12 February 2003
Dear GIS User
TIPS - DEM and Hydrological Analysis
The motivation to write these "tips" came from my e-mail inbox being filled with various questions and requests for the assistance. It has happened regularly since I made ArcView Basin1 Extension available to GIS public via Environmental Systems Research Institute (ESRI) users scripts web site in April 2000. In the majority of cases, the solution referred to the improvement of the quality of elevation grid used for the analysis. It gave me the thought that many users underestimate or do not fully realise and appreciate the effect of DEM quality on accuracy of information generated from the DEM. Other users who perform hydrological analysis using DEMs may hate to read user manuals, instructions and online help - just to get results as fast as possible. Most of them usually trade the time saved such a way, for finding (hard way) that results of their analysis are beyond the reasonable limits.
Obtaining results with the sufficient accuracy, i.e. accuracy that is acceptable in terms of the purpose for which results will be used, requires both the DEM of quality that supports the required accuracy and appropriate tools to obtain the results. Using poor and insufficient source data to create DEM results in DEM of inferior quality. Thus, one can not expect good outcome using such DEM. The hydrologically correct elevation grid with sinks filled and a suitable cell size is the precondition for performing hydrological analysis successfully and for generating meaningful geographical and flow related information from the DEM.
For example, just few sinks in DEM may have disastrous effect on results. The stream network generated from the DEM may be disconnected and have parts missing. Or the path of raindrop traced from the user-defined point may stop in the first sink in its way instead of continuing to an outlet. And so on…. These are just two typical effects of the DEM that has sinks. Different criteria, which apply in special cases of karst topography where sinks are natural occurrence, are discussed below.
Surface of the DEM, in general, differs with the interpolation technique and types of input data used for the DEM creation. The recognised way to create a hydrologically correct DEM is by using TOPOGRID command and its subcommands in ArcInfo GIS. As minimum input data, both elevation coverage and stream coverage is required. The stream coverage must comply with requirements stipulated by the TOPOGRID (arcs oriented downstream, single arcs, and arcs connected in nodes to create network). If one creates the DEM using ArcInfo TOPOGRID command from only a point or contour coverage (or both), this DEM is not hydrologically correct simply because the stream network coverage has been omitted as input for the DEM creation. As result, the stream network on a map will differ from one generated from the DEM. This applies for any technique of DEM creation that omits stream network as input (including DEMs created from satellite imagery and aerial photography). Another factor adding to the inaccuracy of stream network, generated from the DEM, might be small map scales (e.g. 1:250 000 and smaller) and low density of input data used for DEM creation. Because of the small scale, the optimal DEM cell size (inversely proportional to the scale of input data) might be too large.
The TOPOGRID command incorporates a drainage enforcement algorithm for automatic removal of spurious sinks and according my experience, if both stream and lake coverage are included as the input data for the DEM creation, it gives better results in hydrologically correct DEM creation than any other method. The TOPOGRID allows the input of other various data types and user defined tolerances to further enhance the quality of DEM.
The incorporation of a stream network in TOPOGRID command removes sinks which would not otherwise be removed by the automatic drainage enforcement algorithm and corrects remaining drainage anomalies in elevation grids. This is required by hydrological analysis where accurate position of streams generated from DEM is of great importance. ArcInfo online help tells us that stream data always take priority over point or contour data in TOPOGRID command. Thus elevation data points which conflict with strict descend down each stream line are removed, with all conflicts which exceeds the user specified elevation tolerance being flagged for possible correction of errors. In practical terms, the stream coverage obtained by digitising or scanning and vectorising for example from 1:50 000 maps should be to define major drainage lines, leaving the drainage enforcement algorithm to define the lower order streams which are not available in maps. The level of detail is controlled by the amount of input data available and by the specified resolution of the DEM. Experimenting with different cell sizes of DEMs using 1:50 000 data for their creation has shown that the 20-meter cell is optimal. It allows generating contours and streams almost identical with those used as input.
Comparing stream network generated from the DEM created with stream data and from the DEM created without stream data, one may notice that there is great difference in locations of streams. This difference is more pronounced in flat areas. Another important fact is that cross sectional profiles drawn across river valleys differ, sometimes greatly. The general rule is that cross sectional profiles extracted from the DEM that was created without using stream data are flatter than those from the hydrologically correct DEM.
Another important input to TOPOGRID, which should not be omitted, is lakes coverage. Within a lake, the elevation data is usually not available. The procedure assigns the average elevation value of all cells along the shoreline to all output cells within the lake. If one omits for example a dam reservoir, the surface in this no data area will be interpolated by TOPOGRID based on data outside the lake boundary. Then the catchment boundary delineated for the gauging station at a dam wall may pass through the reservoir, particularly if it is situated in mountain area, and generated contours may not follow the real reservoir boundary but intersect it in many points.
In special cases of karst topography sinks are natural occurrence and affect the flow path. Here, depending on intensity and duration of storm events, sinks are filled to different levels or overflow and may determine more then one distinctly different stream network. In such a case there is a need to input point coverage of sinks during DEM creation using ArcInfo TOPOGRID's SINK subcommand. The program will not attempt to remove sinks provided in this coverage. If the sinks overflow in more then one direction, depending on attributes of a particular sink and storm event, modelling to predict the amount of overflow and where the overflow would travel may be considered.
Finally, if one has a hydrologically correct DEM of a suitable cell size with sinks filled, it is recommended to use the stream network generated from the DEM to delineate a catchment from a defined point. It is to ensure that the user-defined point from which catchment is delineated (catchment pour point) is positioned on a river that follows the bottom of a valley. It may not be so, if the point is positioned on a stream captured from a map (and is worse if the DEM is hydrologically incorrect). ArcView Network Analyst prefers the stream network generated from the DEM too. It should guarantee that network is gap-less (uninterrupted streams and all streams connected in nodes).
The point, which is not located on a cell that defines the stream, in the DEM, usually results in a small catchment delineated on the bank of the stream. To place the single user-defined point directly on the stream is quite difficult even for an experienced user, particularly if delineating a large catchment from a grid having a small cell size. To deal with this problem, the user-defined point within scripts or extensions performing the delineation is usually represented by a small grid, say 4x4-cell extent. This should ensure the high-accumulated flow point selection when delineating a drainage basin using the aGrid.Watershed request. It searches within snap distance around the area represented by a grid of 4x4 cell for the cell of highest accumulated flow, and moves the pour point to that location.
The most of ESRI and user codes facilitating hydrological analysis work fine with hydrologically correct DEMs. If results of your analysis are poor, revisit your DEM.
Basin1 ArcView GIS Extension together with instructions can be downloaded from ESRI web site < http://arcscripts.esri.com >. After accessing the site, search by author's name "Petras" to access the application site.
Sinks in DEM can be filled in ArcView by using script "Spatial.DemFill" written by ESRI. The file can be downloaded from the above ESRI web site. Search by a string "Spatial.DemFill".
Dr Ivan Petras, Pr. Eng.
Department of Water Affairs & Forestry
Private Bag X313
Pretoria, South Africa 0001