instead, the Multistar-Powell method was used to model the upland tank model parameters to produce a simulated hydrograph, yielding results no less favourable than those produced using the kinematic wave model. In the absence of comprehensive measured data and physiographic factors, the upland tank model has practical uses. Copyright © 2015 John Wiley & Sons, Ltd. l'erreur à l'occurrence de la pointe de crue (PTE), et l'erreur relative de volume total (EV) pour évaluer et interpréter les données hydrauliques et nos résultats de simulation. Nous avons déterminé que les deux modèles sont applicables; les résultats ont montré l'erreur relative à la simulation du débit de pointe pour le modèle de réservoir de plateau peut être inférieur à 10 %. Les prévisions de ruissellement total à grande échelle généré par des typhons présentaient des erreurs importantes dans les deux modèles. Des événements de précipitations moins intenses produisaient des erreurs relatives plus faibles dans les deux modèles.L'hydrogramm...