A watershed and drainage network from a Digital Elevation Model (DEM) of the Tibetan Plateau.

The goal of this exercise is to delineate a watershed and drainage network from a Digital Elevation Model (DEM) of the Tibetan Plateau. There are many lakes in the plateau, and knowledge of watershed boundary and drainage is important to conduct water balance exercise. I used Spatial Analyst geoprocessing tools in ArcGIS for this exercise. As we will see, watershed delineation and drainage extraction depend on scale a lot. At higher resolution of analysis, lots o details on watershed boundary and drainage lines may be present that may not be needed. At low resolutions, important details could be missing. The key is to strike a good balance by selecting parameters that will produce watershed that closely matches those on the ground. There are many third party sources of maps that can be used for this comparison.

We downloaded a raw or unprocessed DEM of 90 m resolution from SRTM (Shuttle Radar Topography Mission) web site. However, there were issues with the raw DEM that we downloaded from class website, it was producing watershed with many details. So, a filled DEM was provided to us from the lab, so for this exercise we will use the filled DEM.

ArcGIS Spatial Analyst toolbox has many tools for watershed delineation and drainage extraction. To get the process started, I used the flow direction GP tool to compute the direction of flow of water. ArcGIS uses the D8 algorithm to calculate the flow direction. In one of the papers that Prof. Sheng gave us to study that compared flow direction algorithms, we discovered that D4 algorithm was the most accurate and efficient algorithm, but ArcGIS does not implement that, so we will have to use the D8 algorithm, because this is fine for most cases, and is also simple to use.

Then, I used the flow accumulation GP tool to calculate the accumulation raster for Tibetan Plateau. Cells in the flow accumulation raster get assigned a number that indicates the total number of cells actually contributing flow to that cell. Greater the number assigned to cells, this indicates a cell that is more likely to be a part of drainage network.

The next step is to derive a stream network from the flow accumulation raster. A GP tool to reclassify the raster was used to apply a threshold value to the flow accumulation raster. The raster cell values were classified in two classes. One class had cell values between 0 and 500 (which is the chosen threshold value), and this class was assigned a value of NoData. This means that those cells whose values are less than 500 will not be considered in building the stream (or drainage) lines. The other class had cells with values between 500 and maximum cell value. These cells will participate in building the drainage line. Selection of a correct threshold value is important. Several threshold values need to be tried, and the stream line needs to be compared with the actual stream network on the ground. The threshold value selected for this exercise is

500, and probably more values should be tried to finalize the stream line, but that exercise may be time consuming.

Stream links were derived using a GP tool and each link was assigned a unique value and a flow direction. The extracted stream line was then assigned a stream order by using the Stream Order GP tool. The default method by Strahler was used. Now the Stream to Feature GP tool was used to convert the raster stream to vector stream features. For comparison of drainage network, imagery can be used. Satellite image mosaics are available at http://glcfapp.glcf.umd.edu:8080/esdi/index.jsp, and were used as reference.

The final step was to create a watershed basin using the filled DEM. This was compared with watershed data downloaded from http://www.cger.nies.go.jp/db/gdbd/gdbd_index_e.html (Global Drainage Basin Database) and http://hydrosheds.cr.usgs.gov/ (HydroSHEDS).

In conclusion, I find that my stream network is more detailed than the reference image available

at the above websites. This implies that a better threshold value needs to be selected. So there is a need to invest more time in selecting this value, so a more realistic stream network can be delineated. On the other hand, the watershed basin derived from the filled DEM appears reasonable which means that z limit of 20 m used to fill the DEM was a reasonable selection. A DEM of 90 m resolution can at best be described as coarse resolution. Higher resolution will

help in delineating more accurate watersheds and drainage networks. However, selection of z limit for filling the DEM, and threshold value for extracting streams remains a critical step, and must be given due attention.