In moderately to steeply sloping terrain, DEMs from topographic maps can be useful in locating streams. Open terrain where stream can be seen also can be checked with high resolution aerial photos (infrared sometimes helpful). But the best is a DEM developed from high resolution LiDAR. Rivers and streams with high density tree cover and in relatively flat terrain such as coastal plain and river valleys are typically difficult to rely on in all circumstances. With LiDAR, getting to 10 m, 5 m or sometimes even 2 m resolution will help provide more accuracy and precision when georeferencing. One must remember that some stream and river dimension can change with time (e.g., erosion, deposition, landslides, anthropogenic modification, etc.). Braided stream types adjust frequently from excess sedimentation in low gradient terrain unless stabilized by firm vegetation to be anastomosed.

In GIS, there is a routine called change detection. If one has the actual ground truth stream coverage or recent high resolution LiDAR based DEM, one could compare this level of detail with the 100 m, 50 m or other DEMs developed from less intensive methods in developing DEMs.

You might look over some of my researchgate papers as some attempt to address and deal with aspects of stream and watershed delineation. In hydro modeling of stream networks, often the dominant channel is picked up as a single thread. Channels sometimes split, join, braid, have flood overflow channels, sloughs, oxbows, and/or develop sufficient widths to be mapped more than a single thread, varying in width, etc. Your question on quantitatively Identifying location might also extend to network extent, location relative to thalweg or channel centerline. Another approach might use pixels (varying densities as desired) that stream touches or flows through, and using various DEM resolutions, identify pixels properly identified, and then also count the number of errors of commission or omission.

you can select several points of your true river randomly and then compute the orthogonal distance from these point of your delineated streams (using both resolutions). Stats of these distances (i.e. avg, stdev...) will quantitatively measure the reliability of your delineation.

Alternatively you can use this qgis plugin

https://plugins.qgis.org/plugins/QuickSTEP/

regards,

Fulvio

Please refer to the following paper which employed two metrics for assessing the accuracies of DEM-derived river networks against a real one. Hope it helps

Hui Li, Jiayang Zhao. Evaluation of the Newly Released Worldwide AW3D30 DEM Over Typical Landforms of China Using Two Global DEMs and ICESat/GLAS Data. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 11, no. 11, pp. 4430-4440, 2018. doi: https://doi.org/10.1109/jstars.2018.2874361 Chang Liao

Estimated Chang Liao

I would test with morphometric indicators, such as channel order, total distance, sinuosity, gradient, average slope, drainage density, frequency (number of first order channels per surface), bifurcation ratio, etc.,

It would compare quantitatively with the values ​​of the "real" channel.

I hope it is useful.

regards,

Victor

Li Hui Thanks for the comments. The "Two indices, correctness (Cr ) and figure of merit (FM)" might be helpful in this case. I will take a look at them.

But I wonder what delineation method did you use in this study.

DDNs are delineated from the AW3D30 and these two DEMs using a standard GIS algorithm, including sink filling, flow direction determination, flow accumulation, and stream definition [42]. The priority-first-search algorithm is applied to fill the sinks [43]. The flow direction is determined using the deterministic eight-neighbor flow-direction (D8) algorithm [44]. The same CA threshold of 1000 cells is used to define the streams for all the DEMs to facilitate the comparison.

Meanwhile, RDN is manually delineated across the watershed from Google Earth (GE). Chang Liao