Drought simulations 1988

1. Sensitivity to rainfall forcing Convective storms along with orographic enhancement effects dominate spring and summer precipitation in the region, and therefore it is expected that the spatial variability of runoff production and streamflow variability closely reflect rainfall patterns within each of the subbasins within the Monongahela River basin.

Figures 20.4 and 20.5 show the simulated time-series of streamflow at Elizabeth as a function of rainfall distribution at 5- and 1-km resolution, respectively.

1988-5 km

1988-5 km

(a) 01 Apr 15 Apr 01 May 15 May 01 June 15 June 01 July 15 July 01 Aug 15 Aug 31 Aug

(b) 01 Apr 15 Apr 01 May 15 May 01 June 15 June 01 July 15 July 01 Aug 15 Aug 31 Aug

Figure 20.4 Comparison of observed and simulated hydrographs at Elizabeth, PA, between 01 April and 31 August 1988 at 5-km resolution. The model was forced with rainfall fields estimated according to: (a) the hypsometric method; and (b) the modified Thiessen polygon method

(b) 01 Apr 15 Apr 01 May 15 May 01 June 15 June 01 July 15 July 01 Aug 15 Aug 31 Aug

Figure 20.4 Comparison of observed and simulated hydrographs at Elizabeth, PA, between 01 April and 31 August 1988 at 5-km resolution. The model was forced with rainfall fields estimated according to: (a) the hypsometric method; and (b) the modified Thiessen polygon method

1988-1 km

800 600 400 200

1988-1 km

800 600 400 200

01 Apr 15 Apr 01 May 15 May 01 June 15 June 01 July 15 July 01 Aug 15 Aug 31 Aug

Figure 20.5 Comparison of observed and simulated hydrographs at Elizabeth, PA, between 01 April and 31 August 1988 at 1-km resolution. The model was forced with rainfall fields estimated using the modified Thiessen polygon method

01 Apr 15 Apr 01 May 15 May 01 June 15 June 01 July 15 July 01 Aug 15 Aug 31 Aug

Figure 20.5 Comparison of observed and simulated hydrographs at Elizabeth, PA, between 01 April and 31 August 1988 at 1-km resolution. The model was forced with rainfall fields estimated using the modified Thiessen polygon method

The hydrographs in Figure 20.4(a) correspond to rainfall fields estimated using the standard hypsometric method, while the modified Thiessen polygon method was used for the simulation in Figure 20.4(b). Overall, the simulations are better using the rainfall fields generated by the modified Thiessen polygon method independently of the spatial resolution (compare Figures 20.4(a) and 20.4(b), and Figure 20.4(b) and Figure 20.5). This suggests that the information content provided by the existing raingauges is close to that which can be optimally described by the individual polygons.

In late spring and throughout the summer, however, the lack of adequate rainfall forcing results in poor temporal correlation between simulated and observed flows even at 1 km resolution. Because most of the warm-season rainfall is produced by small-scale thunderstorm activity (e.g. spatial scales on the order of 10-20 km2), the raingauge network (Figure 20.1) is too sparse to detect the occurrence of rainfall in regions of the basin where raingauges are located very far apart from each other, thus explaining the missing peaks. By contrast, if one or more of the raingauges are colocated with such events, then the total precipitation can be grossly overestimated (i.e. proportionally to the ratio between the actual area of the rainstorm and the Thiessen polygon area) such as between May 15 and June 1 (Figure 20.4(b)). Consistency between the smaller spatial scales of rainfall forcing and model resolution is therefore necessary to capture the runoffresponse to isolated convective storms.

Nevertheless, model statistics of streamflow closely follow the spatial patterns of those of existing observations (not shown), suggesting therefore that the model captures well the intraseasonal variability of water fluxes during the 1988 drought across the basin. The root mean square error and bias in the streamflow simulations increase with drainage area, and the coefficient of variation is usually higher in the summer than it is in the spring, and it decreases when the watershed area increases, indicating reduced variability due to spatial integrating effects of streamflow propagation.

2. Structural stability Comparison of the 5- and 1-km hydrograph simulations at Elizabeth (Figures 20.4(b) and 20.5) reveals important characteristics relevant to the structural stability of the LHM-3D with regard to spatial resolution. First, the relative contributions of subsurface flow (interflow plus baseflow) to the total streamflow are considerably different. While the streamflow at coarse resolution (i.e. 5 km) is produced as surface runoff (Figure 20.4(b)), the streamflow at the finer resolution (i.e. 1 km) results mostly from the contribution of subsurface flow (Figure 20.5). This implies two different control processes on runoff production: (i) vertical infiltration capacity at 1 km-vertical control; and (ii) magnitude of the lateral gradients of hydraulic head (elevation plus pressure head) and hydraulic conductivity at 5 km-lateral control.

Figure 20.6 compares the ratio of the simulated subsurface flow (Qs) to the total simulated streamflow (Q) at Elizabeth, Enterprise, and Dailey watersheds for both 5- and 1-km resolutions using the rainfall fields estimated via the modified Thiessen polygon method. Dailey is representative of hydrologic processes in the regions of steep terrain on the eastern half of the Monongahela basin, whereas Enterprise is representative of the hydrology of the central and western catchments. The differences in (Qs/Q) are particularly noticeable from late spring to early summer. At Dailey (top panel), the timing coincides with strong reduction in relative subsurface flow after four consecutive days of rainfall in early May that leave the shallow soil column saturated. Subsequently, soil water is quickly redistributed down slope at 1 km, and only at much slower rate at 5 km consistent with the differences in topographic slopes at both resolutions. A similar, though weaker, reduction in subsurface flow contribution takes place in response to the series of storms that occur between May 16 and May 20. After this time, the streams are maintained almost entirely by the incoming subsurface flow at 1-km resolution, while it takes almost a month for this to be the case in the 5-km simulation.

In the densely forested slopes above Dailey, evapotranspiration reaches a peak in late spring (June) during the greening season, lowering soil moisture content in the root zone, decreasing lateral hydraulic head gradients in the hillslopes, and consequently reducing subsurface flow to the streams. Vegetation is therefore a limiting factor of subsurface flow via evapotranspiration. During short-duration intense rainfall events, surface runoff production is controlled locally by the soil's infiltration capacity in the uppermost layer. After the end of June, streams are maintained by subsurface flow at either resolution. This effect is much stronger at 5-km resolution consistent with lower root zone soil moisture over larger areas (i.e. larger grid cells) and gentler topography (low hydraulic gradients).

The (Qs/Q) time-series for Enterprise (mid panel) exhibits lower relative subsurface flow contribution at 1 km than at 5 km in contrast to what happens at Dailey. This is explained by the fact that the soils are significantly deeper and the topography generally smoother in this catchment, and therefore surface runoff is controlled by the infiltration capacity of the upper soil layers. Furthermore, the predominant vegetation types at lower elevations are characterized by shallow root systems,

Enterprise
Figure 20.6 Comparison of the time evolution of the ratio of subsurface flow (Qs) to total streamflow (Q) at 1- and 5-km resolutions km resolutions at Dailey, Enterprise, and Elizabeth in 1988

and thus increased evapotranspiration in late spring and early summer seasons predominantly affects the upper soil layers and therefore has a smaller impact on rainfallrunoff response at Enterprise than at Dailey. The Qs/Q behavior at Elizabeth (bottom panel) reflects the integrated behavior of the Monongahela basin, which is qualitatively similar to that discussed above for Dailey, reflecting the combination of the two types of catchments in the basin.

The changes in the rainfall-runoff response due to changes in the spatial resolution of the model indicate that there is a change in governing physical processes at different resolutions, specifically infiltration capacity, evapotranspiration and root zone soil moisture, and lateral subsurface flow. These results imply therefore a dependency between model resolution and simulated physics. Although it is possible to predict streamflows with comparable skill at both resolutions, the pathways

1993-5 km rw

E mE

to S

1000 800 600 400 200 0

1993-5 km rw

1000 800 600 400 200 0

15 Apr 01 May 15 May 01 June 15 June 01 July 15 July

01 Aug 15 Aug 31 Aug

15 Apr 01 May 15 May 01 June 15 June 01 July 15 July

o to

o to

1000 800 600 400 200 0

(b) 15 Apr 01 May 15 May 01 June 15 June 01 July 15 July 01 Aug 15 Aug 31 Aug

Figure 20.7 Comparison of observed and simulated hydrographs at Elizabeth, PA, between 09 April and 31 August 1993 at 5-km resolution. The model was forced with rainfall fields estimated according to: (a) the hypsometric method; and (b) the modified Thiessen polygon method of water within the basin during drought can be quite different depending on the physiographic characteristics of individual catchments. For example, the Monongahela appears to be more resilient to drought in the 5-km simulation than in the 1-km simulation. On the basis of the observations, however, we know this is not the case (see Figure 20.4(b) and Figure 20.5). In summary, physically realistic simulations therefore require that model resolution be consistent with the spatial scale ofthe dominant hydrologic processes.

0 0

Post a comment