Valley boundary delineation

[cforgaci]

• [x] get Copernicus 30m DEM for Bucharest
• [ ] try existing tools (not necessarily in R) to determine valley edges
• [x] try to get code/workflow description for Vermont River Corridor delineation

[cforgaci]

A stepwise GIS approach for the delineation of river valley bottom within drainage basins using a cost distance accumulation analysis

The cost distance accumulation algorithm described in this paper can be used to delineate the valley bottom.

Steps:

1. Depressionless raster: eliminate localized peaks and sinks

2. Conditioned slope raster: replace values of 0 with small positive values

3. calculate cost distance accumulation within the drainage basin from river center lines

4. extract the river valley bottom from the cost distance accumulation raster based “on calculating an estimate of the threshold cost distance accumulation that corresponds to the boundary of the river valley bottom using wetland areas that are adjacent to rivers.”

   1. extract wetlands through a spatial analysis of proximity between input wetland and river layers.

   2. filter above zero and max. 500 CDA values

   3. calculate the mean of the resulting values to be used as a threshold CDA for extracting the river valley bottom

5. Repeat for all drainage basins involved

ArcPy code available upon request: [gasper.sechu@agro.au.dk](https://github.com/orgs/CityRiverSpaces/projects/1/views/gasper.sechu@agro.au.dk)

[cforgaci]

Valley delineation:

• This is only run when the specific option of cost distance is chosen. In some cases, like flat cities, this might not be helpful and another alternative might be preferred.

• Data:

  • The DEM dataset that is globally available has a 30m resolution. We strart with this. Is this detailed enough? What is globally available is a DSM, not a DTM; is this good enough?

  • How does this work with flat cities? Is high resolution more needed in those cases? After

  • Data availability, considering resolution and quality

    • This is available in the Netherlands and Austria (see data for Graz, for instance) at a high resolution, DSM and DTM, 0.5m. In Italy, (see data for Ferrara, for instance) 1m resolution is available. However, this is not the case with other countries.

    • On European level, 10m is available from EEA/Copernicus. We have to check access to data. We can fall back on this geographic extent if the 30m is not enough.

  • Why are wetlands are used as input to expand river's active area in the valley? Consider using the OSM tag nature:wetland

• Algorithm:

  • Distance accumulation in ArcGIS, also used in the Vermont Env Agency's delineation method

  • A stepwise GIS approach for the delineation of river valley bottom within drainage basins using a cost distance accumulation analysis using ArcPy

• Output:

  • Masked raster or polygon?

Data sources:

• MULTI-MODAL GEOSPATIAL AND THEMATIC DATA TO FOSTER GREEN DEAL APPLICATIONS workshop data

[cforgaci]

This is a secondary task under coridor edge delineation. We explore this further while focusing on the corridor edge delineation.

[cforgaci]

@cforgaci checks why wetlands are used in the delineation method.

[cforgaci]

Wetlands adjacent to rivers are used to calculate "an estimate of the threshold cost distance accumulation that corresponds to the boundary of the river valley". Wetland coverage is limited in urban areas and might not be necessary on a the scale of a city.

[cforgaci]

@fnattino at the useR conference I met someone from the GRASS GIS core team who recommended r.geomorphon for our valley delineation task. Let's explore this too. GRASS GIS can be accessed in R via the rgrass package.

[cforgaci]

@fnattino at the useR conference I met someone from the GRASS GIS core team who recommended r.geomorphon for our valley delineation task. Let's explore this too. GRASS GIS can be accessed in R via the rgrass package.

I am hitting several walls with this:

1. Installation of GRASS GIS has proven to be difficult as I cannot make the environment variables and hence the communicaiton between rgrass and GRASS GIS work. So, I am afraid introducing this dependency is problematic and undesirable.
2. I checked the output of the r.geomorphon algorithm from GRASS GIS and its SAGA equivalent Geomorphons, but I do not manage to detext the valley. This issue might be due to the local scale of the elevation model and because it is not a DTM.

I would not rule out this approach entirely, but I am less confident in it now. Let's discuss this in our next meeting.

[fnattino]

With #14 , I think we have now understood a clearer understanding on the difference between how the costDist function from the rspatial/terra package works and "the Vermont method", which makes use of the PathDistance tool from ArcGIS (now deprecated and replaced by the Distance Accumulation).

Both determine the minimum accumulative cost from a source (the river bed) to each cell location. However, costDist multiplies the values in the "friction" raster cells by horizontal distances, while the path distance tool multiplies the friction by actual surface distances (as determined from the horizontal distances plus the elevation). Now, when using the terrain slope as friction raster, I would argue that the use of the cost distance tool produces an output that is more physically sound than the path distance tool: the output raster can be really interpreted as accumulated elevation across the path with minimum slope. From what I understand from reference [1] below, the friction employed in the path distance tool should instead reflect other types of impedance (in the example, surface roughness) - the terrain shape is already accounted by the fact that actual surface distances (and not horizontal distances are used).

This resources explain the difference between the cost distance and the path distance as implemented in ArcGIS: [1] https://pro.arcgis.com/en/pro-app/latest/tool-reference/spatial-analyst/understanding-path-distance-analysis.htm [2] https://pro.arcgis.com/en/pro-app/latest/tool-reference/spatial-analyst/cost-distance.htm

[fnattino]

But getting back to the use of cost distances for the delineation of the valley edge for CRiSp:

• costDist using the slope as a friction, could be effectively used to delineate the valley (as shown in #14). However, this approach defines the valley on the basis of a parameter that represents the threshold in elevation gain with respect to the river bed. I feel that this is not optimal, as this parameter might assume very different values for cities with different topographies (e.g. for a flat city, even a threshold of few meters would produce very wide valleys).
• the Distance Accumulation tool from ArcGIS would be better suited if one would use the following settings: no friction (cost) raster, use the DEM as input surface raster. The output raster would then represent only actual surface distances from the river bed, and the threshold to define the valley would be a distance threshold that accounts for elevation. This approach would work for both hilly and flat cities and one could use a consistent parameter for the analyses. Unfortunately, I haven't found implementations other than the ArcGIS one.

[cforgaci]

Working with the cost distance approach and the threshold set as in https://hess.copernicus.org/preprints/hess-2020-361/ we approximate the valley edges reasonably well. We still need to test to what extent the resulting valley polygon leads to a correct delienation on the street network.

The downside of this approach is that the cost distance threshold is manually fixed for the entire valley, which means that it cannot account for changes in ridge height along the valley.

If results are good enough, we use this method for now, but in the long term we will consider a different valley detection method that can overcome differences in height along the valley. This will probably work on a smoothed-out raster and should automatically determine the height of the valley.

Replacing the manually set threshdold with this automatic method is also essential for the scalability of delineation, because variation accurs not only along one valley but across cases (different valley depths) as well.