Assessing the effects of non‐stationarity on reservoir yield estimations: A case study of the Southern Okavango Integrated Water Development system in Botswana

Streamflow data used for water resources planning should ideally be stationary, and any non‐stationary behaviour is taken into account. However, with limited time series data, the influence of non‐stationarity is often hidden and can result in unreliable estimates. This paper examines the impact of non‐stationarity on the Southern Okavango Integrated Water Development (SOIWD) project that was carried out with streamflow collected between 1969 and 1989 against an extended time series from 1990 to 2019. To achieve this, (a) the statistics of these inflows and (b) the reliability of satisfying water demands from the proposed reservoirs were tested based on the two periods (1969–1989 and 1990–2019). The results show that average monthly flows for July, August and September significantly change when peak outflows from the delta occur. Given the expected variability of the flow regime, an uncertainty approach utilising flow perturbation from ±5% to ±90% was also used to investigate the system's response to changes in the driving flow conditions. The increasing availability of flow data from 1990 to 2019 has shown that the SOIWD system would not have satisfied the water demand as it would not be operationally viable. This confirms the importance of accounting for non‐stationarity in reservoir yield estimation and reemphasises its importance in hydrological studies.

Reliable predictions are possible only if the time series is stationary, is of adequate length to present better underlying temporal variability, and has a consistent time interval, low noise and historical consistency such that future conditions could be like the past (Cheng et al., 2015).However, data in hydrological studies are very noisy, often of limited length or non-existent, especially for data-scarce regions such as southern Africa and conform to non-stationarity.
Hydrometeorological datasets exhibit non-stationarity due to anthropogenic activities, natural climate variability, extreme events and climate change.
Climate change significantly impacts non-stationarity, and its effects are more evident in the 21st century (IPCC, 2021).As the patterns and characteristics of hydrometeorological parameters shift in their long-term means, extreme events and variability, climate change intensifies the non-stationarity exhibited on these parameters (Milly et al., 2008).Understanding the influence of climate change has been a major thrust in the past two decades (Kubiak-W ojcicka et al., 2023).
Such an understanding will improve policy, water resource management and infrastructural development.Integrating climate change projections, hydrological modelling and stakeholder-induced adaptive strategies.Dam construction in southern Africa has a major development along many rivers to harness water for industrial and household irrigation.Crucial to dam construction is the prior analysis of the yield from the reservoirs to be created.Non-stationarity plays an important role, and with limited data can adversely affect the accuracy of estimated reservoir yields.This paper emphasises the importance of longterm time series data exhibiting non-stationarity in water infrastructure design and decision-making by assessing the Southern Okavango Integrated Water Development (SOIWD).

| Background
Sustainable development is tightly coupled with the availability and access to water resources of desired quantity and quality.Countries in the developing phase are heavily dependent on natural resources to achieve national development.This dependency on natural resources is the case for southern Africa, a region characterised by high climatic variability.These inherent problems hinder the region from coping with sustainable development.Climate change reports (Gumbo et al., 2021) have shown southern Africa as one of the regions to be adversely affected by the impact of a changing climate.Mining and agricultural activities dominate the region and become entry points for sustainable development.These water-demanding activities present availability and access issues, especially in semi-arid areas.This is the case for Botswana.Harnessing water resources through the construction of reservoirs is necessary for the country and requires detailed assessments to ensure that the expected yields are met.
Long-term flow time series representing temporal variations of river flows are necessary for planning and designing reservoirs for water supply purposes (Langat et al., 2017).The lack of such data has been recognised as a significant problem (Mazvimavi, 2003;Oyebande, 2001), and in 2003, the International Association of Hydrological Sciences (IAHS) launched a decadal initiative aimed at improving Predictions in Ungauged Basins (PUB) (Sivapalan, 2003;Sivapalan et al., 2003).Water-related developments such as dam construction and irrigation development, must occur in data-scarce, ungauged situations.Hydrologists are often called upon to generate realistic water resource information to support such developments.
The problem of the ungauged basin is always encountered in practice.
In the endeavour to make precise predictions of river flow in an ungauged basin (and a vast number of basins worldwide are virtually ungauged!),there is muted consensus in the hydrology community that such a feat may never be accomplished.However, the treatment of uncertainty has emerged as a significant constituent of understanding the problem of making predictions in ungauged basins.Identifying potential sources and incorporating uncertainty is critical to the problem of understanding what it means to make predictions in an ungauged basin.Understanding the uncertainties surrounding a hydrological assessment is not a detriment but rather a protocol that gives confidence in the generated results for decision-making.Thus, hydrological modelling has focused on investigating the possible sources of uncertainty, quantifying these uncertainties and assessing their impact on hydrologic predictions.
Several studies were done in southern Africa on the Okavango and Zambezi Rivers (Mazvimavi & Wolski, 2006;McCarthy et al., 2000), Lake Malawi (Shela, 2000) and in other parts of the world (Lindström & Bergström, 2004;Xiong & Guo, 2004;Garcia & Mechoso, 2005) and show that river flows exhibit long-term variations.Southern Africa was generally wet from the mid-1950s to l980 and dry from 1981 to 2000.
As such, any conclusions made about water resources assessments, including results of reservoir yield analyses based on data from the mid-1950s to 1980, will likely differ from those based on 1981 to present data or consideration of the uncertainty related to the flow data.The SOIWD study in Botswana was conducted from 1986 to 1991 and mainly relied on river flow data collected between 1969 and 1989 on rivers draining the Okavango Delta (Scudder et al., 1993;SMEC, 1987SMEC, , 1991aSMEC, , 1991b)).The SOIWD study investigated several options for infrastructural developments for utilising outflows from the Okavango Delta for household, irrigation and mining water supply.This paper seeks to prove non-stationarity as a significant component in reservoir yield estimation using a case of the SOIWD.Two streamflow time series datasets are used (1969-1989) and (1990-2019) to (a) test the statistics of these flows and (b) the reliability of supplying the design yields from two proposed reservoirs during the SOIWD study.This paper will demonstrate the value of long-term hydrological monitoring and increase awareness about establishing and maintaining hydrometric systems for long-term planning.

| SOIWD plan for water resources development
The Botswana Government commissioned SMEC (1987, 1991a, 1991b) 1).This reservoir was planned to store flood flows from July to September, after which the water level would be reduced to enable flood recession cultivation.c. Construction of two dams, each downstream from where the Thamalakane River bifurcates into the Boteti and Nhabe Rivers.One dam was proposed across the Boteti River at Samedupi (Samedupi Dam) and another across the Nhabe River at Toteng (Figure 1).This would create the Maun reservoir with an estimated capacity of 138 Mm 3 , 12 Mm 3 of dead storage and a surface area of 58.6 km 2 .The primary purpose of this reservoir was for municipal water supply to Maun Village, which had a projected water demand of 4 Mm 3 /year by 2010, to make water available for irrigating 1300 ha and to release water to Lake Ngami when needed.The Botswana Government awarded a contract for developing infrastructure proposed by SOIWD in 1990.This contract was terminated in 1991 due to local and international opposition to this project (Scudder et al., 1993).Residents alleged that previous bunding and dredging had destroyed some channels within the delta.They no longer benefit from harvesting aquatic plants for food, basket weaving and construction, fishing and flood recession cultivation from the affected channels.The SOIWD study used hydrometeorological data collected for 21 years .The reliability levels of satisfying the projected demands were estimated based on this data.Using an extended streamflow time series

| Study area description
The Okavango River Basin (ORB), with an area of 156,250 km 2 originates from Angola flowing into Botswana, where the flat topology of the area gave rise to the formation of the Okavango Delta.This river basin has its headwaters in a high-rainfall region within Angola.These headwaters receive a mean annual rainfall of 1300 mm/year, higher than 550-450 mm/year over the delta (OKACOM, 2011).The delta has an area of about 35,000 km 2 , with 6000-12,000 km 2 being inundated yearly, with the distal end of the delta having distributaries draining into the Thamalakane River (Figure 2).The river passes through Maun, which had a population of 25,000 and 65,693 persons in 1991and 2011, respectively (Statistics Botswana, 2015).Mendelsohn and el Obeid ( 2004) and Mendelsohn et al. (2010) adequately describe the Okavango River Basin's features and delta.The northwestern part of Botswana, within which the delta is located, has limited water sources.Therefore, proposals have been made to utilise water within the delta to satisfy water demand for various activities.
Several recommendations were made to improve water drainage into the Thamalakane and Boteti Rivers for irrigation (Wellington, 1949) and construct a reservoir of 105 Mm 3 along the Thamalakane River (Brind, 1954).In time, all these proposals influenced the need to assess the potential of harnessing water resources from the delta for developmental purposes.

| Derivation of annual and monthly flow statistics
A comparison of annual and monthly flow statistics derived from the 1969-1989 and 1990-2019 periods was undertaken to determine if there were significant changes in river flows.Water that contributes to the SOIWD reservoirs drains from the Okavango Delta.Change detection was conducted in this study at Mohembo (inflows into the delta) and Thamalakane (outflows from the delta into the proposed reservoirs) on the time series used for the SOIWD assessment  against a more extended period .The Mann-Whitney and t-tests were used to assess the changes in the average annual flows.The Kolmogorov-Smirnov test was used to determine differences (1969-1989 vs. 1990-2019)  F I G U R E 2 Rivers draining from the Okavango Delta with the Thamalakane River bifurcating into the Nhabe River that flows into Lake Ngami and Boteti River, eventually joining the Makgadikgadi Pans.

| Reservoir simulation
The Matlapaneng Reservoir stores water from July to September, causing inundation of the upstream floodplain to facilitate flood recession cultivation.In October, the reservoir is drained to enable crop cultivation on the upstream floodplains.The operation of this reservoir is described by Equation (1). where.
S 1 (i, j) = volume of water stored in the Matllapaneng Reservoir during year i and month j.
I 1 (i, j) = inflow into Matlapaneng Reservoir estimated from river flow measurements made on Thamalakane River at Maun Bridge.
A 1 (i, j) = surface area of water in Matlapaneng Reservoir estimated from the volume of water stored in the reservoir.
E( j) = mean monthly evaporation rates estimated from pan evaporation measurements at Maun.P(i, j) = monthly rainfall measured at Maun.R 1 (i, j) = release of water to downstream reservoirs.
Since the Matlapaneng Reservoir only stores water from July to September, The priority for the Maun reservoir is for municipal water supply (4 Mm 3 /year), followed by irrigation (7.7 Mm 3 /year).Suppose the Maun Reservoir has sufficient water to satisfy the municipal water demand during the next 12 months.In that case, a release of available water up to 7.5 Mm 3 in October to Lake Ngami.This is described in Equation (4).O 2 (j) = release of water to Lake Ngami, only when j = 10.
Since the water supply to Maun has the highest priority, the release of water to Lake Ngami is only made if there is adequate water to satisfy this demand during the next 12 months.
S 3 (i, j) = volume of water stored within the Boteti River.
α, β and S min are parameters to be estimated using observed monthly flows of the Boteti River at Samedupi and Rakops.Rakops is located downstream of the Sukwane Reservoir.The flooded area, A 3 (i, j), is a linear function of the storage volume along the Boteti River.
The water demand planned to be satisfied from Sukwane was 12 Mm 3 /year for water supply to the Orapa Mine.

| Rainfall, evaporation and infiltration characteristics
The average annual rainfall at Maun, close to the planned reservoirs, is 540 mm/year, while the Class A-pan evaporation is 2010 mm/ year.There is a significant deficit in water availability as simple calculations produce an Aridity Index (AI) of 0.26 in semi-arid condi- Infiltration is another factor that influences the output from the delta to contribute to the proposed reservoirs.The delta has varying infiltration rates mainly because of vegetation cover, soil types and the presence of water (Wolski et al., 2006).During the dry periods, the soil dries up, creating a higher affinity so water and increasing their infiltration.During the wet periods, the contribution from the river will induce inundation within the delta, potentially decreasing the infiltration rates.
As such, during wet seasons, groundwater infiltration may not affect streamflow generation into the downstream reservoirs much, but groundwater losses can significantly affect the streamflow during the dry seasons.Several authors (McCarthy, 2006;Ramberg & Wolski, 2008;Wolski et al., 2006) have written on the groundwater influence on the amount of streamflow generated from the delta.This research, however, is interested in the streamflow that results from the delta, and that which is lost through infiltration in the delta is acknowledged but less important.The losses that affect surface water resources (evapotranspiration) become important to this study.

| Streamflow characteristics
The 2.8% and 2.0% contributed at Thamalakane for the 1969-1989 and 1990-2019 periods, respectively, as shown in Table 1.A considerable proportion of the flow is captured within the delta because its characteristics slow down the river's velocity and allow processes that promote water storage to persist.Several rivers (distributaries) branch out of the main river, and most do not return to the main river as it leaves the delta (El-Manadely et al., 2002).This accounts for the high streamflow levels into and low streamflows out of the delta.

| Statistical relationships with extended timeseries data
At Mohembo, the mean annual runoff for the 1969-1989 and the 1990-2019 period showed a difference of À6.7%.There was a significant decrease of 33.5% at Thamalakane for the same period.The changes in mean annual runoff at Thamalakane are more significant than those at Mohembo because of the presence of the delta.As the water flows into the delta, several processes affect the resulting discharge from the delta (El-Manadely et al., 2002).The 1969The -1989 period was generally wet, as shown by more runoff at both gauging stations.The period that followed, 1990-2019, was generally dry, and the streamflow was less than in the previous period.Using these periods independently will give water-sufficient and water-deprived results .However, this does not provide a clear picture of the available water resources.The whole time series  will consider the long-term variability in streamflow and deliver results that account for both the wet and the dry periods.2. Reliabilities estimated using the observed 1969 to 1989 monthly flows show that all the reservoirs would have supplied water with the level of reliability set during the SOIWD system's design (Table 2).However, results of reservoir yield analysis conducted using flow data up to 2019 show that all the reservoirs would have failed to supply all the water demands with the reliability levels that were the basis of  the design of the system, viz; 99% for Maun water supply, 80% for irrigation and 90% for Orapa Mine.

This conclusion agrees with
The analysis of the 1969-1989 period assumed that the periods with which water supply could not be achieved were rare and could only last a short period within a year.Water resources management in such instances is easier as short periods with a limited water supply can allow for immediate response.It is easier to have contingencies for addressing reduced water supply for a short period, and its impacts on the ecosystem are usually short-lived.This is different when supply cannot be satisfied for an extended period, in our case, decadal.The 1990-2019 period showed that the supply would not be satisfied for a significant period.This would have had significant impacts on the management of the dam.More environmental issues are associated with extended periods where supply cannot be satisfied, affecting the catchment's ecological balance.
A physical option for increasing the reliability of satisfying water demands is to increase the capacities of reservoirs.An increase in the Maun reservoir's capacity is impossible as this will result in flooding of built-up areas.In addition, evaporation losses dominate the water balance of proposed reservoirs.The Maun Reservoir would have an average depth of 2.4 m with average monthly evaporation losses amounting to 3.5 Mm 3 /month, approximately equal to the design annual water demand of 4 Mm 3 /year.The then-design water demand for Maun (4 Mm 3 /year) is minor compared to the total supply capacity of the proposed reservoir (138 Mm 3 ), and the estimated reliability depends mainly on the availability of inflows into the reservoir.

| Uncertainty considerations in reservoir yield estimations
Several uncertainty levels were used, ±5% to ±90%, to determine the failure rates of the reservoirs, as shown in Figure 6.The results show that 1989-2019 had a higher failure rate, 1969-1989 had the least and 1969-2019 had failure rates between the other periods.The flow data showed that 1969-1989 had more streamflow than 1989-2019.
This would explain the failure rates, as shown in Figure 6.Using the period 1969-1989 and 1989-2019  F I G U R E 6 Representation of failure rate at different uncertainty levels (1969-1989, 1969-2019 and 1989-2019) for the different water demands.
[Color figure can be viewed at wileyonlinelibrary.com] for irrigation slightly higher.This is because the water demands for the town and those for irrigation purposes have differences in the assurance of demand.Domestic water use requires a higher assurance of supply than irrigation water demands.
Reliable water supply for various water demands depends on accurate reservoir yield estimations.These estimations are heavily dependent on data of sufficient quality and quantity.However, many uncertainties cloud the reservoir yield estimation process and affect the results' reliability.This study evaluated reservoir yield estimations of a shorter period against an extended version of the time series to understand how long-term variability affects reservoir yield estimations.The research used a case study of the SOIWD, which, without the comprehensive time-series data, proposed reservoirs that would satisfy the demands for Maun, irrigation and mining activities.Based on the 1969-1989 streamflow data, the SOIWD study concluded that the proposed system of reservoirs would achieve monthly reliabilities of 99% for Maun, 80% for irrigation and 90% for Orapa Mine.This was, however, different when using the extended time-series data that started from 1990 to 2019.The reliability levels reduced to 77.8%, 77.0% and 61.4%, respectively.If the SOIWD study had been conducted using river flow data up to 2019, the conclusion would have been that the proposed reservoirs would not satisfy the water demand.This example proves that non-stationarity arising from longterm variability is a crucial component when conducting such assessments.However, data, as in the case of the SOIWD project, would not be available or of limited length.Would that imply that such projects must wait until enough data has been gathered?As this research points out the importance of non-stationarity in hydrological studies, it becomes essential for similar projects to account for the uncertainties introduced by non-stationarity.Without data collected for a more extended period (50 years or more), hydrologists can implement modelling strategies that simulate likely streamflow conditions that account for non-stationarity.When there is limited modelling ability, the project can also ask for expert judgement and stakeholder participation input.This will provide valuable insights into non-stationarity, especially for data-scarce regions like southern Africa.
to investigate the plausibility of constructing reservoirs fed by the Thamalakane and Boteti Rivers for domestic water provision to Maun Village, agricultural purposes (irrigation and livestock) and mining activities.Dubbed the SOIWD, their assessment made the following recommendations: a. Construction of bunds connecting islands to contain flood flows along the lower Boro River and deepening its channel for 28.6 km to increase the conveyance capacity to 20 m 3 /s and outflow by 45 million cubic meters (Mm 3 )/year.b.Development of the Matlapaneng Dam on the Thamalakane River upstream of Maun to create a reservoir with a total supply capacity (FSC) of 24 Mm 3 for improving flood recession cultivation (Figure d. Development of Sukwane Dam across the Boteti River (Figure 1) some 140 km downstream of the Samedupi Dam with an FSC of 66 Mm 3 .The purpose of this dam was to meet the water requirements for Orapa Mine, which had a projected water demand of 12 Mm 3 /year by 2010 and to assist downstream flood recession agriculture.SMEC (1991) concluded, using the available 1969-1989 hydrometeorological data, that the proposed dams were able to satisfy the following water demands: a.The Maun water demand of 4 Mm 3 /year could be met with a reliability of 99% monthly.b.Water for irrigating 1300 ha could be provided from Maun Reservoir with 80% crop reliability.c.The water demand for Orapa Mine (12 Mm 3 /year) could be provided from Sukwane Dam with a 90% reliability compared to the 52% reliability achievable without developments recommended in the SOIWD study.d.The release of water from Sukwane Dam for downstream uses could be undertaken in 75% of the years compared to natural flow is available in 68% of the years.
, uncertainty arises regarding the feasibility of recommendations based on the 21-year time series.The viability of the recommendations drawn from the SOIWD to meet water F I G U R E 1 Location of dam sites and reservoirs proposed for the Southern Okavango Integrated Water Development.demands becomes uncertain.Understanding the uncertainty brought about by an increased record length prompted the research to validate the SOIWD assessment's conclusions in meeting the area's water demands.
between empirical distributions on the flow duration curve of the monthly flows of the Thamalakane River at Maun.The reliability levels of satisfying the demands set in the SOIWD study were determined by simulating reservoir operation using the behavioural analysis method (McMahon et al., 2007a).SMEC (1991a) estimated reliability levels based on the number of months a reservoir satisfies the demand.The same approach is made in this study.
138 Mm 3 capacity of Maun Reservoir.D u (j) = monthly urban water demand for Maun Village.D a (j) = monthly irrigation water demand.
Reservoir are water that spills from Maun Reservoir, R 2 (i, j) into the Boteti River.The Maun and the Sukwane River region has flat terrain with no major rivers joining the Boteti River.The change in water storage along the Boteti River is described by Equation (5).A linear reservoir model (Equations 7 to 9) is used to describe the flow of water along the Boteti River from Maun Reservoir to Sukwane Reservoir: tions.These characteristics validate that the inflow into the reservoirs is comprised mainly of the headwater contributions of the ORB.Higher evaporation reduces reservoir storage(Althoff et al., 2020).Research byHelfer et al. (2012) explains how reservoirs in Australia risk losing about 40% of their storage to evaporation annually.Australia and Botswana are water-scarce countries that depend on stored water to sustain their various demands.Long-term predictions of rainfall(Dey et al., 2019) and evaporation(Konapala et al., 2020) show decreasing rainfall and increased evaporation induced by climate change in the southern African region.These changes directly influence the amount of water collected in the proposed dams.Understanding the influence of land use land cover and climate change on water development projects has become necessary in water resources management.
annual flow patterns for the Okavango River at Thamalakane River at Maun are shown in Figure 3. Comparing the total streamflow collected for both periods, the 1969-1989 period had 6879 MCM, while the 1990-2019 period recorded 3605 MCM.This was a significant reduction (47.6%) from the period used to estimate the yield potential of the reservoirs.Uncertainty in the ability of the river to satisfy the reservoir needs arises.Policy and decision-making rely on the outputs from these assessments.Accounting for uncertainties associated with these assessments gives reliability to the results.According to McMahon et al. (2007b), a more extended streamflow time series (50 years and longer) is ideal for reservoir yield estimation studies.A longer time series can account for the long-term variability in streamflow and give estimates closer to reality.The SOIWD assessment made recommendations based on a 21-year-long streamflow time series and this period estimated that the reservoir yields could satisfy the flow recorded in those 21 years.However, looking at a 51-year time series, the yields for the reservoir could not be satisfied.This result aligns with the recommendations made by McMahon et al. (2007).Of the total inflow recorded at Mohembo for both periods, F I G U R E 3 Temporal variation of monthly flows of the Thamalakane River at Maun and a comparison of the 1969-1989 (dotted line) and 1969-2019 (solid line) flow duration curves.
McMahon et al. (2007a), who expressed the need for more extended time series for reservoir yield estimates.The results also show that a change in streamflow into the delta adversely affects water availability at Thamalakane.The Mann-Whitney and t-tests do not show any significant change in the average annual flow for the Okavango River at Mohembo.The Mann-Whitney test suggests a change in the average annual flows of the Thamalakane River at Maun.However, the t-test shows no significant change.Annual flows of the Thamalakane River at not normally distributed as required for the t-test.The t-test may not have distinguished the difference when the change is relatively marginal.The Thamalakane River had an extended period, 1991-2005, with extremely low flows, unlike the earlier period, 1969-1989 (Figure 3).The Kolmogorov-Smirnov test revealed that the flow duration curve of the monthly flows for the 1969-1989 period was significantly (5%) different from that for the 1969-2018 period.The estimated maximum absolute difference between the two curves was 0.25, while the critical value for accepting the null hypothesis of no difference was 0.10.The observed October 1971 to September 1991 monthly Boteti River flows at Samedupi and Rakops were used to calibrate the reservoir model (Equation (6)).The calibrated parameters for the Boteti linear reservoir model (Figure 4) are S min = 25.0Mm 3 , α = 0.25, and β = 1.182 with a coefficient of determination (R 2 ) = 0.95, and the root mean square error of 5.3 Mm 3 / month.The average monthly flows at Rakops during the calibration period were 13.3 Mm 3 /month.3.4 | Reliability of reservoir yield estimations of the SOIWD The simulated changes in volumes of water stored in the Maun and Sukwane Reservoirs from October 1969 to September 2019 are presented in Figure 5.These two reservoirs would have been empty most of the time from August 1993 to October 2004, therefore unable to satisfy the water demand.The reliability of satisfying water demands from reservoirs proposed by the SOIWD study estimated by a behavioural analysis on a monthly interval is presented in Table U R E 5 Simulation of water storage in the (a) Maun and (b) Sukwane Reservoirs.
Comparison of the annual statistics of the Okavango River at Mohembo and Thamalakane River at Maun.Calibration of the Boteti River linear reservoir model to predict river flow movement between Samedupi and Rakops close to the Sukwane Reservoir.Pred indicates predicated and obs indicates observed.[Color figure can be viewed at wileyonlinelibrary.com] to estimate reservoir yields would give lower and higher failure rates.But considering the whole time Reliability in terms of percentages of months when water demands are satisfied from the SOIWD reservoirs.In brackets are the design water demands for the system.
serieswould provide failure rates informed by adequate and inadequate streamflow periods.Using this time series would address issues dealing with long-term streamflow variability.The highest failure rates were recorded for the Orapa Mine, with that for Maun and irrigation exhibiting almost similar failure rates, with that T A B L E 2