Please have a look at some interesting PDFs....

Generally I would suggest it is very difficult to use and rely on GIS or associated models without also incorporating some precipitation, flow, geology, soil, and stream data. Some countries have probably fair to good coverages to work with to apply to various problems, but one must remember, torrential flood risk is a life threatening and property damage phenomenon that deserves review of risk maps or zonation by certified and licensed professionals to help cover the liability issues that can be involved. That being said, I would suggest that LiDAR coverage in its ability to develop high resolution DEMs is a very helpful tool for experienced hydrologists, geologists, geomorphologists and civil engineers to estimate stream channel, slope, gradient, sinuousity, landform detail, etc. At best, a successful hazard model needs some degree of ground truthing and combining appropriate streamflow, rainfall, channel morphology, flood elevation indicators, and for torrential conditions, there may be data or observations suggesting that past torrents have left indicators of movement of boulders, areas with excessive sediment or woody debris deposits, and other indicators. GIS can be such a convincing tool by the quality and detail of outputs and products, that it can mask weaknesses. It is important to publish the limits to help others understand uncertainty involved.

A couple more ideas. Channel gradient and valley gradient can be important, considering if you have found areas with high rainfall intensity, such as mountains that face incoming moist air masses, and the elevation change forces uplift. Cchannels can have high gradient change, but if the gradient change is falls(abrupt elevation drop), most of the energy is lost. Cascades have continuing steep channel and narrow valley with high grade, able to generate standing waves, move boulders, etc. These produce and deliver torrential flow, with ability to move boulders if enough flow, undercut streambanks if not well ptotected or vegetated, perhaps expose bedrock, bank trees maybe tipping or fall into stream as undercut, so there should be indicators.

Braided channels are evidence of high sediment loading, and the channel has lost capacity to move sediment and water, so flooding may be more frequent than expected. Braided streams are often lower gradient and not necessarily torrential, but may be present below torrential conditions, where grade abruptly reduces, forcing sediment and debris to deposit. When a braided stream has a narrow confined valley, the cross section area is confined with limits on floodplain, so velocity during flood stage would be higher than braided stream in a broad valley. If you can generate estimated flows for 50 or 100 year flood, you can estimate that the channel is apt to carry the 1.5 year flood, so the remainder of the flow must be carried across the valley, unless the channel is entrenched like a gully. HECRAS could be used with some basic channel and valley morphology to estimate flow depths, cross section area, and mean velocities. Using the rule of 10 for wading safety, any conditions where velocity in feet per second times depth in feet equal or exceed 10 is unsafe to wade. As depth increases, also a bouyancy factor develops. Whether torrential or not, this simple rule may be enough to help estimate hazard zones. Driving through water is especially succeptable to buoyancy issues as the water depth increases, so the factor of 10 for waders does not apply. Again, hazard conditions for driving through water are much less than torrential flow, and for wading. Nonetheless, a hazard that publically is not well understood.

I believe William F. Hansen has given you a very good overview of th factors involved when studying a flood event. As he pointed out, it is important to couple methodology. GIS tools alone won't give you as much information as when it's coupled with numerical modelling (e.g. HEC-RAS) and statistical analysis. It is nearly impossible to have a good picture of what's going on in a flood without analyzing it with diverse approaches.

I've recently learnt about the wading safety rule, by the way, it is used in Civil Engineering and I consider it to be a good criteria when it comes to establishing hazard classes for GIS representation of results.

Also, make sure you have data on sediment and wood transport, as this is key to the flood's impact. These are a few references that might help you. GIS is used there:

Thesis Streambank erosion and channel widening: implications for flood hazard

Article Potential large woody debris recruitment due to landslides, ...

Chapter Factors controlling streambank erosion and channel widening:...