Template:GUM16

 

16 - Building A GSSHA Model

GSSHA User's Manual

To this point this manual has described the theory, processes, solutions, and various inputs and outputs that can be used in the GSSHA model. This section is intended to provide the user with a step-by-step instruction on how to build a model beginning with a blinking curser. While there are many ways to construct a model with GSSHA, significant experience by the authors suggest that the following methodology is a prudent, if not the best way, to construct a GSSHA model.

GSSHA has been closely linked to the WMS v6.1 and higher, and it is strongly recommended that this software be used to develop the needed inputs for the model. As listed in Section 13, Additional Inputs, many required and optional plain vanilla tables must be created outside the WMS framework, and these can be created with a simple spreadsheet or word processing software. GSSHA is intended to be used with the Mapping Table and the related index maps described in Section 12. This is the method employed by WMS 6.1 and higher. Experience with the CASC2D model indicates that this method of assigning parameters is superior for the typical case where parameters must be assigned based on reclassification of common index maps, such as land use, soil-texture, and vegetation (Downer et al., 2002a). In addition to the WMS and a spreadsheet to build table inputs, a GIS is usually needed, or at least helpful, in developing many of the data layers that the GSSHA model requires. While the steps required to build a GSSHA model will be presented here, WMS how-to information is contained in the WMS User Manual (Nelson, 2001) and the WMS for GSSHA Primer (Downer et al., 2002b).

An essential element in successfully developing a complex GSSHA model is to start simple, get parts or processes of the model to run, and then build upon success. It is important to follow the methodology described here. WMS can be used to build a complex model all at one time. Such a model could have millions of cells, a complex stream network, unsaturated zone calculations, saturated groundwater flow, etc. Such a model would also have little chance of ever working properly because the amount of information and possible problems is overwhelming.

16.1 Delineating the Watershed

The first step in building a GSSHA model is to delineate the watershed. The watershed is delineated from the DEM. DEM data of various resolutions can be obtained from the USGS and EPA Basins data bases, accessable through links provided by the geospatial data access website, http://www.xmswiki.com/xms/GSDA:GSDA . 90m resolution data are available for all of the United States; 30m resolution data are available for most areas, and 10 m data are becoming available. While models are not routinely run with grid sizes finer than 90m, the 10m data has a much better vertical resolution, typically 0.1 m as opposed to 1.0 m for the 90m resolution data. Unless the GIS or GUI cannot digest the large number of data points, the 30m data will provide better watershed and stream delineations. These data are typically available as 7.5 minute quad sheets. When the watershed overlaps two or more maps, then the overlapping sheets need to be put together and any discrepancies between the different maps resolved. The WMS software has tools available to accomplish these tasks. With the DEM covering the watershed in hand, WMS can be used to delineate the watershed above any give point in the basin, basin outlet, based on calculations from TOPAZ (Martz and Garbrecht, 1992). The TOPAZ model also determines the stream network from the DEM data. This may or may not be useful. If a watershed boundary has been predetermined, the watershed polygon can be used to “cut out” the appropriate DEM data, or can be imported for use as the watershed boundary in WMS. If a watershed boundary is imposed on the DEM then it is likely that the DEM or resulting grid will have to be edited to force the cells in the predefined watershed to all drain toward the basin outlet.

16.2 Selecting a Grid Size

Selection of an appropriate grid size is important in successful modeling of a watershed and has been discussed in Section 4.2 of this manual. Key to successful modeling of processes at the cell level is that the cell size be smaller than the size of the essential feature of the landscape involved in the model. For example, Ogden et al. (2000) used a 30m grid size to model the Fort Collins, Colorado, flash flood of July 28, 1997. This small grid size was used because the urbanized Spring Creek catchment in Fort Collins has a number of small-scale landscape features such as: roads, parking lots, buildings. Had finer scale DEM data been available at the time, an even smaller grid cell may have been employed. Conversely, Doe and Saghafian (1992) successfully modeled the Taylor Arroyo watershed near Trinidad Colorado with a 300m grid. While the climates are similar, the Taylor Arroyo watershed is essentially completely undeveloped. 300m grid cells are not large compared to soil texture and vegetation complexes in the watershed. It’s possible that even larger grid cells could be used. The grid size also has an effect on slopes, which may be important for overland sediment transport calculations as discussed by Sanchez (2002). It is important to note that the grid size must be chosen in view of the data available to support providing input data. The grid size should never be smaller than the DEM resolution.

16.3 Overland Flow Routing

The simplest GSSHA model consists of a grid with elevations from the DEM and roughness values assigned, allowing overland flow routing to be simulated. Overland flow routing should first be attempted with uniform values of rainfall and overland roughness and a small time step, on the order of 10 s. Spatially varied maps of depth should be output every few time steps. As described in Section 14.6, these maps are useful for locating problem areas in the watershed, and comparing areas that water ponds to independent topographic data. If the overland flow module will not run with a small time step and the very stable ADEPC overland flow routine, the depth maps should be consulted to identify potential problems in the watershed. The elevations in the watershed may be smoothed using algorithms in the WMS software, or the elevations may also be manually edited. If water is ponding along the edge of the watershed, these cells will either have to be removed from the grid or raised in elevation. Another potential solution to making the overland flow module run is to increase the grid size, which will reduce the Courant number, and smooth the elevations in the model.

Once the overland flow routine will run with uniform roughness parameters, initial roughness parameters and retention depths can be spatially distributed according to index maps of land use and vegetation.

16.4 Infiltration

With a working overland flow model in hand, the next step is to select an infiltration routine and assign the needed initial parameters. In selection of an infiltration routine, the source of runoff and streamflow should be a major consideration. The GA, multi-layer GA, and GAR models are only valid selections if the primary mechanism generating stream flow is Hortonian (Horton, 1933) infiltration excess runoff. Downer et al (2002a) explain the pitfalls of trying to apply the GA based methods to watersheds where the Hortonian runoff mechanisms is not dominant. In this case, the more general RE solution should be used to calculate infiltration. If a shallow water table is present, then the effects of the water table will also have to be included in the model. This will be discussed later. Conversely, if Hortonian flow is the dominate stream flow producing mechanism, solving RE will likely yield only small benefits over solving a less general GA approximation (Downer and Ogden, 2003a).

For single events the GA model or multi-layer GA model will suffice. For continuous simulations the GAR method is used. A single event model using GA may later be changed to GAR by changing the infiltration option and supplying the additional needed parameters. With a method of calculating infiltration selected, the appropriate initial parameters are assigned using the Mapping Table and a combination index map of soil-texture, land use, and vegetation. When using RE the appropriate vertical grid size is important (Downer, 2002a) and the effect of the grid size on runoff should be investigated to determine an adequate resolution. This model should also be run with the uniform rainfall event. As infiltration will tend to reduce the amount of runoff, the model with infiltration will likely run on the first attempt, and the time step can likely be increased without changing the shape of the predicted outlet hydrograph.

16.5 Channel Routing

While small watersheds with no well defined channel may be simulated without channel routing, large basins or basins with a defined channel almost always need channel routing to accurately reproduce the outlet flow. There are many possible ways to locate the stream. The location of the stream network may be surveyed, come from USGS .dlg files or other sources of digital stream network, or the stream delineation provided by TOPAZ may be used. In any case, as with the overland flow, it is best to start with a simple channel network and add complexity. As a first approximation, the main stem and only major tributaries should be included in the stream network. Once the simple network is running, additional stream segments can be added. As the stream segments begin to represent smaller and smaller tributaries, the effect of adding additional streams will begin to diminish.

As discussed in Section 5.1, ideally the steam network comes with surveyed cross-sections and thalweg elevations. Also discussed in Section 5.1.5 is a procedure for taking stream bottom elevations from the DEM. As discussed in this section, it will be important to smooth the thalweg elevations to create a realistic channel profile and perhaps edit the grid elevations to ensure that overland flow can enter the stream. Whether to use break-point cross-sections or trapezoidal approximations generally depends on the availability of surveyed cross-sections.

16.6 Single Event Calibration

A GSSHA model with overland flow, infiltration, and channel routing represents a fairly complete model, and this model can be used to determine appropriate time steps, RE cell sizes, and channel routing parameters, i.e. channel roughness. While this step is not essential, it is useful. For the single event calibration the user should select one storm event from the observed data that provides a reasonably well defined outlet hydro-graph. Overland flow and infiltration parameters can be adjusted to produce the approximately correct volume of flow at the watershed outlet. The in-stream channel roughness is tuned to match the hydrograph peak and shape. This initial single event calibration can either be done manually, or with an automated calibration process, such as the SCE method. With a single event the SCE method will converge in a short period of time, likely overnight. While the overland flow and infiltration parameters from this effort are of limited value (Senarath et al., 2000), the values of in-stream roughness should be approximately correct. Also, this calibrated model can be used to determine what model time step can be used and the appropriate cell size for RE solutions.

Assuming the model has been calibrated at some small time step, say 10s, the calibrated simulation can be repeated with increasingly larger time steps until either a) the model crashes b) the outlet hydrograph begins to oscillate, or c) the shape of the outlet hydrograph begins to significantly change in shape. When any of these occur, the time step is too large and should be reduced until the problem disappears. It is possible that very large time steps, several minutes, can be used in the simulation without significantly affecting the results. This will significantly reduce execution times and may be especially important when using an automated calibration process over an extended simulation period. As this exercise will demonstrate, the model will produce almost exactly the same results for time steps below some critical value, so that using time steps much smaller than this critical value will not result in improved results, only longer simulation times. This optimal time step is then used in subsequent calibrations and simulations.

The same procedure can be used to determine the appropriate cell size to use in the RE solution (Downer, 2002a). Starting with very small cell sizes in the top 10 cm of the soil column, 1 – 10 mm, the cell size is increased until the volume of runoff begins to significantly deviate from the original results. This theshold cell size is used in subsequent calibrations and simulations.

16.7 Long-term Simulations

Long-term simulations with ET calculations, as described in Section 9, are needed to model longer periods with multiple storm events. Long-term simulations must also be performed to properly simulate soil moistures in the unsaturated zone, saturated groundwater movement and stream interaction. As described by Senarath et al (2000) long-term simulations are also necessary to properly calibrate any GSSHA model, even if it to be used only to simulate single events. ET parameters are assigned with the Mapping Table, and related to the combination land use/soil-texture/vegetation index map. The selection of appropriate root depths for such crude indexes can be difficult (Downer and Ogden, 2003a) and these values are most properly thought of as effective values that are determined through calibration. Although the method to simulate the seasonal effects in GSSHA is crude, it has been shown to be effective (Downer and Ogden, 2003a) and the SEASONAL_RS card should be included in the project file if simulations are to be conducted outside the summer growing season, May-September.

The rainfall file and hourly hydro-met data file should be constructed to cover the calibration, verification, and simulation scenarios period. This period may be weeks to years, depending on the available record. To properly calibrate a model, a period with overlapping rainfall and streamflow measurements from several storm events that produce stream flow should be selected. The hydro-met data should start just after the last rainfall event before the simulation or calibration period, and saturated or near saturated initial moistures are assumed. The model should be run in continuous mode with the entire calibration period to locate and fix problems in the input files and assure the model will run for the entire period with the initial parameters, time step and grid size.

16.8 Saturated Groundwater Modeling

Whether to simulate saturated groundwater depends on the properties of the watershed to be simulated and the availability of sub-surface information. For intermittent streams, lateral groundwater flow is not likely an important consideration. For streams with significant baseflow, and for streams where there is an obvious contribution of groundwater flow or known or suspected saturated source areas during rainfall events, groundwater simulations will likely be required to capture the shape and volume of discharge hydrographs (Downer et. al., 2002a, Downer et al. 2002c).

When saturated groundwater is to be simulated the information on the bedrock elevations and properties of the saturated groundwater media are required. In addition, boundaries must be supplied along the watershed boundary. Unlike in surface water flow, where the boundary condition is obvious and simple, assignment of groundwater boundaries can be difficult and the boundary conditions can dominate the groundwater flow solution. If there is a known groundwater divide along all (unlikely) or part (more likely) of the watershed boundary then a no flow boundary can be used on this part of the watershed boundary. Otherwise, head boundaries must be imposed. Imposition of improper head boundaries will lead to bad simulations despite an otherwise good model. Head boundaries along the edge of the watershed should come from measured well data or from a good regional groundwater model that includes the watershed of interest.

If channel routing is being performed, the streams should be defined as RIVER_FLUX boundaries and the flux between the stream and the groundwater will be computed every stream routing time step. For this calculation to be meaningful, the thalweg elevation and the grid land surface elevation must be correct, or at least the difference in thalweg elevation and land surface elevation must be correct. For this reason, land surface elevations in the cells corresponding to stream nodes may need to be adjusted. This is almost a requirement if smoothing of the thalweg elevations has occurred. If groundwater/stream interactions are to be simulated, it is best to begin by editing the land surface grid elevations to produce a smooth channel profile, and then subtract the incision of the channel in the grid cell to determine the thalweg elevations to be used in the channel input file.

Initial water surface elevations may be interpolated from well data or may be some assumed elevations. In either case it is usually necessary to run the groundwater simulations for an extended period to produce realistic initial values of groundwater elevation to be used in simulations. For instance, if the simulation period of interest is May though September, the model should be run from January to May to produce a starting WATER_TABLE file. Depending on the outcome, this procedure may have to be repeated multiple times, where the ending water surface elevation becomes the starting water surface elevation for the next attempt. A proper initial water surface has been established once the baseflow is on the correct order of magnitude, and maps of groundwater elevation are both smooth and agree reasonably well with well observations.

16.9 Calibration and Verification

The complete model should be calibrated and verified to an extended period of data while operating in the LONG_TERM mode. For flow models, the model should be calibrated to observed discharges at the outlet and any interior points. The OPTIMIZE project card can be used to provide peak discharge and discharge volume for individual events and the entire simulation at the watershed outlet and at any desired internal locations. These can be used to calculate a cost function, and parameter sets that produce the smallest cost function, minimum error as defined by the cost function, can be determined either manually or preferably with an automated method, such as the SCE method. Typical calibration parameters listed in order of importance for each process include:

Overland Flow

Surface roughness

Retention depth

Infiltration

Saturated hydraulic conductivity (all methods)

Suction head (GA, multi-layer GA, GAR) or bubbling pressure (RE)

Initial moisture

Porosity

Channel Flow

Roughness coefficient

Evapo-transpiration

Root depth

Canopy resistance

Soil Moisture

Pore distribution index

Wilting-point soil moisture

Groundwater Flow

Saturated hydraulic conductivity

Porosity

To use the SCE method, the number of parameters to be calibrated should be kept to a minimum, typically less than 16 (Senarath et al., 2000). To reduce the number of parameters, the proportions of initial estimates of parameters for different index types can be adjusted as a set, i.e. adjust all values of some parameter, saturated hydraulic conductivity for example, by the same fractional amount.

The model with the calibrated parameter set should be tested against independent verification period, such as a split-sample test (Klemes, 1986). Once the model demonstrates the ability to predict discharge, or other variables of interest, for the verification period, it can be used with confidence to analyze model scenarios and make predictions under varying hydrologic conditions.

16.10 Sediment Transport

Once a working hydrology/hydraulics model has been calibrated and verified, sediment transport can be added to the model. Overland sediment transport parameters are derived from the soil-texture/land-use index maps. Users should consult the manual and sediment transport textbooks (such as Yang, 1996) to learn about using Yang’s method (Yang, 1973) to assign appropriate parameters for the in-stream sediment routing. As with the hydrology/hydraulics portion of the code, the sediment model should also be calibrated and verified to observed data.