On the face of it the modelled hydrograph plots you provide look quite sensible. In the first, the ratio of the flood peaks mirror those of the rainfall peaks, where both of these are preceded by 12 or more hours of continuous rainfall. In the second plot, the first rainfall peak is over twice the size of the second, yet the observed response is greatly dampened compared to the modelled. You would appear to have adopted a time step that is suitably short compared to the period of hydrograph rise, so that’s not a problem. You don’t state what loss model you are using but from the plot it seems that you are using a IL/CL model, and it may be that you will get better agreement from adoption of an IL/PL model (though of course that does introduce problems when using the model for design purposes). You could experiment with an IL/PL model to help satisfy yourself that the rainfall data is consistent with the observed hydrograph.
My only other suggestion is to revisit model configuration. There is very little lag between the rainfall and the hydrograph centroids, so this is either a very small catchment or an urban one (or both!). Thus it may be that you need to review the length and nature of a flowpath that may be directly connected to the catchment outlet.
Anyway, my overall impression is that there are a number of things you could look at, but there is nothing obvious here to “tweak” – your difficulties reflect some portion of the catchment that is behaving differently to your understanding, or else inconsistencies in the data.
Don’t you love hydrology?