3.1.1 Identifying and Correcting Problems with Existing Codes and Methods
Writing computer programs is art. To code a program so that it behaves exactly as the author intends, under all conditions, efficiently, requires careful planning and much experience. Yet, even the most carefully planned, well-designed programs are bound to have errors somewhere. Convention states that it is up to the programmer to make sure that his or her program is bug free before unleashing it on the world; but this is nearly impossible, as is painfully obvious in many commercially available programs on the market today. Much of the time, it takes an outside source to stumble across some of the more subtle errors in a program which might cause it to behave incorrectly or inconsistently under certain conditions. Though it was not one of the research objectives initially, this editing process soon became an important component of the study as work concerning other objectives uncovered problems with the codes used in the model.
3.1.2 Effect of Watershed Subdivision on Model Results
In the two studies previous to this, both Khan, 2001, and Emerick, 2003, divided the watershed area into several smaller sub-areas (sub-watersheds). In his research, Emerick looked briefly at
the effect that the number and size of subdivisions have on the results by comparing the watershed hydrographs generated for the same precipitation event using two different sub- watershed arrangements (Emerick, 2003). Up to this point however, nobody has examined this effect in detail. In this study we examine how increasing the number of sub-watershed areas used to represent the whole affects the hydrograph at the watershed’s outlet predicted by the model. What effect does increasing the number of sub-watersheds have on the peak flow rate (Qp) and time to peak (Tp) of the hydrograph predicted by the model? What number of subdivisions will produce the best results?
3.1.3 Effect of Varying Muskingum Coefficient X on Model Results
The X coefficient in the Muskingum method acts as a weighting factor on storage within the reach. Values for X are specific to a particular section of reach, and depend on the shape of the wedge used to represent storage within the reach as a flood wave passes through. Thus, selecting a good value for X to accurately represent a reach requires accurate information about the characteristics of that reach, such as channel geometry. As Emerick points out, this is difficult to do with our model because we know practically nothing about the reaches in the watershed other than channel slope (Emerick, 2003). In his research, Emerick found that using a low value of X = 0.05 yielded results that compared best with the hydrograph developed from the USGS stream gauge data at Little Pine Creek for a given precipitation event. Emerick also suggests that the value of X selected has little effect on the resulting hydrograph. In this study we test this assertion by varying the Muskingum X value and comparing the resulting hydrographs predicted at the watershed’s outlet.
3.1.4 Examining the Threshold Used to Define Channel Flow
In the model we distinguish between cells where water flow occurs in stream channels, and cells where water flows across the land surface (overland), by computing the number of cells that contribute water to any particular cell in the watershed, then establishing a threshold value for the number of contributing cells required to initiate channelized flow. Previous research considered the area defined by 100 contributing cells as a sufficient threshold value. 100 cells
correspond to an area of approximately 22 acres. By our definitions then, it takes 22 acres of land to generate enough runoff for flow to become channelized. This seems a bit large, considering the abundance of small creeks and rivulets in the wooded areas of Western Pennsylvania. Since our model considers overland and channel flows separately when computing runoff, increasing the number of channel flow cells by decreasing the contributing area required to initiate channelized flow might influence the shape of the hydrograph at the watershed outlet. We examine this idea by lowering the threshold number of contributing cells required for channel flow, and comparing the resulting hydrographs.
3.1.5 Examining the Method Used to Compute Channel Time of Concentration Values
As was the case with correcting errors discovered in other existing codes, this objective arose during other work, when we noticed a strange pattern in the time of concentration (TOC) values for channel cells predicted by the model. Time of concentration is the time required for water flowing from a particular cell in the watershed to reach the outlet. Our model employs a module called the isochrone (equal travel time) code to compute these values. While examining a close up view of a raster file output by this code, we noticed that the TOC values for cells defining a channel reach did not always decrease from upstream to downstream as expected. There were many instances where TOC values would increase for a short distance, before decreasing again. In other cases we encountered isolated cells with lower TOC values than their upstream and downstream neighbors. Whether increasing or decreasing, any such anomaly in the TOC values along a stream channel is undesirable because is does not accurately represent the natural process of stream flow. Out of this observation came the objective of identifying the cause of these anomalous TOC values for cells within the channel reaches, and correcting the problem.