Can use Satellite Imagery of the same location at  different points to study to observe historical trends. Factors include water (flood), wind and even human activity. The soil structure on the banks also plays a role. Where there is limestone present, erosion is faster, and in sandy soils, the erosion is more. River Training has been tried through construction of embankments, dykes, groynes & spurs. 

It depends on your research question. If you want to do a quantitative study of a particular river bend with a view to identifying rates of erosion in a specific place (for example), then you'd need to use a more complicated model, something like a computational fluid dynamics model. If you want to do something that is a bit more conceptual to understand something about how rivers of a given type behave given certain drivers then a landscape evolution model (like CAESAR) would probably be suitable.

Parameters will vary depending on the model, the more complicated the model, the more specific the parameters need to be. For example in a simplified model it may only need grain size distributions/summaries of the bank material and bedload, and from that it might infer/summarise things like bank strength. WHereas in a more complicated, physically explicit model, you may need measurements of cohesive bank strength. Likewise depending on model you may need hydraulic geometry, slope and discharge and the model might infer/calculate transport parameters like near bed shear stress, whereas other models might not specify the flow, rather they would need the shear stress specified as a parameter.

In very simple terms though you need to know how resistive the bank is to erosion and how much force/energy the flow of the river has. At it's most basic level its a physics problem of force vs resistance. It just depends how much this problem is simplified or complicated (i.e. spatially explicit) as to how these parameters would be specified in the model

Dr. David Rosgen has developed a BANCS Model, that has to be validated to your conditions, which may be useful.  It uses boh qualitative and quantitative measures.  For detailed work, total station or ground based LIDAR are needed, bank pins, bank heights, bank stress, substrate type, vegetation type, root density and depth, etc.  For extensive work and time series, remote sensing with aerial photos may be helpful if view of stream is not obscured.  LIDAR from remote sensing is also useful, with some recognized limits.  Rosgen's various references and books are available on his website www.wildlandhydrology.com.  Rivermorph is a software that has been used to store  various typs of stream, channel, bank and other data, also useful.