DECLARATION 2: PUBLICATIONS
7. A COMPARATIVE PERFORMANCE ASSESSMENT BETWEEN THE FINAL
7.2 A Brief Overview of the ACRU and SCS-SA Models
7. A COMPARATIVE PERFORMANCE ASSESSMENT BETWEEN
section provides a brief overview of each model and explains how design flood estimates are determined in each case.
In the ACRU model historical time series of observed daily rainfall and additional climate data, such as temperature or A-pan evaporation, are input to the model together with soils and land cover information to simulate streamflow on a daily basis. Streamflow in the model comprises of both stormflow (surface runoff) and interflow/baseflow. The soil water budgeting routines of the ACRU model explicitly account for antecedent soil water conditions on a daily basis.
Rainfall adds water to the soil water store and evapotranspiration depletes water from the soil water store. The antecedent soil water content directly influences the simulated daily streamflow response, e.g. if a rainfall event on a particular day is preceded by another rainfall event on the previous day, and with that amount of rainfall exceeding the amount of evapotranspiration, the streamflow response on the day will be higher than that of the previous day since the soil water store is closer to full capacity and therefore more streamflow is generated. To estimate design streamflow volumes and design peak discharges, the AMS are extracted from the simulated daily values and an extreme value distribution is fitted to the AMS to estimate the design values. Further details on the computation of streamflow and peak discharge in the ACRU model are provided in the previous chapters.
The SCS-SA model, adapted for South African conditions by, inter alia, Schulze and Arnold (1979), Schmidt and Schulze (1987a) and Schmidt and Schulze (1987b), from the SCS model developed by the Soil Conservation Service of the United States of America (SCS, 1956), is a deterministic event-based model that converts a design rainfall depth into a design stormflow volume (assumed to be surface runoff volume only) and a peak discharge estimate. In the most basic implementation of the SCS-SA model, the stormflow response is simulated based on a single fixed parameter representative of the average catchment stormflow response characteristics, i.e. the initial catchment Curve Number (CN-II; Schmidt and Schulze, 1987a).
Therefore, antecedent soil water conditions are not initially accounted for. For South Africa, two approaches were subsequently developed to adjust CN-II to account for antecedent soil water conditions, namely the Median Condition Method (MCM) and the Joint Association Method (JAM). The MCM is used to adjust initial CNs, i.e. derived from soil properties and land cover / management practices, to a final CN using the Hawkins (1978) equation. The Hawkins (1978) equation computes the water balance to calculate the change in storage within
a soil, and in the SCS-SA model this water balance was computed for a 30 day period leading up to the five largest independent rainfall events from each year. The change in storage was simulated using the ACRU model for 712 homogeneous hydrological response zones and 27 combinations of soil and vegetation properties (Schmidt and Schulze, 1987a). In terms of the MCM, the 50th percentile (median) change in soil water is used to adjust CN-II to a final CN.
One of the limitations of this approach, however, is the inherent assumption that the T-year return period rainfall event produces the T-year return period flood (Schmidt and Schulze, 1987a). The JAM, on the other hand, performs a frequency analysis on the simulated flows from the five largest events in each year of record, and therefore accounts for the joint association of rainfall and runoff, where the second, third or fourth largest rainfall event in each year may produce the largest flood.
It is important to note that for both the basic implementation of the SCS-SA model with CN-II, i.e. no antecedent soil water adjustment, as well as for the MCM, there are several options available to estimate design rainfall. These include: (i) by rainfall station search, (ii) from the hydrological response zone's representative station, (iii) user entered values for selected return periods, and (iv) design rainfall estimated using a regional, scale invariance approach (Smithers and Schulze, 2002). Refer to Schulze et al. (2004) for further details relating to each approach.
In the development of the MCM and JAM, however, the change in soil water used to adjust CN-II was calculated using rainfall data from the hydrological response zone's representative station. Since the methods were developed prior to 1987 the rainfall records were relatively short, approximately 20 years (Schulze et al., 2004). Therefore, when applying the MCM any of the four options listed above may be used to estimate design rainfall, however, the CN adjustment of CN-II is based on the median (50th percentile) soil water change calculated for a specific land cover and soil combination using the rainfall data from the hydrological response zone's representative station. When applying the JAM, the user does not have an option as to which method to use to estimate design rainfall, since the method does not use design rainfall estimates. This is because a frequency analysis was performed on the simulated stormflow volumes, as obtained from the five largest rainfall events in each year of record, i.e. for the length of record available for the hydrological response zone's representative station. In each case the actual soil water change prior to each event was used to adjust CN-II to a final curve number which is used to calculate the stormflow response to design rainfall. The 50th, 80th, 90th and 95th non-exceedance percentiles, which correspond to the 2, 5, 10 and 20 year return
periods, were recorded and the results stored in summary tables for each homogeneous zone for a range of CN-II values representing each soil and land cover combination simulated (Schmidt and Schulze, 1987a; Schmidt et al., 1987). The non-exceedance percentiles are specific to the rainfall records used in the development of the JAM and are not directly comparable to results obtained from design rainfall estimates derived from other sources, i.e.
which may use different rainfall stations and have different record lengths. In addition, the return period stormflow values calculated from non-exceedance probabilities are not equivalent to return period stormflow values calculated from design rainfall estimates, i.e. as obtained from an extreme value distribution fitted to the AMS of daily rainfall. Therefore, particularly for the higher return periods, i.e. the 20-year return period, large increases in the stormflow volume and peak discharge quantiles occur when using the JAM compared to when the MCM or CN- II is used (Schmidt and Schulze, 1987a). Examples of this are provided in the results section of this chapter. Therefore, as recommended by Schmidt and Schulze (1987a), the JAM should be used for lower return period events (2 – 10 years), and the results for these lower return periods may be compared to those obtained from the MCM method, to identify if possibly the 20th or 80th percentile antecedent soil water change should be used to adjust CN-II, instead of the median (50th percentile).