D Hughes, D Watkins, F Münster & B Cobbing

Institute for Water Research

Rhodes University

November 1998



It had already been recognised in 1996 that the volume of water (expressed as a percentage of mean annual runoff - MAR) that was required for the environmental reserve, generally decreased as the desired future state (now referred to as the Management Class) moved from category A down towards category D. However, there exists a large amount of scatter of values within each management class which could be related to :

    • The subjectivity inherent in setting any IFRs due to the complex processes involved and the lack of any monitored information to confirm the results after a development has taken place.


    • Differences in the flow regimes of the rivers concerned.


    • Regional differences in riverine ecological processes or the specific ecological significance of a river reach or cross section.
    • Differences in geomorphological characteristics of the channels.


The first point is difficult to handle in any scientific study and can only be minimised at this stage of the development of the BBM by selecting those IFR results in which a reasonably high degree of confidence has been expressed by the specialists involved. The last two are difficult to address without more detailed analysis of eco-hydrological relationships. While the second and third points are considered by the present authors to be not independent, but neither are they considered to be totally dependent. This initial document only addresses the differences that might be due to differences in the flow regime characteristics of the rivers.


This component of the preliminary reserve programme is concerned only with attempting to derive relationships between the past IFR results and the natural hydrological (or flow) regime of the rivers. It was therefore necessary to accept at the start of the project that no precise relationships would be found because other factors had to be initially ignored. A purely numerical statistical approach would have been unlikely to yield any useful results and therefore a more 'conceptual' approach had to be adopted. This 'conceptual' approach has largely involved identifying those components of the flow regime that can logically be considered to have some bearing on the flows that would be required to maintain the ecological functioning of a river system. The main hypothesis is that rivers which have more variable regimes (in several senses) are likely to require less water to function ecologically at a specified level, than those which have more regular regimes and therefore more constant and reliable habitats.




2.1 Information available from past IFRs


The study was constrained to be able to evaluate any relationships based on the information that could be readily obtained from past IFRs in which a reasonable degree of confidence has been expressed in the results. The following relevant information was generally available :


    • Total flow volume for the four building block components (i.e. maintenance low flows, maintenance high flows and freshes, drought low flows and drought higher flows and freshes).


    • Monthly distributions of the four building block components.


    • Present ecological state (A to F) and future management category (A to D).


In many cases time series of daily flow data that were used in the IFR workshop to check the range of flows set for the IFR were available, but not in all cases. This was because the authors were not involved in all past IFR's and it would have been impossible in the time frame of the project to contact the relevant specialists and obtain all the data. In some cases, only monthly data were used in the workshops. It should therefore be recognised that for some sites, the comparisons between the results and indices of the natural hydrology were not really possible as no daily time series were available. In such cases, some indices based on natural (or naturalised) monthly time series were used coupled with other indices based on daily time series from nearby flow gauges that do not necessarily represent natural conditions. This issue will be referred to later.


2.2 Constraints related to future extrapolation


If any method developed is to have value in terms of the preliminary reserve project, it must be based on indices of flow regime characteristics which can be quantified using readily available information for the whole of South Africa. The most widely available source of information is the WR90 publications (or electronic data) which contain data at the quaternary scale and with a monthly time resolution. If this project makes use of any indices that are based on daily time series data (from whatever source), it must also provide a solution to estimating their value at sites where daily data are not readily available. A parallel study (being conducted largely by the fourth author) is therefore investigating the regionalisation of some indices (or flow regime characteristics) that have to be currently estimated from daily data.


One of the problems that will always be present is the fact that the information that is widely available, nationwide, is based on quaternary catchments. It is extremely difficult to extrapolate WR90 data to smaller catchments due to the complexities of hydrological processes that affect scale variations and the fact that gauged data for smaller catchments are not widely available.




The following hydrological indices were extracted from the daily data that were available for each IFR site that had formed part of a previous workshop. In each case an explanation of the reasons for considering them important is given.


    • Annual coefficient of variation (standard deviation of annual flow volume / MAR). This represents the annual variation in total volume of runoff, can be considered a coarse representative of flow reliability and it might be expected that rivers with high annual CV's would have lower IFR requirements. It is also readily estimated (for catchments at the quaternary scale) from data contained within WR90.


    • Seasonal monthly coefficients of variation based on the main summer months (JFM-CV; January, February and March) and the main winter months (JJA-CV; June, July and August). These are readily estimated from WR90 data but might be expected to have different effects in the different rainfall regions (i.e. the main summer rainfall region, winter rainfall region and all-year rainfall region). The JJA-CV, for example, may be useful in the context of lowflows (maintenance and/or droughts) in summer rainfall region rivers, where the requirements might decrease with increasing CV. However, lower wet-season CV's might suggest higher (as well as more reliable) baseflow contributions, which could also have a bearing on the lowflow IFR. Wet season CV's could also be the only readily available index which has a bearing on the variations in high flow requirements.


    • Daily CV's for winter and summer seasons have also been extracted from the daily data available. These indices could be more useful than monthly based CV's, particularly in the wet season and in relation to the setting of the high flow requirement of the reserve. However, these are not likely to be generally available, unless they can be regionalised.


    • The authors of this document consider that the separation of total flow into baseflow and higher flow components based on traditional hydrograph separation procedures could yield an important hydrological index. This is based on the assumption that if the natural regime has a high baseflow component, it might be expected that this would be reflected in the ecological functioning of the river and consequently, the IFR results. However, an index of baseflow contribution (BFI) really requires daily data to derive (and is generally different between natural and impacted regimes). Therefore, while a conventionally derived BFI has been used, it has also been necessary to investigate methods of regionalising the index using relationships with monthly WR90 data.




Wherever possible, the same daily time series that were used during the workshops were used in this study (a mixture of observed, simulated and patched or extrapolated data), in other cases the best available data to represent natural conditions were used. Unfortunately, in some cases these data were from gauges located on rivers with large scale upstream impacts and could not be considered to be representative of natural conditions. Table 1 provides some more details about the source of the daily time series used and it is clear that for some sites the data can be considered more reliably representative of natural conditions than others.


Table 1 Source of daily time series of flows


IFR Site or River Source of data & comments
Sabie/Sand sites Data available from the workshop, which was based on VTI model simulations of natural conditions.
Komati River Data available from the workshop, which was based on VTI model simulations of natural conditions.
Mkomazi River Data available from the workshop, which was based on patched and extrapolated observed data using natural MAR's.
Luvuvhu River Data available from a VTI model simulation of the upper reaches of the Luvuvhu.
Tugela River sites Data available from the refinement workshop using data patched with a correction for present day impacts.
Letaba River sites Data available from B8H010 (IFR1), B8H017 (IFR2) and B8H008 (IFR3), all records being patched to the same length of 36 years. All records appear to be impacted by abstractions.
Mogalakwena sites Data available from gauge A6H009 (near confluence with Limpopo). 16 years of data that are unlikely to be very natural.
Bivane River sites Data available from gauge W4H004 on the Bivane River close to IFR1&2. Some 20 years of data that appear to be close to natural conditions.
Mvoti River sites Data available from U4H002 (30 years) which is located upstream of IFR1 and is affected by afforestation and irrigation abstractions. These data are not very suitable for the downstream IFR sites.
Berg River sites Data available from G1H036 (18 years) which lies between the two IFR sites. The records are affected by afforestation and irrigation abstractions.




In general terms the procedures followed during this study have involved attempts to identify suitable hydrological indices to relate to the four components of the BBM (maintenance/drought, lowflows/highflows), expressed either in volumetric terms, as a percentage of natural MAR or as a percentage of another component. The next step was then to attempt an approach that allows the monthly distribution of the component to be defined. Parallel steps have necessarily involved trying to explain some of the anomalies, making use of site specific information related to known differences in the ecological functioning of the systems, or special circumstances that prevail at individual sites or river reaches.


Figures 1 and 2 illustrate the general relationships between management category and the total requirements set for maintenance and drought years, expressed as a percentage of natural MAR. Given that the requirements are expected to generally decrease with an increasing value (A D or 1 7) of the management category, four sites are immediately noticeable as anomalies (Sabie 6 and 8, Berg 1 and 2). As far as the drought requirements are concerned, the main anomalies are the two Berg sites.


The following sub-sections discuss the results of the attempts to account for the variation within each management category


5.1 Maintenance low flows


These are expected to INCREASE with increasing values of the baseflow index (BFI - always less than 1), but DECREASE with increasing values of flow variability. It therefore follows that if indices of variability and baseflow response are to be combined they would have to have opposite effects. Figure 3 illustrates the best results obtained in the study thus far and uses a combined index of the sum of the maximum monthly CV's for the summer (JFM) and winter (JJA) periods, divided by the BFI. Non-linear estimation equations have been derived that fit through many of the points and that appear to be sensible from a purely hydrological point of view (figure 4).


The A/B line is difficult (!!) to locate as there is only one example of this category. The B line passes successfully through, or close to, four of the 5 high confidence sites, while Sabie 8 lies above the line, Bivane 1&2 falls lower and Luvuvhu 3 (low confidence) is way off the line. It is possible to suggest that the lowflow IFR for Sabie 8 (Sand River) was set quite high to ensure that this sand bed river remains perennial, while the Bivane site lies within a gorge and requires less water. The B/C line passes close to four of the sites, but Sabie 6 lies a long way above (high requirement) and Tugela 2 quite far below. The Sabie site (Mutlumuvi) is probably high to account for the need to rehabilitate this site after the collapse of Zoegnog Dam.


There are eleven sites falling into category C, four of which are on the Mogalakwena River, where the lowflow requirement is negligible (it is a temporary river). The line has been drawn more-or-less parallel to the other lines and apart from the two Mvoti sites, most of the others fall quite far above. Three of these are the Letaba sites where the CV data have been drawn from the naturalised monthly data used in the workshop, while the BFI data have been based on the observed records. The latter are therefore expected to be too low, and if corrected would bring down the value of the combined index used in the graph, making the points fall closer to the C line.


A C/D line has been drawn in, but does not pass close to the only C/D category site. The line is therefore positioned on the basis of extrapolation from the other lines. It should be noted that the only Western Cape river represented (Berg) has requirements that are far greater than these extrapolations suggest. The flow data used to evaluate the hydrological indices for the Mvoti 3 site is a long way upstream and it is possible that the flow regime at site 3 would be less variable with a higher baseflow contribution. This point would then move vertically downward toward the C line and could be illustrative of scaling variations between large and small catchments or rivers. Problems have also been identified with the hydraulics at this site.


5.2 Drought low flows


Attempts were made to relate drought low flows in a similar way to maintenance low flows, but there were less clear relationships. A second attempt related drought lowflows, expressed as a % of maintenance lowflows, to the sum of the monthly CV's for winter and summer. The relationship was far from clear and did not really provide a very useful method. Subsequently, discussions with various ecological specialists suggested that it does not really make sense to define different drought low flows for different management classes (i.e. there is really only one drought condition that applies to all management classes). Until a better approach can be proposed, drought flows are therefore taken as equivalent to the 'D' class flows (as shown in figure 4)..


5.3 Maintenance high flows


There is a relatively narrow range of values of maintenance highflow requirements (expressed as % MAR) but there does not seem to be any relationship with the available hydrological variables within the various management classes. This result could have been anticipated given the high degree of variability in site specific circumstances surrounding the setting of high flows (related to the channel size and shape, riparian vegetation needs and geomorphological needs in terms of sediment transport, amongst others).


In an attempt to obtain at least an initial estimation approach for high flows, the total IFR requirement was plotted against the same variable used in figure 3 (i.e. the sum of the monthly CV's for winter and summer divided by the BFI). The result (figure 6) is a set of relationships (for each management category) with a relatively high degree of scatter (outliers due to low flows and due to high flows are combined) but which could be of value. The lines for each management category follow the same approximate pattern as those in figure 4 (repeated as bold lines in figure 6), but are straighter and of course above the lowflow lines. If these lines are used in conjunction with figure 4 the result is values for the total maintenance IFR and the lowflow component, from which the high flow component can be derived. Because one set of lines is straighter than the others , applying these relationships gives variable highflow contributions for each management category of between 5 and 8 for C sites, 5 and 10 for B/C sites, 5 and 10 for B sites and approximately 8 to 15 for A/B sites (all values given as % MAR). In general terms, these figures do correspond to the experience from past IFR's, although there are quite a few outliers. The main outliers are Sabie 6 and 8 (related to the movement of sediement and rehabilitation), the Berg River sites (already identified under the lowflow section) and the Luvuvhu sites (low confidence results in general).


5.4 Drought high flows


The drought high flow requirements range from very small % MAR values to less than 3% and while it has not been possible to discover a suitable estimation relationship, this component is not of great importance.




Mean seasonal distributions of total flows and separated baseflows have been compiled for all of the IFR sites using the available data referred to in Table 1 and compared with the seasonal distributions of IFR lowflow and highflow requirements (where available). Some general observations could be made.


    • The largest differences between the natural and modified (i.e. IFR) lowflow regimes occur in the wet season, regardless of the IFR requirement. This is effectively the same as stating that the environmental reserve requires a higher proportion of the natural flows in the dry season than in the wet season.
    • As the IFR requirement decreases relative to the natural flows, a greater proportion of the wet season flows can be released for other uses, relative to the dry season. If the proportions remained similar, there is the possibility that dry season flows would almost disappear for rivers with very low requirements.


    • While the shape of the natural flow distribution cannot be neglected, it appears to be of less importance in defining the final lowflow IFR distribution than other factors.


  • There are apparently fewer consistent relationships between the monthly distributions of high flows at the various IFR sites than there are for the lowflows.

6.1 Monthly distributions of lowflows


An attempt was made to determine generalised distributions and sets of weighting factors that could be estimated from the total maintenance or drought lowflow requirement and then applied to the distribution of natural baseflows. This was found to be a practical approach and gave reasonable results for most of the sites included in the analysis for both maintenance and drought low flows. The method relies upon having natural baseflow distributions available and the assumption was made that these would be supplied from a regionalisation of natural baseflow characteristics (part of a parallel study to this one). The curves are all based on a single master distribution which is then coupled to a weighting equation for each month of the year which is controlled only by the total value for the year as a whole. The correct curve to use is determined, iteratively, by changing the total value until the total of the monthly IFR lowflows equals the correct value (set in the workshop or extracted from figure 4). Approximate, non-dimensional, monthly distributions of baseflow contribution (as % of mean monthly total runoff) for all the regions of South Africa have now been compiled.


6.2 Monthly distributions of high flows


While there are a number of differences between the results for the various sites, some generalisations can be made. Some of the differences appear to be related to whether or not large flushing flow events were specified in the workshops, or whether the assumption was made that these would occur naturally (i.e. would not be affected by the water resource development) and were therefore not specified.


The estimation approach that was finally adopted initially involved expressing the mean monthly natural high flows (total flows - baseflow) as a percentage of their total for the year. If the IFR high flow volume requirements are similarly expressed as a % of their total for the year, then useful patterns seem to emerge (for the summer rainfall regions).

  • The IFR % values for October and November are generally very similar to the natural high flow % values.
  • There are very few sites where high flows have been quantified for April to September.
  • The March IFR % value varies between 0.5 and 1.0 times the natural high flow % value (and has been generalised as 0.75).
  • If the natural high flow peak occurs in February, the highest IFR high flow requirement is in February, while the December requirement is approximately twice the January requirement. If the natural high flow peak occurs in January, the highest IFR flow requirement is in January and the December and February requirements are often similar.
  • The highest requirement (in January or February) can be associated with the natural peak relative to the natural total volume over the three wettest months.


The above 'rules' have been formalised and modified for the other regions of South Africa (winter rainfall, various aseasonal regions and some summer rainfall regions with different flood seasons). They can be considered to reproduce the actual IFR results reasonably well across the range of sites for which data are available. Inevitably, the final volumes for each month are very dependent on the estimation equations for total and lowflow maintenance volumes, from which the total high flow volume is derived.




The estimation procedures discussed above provide estimates of the monthly distributions of maintenance and drought lowflows, as well as maintenance high flows. It is also necessary to have procedures available to specify when maintenance (or above) and when drought conditions should apply. This introduces the concept of assurance (i.e. how often are maintenance, or above, flows required). There is very little past experience of the use of this concept within IFR workshops, although recently the IFR model has been used to specify a time series of IFR flows from which the assurance information can be extracted.


In the absence of further information, a regionalised set of monthly rule curves has been generated based on the flow duration curve characteristics of the natural flow regime. Thus, a river with a relatively reliable flow regime would be expected to have rule curves that suggest high assurance (about 70%) of maintenance flows, maximum flows not a great deal higher than maintenance (some 20% higher, for example) and a relatively short duration of time at drought conditions. A more 'flashy' river would be expected to have lower assurance of maintenance, a higher maximum (about 50-100% higher, giving a more variable regime) and a longer time at drought conditions. These rule curves are the equivalent of flow duration curves (in that they are based on plots of % assurance versus flow) and have been based on a generic, smooth curve algorithm with 4 parameters). To generate a time series of monthly IFR flows the following steps are followed.


    • Identify the flow value for a specific month in the time series of natural flows.


    • Determine the % time equalled or exceeded on the calendar month flow duration curve.


  • Use the same % time with the assurance rule curve and determine the total flow requirement of the IFR.


As already noted, a parallel study has been carried out to try and derive regional relationships for the natural flow regime variables required by the proposed estimation methods. These variables and the currently available methods of estimating them are as follows :


    • The baseflow index (BFI), a value representing the average proportion of total flow that occurs as baseflow. Good relationships have been derived (based on some 70 time series of flow data) between BFI and Q75 taken from the 1-month annual flow duration curve using natural data. WR90 contains monthly time series (1920 to 1989) for all the defined quaternary catchments in South Africa, from which a duration curve can be compiled and the Q75 value estimated. Initial results indicate that the relationship is different for Western Cape rivers than the rivers of aseasonal and summer rainfall regions. There is also a suggestion that the nature of the relationship varies with catchment size.


    • Monthly CV's for the main wet and dry season. The values used in the estimations of IFR lowflows and total flows are the maximum CV for the months of January, February and March (JFM) and the maximum CV for the months of June, July and August (JJA). Clearly, the main wet and dry season months are different for the winter rainfall region, but as they are combined in all the estimation methods (figures 4 and 6), the issue is not relevant. Estimates of these values can be obtained from the data provided in WR90.


    • Monthly distributions of natural total flow and baseflow. Total flow distributions can be obtained from WR90 data directly, while distributions of baseflow have been regionalised by the fourth author. The results are more than acceptable for several summer rainfall regions and there is a high degree of consistency in the non-dimensional (values expressed as %MAR) shapes of the distributions. The distributions are less consistent for the other regions of the country (winter rainfall and aseasonal regimes).


A Windows95 computer program has been developed that allows all the estimation approaches discussed in this document to be applied. Associated with the program is a data file that includes all the parameters of the estimation equations, as well as the parameters of the regionalised equations and distributions. The program also allows the user to enter the scores for the criteria used to estimate the habitat integrity (according to the method developed by Kemper and Kleynhans as part of the preliminary reserve project) and provides a facility for the initial estimate of the management class. The program therefore represents a Decision Support System for applying the hydrological based estimation procedures of the preliminary reserve approach.



While some useful results have been achieved at this stage of the project, there are still a number of issues that have to be resolved.


    • The amount of IFR workshop data is very limited and this does not make it very easy to determine generalised relationships. Many of the rivers in the sample analysed are within similar hydrological regions and it is difficult to judge whether the derived relationships would hold for other areas. This point is very relevant given that the two Berg River sites are always anomalous.


    • The relationships derived thus far are tentative and require further work to check and improve.


    • There needs to be a much greater specialist ecological input into checking the usefulness of the relationships and identifying anomalies that need to be accounted for in some way.


    • Although the last point referred to 'anomalies', certain sites may only be seen as such because the data set is limited in size. Given more data, the current 'anomalies' may in fact be part of a different group of sites with their own unique characteristics.


    • There is a need to investigate or conceptualise the likely effects of river/catchment size on the relationships and to attempt to incorporate more ecological relationships.


  • The development of relationships between hydrological characteristics and the environmental reserve requirements should be a continuing process that can accommodate additional information as it becomes available.


Figure 1 Maintenance flows set for the rivers in different management categories (A/B=2, B=3, B/C=4, C=5, C/D=6, D=7)


Figure 2 Drought flows set for the rivers in different management categories.



Figure 3 Maintenance lowflow requirement (expressed as % natural MAR) versus the combined CV/BFI index. The top part of the diagram has river name labels, while the lower diagram has management category labels.


Figure 4 Non-linear estimation relationships for maintenance lowflow volumes based on management category.


Figure 5 Relationship between combined CV and baseflow index and total maintenance IFR requirement.


Figure 6 Lines representing the relationships for maintenance lowflows (bold lines - taken from figure 4) and maintenance total flows (thinner lines).

Last Modified: Thu, 10 Sep 2015 11:57:39 SAST