Watershed Analysis of the Tibetan Plateau

          The Tibetan Plateau has some of the world’s largest group of high altitude endorheic lakes and it is very important to analyze water balance in these lakes. This lab introduces us on how to perform a watershed analysis in order to derive topographic features such as stream features. A watershed is an area that drains water to a common outlet as concentrated drainage. In order to create a watershed analysis the step that need to be taken are as followed: make a fill DEM, derive a flow direction raster, derive a basin, derive a flow accumulation raster, derive a stream network, stream order and then convert that into a feature.

          The first step in creating a watershed analysis is getting data. Luckily for this lab, the data was provided on the S: drive. The data we need is a DEM and a lakes layer of the Tibetan Plateau. Also provided was a Landsat image for later comparing an analysis with the final map. The DEM might have some error that will affect the outcome of the watershed so we need to create a DEM that has filled in any depressions. Located under hydrology under the spatial analysis section in the toolbox in ArcMap, I used the fill tool to create a DEM that has its depressions filled. I set the z value as 15, instead of using the default. This puts a limit on how many meters can be filled up. Now that I have an “error free” DEM, I can create a flow direction raster with the input being the filled DEM. This shows the direction water will flow out of each cell. There is a special formula using an eight-neighborhood grid in calculating which way water will flow. You take the elevation number in the middle and subtract that number from the numbers around it. The number with the highest remaining number is the direction of where the water will flow. However the different cells are weighted differently. The immediate neighbor cells are divided by 1 and the corner neighbors are divided by 1.414. This is how the water flow direction is determined in ArcMap.

          The next task to do in the analysis is to create a basin analysis. By using the direction flow raster as the input feature, I created a basin raster using the basin tool found under the hydrology tab. However, we do not need it to be a raster so from there I changed the raster basin into a polygon feature and made the color hollow with black outline to show where the drainage basins are located and where the water flows into. Next, I created a flow accumulation using the flow direction raster as the input. The flow accumulation raster tabulates for each cell the number of cells that will flow to it. From the flow accumulation raster I can now create a stream network by applying threshold value to it. I chose to use a threshold value of 500 (meters). In order to do this I went into the reclassify tool in the spatial analysis. I chose to reclassify the flow accumulation into equal interval and one class. I kept the maximum value at the value it has and then I changed the zero to 500. So it read 500-1469259. Before I set the output and clicked okay, I made sure that the “change missing values to NoData” was checked. I now had a reclassified flow accumulation and a stream network. Next, I created a stream order, which is a method for identifying and classifying types of streams based on their number of tributaries. I used the stream order tool under hydrology tool and used the reclass of flow accumulation as the input stream raster and input the flow direction as well. The method for stream order I used was Strahler’s method. So far I have made a fill DEM, a flow direction, a basin polygon feature, flow accumulation, and a stream order.

          In order for the stream network (the reclassified flow accumulation raster) to look like kind of real streams I had to convert raster to a polyline feature. Now that I have all the necessary features and raster data to see the watershed and do an analysis, I can compare it to thedata I downloaded offline from the http://hydrosheds.cr.usgs.gov. From this website I downloaded 30 second DEMs of the research area of the Tibetan Plateau and also flow direction, basin polygon and stream layers. The comparison and analysis is explored in the next paragraphs.

          From the HydroSHEDS website I downloaded the available data sets: DEM, flow direction, streams, and the basins. In the top output maps I have my final watershed analysis and the HydroSHEDS analysis on the first page, showing the flow direction, basins and streams combined, along with the landsat image to compare. However, I will discuss the maps in detail step by step on the second page.

First maps to compare on the second page are the filled DEMs.(I’m not sure how far you can zoom in) If you look closely, you can already see that my fill DEM is more detailed than the HydroSHEDS. This may because my fill DEM is at a lower second. The fill DEM I downloaded was a 30 second fill DEM, whereas my fill DEM is probably a 15 second of maybe even a 3 second. Also you can tell that the resolution of my fill DEM is at a lower resolution than HydroSHEDS DEM. This may be because the seconds are lower and more detailed. The resolution and quality of the fill DEM will affect the detail of the final watershed analysis. Also, there could be some differences in the z limit that was used creating the fill. The next map I produced and compared was the flow direction. Here, it looks as if my map is more detailed and HydroSHEDS flow direction is more generalized. This is obviously because the 30 second fill DEM is more general than my fill DEM, resulting in a more general flow direction. Furthermore, there are discrepancies using GIS when it comes down to what algorithm was used in computing the flow direction. It could have been the D8 algorithm which uses the eight neighboring cells, or the D-infinity algorithm could have been used, which uses a infinite number of directions. This could cause problems on choosing which algorithm to use.

           The basins in each map differ from one another. My basins map is again, more detailed while the HydroSHEDS basins are more generalized. Again, this is most likely because the fill DEM to create the flow direction was a 30 second fill DEM and less detailed than my fill DEM. Therefore it created a less detailed flow dire3ction raster. As a result, the basins were also less detailed. Therefore, by having a more detailed fill DEM, my resulting maps were also more detailed. The streams polyline feature is another example of a more detailed result of the lower resolution DEM. As you can see my streams feature is more detailed then HydroSHEDS streams. This is because HydroSHEDS streams were created from a 30 second DEM and mine was created from a 15 second of maybe even smaller a 3 second. This is the reason for some discrepancies. Furthermore, when creating the streams by reclassifying the flow accumulation, I had to put a threshold value. I put 500, but the HydroSHEDS user could have put a different threshold. This could also lead to the differences between my results of the streams and HydroSHEDS streams.

          As you can see there are many discrepancies when doing watershed analysis and it all depends on what data the map make decides to use and what he/she decides to put as the z value or threshold value. This will make the maps look different and unique. Furthermore, we can use the landsat image as an informational tool in validating the extracted drainage networks. As you can see in the image, shows in red the lakes in which the water drains into. It also shows the contours of the land showing how the water drains down into those lakes.

By

Chelsea Kemp

May 9, 2011