A Semi-Implicit, Three-Dimensional Model for Estuarine ... - USGS

104 A Semi-Implicit, Three-Dimensional Model for Estuarine Circulation
All of the error measures presented in figure 5.4 for solutions with Δx ≥ 1,000 ft (305 m) were obtained using the four-point
interpolation scheme given by equation 4.10. For solutions with Δx < 1,000 ft (305 m), two-point interpolation was used because
it gave results identical to those from the higher-order interpolation. Only the solutions using the two largest values of Δx (5,000
and 10,000) were affected significantly when the low-order interpolation was used; for these solutions there was additional
attenuation in the routed hydrograph.
The two-level model was tested again to determine the behavior of errors as the time step Δt was varied over a wide range
while keeping the space step fixed at 2,000 ft (610 m). The list of time steps and the ranges of Courant numbers for the test runs
are given in table 5.4. The graphs of the four error measures from table 5.1 are presented in figure 5.7 using three curves that are
labeled by the number of iterations used in the solutions (similar to fig. 5.4). Because the time base of the input hydrograph is 150
minutes, a time step of 15 minutes corresponds to ten points in time to resolve the hydrograph.
An inspection of the graphs in figure 5.7 reveal the following information regarding the numerical errors:
1. The magnitude of the errors generally increases with Δt.
2. There is much less error in the solutions using two iterations than in those using one iteration. The time step must
be reduced to 1 minute or less to keep the errors for solutions using one iteration to a reasonable level; a time step
of 10 minutes or less is acceptable if two or more iterations are used.
3. Mass preservation and phase errors are reasonably small (< ± 1.4 percent) for all runs using Δx = 2,000 ft (610
m). Mass preservation errors are negative (indicating a loss of mass) for the solutions using two iterations, but
positive otherwise; phase errors generally are positive (lagging) except for the three solutions using two iterations
and Δt ≥ 5 minutes, which have negative (leading) phase errors. In general, mass preservation and phase errors
are affected more by a large Δx than by a large Δt, as suggested by comparing results in figure 5.7 with those in
figure 5.4.
Regarding stability, the following information was found in preparing the test simulations for figure 5.7:
1. No stable solutions were obtained using one iteration with a time step greater than 5 minutes and gravity-wave
and advection Courant numbers greater than 1.9 and 0.34, respectively.
2. The solutions using two iterations were stable up to a time step of 15 minutes, but the solution for Δt = 15 minutes
was spoiled with severe oscillations (see fig. 5.8).
3. The solutions using five iterations were stable up to a time step of 20 minutes, the largest attempted.
Table 5.4. Ranges of Courant numbers for hydrograph problem simulations using a
space step (Δx) of 2,000 feet.
Δt, in minutes Crg Cra 0.5 0.13–0.19 0.02–0.03
1 0.27–0.38 0.04–0.07
2.5 0.67–0.95 0.11–0.18
5 1.3–1.9 0.22–0.34
10 2.7–3.8 0.44–0.68
15 4.0–5.6 0.67–1.0
20 5.3–7.5 0.89–1.4