Do you mean that the depths are at single points along the river, as opposed to along transects?
If that's the case, you could still get a basic interpolation of depth using a TIN, but the resulting surface will be V-shaped, and likely not accurate. Still, it could give you a good sense of depth along the river, if not across. To get a more accurate shape to the bottom, you'd need more points added.
The other thing to keep in mind is the shape of the river. A TIN probably won't generate correctly around a bend if the river is quite sinuous, unless you've got points at fairly frequent intervals in those spots.
HEC-RAS, while it is made to work with transect data, has the ability to interpolate depth across a transect and along the river. Its particular advantage here is that it treats the river or transect as being a one-dimensional linear object for interpolation, and the results of that are "projected" onto the two-dimensional shape. But you have to ask, to quote someone I know from Florida, "Is the juice worth the squeeze?" Learning how to use the USACE's hydro-modelling software for data so sparse may not be.
In my opinion, you probably ought to see if there are ways to augment the data you've got. What area are you working in? You might check if there are any bare-earth (not hydro-flattened) DEM or LiDAR datasets available to try and fill in some missing data points.
And if there aren't other sources of data for the river's shape, what level of approximation would be acceptable to your project and the integrity of the results?
- Josh Carlson
Kendall County GIS