The three QPF storms were chosen to represent landfalls in three locations on the island: Soulik (2013) on the northern third, Morakot (2009) on the middle third, and Tembin (2012) on both the southern third and moving parallel to the coast. Simulations were run for all initialization times where the typhoons began located upstream of Taiwan and crossed over within seventy-two hours, resulting in fifteen model runs. The cross-track errors divided rather evenly between being large while approaching the island, becoming large while interacting with it, and being small throughout.
When the track is simulated well, COAMPS-TC is accurate in reproducing the pattern of observed heavy rainfall but the amounts can vary from gauge observations. These differences can be very large (>1000 mm), but when a threshold for debris flow is used the model has very high probabilities of detection (>0.95) with the more significant operational forecast problem being higher false alarm ratios. Replacing the set threshold with results from the HL-RDHM model is also discussed.
Following an ingredient-based framework, the rainfall rates and durations were examined for stations with large errors. Since the rates in COAMPS-TC were often higher than observed, simulations were run for three independent typhoons to examine the effects of the terrain on the intensity. Sinlaku (2008), Jangmi (2008), and Nanmandol (2011) also represent landfalls in three regions of Taiwan. In these cases, terrain effects appear to be more important than environmental factors (e.g. wind shear) in weakening the storms near landfall. Since this deintensification process is quite complicated, these simulations will likely serve as a guide to setting up idealized simulations to study the basic model dynamics.