Hydrology for the Water Management of Large River Basins (Proceedings of the Vienna Symposium, August 1991). IAHS Publ. no. 201,1991.

THE "ACRU" MODELLING SYSTEM FOR LARGE CATCHMENT WATER RESOURCES MANAGEMENT

K. C. TARBOTON & R. E. SCHULZE Department of Agricultural Engineering, University of Natal, Pietermaritzburg, South Africa

ABSTRACT The ACRU agrohydrological modelling system was applied to the critical runoff producing Midmar subcatchment of the Mgeni river basin, which supports the water needs of more than three million people and supplies water to industry and agriculture producing 20% of South Africa's Gross National Product. Projec­tions that the water resources of the Mgeni will be completely utilized by the year 2005, together with increasing urban, industrial and agricultural development make management of the Mgeni's water resources imperative. Verification of the ACRU modelling system on gauged subcatchments within the Midmar catchment is presented, then two scenarios are used to assess the impacts of agricultural development, in the form of increased afforestation and the proliferation of farm dams, on the catchment water resources. The potential of ACRU for use by managers and planners in the reconciliation of increasing and varied demands on limited water resources is illustrated in its ability to assess objectively the impacts of potential development scenarios.

INTRODUCTION

The 4387 km2 Mgeni catchment located on the east coast of South Africa (Fig. 1) supplies water to 3.6 million people and supports industry and agriculture producing 20% of South Africa's Gross National Product. Unique problems faced in this catchment include exceedingly rapid urban and agricultural development, water resources which are close to being completely utilized and a lack of knowledge of how development and land cover changes impact water resources. It has been predicted that the population in the area presently supplied could increase to between 9 and 12 million by the year 2025 (Home Glasson Partners, 1989) and that the water resources of the Mgeni catchment will be fully utilized by the year 2005. Competition for water from the increasing population concomi­tant with increased agricultural, urban and industrial development make effective water resource management within the Mgeni catchment imperative.

219

K. C. Tarboton & R. E. Schube 220

Fig. 1. Mgeni Catchment Location.

Agricultural development, predominantly in the upper reaches of the Mgeni catchment, impacts the water available to the large urban supply reservoirs while urban development, predominantly in the lower reaches of the catchment, increases the demand on the reservoirs. In this paper the Agricultural Catchments Research Unit (ACRU) modelling system is applied to the upper reaches of the Mgeni catchment in order to assess impacts of likely agricultural developments on water resources. Information obtained from hydrological simulation provides water resources managers with objective answers on how agricultural develop­ment, in the form of land cover changes, impacts water resources, thereby enhancing their ability to manage catchment water resources.

Developed in the Department of Agricultural Engineering at the University of Natal, Pietermaritzburg, South Africa, ACRU is a multipurpose daily soil water budgeting model capable of simulating runoff, reservoir storages, sediment yield, irrigation demand and supply, land use impacts and yields for various crops. ACRU is a physically based conceptual modelling system that idealizes and conceptualizes physical processes in the sequences that they would occur in nature. The ability of ACRU to evaluate the influence of the above broad range of processes on catchment response makes it an effective aid for decision-making when assessing various impacts on water resources.

As a pilot study to simulating water resources throughout the Mgeni catchment, the catchment upstream of the Midmar dam (Fig.l) was chosen for investigation since a substantial portion of the mean annual runoff (MAR) of the entire Mgeni catchment is generated within this area and, being subject to little industrial development it provides good quality water (Breen, Akhurst & Walmsley, 1985). Agricultural development, however, in the form of land cover

221 The "ACRU" modelling system

changes and the construction of numerous small farm dams continues. It was shown by Maaren & Moolman (1985) that there was a significant reduction in streamflow with time due to the construction of small reservoirs within the catchment, with this effect being accentuated during dry years. A further practice causing land cover change is the afforestation of land previously uncultivated or under annual crops. In South Africa the Forestry Council in its Strategic Forestry Development Plan (1989) called for an increase in the area under forestry by some 380 km2.year'1 from 1990 until the year 2010. This implies that there is likely to be a doubling of the present area under forestry over the next 20 years, much of it in catchments such as Midmar with a MAP generally exceeding 850 mm.

In this paper the ACRU agrohydrological modelling system is evaluated for its ability to simulate streamflow from the rural catchments upstream of the Midmar dam in the Mgeni catchment and used to assess the impacts of increased afforestation and the proliferation of farm dams on the catchment water resour­ces. Information on present land cover and the number of farm dams within the Midmar catchment is given and the effect on water resources that these dams have already had, is shown by simulating the potential streamflow without the dams. Implications of land use change in the form of increased afforestation are shown by investigating the local and regional effects on streamflow, of a doub­ling of the area under forestry in the Midmar catchment.

ACRU MODELLING SYSTEM

Structure

The ACRU agrohydrological modelling system (Schulze 1989) is structured on daily multi-layer soil water budgeting (Fig.2). Rainfall and/or irrigation not abstracted as interception by the vegetation canopy or stored on the surface is partitioned into stormflow and effective rainfall that enters the topsoil horizon.

Saturated drainage from the topsoil to subsoil horizon takes place when soil water in the topsoil exceeds field capacity. Similarly, when the soil water in the subsoil horizon exceeds field capacity, drainage to the intermediate and groundwater stores occurs, from which baseflow is generated. Unsaturated soil water redistribution between the two horizons occurs according to their relative soil water contents. Runoff comprises stormflow, in the form of quickflow, delayed stormflow and baseflow. Total evaporation takes place simultaneously from previously intercepted water as well as from the soil horizons in the form of soil evaporation and plant transpiration, depending on plant growth stage, root distributions and horizon water contents.

The nature of this model implies that the model is not a parameters fitting or optimizing model but parameters are replaced by variables estimated entirely from physical features of the catchment. ACRU has been designed as a multi­level model with multiple options available in many of its routines depending on the level of sophistication of available input data or the type of output required.

An important option in areas of complex land cover and soils is that ACRU can operate either at a point or as a lumped or as a semi-distributed cell type model.

K. C. Tarboton & R. E. Schulze 222

Fig. 2. General structure of the ACRU agrohydrological modelling system.

In distributed mode each subcatchment can generate individually requested and different output. To facilitate land cover or management changes over time, be they gradual changes such as forest growth or expanding urbanization, or abrupt changes such as clear-felling or reservoir construction, a dynamic time series dependent input option is available.

Model input

Model input includes information on catchment location/position, daily rainfall, evaporation, details on soils, vegetation/land cover, irrigation (including schedul­ing) and reservoir dimensions. To alleviate the problem of simulating either

223 The "ACRU" modelling system

distributed or complex catchments requiring extensive input information, inputs to ACRU are by way of an expert system type program-interface called the menu-builder, which leads the user interactively through alternative decision paths. Decision support is by means of interactive data and information input whereby the menubuilder prompts the user for information, gives instructive explanation via a help facility and supplies default values where required. For example, soil input decision support offers two levels of sophistication. At the lower level, only soil texture and depth need be estimated with relevant hydrological soil properties being read from default decision support tables while at the higher level, poros­ity, field capacity, wilting point and drainage response rates between soil layers as well as texture and depth would be requested as inputs for each layer. Similar decision support at two levels of sophistication is also available for other inputs.

A series of screening tests and error traps check input for validity and acceptable range. Depending on the severity of the input error either a warning or an error massage is issued requesting re-entry of the information.

Model output

Output options at daily, monthly or annual levels include simulation of the hydrological soil water budget, runoff, reservour status, sediment yield, crop yield and irrigation demand and supply. The soil water budget output shows rainfall, interception, effective rainfall, soil water contents in each layer and fluxes between layers. Runoff components for example, which can be output include daily storm and basefiows, design flows, peak discharges with an option for extreme value distribution analysis. Graphical output options are available for some of these components. Reservoir status output includes details on overflows, seepage, abstractions and inter-basin transfers. Crop yield models incorporated in ACRU at present include maize, wheat, sugarcane and primary productivity models. Irrigation output includes details on scheduling methods, irrigation requirements and losses.

When observed data are available a statistical summary can be requested to compare observed and simulated values for various outputs including streamflow, total evaporation and soil water status.

APPLICATION OF ACRU TO MIDMAR CATCHMENT

The ACRU modelling system was applied to initially the 912 km2 Midmar catchment, disaggregated into 19 subcatchments as shown in Fig. 3 to obtain a base simulation using present land cover information and to evaluate its perform­ance in terms of simulated and observed streamflow and the correlation between these values. The impacts of agricultural developments were then evaluated using two development scenarios.

To assess the impacts of farm dams on water resources all present reservoirs upstream of the Midmar dam were hypothetically removed for simulation under scenario 1. Areas under irrigation remained unchanged but irrigation was extracted from streamflow rather than from reservoirs.

K. C. Tarboton & R. E. Schulze 224

Fig. 3. Midmar suhcatchment discretization and system layout.

Impacts of increased afforestation were assessed in scenario 2 by assuming that the subcatchments 1, 6, 8 and 12 were completely afforested to Eucalyptus grandis except for the wetlands and areas under reservoirs. This was equivalent to an increase in afforestation of 11 881 ha over and above the existing 11 845 ha of forestry within the Midmar catchment, or a 100% increase (i.e. a doubling) in afforestation. Daily rainfall from 6 stations was used with a subcatchment rainfall adjustment factor based on its median annual rainfall compared with that of the rainfall station.

Present physiographic, land cover, reservoir and irrigation information used as inputs for the base simulation are shown for each subcatchment in Table 1.

Within the land use groups represented in Table 1, grassland was subdi­vided into different categories according to its grazing condition, with different hydrological properties of interception, leaf area index, and root distribution for each category. Similarly forestry was subdivided into indigenous forest and the major commercially grown species, Eucalyptus grandis, Pinus patula and Acacia mearnsii. Streamflow gauging weirs at the outlets of subcatchments 6 and 12, namely weirs U2H007 and U2H013 respectively, with observed daily streamflow data enabled comparison of observed with simulated streamflow data.

RESULTS AND DISCUSSION

Nomenclature

In considering the results the following nomenclature was used. Streamflow refers to the total water flowing out of a particular subcatchment, and consists of runoff generated from that subcatchment plus overflow, seepage and controlled discharge from any reservoirs which may be found within the subcatchment plus

225 The "ACRU" modelling system

Table 1. Pysiographic, rainfall and present land cover information for the Midmar catchment.

Subcatchment Median Mean Land cover Number Area annual altitude Grassland Wetland Forestry Dryland Irrigated Urban Dams

1 2 3 4 5 6 7 8 9

10 11 12 13 14 15 16 17 18 19

Total

(km2)

38 111

61 53 55 29 30 40 29 86 93 25 31 27 46 57 29 16 56

912

rainfall (mm)

1031 928 947

1026 966 967 891 887 981

1016 1011

957 1016 966 932 903 839 842 841

952

(m.a.s.l)

1151 1454 1378 1254 1409 1214 1848 1779 1712 1530 1430 1207 1287 1142 1150 1298 1279 1236 1089

1328

(%)

64 67 64 45 63 70 81 64 85 81 72 67 50 38 48 67 50 73 54

64

(%)

0 1 0 0 0 0

14 6 0 0 0 0 0 7 0 0 0 0 0

1

(%)

9 9 9

32 20 9 0 1 1

11 24 10 21 30

7 18 15

7 5

13

agric.

(%)

6 9

12 12 15 16

0 15

6 5 2

18 21 16 34 11

4 0

30

12

agric.

(%)

18 13 13 73 1 4 0

12 4 2 2 5 7 8

11 4

12 0 6

7

(%)

0 0 1 3 0 0 0 0 0 0 0 0 0 0

>.5 >.5

19 20

5

2

(%)

3 1 1 1 1 1 5 2 4 1

>.5 >.5

1 1

>.5 >.5 >.5

0 >.5

1

No

29 56 55 44 40 22 13

6 16 42 19

3 15 23 24 9

12 1

30

459

the streamiiow contribution of all upstream subcatchments. Runoff refers to the water produced from a particular subcatchment and consists of stormflow and baseflow. When there are reservoirs within a subcatchment a portion or all of the runoff (depending on the spatial location of reservoirs within the subcatchment), is routed through the reservoirs before it contributes towards the streamflow.

Base simulation evaluation

Selected statistics for daily and monthly streamflow simulation over the 16 years from 1971 to 1986 using present land cover information are shown in Table 2. At both gauging weirs U2H007 and U2H013 the total simulated streamflow was 11% less than the observed streamflow for the simulation period. Totals of monthly and daily observed and simulated streamflows for U2H013 are different because of the way the model handles missing data. If observed streamflow data are missing for a particular day, that day is omitted from the statistical analysis for purposes of assessing daily model performance while the entire month is omitted from the monthly analysis of performance. Good correlation between ob­served and simulated streamflows is shown by the high correlation coefficients (Table 2) for both daily and monthly simulations at both gauging points. Accor-

K. C. Tarboton & R. E. Schulze 226

Table 2. Satistics of performance of ACRU for daily and monthly totals of daily streamflow simulation.

Statistic

Total observed streamflow (mm) Total simulated streamflow (mm) Mean observed streamflow (mm) Mean simulated streamflow (mm) Correlation coefficient Students' 't' value Regression coefficient Base constant for regression (mm) Variance of observed values (mm) Variance of simulated values (mm) Coefficient of determination Coefficient of efficiency

Daily simulation U2H007

2461.08 2170.71

0.42 0.37 0.75

87.21 0.64 0.10 0.59 0.42 0.57 0.55

U2H013

3993.61 3561.47

0.68 0.61 0.76

87.89 0.83 0.05 1.17 1.38 0.57 0.96

Monthly simulation U2H007

2461.08 2170.71

13.83 12.20

0.88 24.84 0.64 3.29

439.07 234.25

0.78 0.76

U2H013

3945.97 3527.80

20.77 18.57

0.92 31.12 0.86 0.75

656.28 576.70

0.84 0.90

ding to Students' V test correlation is significant at the 99.5 percentile level in all cases.

The coefficients of determination and efficiency show a slight systematic error in the underestimation of the streamflow at gauging weir U2H007 while no systematic error is detected at U2H013. Fig. 4(a) shows the close correlation between observed and simulated annual streamflows over the 16 year simulation period.

Plots of daily observed and simulated streamflows at weir U2H013 for January 1986 and monthly values for the year 1985 are shown in Fig. 4 (b) and (c). The underestimation of observed streamflow, revealed in the statistics, can be seen in the daily plot which also shows the observed streamflow peak to lag the simulated streamflow peak indicating that simulated catchment response time is less than actual response time. Research to improve routing routines to rectify this effect is presently being undertaken. Use of current land cover data which reflects greater agricultural development and demands on water resources than in 1971 could be the cause for the underestimation in streamflow. By using the dynamic input file option which caters for changing land cover this problem could be rectified.

It should be noted that ACRU is a not a parameter optimizing model and that input variables are changed only when there are physical reasons for doing so. This implies that the model is transferable to other catchments in which differences in variables are due to catchment specific physical characteristics.

The potential of ACRU to assess water resources is shown in Fig. 4(d) by the frequency analysis, which shows percentiles of non-exceedance, of simulated streamflow into the Midmar dam. For example, the eightieth percentile represen­ted by the line labelled 80 % shows the monthly streamflow into the Midmar dam that one could expect not to be exceeded 80% of the time or conversely to be exceeded on average once in every 5 years, for each month of the year. This

227 The "A

CR

U" m

odelling system

x> n> <o 13

£ m

]

-o O

J

> i) CO

-O

O

/J / /

y / ^/

^^

X

^̂"̂

\

// 1 /' " //' -//

-/'

-1

---

1 i:

£ T

D

— Simulated served -

J3

O 1 " =3S^~T

<<

< // V \\ \\

li i /

I /

1

\\ yf

:>;:==•

/ / "•- \ V

..

-: •

_

•

S O

S

o

K, C. Tarboton & R. E. Schuhe 228

could aid a water resource manager in establishing the probability of receiving certain quantities of water into the reservoir at certain times of the year in order to establish reservoir management strategies.

Scenario 1: All reservoirs removed from simulation

In the Midmar catchment there are presently 459 small reservoirs with a total surface area of 874 ha. For the purpose of this study the Midmar dam was not considered part of subcatchment 19 but was the body receiving all streamflow from subcatchments 1 to 19. The effect of removing all the reservoirs from the simulation is shown in the subcatchment runoff and streamflow values under scenario 1 in Table 3.

Table 3. Median annual runoff and streamflow for base simulation and under agricultural development scenarios.

Subcatch- Base simulation Scenario 1: Removal of all dams Scenario 2: Doubling of forestry ment Runoff Streamflow Runoff Streamflow Runoff Streamflow Number (mm) (mm) (mm) (% change) (mm) (% change) (mm) (% change) (mm) (% change)

144 -37

119 134 + 7 140 +18 40 -68 105 -12

37 -72

1 2 3 4 5 6 7 8 9

10 11 12 13 14 15 16 17 18 19

229 98

105 107 114 125 225 130 277 194 149 125

97 117

48 227 180 112 103

212 130 126 128 132 134 210 154 273 201 155 130 111 140

62 234 185 112 113

-3 + 33 + 20 + 20 + 16 + 7 - 7

+ 24 - 1

+ 4 + 4 + 4 + 14 + 20 + 30 + 3 + 3

0 + 10

187 130 + 4 187 0 75 -50 164 -12

158 113 +10 168 +6 142 -10

For all subcatchments except 1, 7, 9 and 18 (which had no reservoirs) there was an increase in subcatchment runoff of between 3 and 33% under scenario 1 (i.e. removal of all dams). Subcatchments 1, 7, and 9 are the three subcatchments with the greatest areas under reservoirs, their percentage covera­ges by reservoirs being 3, 5 and 4% respectively (Table 1). Subcatchment 7 has no irrigated agriculture while subcatchment 9 has 4% under irrigated agriculture. Greater runoff from subcatchments 7 and 9 with existing reservoirs could be due to the large water surfaces being more effective in producing runoff than vegeta-

229 The "ACRU" modelling system

tive surfaces. Evaporation losses from these reservoirs would also be low due to the relatively high altitudes (1848 and 1712 m) of the subcatchments. Demands for water from the reservoirs are small, due to zero to little irrigated agriculture. Although subcatchment 1 has 18% of its area under irrigated agriculture, this is the subcatchment with the highest rainfall (1031 mm), and its reservoirs would thus be assumed to be predominantly full and therefore more effective runoff producers than was the same subcatchment when simulated without reservoirs under scenario 1. Streamflow from subcatchments 1 to 6 was increased by 18% from 119 to 140 mm when reservoirs were removed. Streamflow from sub­catchments 7 to 12 remained unchanged due to the decrease in runoff from subcatchments 7 and 9 balancing the increases in runoff from the other cells. The nett effect was an increase of median annual streamflow into the Midmar dam from 158 to 168 mm. This implies that the present reservoirs upstream of the Midmar dam have the effect of reducing median annual streamflow into the dam by 6 % or the equivalent of 6 10* m3.

The seasonal variation in runoff from subcatchment 8 under scenario 1 is compared with that of the base simulation for a dry (10%) and the median (50%) year in Fig. 5. The variability of runoff when reservoirs are removed is greater in dry years, with more runoff in the wet months and less in the dry months than under the base scenario with reservoirs. For both the median year and in wet

(a) Runoff (mm) f 10% Base simulation 10% Scenario 1

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Months

(b) Runoff (mm)

n—50% Base simulation — 5 0 % Scenario 1

Jan Feb Mar Apr May Jun Jul Aug Sep Ocl Nov Dec

Months

Fig. 5. Seasonal variation in runoff from subcatchment 8 for base simula­tion and under scenario 1 with farm dams removed in (a) dry years and (b) year of median runoff.

K. C. Tarboton & R. E. Schube 230

years (not shown) runoff from the base scenario is less than under scenario 1 (i.e. removal of all dams) in the summer rainfall months from October through to March because the reservoirs intercept a portion of the runoff. During the winter months with very little rainfall, runoff comprising predominantly seepage and controlled discharges, is more from the base scenario with reservoirs.

Scenario 2: Doubling of forestry in the Midmar catchment

Impacts of afforestation on runoff from subcatchments 1, 6, 8 and 12 are shown in Table 3 under scenario 2. Marked local reductions in runoff were simulated for the subcatchments in which the afforestation occurred, with the magnitude of runoff reduction being related to the MAP of the subcatchment. An increase in percentage afforestation from 9 to 97% (i.e. entire catchment minus 3% dams, Table 1) in the relatively wet subcatchment 1 (MAP 1031 mm) resulted in a 37% decrease in runoff while a similar increase in afforestation from 1 to 93 % in the relatively dry (MAP 887 mm) subcatchment 8 resulted in a 72% decrease in runoff. Upstream of subcatchments 6 and 12 the increased areas under afforesta­tion were respectively 124% from 4845 to 10 833 ha and 169% from 3497 to 9391 ha and the streamflow reduction from these subcatchments was 12% in both cases. By doubling the area under afforestation in the Midmar catchment from 11

Runoff [mm] 100

80

60

10

20

0 'an Feb Mar Apr May Jun Jul Aug Sep Ocl Nov Dec

80

60

40

20

0 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Months

Fig. 6. Seasonal variation in runoff from subcatchment 8 for base simula­tion and afforestation scenario 2 in (a) year of median runoff and (b) wet year.

-5096 Base simulation — 50% Scenario 2

(a)

90% Base simulation - - - 90% Scenrio 2

231 The "ACRU" modelling system

845 to 23 726 ha the median annual streamflow as simulated at the outlet of subcatchment 19 into the Midmar Dam, was reduced by 10% or 9.2 106 m3.

Seasonal variation in runoff from subcatchment 8 under the afforestation scenario 2 is compared with that of the base simulation for the median (50%) and wet (90%) years in Fig. 6. Subcatchment runoff is less under afforested condi­tions throughout the year for median, wet and dry (not shown) years. Runoff in the dry months of the year is close to zero even under wet conditions at the 90th percentile of non-exceedance (Fig. 6(b)) and the duration of the period with close to zero runoff is extended for the year of median runoff when compared to the dry period duration in the 90% wet year, as illustrated in Fig. 6(a).

CONCLUSIONS

The ACRU modelling system has been shown to mimic observed streamflow for the 912 km2 Midmar catchment satisfactorily with good correlation between simulated and observed streamflows at annual, monthly and daily levels. It was suggested that the dynamic input file option available in ACRU could be used as a means to rectify the small underestimation of streamflow, assumed to be due to the use of current land cover data. Successful assessment of the current water resources of the Midmar catchment using the ACRU modelling system indicates its potential for application to the entire Mgeni catchment and other large catch­ments.

Agricultural development assessed in terms of local impacts on runoff of present reservoirs within each of the 19 subcatchments were assessed, illustrating the variation in impact of reservoirs depending on subcatchment altitude, rainfall, land cover and reservoir demands. Simulated runoff from catchments without reservoirs was shown to be more variable than that from the same catchments with reservoirs. Present reservoirs upstream of the Midmar dam were shown to have the overall impact of reducing median annual streamflow into the dam by 6 % or 6 106 m3. ACRU was able to quantify the marked local reduction in runoff from affected subcatchments under a potential afforestation scenario. The degree of runoff reduction appeared to be related to the relative MAP of the areas affor­ested. Runoff from afforested subcatchments was very low even during wet years and the duration of the period of low runoff was extended into the summer rainfall months. The regional impact of a doubling of forestry in the Midmar catchment was a 10% or 9.2 106 m3 reduction in streamflow into the Midmar dam.

The problem of rapid agricultural development in the Mgeni catchment was addressed by assessing quantitatively and objectively the impacts of potential development scenarios on a critical portion of the catchment. This hydrological simulation process provides water resources managers with knowledge needed to make informed proactive planning decisions. The ACRU modelling system was shown to be an effective modelling tool because of its ability to simulate present water resources and impacts on present water resources, and its potential for transferability to other large catchments.

K. C. Tarboton & R. E. Schuhe 232

ACKNOWLEDGEMENTS

We should like to thank the Water Research Commission for funding the project entitled "Development of a Systems Hydrological Model to Assist with Water Quantity and Quality Management in the Mgeni Catchment", from which this paper emanated. Computing facilities provided by the Computing Centre for Water Research are acknowledged gratefully.

REFERENCES

Breen, C. M., Akhurst, E. G. S. & Walmsley, R. D. (1985) Water quality management in the Mgeni catchment. Natal Town and Regional Planning Supplementary Report, 12.

Home Glasson Partners (1989) Water Plan 2025. Umgeni Water Board, Pietermaritzburg. Linacre, E. T. (1984) Unpublished manuscript. School of Earth Sciences, Macquarie University,

Sydney, Australia. Maaren, H. & Moolman, J. (1985) The effects of farm dams on hydrology. Proc. S. African Nat.

Hydrol. Symp. Dept. Agric. Engng., Univ. Natal, Pietermaritzburg, ACRU Report, 22, 428-441.

Penman, H. L. (1948) Natural evaporation from open water, bare soil and grass. Proc. Roy. Soc. A193, 120-146.

Schulze, R. E. (1989) ACRU: Background, concepts and theory. Dept. Agric. Engng. Univ. Natal, Pietermaritzburg. ACRU Report, 35.