I know there is an increasing emphasis to use models to provide meaningful reconnaissance of conditions that may lead to benefit various types of hydrological and geomorphological subjects.  A lot of flooding issues could be assessed with timely aerial photos to see channels adjustments, etc.  A lot depends not only on the quality of the photos and DEMs, but the knowledge of being able to read between the lines, ie, interpret what you see.  For may areas with forest canopy, LiDAR is the game changer on capacity for surface elevation detail.  Flooding issues should address land use but also with understanding of channel type and capacity, sediment movement and shifts in vegetation types, density, impervious surfaces, etc.  I would suggest you might look at some of Dave Rosgen's information on www.wildlandhydrology.com including free references on stream types and their flooding potential.  Also if you have extensive flooding due to gullies that have extended the channel system and lost contact with the floodplain, rehydrating bottomlands and forestlands or grasslands would help detain and store water and sediment to more natural processes.  As you suggest, a lot can be learned from remote sensing, but also a lot depends on the analyst(s) capabilities to interpret fine detail and change.  If you have headwater wetlands and valleys that have been channelized or gullied, you may find some opportunity to restore hydrology to more natural conditions and probably help restore various habitats.  Building headwater dams is some times considered, but they are costly to build and maintain, and if not done properly such as with overflow protection such as relief channels designed for severe storms and measures to lower stored water prior to severe storms, dam failure can cause severe impacts that are difficult to recover from.  I have a paper on a mitigation project (report 2013 or 2014) that involved removing a headwater dam that was poorly maintained, and it has some LiDAR photos that we did not have when we started the analysis.  I have also co-authored a couple of other papers and powerpoint uploaded to researchgate using LiDAR you might find interesting.  Since you are working on a public protection issue, you might find some community or agency support to help analyze and provide options to address these issues. 

I would think being able to determine the accuracy and precision needed for a 3D model to be useful would in-itself be a challenge.

In principal it is possible with some additional information necessary regarding interior orientation of the images and georeferencing.

What is it?

I very agree with William F. Hansen. Remote sensing technique can provide valuable information for topologic map and change detection. For your purpose, you should employ your professional hydrologic knowledge to cope with these remote sensing information. Nowadays, dense image matching has achieved a great progress. After rigorous camera calibration (internal and external), dense multi-view stereo can generate a 3D model that is comparable to LIDAR.  And the camera is cost-effective, maybe it is important for your monitoring system. But image matching can't  penetrate vegetation, and usually fail in  repetitive texture and weak texture, while LIDAR doesn't have these problems. So you should choose a proper remote sensing technique according to your actual situation.   

I think it would be an interesting topic.Dense matching produces more detailed point clouds with an aim of providing better accurate description of features.Factors that will affect successful and accurate dense matching in this case include presence of scratches and dirt ,illumination,texture,Ground sampling distance of your images.A consideration of these factors would be important in deciding which method to generate your point clouds with.

What is the accuracy and precision needed for a 3D model to be useful? 

Yes, it is possible. However, you must have proper orientation parameters (interior and exterior) to get accurate results.

In principle, the dense image matching should work with scanned analogue images (as it works with any images), if the images just have enough overlap between each other. The images may have various distortions due to scanning process and camera interior orientation, and all these will affect negatively to the final point cloud accuracy.

You would need either image exterior orientation info or ground control points to transfer the point cloud to real world coordinates. Or you could match the point cloud to an existing more recent point cloud.

But then if the point cloud created through this process would be usable and accurate enough for your hydrological and geological analysis is another question.

I have ground control available for each photograph, my proposal has now been submitted with a view to pursuing the process using Hirschmuller's Semi-Global Matching Algorithm. Thank you all for your informative and encouraging responses. I will no doubt be in touch again when I begin generating the point clouds. Again thanks for your responses.

With any available information about the analogue/scanned images, control points, and feature matching, you should be able to estimate the essential/fundamental matrix or the relative/absolute orientation which may then be used for image rectification. The rectified images can then be fed to the Hirschmuller's Semi-Global Matching Algorithm or other algorithms.