DECLARATION 2: PUBLICATIONS
6. IMPACT OF MODEL CONFIGURATION AND PARAMETER ESTIMATION
6.4 Single versus Incremental UH Approach
This section outlines the methodology applied and results obtained for Objective 1 – Identify if the incremental UH approach with the Schmidt and Schulze (1984) estimated lag and synthetic daily rainfall distributions (Weddepohl, 1988) consistently performs better than the single UH approach, also using the Schmidt and Schulze (1984) estimated lag time, for all verification catchments (Section 6.2). These include operational catchments where short duration rainfall data are not available, which is generally the case when estimating design floods in practice in South Africa, due to the scarcity of short duration sub-daily rainfall data in South Africa.
6.4.1 Methodology
The results obtained from the assessment of the CSM system developed, as documented in Chapter 4, i.e. with revision to the volume used in the peak discharge computation (UQFLOW OTD), and applying the single UH approach, are compared to those obtained when applying the incremental UH approach with the synthetic rainfall distributions (Weddepohl, 1988) applicable to each catchment, as detailed in Table 4.1. In both cases the same input information from Table 4.1 was used, and only the peak discharge computation procedure was changed.
6.4.2 Results and discussion
In terms of overall model performance as indicated by the NSE values for all verification catchments, as summarised in Table 6.1, it is evident that the incremental UH approach performed better than the single UH approach (higher NSE values) for nine (9) catchments and with slightly lower NSE values at V1H032 and X2H027. Catchments V1H032 and X2H027 are considerably larger than the other catchments and therefore the results may suggest that the performance of the incremental UH approach deteriorates with catchment size, i.e. for catchments outside of the recommended size range (< 50 km2) defined for the ACRU model (Schulze, 1995). The results, however, for these two catchments are only slightly worse than those obtained from the single UH approach, whereas for the remaining catchments, in most cases, substantial improvements were obtained when using the incremental UH approach compared to the single UH approach. Therefore, in general the incremental UH approach provides better results compared to the single UH approach. The general poor performance of the model with predominantly negative NSE values, for both the single and incremental UH approaches used to simulate the peak discharge, is attributed to (i) the simulated stormflow volume on any given day not being representative of the observed stormflow volume for that day, (ii) variations in the sub-daily temporal distribution of daily rainfall from day-to-day, and (iii) variations in lag time from day-to-day, as detailed and discussed in Chapters 4 and 5.
Therefore, on a day-to-day basis the simulated versus observed comparisons are relatively poor, however, the predominant or most typical conditions are accounted for. Recommendations have been made in Chapter 5 to further improve on these results and incorporate or develop methods to more adequately account for these variations on a day-to-day basis.
Table 6.1 Comparison of NSE results between observed versus simulated daily peak discharges when applying the single and incremental UH approaches
Catchment Area (km2)
NSE Daily Peak Discharges -
Single UH approach
NSE Daily Peak Discharges - Incremental UH
approach
U2H020 0.26 -1.89 -1.20
V7H003 0.52 -1.12 -0.49
G2H010 0.73 -23.70 -3.14
V1H005 0.98 -10.53 -7.47
V1H015 1.04 -1.24 -0.41
U2H018 1.31 -10.02 -5.59
W1H016 3.30 -0.70 0.27
X2H026 13.82 -6.57 -4.68
A9H006 16.00 -1.43 -0.83
V1H032 67.80 0.17 -0.01
X2H027 77.16 -3.91 -4.49
A comparison of the MRE between observed and simulated design peak discharges, for return periods ranging from 2 to 100 years, when applying both the single and incremental UH approaches is shown in Figure 6.1. The results, similar to the NSE values, indicate that improved design peak discharges are obtained for all verification catchments (lower MRE values) when using the incremental UH approach, except once again for catchments V1H032 and X2H027. The results for catchment V1H032, however, are very similar when applying the two approaches, i.e. the results are practically identical, with the single and incremental UH approach results sharing the same plotting position in Figure 6.1, and the results obtained when applying the incremental UH approach are only slightly worse for catchment X2H027 compared to when the single UH approach is applied. The MARE was not presented here since the values are identical to the MRE values, i.e. both methods consistently overestimate the observed design peak discharges. The significant differences between the results obtained for the Lambrechtsbos B (G2H010) Catchment (Figure 6.1), are related to the ability of the incremental UH approach to account for the distribution of daily rainfall. The Lambrechtsbos B (G2H010) Catchment falls into rainfall intensity Region 1 associated with low intensity rainfall uniformly distributed throughout the day. For the single UH approach the rainfall intensity is not accounted for and consequently the storm duration is assumed to be equal to the catchment response time, i.e. lag time, which for this catchment is very short resulting in significantly higher peak discharge simulations. This, once again, indicates the sensitivity of
the peak discharge simulations to the distribution of daily rainfall used and the importance of adequately accounting for the distribution of daily rainfall.
Therefore, from the NSE and MRE values obtained above it may be concluded that in general the incremental UH approach provides better results, and should therefore be used as the default option in the CSM system. Consequently, the incremental UH approach will be used in all subsequent investigations and assessments in the sections to follow. In addition, there is room for more improvement in the results when using this approach, if the actual distribution of daily rainfall, or an improved method of disaggregating the daily rainfall into a hyetograph on a day- to-day basis, is developed and used. Furthermore, relationships between rainfall intensity and catchment lag time were shown in Chapter 5, therefore, lag time may possibly be adjusted based on the distribution of daily rainfall in future development of the system.
Figure 6.1 MRE between observed and simulated design peak discharges (2 – 100 year return period) when applying the single versus incremental UH approach