2-plane sub-basins
Elevation zones
Vegetation
Aspect
Permissible minimum HRU size
Compiled by D. Garen, 19 June 2013 (Source: HRU_delineation.docx)
Prerequisites
1) Watershed delineation and spatial data layer preparation via BAGIS is already complete.
2) DEM smoothing is done if necessary.
3) Vegetation grid (e.g., west_covtype from GIS Weasel) is imported into project.
HRU delineation steps
The concept recommended here is to use four criteria or rules to delineate HRUs: sub-basins, elevation zones, vegetation classes, and topographic aspect. The intersection of these four rules delineates pixels with unique combinations of the four characteristics, which constitute HRUs. These delineations can be spatially non-contiguous. Very small HRUs are eliminated and combined with neighboring HRUs. The steps in this process are described below.
1) Sub-basins
To account for the spatial variability of meteorological forcings (precipitation and temperature), the watershed should be partitioned into sub-basins. The number of sub-basins is arbitrary and depends on the size of the watershed, but it could range from one (for a very small watershed) up to perhaps ten for a large watershed. (Of course, PRMS is not designed to handle really large watersheds.) A balance must be struck in selecting the number of sub-basins, in that more sub-basins gives better spatial representation but it also causes more HRUs to be delineated.
Sub-basins can be delineated using the “contributing area” rule type in the Define Zones tool in BAGIS-H. This requires as input a threshold number of cells as a minimum sub-basin size. An approximate guide to the number of sub-basins corresponding to different values of the threshold can be obtained by creating the “Threshold-Link LUT”.
It is recommended to select a threshold number of cells that will result in perhaps ~20 or 25 sub-basins. Once this is created, the sub-basins can be combined into fewer sub-basins by examination and then put together by reclassing the areas using the “Raster Reclass” rule type.
2) Elevation zones
To account for the dependence of meteorological forcings with elevation, the watershed should be partitioned into several elevation zones. The number of zones is arbitrary, but somewhere around four to six categories is reasonable. The elevation interval needed to achieve this depends on the relief within the watershed, but 250 meters is a good all-around interval. Sometimes the highest elevation zone does not have very much area, so it can be preferable to combine it with the next lower category.
An elevation zone grid can be obtained by using the Define Zones tool in BAGIS-H and creating a “Raster reclass (continuous data)” rule.
3) Vegetation classes
The current standard vegetation data set is the National Land Cover Data (NLCD). This was used in the GIS Weasel and called “west_covtype” in its “data_bin”. This grid contains 16 classes in the lower US (and four additional ones unique to Alaska). These 16 can be reclassed to seven categories by combining several similar classes together. This reclass is as follows:
11, 12 ==> 1 (water)
21, 71, 81, 82 ==> 2 (grass)
41, 42, 43 ==> 3 (forest)
52 ==> 4 (shrub)
90, 95 ==> 5 (wetland)
31 ==> 6 (barren)
22, 23, 24 ==> 7 (developed)
The reclassed vegetation grid is created via a reclass rule in the Define Zones tool. This reclass table can be entered manually, or it can be stored and read in for repeated uses so that it does not have to be re-entered each time.
4) Aspect
Topographic aspect is important in affecting the amount of solar radiation striking the earth’s surface, which is a major driver of snowmelt and evapotranspiration. A four compass direction plus flat categorization is used here.
Sometimes, watersheds have large nearly flat areas. In such areas, the slope is very small, but the GIS algorithm will still assign an aspect to these pixels. This, however, is unlikely to be hydrologically significant. Therefore, it is recommended to mask out these areas and consider them to be flat. This is accomplished by selecting a threshold slope -- say, 10 degrees or 15-20 percent slope -- below which slope, and therefore aspect, will be considered to be flat. A grid can then be created with two categories -- one less than the threshold and one above the threshold -- using a Raster Reclass rule within the Define Zones tool.
Then, the aspect can be reclassed into the four major compass directions using the “Template - Aspect” function in Define Zones. Use four zones and set “Filter iterations” to zero to turn off smoothing (not necessary and can create undesirable artifacts). Before entering this dialog box, set as a “parent layer” the two-zone slope grid created as described above. Also, apply this only to zone 2 of the two-zone slope grid (with zone 2 being the zone with slope greater than the threshold). Doing all of this will create an aspect grid of five categories -- four compass directions plus flat (with all pixels in slope zone 1 being added to the flat category).
5) Combine the rules
The four grids created in steps 1-4 above can be combined in Define Zones to create four rules for HRU delineation. Each of the four grids contains discrete (integer) values, so each is used with the “Raster reclass (discrete data)” rule type. There is actually no reclassing to be done, so when a rule is created, the reclass is just a formality, copying the class numbers from “old values” to “new values” (using the “Auto” button in the dialog box). When the four rules are defined, they are executed using the “Generate HRUs” button. Make sure “Allow non-contiguous HRUs” is checked. This will create the first iteration of an HRU grid.
6) Eliminate small HRUs
This process can generate several hundred HRUs, some of which are very small or even only one pixel in size. Small HRUs can be eliminated and combined with neighboring HRUs using the “Eliminate” tool under Refine HRU.
In the dialog box, first select the HRU grid just created. Then, under “Eliminate Threshold”, select “Area of zone”. A threshold area can then be entered. This minimum HRU size is arbitrary and is dependent on the size of the watershed, but a minimum such that the result gives something in the neighborhood of 100 or fewer HRUs is desirable. A threshold of 5 km2 might be a good starting value. For non-contiguous HRUs, select the “Length rule”, and make sure the “Allow non-contiguous HRUs” box is checked. This process will create a new HRU grid with the small HRUs with a total area less than the threshold blended into neighboring larger HRUs.
It is likely that the eliminate tool will not get rid of all small HRUs. This is an artifact of how the eliminate function works. It is therefore often necessary to run the eliminate tool iteratively (e.g., two or three times) to get all small HRUs out of the grid.
7) Renumber HRUs
The processes up to this point do not result in sequential numbers for the HRUs, that is, some numbers are skipped. For PRMS, the HRU numbers need to start with 1 and run sequentially, with no gaps.
To renumber the HRUs sequentially, use the tool “Assign sequential HRU ID numbers” under Refine HRU. All that is needed is to specify the HRU grid. The tool will replace this grid with one containing sequential HRU numbers.
8) Completion
These seven steps complete the process of HRU delineation. The result is then ready to pass on to BAGIS-P for PRMS parameterization.
Resampling for use in detrended kriging (DK) preparation of meteorological forcings
The detrended kriging (DK) program is used to spatially interpolate station meteorological forcing data (daily precipitation and maximum and minimum temperature) over the watershed grid and to average these values over the appropriate pixels for each HRU. The spatial resolution of the grids used in BAGIS and BAGIS-H is quite high, around 20-30 meters. This is much higher and results in far too many grid cells to be conveniently handled in the DK program for creating HRU forcings. The HRU and DEM grids, therefore, need to be resampled to a coarser resolution for this purpose. Also, once they are resampled, they must be exported as ASCII grids for use in DK.
The resampling and conversion to ASCII can be accomplished using routines in ArcToolbox. The resampling tool is found under Data Management Tools ==> Raster ==> Raster Processing ==> Resample. To be specified are the input grid, an output grid name, a cell size (100 meters recommended), and a resampling technique (use MAJORITY for the HRU grid and BILINEAR for the DEM grid). Also, for the DEM grid, the extent must be matched to the HRU grid. This is done in the “Environments” dialog box by opening “Processing Extent” and specifying the HRU grid in the “Extent” box.
The conversion to ASCII can be accomplished using another routine in ArcToolbox. This is found under Conversion Tools ==> From Raster ==> Raster to ASCII. Only the input grid and an output file name are to be specified. Both the resampled HRU and DEM grids need to be converted to ASCII.
Conclusion
BAGIS-H is an ArcGIS-based toolkit that provides a user-friendly and streamlined interface and dialog for executing ArcGIS commands focused on the specific purpose of HRU delineation. Nevertheless, users still need to have an “intermediate” level of fluency with GIS concepts and specifically with the ArcGIS software. While BAGIS-H tools are easier to operate than the underlying ArcGIS commands, users still need to understand what is happening, and they will still need to perform some basic ArcGIS manipulations manually, such as changing grid symbology, exploring grid values, zooming in and out, adding data, etc. This also includes the operations described above for preparing DK files.
The tasks outlined above represent the recommended standard procedure for the use of BAGIS-H in delineating HRUs for the application of PRMS at the NWCC. Many variations are possible, so the hydrologist has considerable leeway in using other techniques and criteria in customizing the delineation to any particular watershed. Doing this will probably involve using other BAGIS-H tools not mentioned above and/or other manual manipulations in ArcGIS.