1. Introduction
Catchment area and watershed delineation is a common task in hydrology. Accurate drainage boundaries are essential for accurate budgets and morphological characteristics are considered to be crucial key issues particular in flat terrains. The manual delineation of a catchment based on a topographic map with contour lines is a difficult task for flat terrains. However, in combination with a thorough field survey or evaluation of satellite images reliable catchment delineation will result. Using a digital terrain model and a computer algorithm has the advantage that the result is independent from human decisions and being less time-consuming. The accuracy of the result is depending on both quality and type of DTM and the computer algorithms used. The quality of the DTM is mainly controlled by the spatial solution and the precision of the altitude for each pixel. Thus it is likely that a DTM with a course resolution will have difficulties to replicate hydrological patterns in particular in flats landscapes and be the reason for a false or biased watershed. The second critical factor comes with the algorithms used for delineating the watershed. Delineation tool has been developed since the 1980’s. A brief review is given in the following paragraphs.
Developed an algorithm for defining the path of steepest slope in downstream or upstream direction by using TIN (Triangular Irregular Network) data. This algorithm allows for the definition of basin and sub-basins to any point on a river course. However, most digital terrain data are not provided as TIN but as raster based digital terrains models (DTM) [1].
Developed software tools to derive morphologic information from raster DEM that have proven to be useful in hydrologic applications. First part of the analysis is a conditioning phase that generates three data sets: a DEM with depressions filled, a data set indicating the flow direction for each cell, and a flow accumulation data set in which each cell receives a value equal to the number of cells that drain to it. The original DEM and these three derivative data sets can then be processed in a variety of ways to optionally delineate drainage networks, overland paths, watersheds for user-specified locations, sub-watersheds for the major tributaries of a drainage network, or pour point linkages between watersheds. The efficient derivation of watersheds for large numbers of stream sediment and hydrologic geochemical samples also presents an extremely useful potential application and a software development challenge [2].
Generated a set of recursive algorithms performing the actual topographic feature extraction and synthesis into a full basin model. He improved a framework for automatically watershed extracted from a DEM. The method is based on (Band, 1986) where the author presented an algorithmic approach to automatically extract the drainage lines and divide the networks of watershed from DEM. The present framework is conducted to support and serve the hydrologic models parameters for any sort of landscape investigation when topography plays an important role by using a set of recursive algorithms which describe a new implementation. This approach allows producing more details with respect to surface form in addition to manage, store and retrieve drainage area attribute. This method supports the parameterization of distributed components for runoff models and various hydrological applications [3].
Presented a set of ten algorithms to automate the determination of drainage network and sub-catchments from DEM’s. Main purpose of the algorithms named DEDNM was to parameterize rapidly drainage network and subwatershed properties from DEM’s for subsequent use in hydrologic surface runoff models. The algorithms perform DEM aggregation, depression identification and treatment, relief incrimination of certain areas, flow vector determination, watershed boundary delineation, drainage network and sub-catchment definition and systematic indexing, tabulation of channel and sub-catchment properties and evaluation of drainage network. The algorithms were developed using raster DEM’s with a spatial resolution of 30 × 30 m and 1 m in elevation, similar to those distributed by the USGS for the 7.5' × 7.5' topographic quadrangles. Even though the development of the algorithms focused on problems encountered at this DEM resolution, their application is not restricted to that resolution [4].
A major improvement with respect to single flow direction algorithms was introduced by [5] presenting a multiple flow direction techniques through distributing the upslope area among all possible directions and thus producing a more realistic picture of surface water flow.
Garbrecht and Martz [6] performed an automated extraction of channel network and sub-watershed characteristics from raster DEM by using DEDNM [4]. This model can process DEM data of limited vertical resolution representing low relief terrain. Such representations often include ill-defined drainage boundaries and indeterminate flow paths. It is similar to other models that are based on flow routing concepts, but it includes enhancements for processing low relief landscapes where the rate of elevation change. This model should be applied to subareas that are homogeneous. In general, the close agreement between the various parameters describing the overall network and sub-watershed characteristics demonstrates the ability of DEDNM to over-come the problems associated with ill-defined drainage boundaries and indeterminate flow paths in low relief terrain.
Tarboton [7] presented an algorithms based on representing flow direction as a single angle taken as the steepest downwards slope on the eight triangular facets centered at each grid point. The upslope area is then calculated by proportioning flow between two down slope pixels according to how close this flow direction is to the direct angle to the down-slope pixel. This procedure offers improvements over prior procedures that have restricted flow to eight possible directions or proportioned flow according to slope. The new procedure is more robust than prior procedures based on fitting local planes while retaining a simple grid based structure.
Martz and Garbrecht [8] presented two new algorithms which are based on a deductive, but qualitative assessment of the most probable nature of depressions and flat areas in raster DEM’s. The algorithms have proved to be robust and able to handle all types of actual and hypothetical topographic configurations. The algorithms also provide results that are intuitively more satisfactory and more realistic than other methods. The method proposed to define drainage over flat area in a DEM using information of the surrounding topography and allows for flow convergence within the flat area.
Turcotte, et al. [9] showed numerous limitations of the widely used D8 approach and highlighted that these limitations could be overcome by the proposed approach, where D8 leads to a rather coarse drainage structure when monitoring or gauging stations need to be accurately located within a watershed and it is also unable to differentiate lakes from plain areas. Therefore using a digital river and lake network (DRLN) as input in addition to the DEM has been developed allowing for the definition of a drainage structure which is in agreement with the DRLN. It also led to a better match between observed and modeled flow structure where the results of the proposed approach clearly demonstrated an improvement over the conventionally modeled drainage structure.
Jones [10] presented a new “drainage enforcement” that insures drainage continuity through flat areas and out of depressions. This algorithm is based on the priority-firstsearch weighted-graph algorithm. Relatively simple methods for defining internal basins and incorporating digitized stream data are also discussed.
Zhang [11] proposed a new delineation approach that mainly based on D8 algorithm for determination of flow field over the rugged terrain and in association with D∞ algorithm for the relative flat area to fully take the advantage of D8 and D∞ in watershed delineation. The D∞ is based on representing flow direction as a single angle taken as the steepest downwards slope on the eight triangular facets centered at each grid point.
Osma-Ruiz, et al. [12] described a new algorithm to calculate the watershed transform through rain simulation of grayscale digital images by means of pixel narrowing. The efficiency of this method is based on limiting the necessary neighboring operations which is the most expensive computation in the context to compute the transform to the outmost and in the total number of scanning’s performed over the whole image. Experiments demonstrate that the proposed algorithm is able to significantly reduce the CPU time of the fastest known algorithm without involving any loss of efficiency. It generated an about 31% improvement using various image sizes in comparison with the Sun, Yang and Ren algorithm which called SYR, where the new algorithm achieves linear running time with respect to the size of the input images.
Danner, et al. [13] presented the TERRASTREAM software package and the experimental results on real elevation point sets show that the implemented approach handles massive multi-gigabyte terrain data sets. A data set containing over 300 million points and over 20GB of raw data was processed in less than 26 hours on a system, where most of the time (76%) is spent in the initial CPU-intensive DEM construction stage. the author performed experiments on a Dell Precision Server 370 (3.40 GHz Pentium 4 processor) running Linux 2.6.11. The machine had 1 GB of physical memory.
Kenny, et al. [14] developed two separate routines, one for sinks and one for flats, for establishing flow direction in an un-filled DEM environment. Each sink and flat is analyzed in sequence and flow directions are resolved iteratively, utilizing the surrounding terrain, the morphology within an unfilled DEM and the recognized flow patterns translated from surface hydrology features. In comparative analysis with five commonly employed sink flow routing algorithms the proposed sink routing routine resulted in the least alteration to both the elevation and flow direction surfaces.
The backbone of many GIS performing watershed delineation from a raster DEM is either the D8-algrithms [15] or the D∞ [7], both extracting a potential drainage network. While D8 only extract the drainage network at specific 8 directions, D∞ work along infinite directions (Figure 1). However, both D8 and D∞ face a problem, if more than one lowest neighboring cell is identified: this problem can be solved either by a predefined direction or randomize single flow approach. On contrary to single flow direction algorithms multiple flow direction techniques [5] produce more realistic water flow pattern. The second general problems are depressions which can be solved to some extend by depression filling algorithms (Figure 2).
Almost all software packages performing watershed extraction and catchment delineation tasks are based on similar techniques and create certain outputs:
• Automatically filling of spurious small depressions in DTM and setting thresholds to leave larger/deeper depressions unfilled. In the latter case placing a drain (null cell) at the bottom of unfilled depressions to model their internal drainage.
• Computing vector flow paths, watersheds, basins, and ridge lines.
• Controlling drainage network density and basin size using flow accumulation thresholds for outlet, upstream limit, and branching points.
Figure 1. Flow direction and Watershed boundaries based on D8 & D¥.
Figure 2. Flow direction a: Randomized; b: Non randomized and fill depression a: Sink unfilled terrain; b: Sink filled terrain; c: Flooded terrain.
• Computing upstream catchment and downstream flow paths for specific locations by manually placing seed points at a desired outlet.
• Compute the geomorphic characteristics; hydrologic attributes of flow paths and catchment polygons of the DEM cells.
• Creating segmented flow path network using elevation or flow accumulation values Other software like HydroSHEDS [16] or ILWIS [17] offer additional operations like void-filling by means of different techniques (filter, neighboring analysis, etc.), hydrologic conditioning (e.g. removing vegetation cover), weeding of coastal zone, stream burning, filtering (smoothing), molding of valley courses, carving through barriers, and manual correction.
Overall two different strategies can be followed up: one strategy is to evaluate a giver raster DEM in total. Sub-catchments which are at the boundary of the DEM are inevitably biased; a proper selection of the frame in particular in the downstream area is thus a prerequisite for a correct delineation of the entire catchment. A second possible strategy is using a manually positioned seed point at the proposed outlet of a catchment area. This is with respect to mathematical complexity the easier task in comparison to a complete terrain analysis. However, concerning practical applications both approaches have advantages and disadvantages and the users would be always well advised to have both options at hand.
2. Objectives
The main focus of this study was evaluating the possibility to delineate catchments from flat and arid areas by means of DTM avoiding hard techniques like river burning or other manual hydrological DTM corrections. Thus it should be shown whether is possible to apply the technique on areas where no river maps are available. Three GIS packages were tested for a considerable large region of interest (~100,000 km2) utilising two DEM: the 90 m and 30 m SRTM (Shuttle Radar Topography Mission) data set in addition to the ASTER 30 m covering the same ROI. For comparison a thorough field survey and manually catchment delineation was available [18] Software used was Arc Hydrotools, TNTmips and RiverTools.
3. Methodology
Figure 3 illustrates how the three GIS used perform the