Datasets on alien plant invasions were obtained from a range of sources, updated and combined to produce a dataset for the whole area covered by the catchments in the WCWSS. e spatial datasets were as follows: 1. Invasion mapping done for an assessment of water availability in the Berg Water Management Area (WMA 19) carried out by Aurecon (supplied by Cheryl Beuster of Aurecon) and known as the Berg WAAS (DWAF, 2010) .

is dataset was edited to correct species attribute data as veri ed in the eld, mainly where pines were listed when the actual dominants were Populus species, Eucalyptus species and Acacia species. e plantation dataset from this same study was checked to con rm whether or not these areas were plantations and not unmanaged invasions. 2. Data from CapeNature based on mapping of the invasions in the management compartments in the nature reserves that overlapped with the study catchments (supplied by erese Forsyth). is mapping was done in 2010/11. 3. In the Upper Berg (G10A) information was taken from mapping done by the CSIR for the Working for Water programme in 2000. is was cross-checked with data on the pre-treatment state of areas cleared under Working for Water contracts. Data extracted from the Working for Water Information Management System were used to re ne the information on invasions in both the Franschhoek River and Berg River catchments. 4. Data for invasions in the lower catchment from mapping done for local municipalities in the early 2000s for the CAPE Fine-scale Conservation Planning Studies, and the West Coast invasion mapping done for the CAPE project in 1999. ese datasets were cross-checked and compared with Google Earth images to re ect the state of invasions in the late-2000s to bring them in line with the Berg WAAS. 5. Additional polygon data were mapped in Google Earth from images dating from the early to middle 2000s to avoid including more recent clearing operations.

e species data for the combined set was edited to ensure that the species names and density values were consistent. Each of the invaded units (polygons) was identi ed as being riparian or non-riparian, or one where groundwater could be accessed by plant root systems. is information was required for the estimation of the hydrological impacts as an indication of relative water availability (Le Maitre et al., 1996; Le Maitre et al., 1999, 2013) . Where necessary, invaded polygons were sub-divided to de ne the riparian sections more accurately. e datasets were then combined with data on the sub-catchment boundaries of the WCWSS to provide information on the extent and state of invasions in each of the sub-catchments in dryland, riparian and groundwater settings.

e existing invasions were then projected forwards for 45 years using a sigmoidal curve for the spread as has been found in many studies of invasions (Le Maitre et al., 2002; Hengeveld, 1989; Birks, 1989) , a spread rate of 10% per year (Van Wilgen and Le Maitre, 2013) and a densi cation rate of 1% per year.

e use of this model results in slow initial invasion followed by a rapid increase which then slows as the invadable land becomes fully occupied. e National Land Cover 2000 dataset (Van den Berg et al., 2008) was used to extract the remaining natural (i.e. invadable) areas in each sub-catchment by excluding all the transformed land classes (e.g. cultivated, forest plantation, urbanised). e river lines from 1:50 000 topographic map series were bu ered to create a layer which de ned the riparian zones. Areas mapped in the land-types (LTSS, 1972-2002) as having deep sandy soils were used to de ne areas where groundwater would be accessible to plant root systems (Le Maitre et al., 2013) . is information was used to divide the potentially invadable areas into drylands, riparian and groundwater access for the spread modelling.

e projection of the invasions was then done for each subcatchment in a spreadsheet.

e approach taken in this study is the same as was used in previous studies of the e ects of ‘No management’ on invasions (Le Maitre et al., 2002; Van Wilgen et al., 1997) . We selected the year 2000 as the base year for the simulations and projected invasions forwards to 2045 because 45 years generally is used as the life-span of the infrastructure in assurance of supply studies like this one. e 2000 invasion data were used to establish the ‘Baseline’ scenario, with the projected invasions in the year 2045 providing the ‘No management’ scenario and removal of all invasions the ‘E ective management’ scenario.

One of the criticisms levelled at the original ow reduction model developed for invasions (Le Maitre et al., 1996) was that the reductions were estimated as the equivalent depth (i.e. mm), which can result in overestimations. is was addressed by revising the approach to estimate proportional reductions (Dzvukamanja et al., 2005; Le Maitre et al., 2013) , like those formerly used to estimate reductions a er commercial a orestation (Scott and Smith, 1997) . is meant that reductions caused by dryland invasions could be estimated by matching the plant growth and growth form characteristics to those used in the plantation stream ow reduction models (Table 1).

Whilst this approach can be used for dryland invasions, it is likely to underestimate water-use in riparian invasions or where groundwater is accessible (Le Maitre et al., 2015) . In riparian and groundwater settings evaporation from vegetation is driven mainly by the available energy and can exceed the annual rainfall if su cient water is available (Dye and Jarmain, 2004; Everson et al., 2014; Le Maitre et al., 2015) , providing the plants’ hydraulic conductivity is high enough and they do not regulate their transpiration by closing stomata when internal moisture stress or vapour pressure de cits are high (Manzoni et al., 2013; Calder, 1991; Whitehead and Beadle, 2004; Jarvis and McNaughton, 1986) .

e evaporation from riparian or groundwaterlinked stands of invading plants can be estimated using micrometeorological techniques or remote sensing. e CSIR obtained high-resolution, remote-sensing estimates of evaporation (Et) for two sample areas in the upper part of the Berg River catchment from the eLEAF Competence Centre in e Netherlands – these being Franschhoek to Klapmuts, and the Berg River oodplain for 5 km upstream of Hermon (Le Maitre et al., 2016) . e Et was estimated using the proprietary Surface Energy Balance (SEBAL) model (Bastiaanssen et al., 1998) as part of a study of irrigation water-use in the wine and deciduous fruit growing areas of the Western Cape, known as Fruitlook (http://fruitlook. co.za/). eLEAF provided estimates of the weekly evaporation from 2 October 2013 to 29 April 2014 and for individual cloud-free days which coincided with satellite passes from May to September 2014. e data are at a resolution of 20 x 20 m which is suitable for estimating annual evaporation from fairly narrow strips of riparian vegetation.

Riparian invasions which overlapped the two sample areas were selected from the mapped data and screened for those dominated by particular species and having natural riparian vegetation in good condition. Invaded areas situated on steep southerly aspects were avoided as they are subject to topographic shading e ects which are a problem for energybalance-based methods like SEBAL. Evaporation data for each of the sample periods was extracted and used to calculate the monthly and annual evaporation. e weekly data were simply summed as they provided a continuous record. e single day data for the winter period were multiplied up by the number of days in the month to calculate the corresponding monthly totals (Table 2). is approach could overestimate the monthly total as there are many cloudy days, but a comparison with measured evaporation for the same days and for the month at a site near Hermon (Dzikiti et al., 2016) suggested that the overestimate for those months was not signi cant. However, the SEBAL data were found to overestimate evaporation by about 20% compared with ground-based measurements. Unfortunately it is di cult to correct for this for all the invaded areas without more ground-truthing at other sites, especially in montane environments. However, what really matters for the modelling is the di erence in the evaporation from invaded riparian and natural riparian areas which is less in uenced by this systematic error. e values that were obtained are in line with those from other studies for similar species, for example Acacia mearnsii (Dye and Jarmain, 2004) , especially given that subsequent work has found much greater interception losses in stands of this species (Everson et al., 2014) .

Flow reductions for invaded upland areas were estimated from taxon-speci c stream ow reduction factors (Table 1). In areas where groundwater was likely to be accessible to invading plants (e.g. deep sandy soils), the upland ow reduction factors were increased by 20% to allow for the greater water availability (Van Wilgen et al., 2008) . Riparian IAPs have relatively direct access to water, both in the riparian soil and owing past from upstream. e impacts of riparian invasions were estimated using data on the actual evapotranspiration for the di erent taxa (Table 2). A spread-sheet was set up for each sub-catchment to generate a time series of riparian stream ow reduction as follows: (i) Multiply the invaded riparian area in the sub-catchment for particular taxa by the 12 relevant mean actual monthly evapotranspiration values (Table 2); (ii) determine the 12 incremental mean monthly evapotranspiration values from each taxon a er accounting for the mean monthly evapotranspiration from ‘natural’ riparian vegetation (indigenous montane); and (iii) for each taxon generate a time series of dynamically-varying incremental monthly evapotranspiration values by inverse weightings derived from the sub-catchment monthly rainfall sequence, standardised to reproduce the 12 mean incremental values derived in step (ii) above. is inverse procedure ensured that, during any high-rainfall month, the incremental evapotranspiration was less than the equivalent value in Table 2 for that month and vice versa, while preserving the 12 long-term means.

e magnitude of the riparian stream ow reduction in several catchments was so small that it became necessary to add the riparian SFRs for several catchments together. A total of seven riparian stream ow reduction locations were used in the yield model to assess the ow reductions evenly throughout the WCWSS. Indigenous riparian lowland

(Hermon) (regenerating after clearing in 2010) 120 102 73 35 25 27 26 34 53 95 105 125 820

e full set of con gurations of the Pitman (WRSM2000) rainfall−runo catchment model for all the sub-catchments used in the Berg WAAS (DWAF, 2010) was de-archived and checked for functionality. e invaded area coverages were intersected with the individual sub-catchment boundaries used in the WCWSS WRYM system yield model (herea er WRYM) and the individual IAP taxon areas, and their densities, quanti ed on a sub-catchment basis. ese data were then used as input to the Pitman catchment model for the generation of monthly SFR sequences per sub-catchment for each of the three scenarios. e information on the invasions and stream ow reductions was included in the Pitman model con guration les and the model was then run to generate sequences of monthly stream ows for all sub-catchments for the three invasion scenarios. ese stream ows were used to populate the latest con guration of the WRYM (DWS, 2014) .

e WRYM model con guration was modi ed to accommodate a large number of additional SFR ‘demand’ nodes necessitated by the updated IAP mapping and the scenarios. e Pitman-generated SFR sequences were input into the system model as ‘demand’ les at each of the aforementioned nodes.

e historical time period covered by the WRYM simulations is 77 hydrological years (1928/29–2005/06). e system model was subsequently run in stochastic mode to determine yieldassurance relationships for the three scenarios. For this purpose, and for each scenario, 201 di erent sets of equally likely stochastic natural stream ow sequences, based on the historical sequence, were generated for all the stream ow input nodes.

In the ‘Baseline’ scenario, the condensed area of alien plant invasions was 22 190 ha or 5.5% of the natural vegetation of the WCWSS (Table 3), ranging from 8.6% in the Berg River to 3.3% of the Riviersonderend catchment ( e condensed area is the equivalent dense area (i.e. 10 ha with 50% cover ≡ 5 condensed ha). Most of the invasions were not in in riparian areas or in areas where groundwater is accessible, with the latter (deep sandy soils) being most prevalent in the Berg and Riviersonderend catchments. e Berg River catchment has the most extensive riparian invasions (16.2% of the total) which occur along the Berg itself and along most of its tributaries, and are dominated by eucalypts and black wattle. Almost all the remaining natural vegetation in the WCWSS is invaded to some degree (Fig. 4). e most densely invaded areas are situated in the upper Berg and upper Riviersonderend catchments, particularly between Villiersdorp and Paarl (Fig. 4). is is important because this is also the portion of the WCWSS which gets the highest rainfall and generates most of the runo (Fig. 2).

By 2045 with ‘No management’, the condensed invaded area equates to about 112 000 ha or 28.1% of the remaining natural vegetation (Table 3). e Berg River is the most invaded, with the condensed area being 50.8% of the natural vegetation, in other words a mean density of about 50%. e Riviersonderend catchment is the least invaded at 10.6%, mainly because it was less invaded than the Berg initially, but the invasions are concentrated in high runo areas. e most extensive invasions are by pines which have a condensed area of about 11 000 ha in 2000 and a projected 70 000 ha in 2045. Flow reductions

e mean annual stream ow reductions for upland and riparian invasions in each of the sub-catchments are substantial, especially with ‘No management’ (Table 4). e greatest reductions for ‘Baseline’ upland invasions are found in the Berg and Riviersonderend sub-catchments with the total reductions for upland invasions coming to 71.0 mill. m3∙a−1. Reductions due to riparian invasions are substantial in the Berg River, but much lower in the other catchments. With ‘No management’ the total reductions due to upland invasions increase substantially to 307 mill. m3∙a−1, especially in the Berg River and Riviersonderend where they could increase 4.5 and 3.0 times, respectively (Table 4). Under ‘No management’, reductions due to riparian invasions also increase substantially to 16.8 mill. m3∙a−1 (3.1 fold), with most of the reductions being found in the Berg River catchment. e relatively low increases in the Riviersonderend catchment are due mainly to the relatively low density of invasions in the ‘Baseline’ state (Table 3) which results in relatively low rates of increase in the density.

e greater density of the invasions in the Berg River catchment leads to a more rapid increase in density and greater reductions under ‘No management’, emphasising the importance of controlling invasions at an early stage.

Stochastic yield reductions

e stochastic analysis of the WCWSS outlined above demonstrate that the stream ow reductions for the ‘Baseline’ and ‘No management’ scenarios result in signi cant impacts on the WCWSS yield over the range failure recurrence intervals (Fig. 5). e yield for the 1:50 year recurrence interval of failure (98% assurance) is widely used in long-term planning for domestic supplies. At this level of assurance, the yields for the WCWSS are 580 mill. m3∙a−1 for the ‘E ective management’, 542 mill. m3∙a−1 for the ‘Baseline’ and 450 mill. m3∙a−1 for the ‘No management’ scenarios. e ‘Baseline’ reduction of 38 mill. m3∙a−1 is equivalent to losing two thirds of the capacity of Wemmershoek Dam every year, or 6.6% of the WCWSS yield of 580 mill. m3∙a−1 at 98% assurance of supply. However, under the ‘No management’ scenario, the reduction increases to 130 mill. m3∙a−1, which is 22.4% of the WCWSS yield, equivalent to losing the capacity of the Berg River Dam each year.

We examined the spatial variability of stream ow reduction impacts on yields within individual components of the WCWSS for the current and future invasion scenarios by determined the so-called ‘historical rm yields’ (HFYs) of three dams which are not a ected by inter-catchment transfers for the period 1928–2005 ( e HFY is determined by increasing the total abstractions in the modelled system by trial-and-error until the modelled system narrowly fails one year during the modelled historical period; i.e. the modelling is not performed in stochastic mode). is approach ensured that the modelled HFY impacts would be solely due to the modelled invasions in each dam’s catchment, i.e., that interpretation of these modelled impacts would not be confounded by inter-catchment transfers. e three dams selected were Wemmershoek Dam in the Berg catchment, Eikenhof Dam in the Palmiet catchment and a hypothetical dam at the location of eewaterskloof Dam in the Riviersonderend catchment. e reason for conducting this exercise for a hypothetical eewaterskloof Dam is that, for the purposes of accommodating winter transfers of surplus in ows from the Berg River Dam into eewaterskloof Dam, the operational full supply capacity (FSC) of the existing dam is about 200% larger than what the upstream catchment could sustain. e spatial di erences of proportional impacts on yield due to invasive alien plant invasions (Table 5) are related primarily to a combination of two factors: the extent of invasions representing the ‘Baseline’ condition and the extent of invadable areas for the ‘No management’ condition.

We also examined the long-term spatial variability of stream ow reduction impacts on average annual water supplies to individual water use sectors within individual components of the WCWSS for the ‘Baseline’ and ‘No management’ alien plant invasion scenarios for the period 1928–2005 hydrological years (Table 6). Given the close operational interactions between eewaterskloof Dam and Berg River Dam, the outputs of the various modelled supply channels from these two dams to individual users were combined for the purposes of these calculations. For the same reason, the individual simulated supply volumes from Voëlvlei Dam and the Misverstand Weir on the Berg River were also combined. Furthermore, the various supply volumes for minor schemes and farm dams were combined by sub-region to enhance interpretation of the spatial variations of IAP impacts on average annual water supplies. For the complete WCWSS, the total mean annual impacts of IAPs on urban supplies (in absolute volume terms) far exceed the corresponding impacts on irrigation supplies (Table 6). e greatest modelled impact of IAPs on urban supplies (in absolute volume terms) is on water supplied from Wemmershoek Dam. Under ‘No management’ conditions IAPs could be expected to consume almost half of Wemmershoek’s 98% assurance yield. e greatest modelled impact of IAPs on irrigation supplies (in absolute volume terms) is on water supplied from the combined eewaterskloof and Berg River Dams.

DISCUSSION

is study clearly demonstrates that the reductions in stream ows as a result of the greater water-use of invading alien shrubs and trees can have substantial impacts on dam and system yields, even for the WCWSS which comprises several large dams. is con rms the ndings of previous studies (Le Maitre and Görgens, 2003; Cullis et al., 2007) . In this case the invasions are mainly in the headwater catchments (Fig. 4) and so have a signi cant impact on stream ows throughout the WCWSS and thus on the yields of each of the parts of the system and the whole WSS.

e fact that the reductions under the ‘Baseline’ invasions were already substantial, as well as the magnitude of the projected reductions, demonstrates why authorities need to take immediate action to clear catchments now rather than delaying action. If no actions are taken, or if they are delayed, then invasions rapidly increase in density and the costs of clearing increase. Although standard discounting approaches may make it seem less costly to develop alternative water sources, this is misleading. e impacts of alien plant invasions are not stationary so, unlike the nite yields from alternative schemes, the yields from invaded bulk water systems will decrease over time (Fig. 1) while the cost of clearing increases. e WCWSS is particularly vulnerable because all its water is supplied by invaded catchments, so that all parts of the WSS will experience reduced in ows and yields (Fig. 1), decreasing assurance of supply. All of the water that is available in the WCWSS catchments is already allocated to users or the ecological reserve, so the only way to increase the assurance of supply is by clearing invasions. Given this, there appears to be an inconsistency in the actions of the Department of Water and Sanitation. One the one hand, they use the provisions of the section of the National Water Act on Stream Flow Reduction Activities to restrict a orestation because of its impacts on stream ow (Dye and Versfeld, 2007) . So, because the water resources in the WCWSS are already fully allocated, they would not have allowed further a orestation in the catchments supplying the WCWSS. Yet the Department seems to be unwilling to acknowledge that invasions by the same species used for forest plantations could be having similar impacts on runo and, as we have demonstrated, the yields. In other words, they should be actively supporting clearing but are currently not doing so and it is not clear why this is the case.

Developing alternative schemes merely transfers the cost of dealing with invasions to future generations as there are no indications that technological advances will radically reduce clearing costs in the future. The severe water shortages currently being experienced by Cape Town and the neighbouring towns within the WCWSS demonstrate clearly that maintaining or increasing the yields of the existing WSS is critical for both immediate and long-term water security. The estimated current-day (Baseline) yield reduction of 38 mill. m3∙a−1 could have provided Cape Town with about 54 days of water per year at 700 ML∙d−1 had the invasions been cleared.

e results of the long-term studies of the hydrological impacts of plantations showed clearly that the reductions in low (dry season) ows are greater than those on annual ows (Scott and Smith, 1997) , indicating a net depletion of catchment water storage. e depletion of catchment storage has been con rmed by other long-term studies where stream ow took time to recover a er clearing (Scott and Lesch, 1997; Everson et al., 2014) . ese ndings suggest that the e ects of invasions are likely to be greatest during droughts, just when the maximum yield is required from the WSS. In addition, most of the invasions are in upland environments so, even if the invading plants are cleared, the depleted catchment storage will take time to replenish and reach levels where stream ows normalise again. is, in turn, makes a strong case for not waiting until droughts are underway before prioritising clearing.

At the moment, clearing in the WCWSS catchments is being funded almost entirely by the Extended Public Works Programme rather than directly by the water-users in the WCWSS. Although users are paying for their water, the funds raised by these levies are not being directly applied to environmental management in these catchments. Dedicated funding and monitoring of the clearing operation by those who pay is the best way to ensure that this is achieved.

e simple message is that the sooner the water-users in the WCWSS invest in the clearing, including the use of biocontrol, the less it will cost them in total and the more they will bene t (Van Wilgen et al., 1997) . By putting measures in place to ensure that the clearing is e ective (Kraaij et al., 2017) , they can ensure that this is a wise investment which will secure water supplies for them and for future generations. Although this study focused on the WCWSS, the same reasoning would apply to all water supply systems where the catchments have been invaded by species with similar hydrological impacts.