Analysis of streamflow and rainfall trends and variability over the Lake Kariba catchment, Upper Zambezi Basin

Water resources management and planning are wholly anchored on the past observation of hydro-meteorology data records that are projected in the future. The time series of the data collected determines the hydrological behaviour of the catchment or river system (Forootan 2019). To sustainably manage water resources requires an understanding of the hydrological regime of a river system that eventually affects the streamflow. Indeed, streamflow is variable and identification in the variability of streamflow data is vital in water resources planning as it moderates the consequences related to overestimation or underestimation of design parameters and water rights allocations way above the available water in the river system (Arrieta-Castro et al. 2020). Detecting any change in streamflow is one of the important parts of establishing hydrological variability. This stream flow variability with enough records establishes a trend of a river system at a point of monitoring (Burn & Hag Elnur 2002). Hence, trend analysis is dependent on the duration of the streamflow data being analysed (Dixon et al. 2006). Trend analysis has been used to determine any changes in streamflow or any climatic parameters. The assumption that records may repeat in the future is not valid anymore due to the trend (IPCC 2007). Various statistical methods are available now to determine the trend. These methods are distinguished based on the approach being used: parametric and non-parametric. A decline in streamflow has generally been established in southern Africa, with concerns about its availability to various competing needs of hydropower production, irrigation and water supply among others. The decline in streamflow has been attributed to insufficient rainfall, an increase in population, land cover changes and hydrological regulation (Banze et al. 2018; Winton et al. 2021; Ndhlovu & Woyessa 2022).

Several studies have been done on trend analysis in streamflow and rainfall. Elouissi et al. (2017) and Benzater et al. (2019) analysed the impact of climate change on the rainfall pattern of not less than 41 stations within the Macta watershed in Algeria, with a downward trend being established together with the corresponding deficit within the time series. Şen (2017) employed statistical procedures to determine trend constituents in any hydro-meteorological time series. Ndhlovu & Woyessa (2021) assessed the impact of various climate change prediction scenarios on stream flow data. It was determined that the climatic component of temperature showed an upward trend with considerable ambiguities in precipitation. Other studies have found no pattern of change in streamflow as rainfall varies spatially and temporally (Hannaford 2015). Gebremicael et al. (2013) analysed the temporal and spatial changes in rainfall and streamflow in the Upper Tekeze Atbara River Basin in Ethiopia using the Mann–Kendall and Pettitt test. The result indicated that rainfall did not exhibit statistically significant changes; however, streamflow patterns exhibited a statistically significant increasing and decreasing trend. Similarly, Tigabu et al. (2020) and Hassaballah et al. (2019) revealed that streamflow was increasing without a significant increase in rainfall change patterns. On the other hand, Orke & Li (2021) showed an increasing rainfall has resulted in decreasing streamflow and a shortage of water resources in the watershed in the same basin. The lack of agreement on trends in the same catchment was probably due to differences in timescales as Orke & Li (2021) used a long-term record (more than 30 years). More than 30 years of recorded rainfall and streamflow data are reasonable for applying trend statistical analysis as demonstrated by other studies such as Love et al. (2010) and Gebremicael et al. (2013).

One of the major basins in southern Africa is the Zambezi, of which the Kariba catchment is part. The hydrology of the basin has existing and potential hydropower development, water supply, irrigation and other ecological processes (Beilfuss 2012). The Intergovernmental Panel on Climate Change (IPCC) has categorised the Zambezi as a river basin displaying more dire potential effects of climate change among 11 major African basins as it has been established that a minimal change in rainfall has a bigger change in runoff over the basin, hence being classified as highly sensitive to any changes in climate (Pachauri & Meyer 2014; Chisanga et al. 2022).

Lake Kariba, whose catchment this research is studying, was principally built for hydropower generation, with time, the catchment now serves multiple uses that include tangible fisheries, agriculture, urban water supplies, transportation, energy and intangible influence derived from tourism (Tumbare 2008). Each use requires a certain quantity and quality of water. Nonetheless, there is no information regarding a systematic trend valuation and analysis of streamflow over the catchment regarding water requirements for each of these uses. This has led to sustainability threats ranging from reduced water availability for hydropower generation to water pollution from aquaculture, agrochemicals as well as urban and industrial activities, invasive weeds and oil leaks from boats (Tumbare 2008).

The effects of the various types of water for power generation and other uses have become a source of concern based on recent trends of low flows that the Kariba catchment has been experiencing (ERB 2019). The causes of these low water levels have been attributed to anthropogenic activities, natural causes, climate change and mismanagement (Fathian et al. 2015). We hypothesize that the data length (both stream flow and rainfall) will impact the trend detection analysis. Therefore, the objective of this paper was to carry out trend analysis on 14 gauging stations located in the Kariba catchment using the entire set of records with varying data length and identify change points, if any, from 1924 to 2022. The results of this study will be of use in the planning and operation of Kariba Reservoir and any future hydraulic infrastructure development in Kariba catchment.

The Kariba catchment lies within the tropical African region south of the equator. It has three distinct seasons with small onset variations. These are the cool and dry season (April to August), warm and dry season (September to October) and warm and wet season (November to March) (SADC 2018). Intensive solar radiation provides the energy source for evaporation and for heating the Earth's surface. A temperature gradient, which builds up between the equator and the poles, drives the general circulation of the atmosphere. The average annual rainfall for the whole catchment is approximately 970 mm, with variations ranging from 1,200 mm in the northern part of the catchment to about 500 mm in the southern part (SADC 1998). Table 1 shows the location of gauging stations, the type of data collected, the duration of the datasets and the type of gauge used to collect the data. It should be noted that different entities collect the data on the Zambian side, the Water Resources Management Authority (WARMA) is mandated to manage the country's water resources, whereas, in Zimbabwe, the Zimbabwe Water Authority (ZINWA) has a similar mandate. However, the duration of rainfall data is less than 25% of the duration of river records.

Due to a lack of consistency in meteorology and climate data in the Lake Kariba catchment that corresponds to the records of stream data under consideration with most gauging stations with less than 15 years of data, while underscoring the importance of surface measurements, which many times are just point measurements, have non-uniform coverage and often have data void regions (Beven 2012). To overcome this challenge of field measurements, Earth Observation System (EOS) techniques, such as the use of reanalysis products, become useful (Schultz 1996; Farr & Kobrick 2000). Reanalysis can be defined as a scientific procedure of mounting a complete record of weather and climate variables and their variations over time (Wang-Erlandsson et al. 2016). The ability of the reanalysis products to simulate close-to-reality situations has been well documented in the literature though appropriate consideration must be undertaken when using the reanalysis datasets (Wang-Erlandsson et al. 2016; Muñoz-Sabater et al. 2021). The major advantage of this dataset is that it covers the entire globe and encompasses a longer period. For this present study, the TerraClimate and ERA5-Land precipitation dataset was used on selected stations to validate the trends that will be observed in the stream or river data. TerraClimate is a dataset of monthly climate and climatic water balance for global terrestrial surfaces from 1958 to 2022 with a temporal resolution of 4 km (1/24th degree) (Abatzoglou et al. 2018), while the land component of the fifth generation of European ReAnalysis referred to as ERA5-Land has the spatial resolution of 11.1 km at the time scale of daily spanning a period of 1963 to 2023. The dataset was obtained from the Climate Engine website (https://climateengine.com/).

A Pettitt's test was done on the monthly time series values with the following assumptions, H₀: Data are homogeneous and H_a: There is a date at which there is a change in the data. If the computed p-value is greater than the significance level α = 0.05, one cannot reject the null hypothesis H₀. The change point detection method is important in recognising jumps in the time series that can be attributed to climate change (Palaniswami & Muthiah 2018)

The statistics analysis was performed with a confidence level of 0.05 using an add-in software XLSTAT in Microsoft Excel. A Mann–Kendall trend test was carried out on the monthly average, minimum and maximum values for the 14 gauging stations in the Kariba catchment for various durations based on two hypotheses (two-tailed test): one is null (H₀) and the other is the alternative hypothesis (H_a). The H₀ elucidates that there is no trend in the time series while the H_a denotes the presence of a trend in the time series. Therefore, the trend identification in the time series will be based on the condition of a 5% significance level, with p-value being ≤α = 0.05, which translates to the acceptance of the H_a, which signifies the presence of a trend in the time series and with p-value being ≥α = 0.05, signifying the absence of a trend. Further analysis of point and areal rainfall reanalysis data obtained over the Kariba catchment was subjected to similar statistical analysis using the Pearson correlation coefficient on selected gauging stations that had statistically significant trends in stream time series. The Pearson correlation coefficient indicates the strength of the relationship between two values, either positive or negative, in this case, the stream measurements and rainfall data. The theory and application of this method has well been documented in the literature and statistical textbooks (Kothyari & Singh, 1996; Sahat et al. 2020)

This section outlines the results of various analyses of mean, minimum and maximum monthly stream flow for all 14 gauging stations and selected stations for rainfall reanalysed dataset. The results presented are from the Mann–Kendall test for trend analysis, Sen's test method for the magnitude of trend determination and the Pettitt test method for determining abrupt changes in data records. A further result of the correlation between the stream flow/water level and rainfall on the three stations that had statistically significant trends are presented, these are the Zambezi River at Lukulu, Senanga and Victoria Falls.

Table 2 shows the results of the trend analysis for monthly mean values for all the 14 gauging stations. In general, all the 14 gauging stations exhibited some trends with four stations exhibiting statistically significant trends whose p-values have been highlighted in Table 2. Nine stations showed decreasing trends, while five showed upward trends with Nanas Farm gauging station being the only one presenting a significant upward trend.

Table 3 shows the results of the Pettitt test that was done on all 14 gauging stations to determine abrupt changes in the long-time mean values and what time the change occurred. Statistically significant changes in the time series were observed at Lukulu, Senanga, Nanas Farm, Victoria Falls, Kalomo and Ume gauging stations.

A trend test on monthly minimum values was done as shown in Table 3 giving summary results for all 14 gauging stations. Ten gauging stations indicated a decreasing trend while four showed an increasing trend. Table 4 indicates that seven stations indicated a statistically significant trend at a 95% confidence level, of which five stations showed a decreasing trend and two increasing trends. The station with a statistically significant trend has p-values less than 0.05 and are in bold.

Table 6 gives a summary of the results of tests using the Mann–Kendall trend test on maximum monthly values. Eight gauging stations exhibited a decreasing trend, while six showed an increasing trend with one station (Zambezi River at Senanga) showing a statistically significant decreasing trend in maximum values.

The results of Pettitt's test done on the maximum monthly time series values are shown in Table 7. Four gauging stations showed abrupt changes in the long-term maximum values: the Zambezi River at Lukulu, Senanga, Victoria Falls and the Gwayi River at Deka Road Bridge.

Table 8 gives a summary of results of the Mann–Kendall trend test on the TerraClimate monthly and ERA5-Land daily precipitation data for Kariba catchment and selected point gauging stations. The results show intense rainfall variability across the catchment, with a decreasing trend on the Kendall's tau negative values, with Chavuma and Lukulu gauging stations showing a statistically significant decreasing trend in rainfall under the ERA5_Land dataset although, overall, the catchment received a minimal increase in rainfall for TerraClimate and ERA5_Land datasets.

The results of Pettitt's test done on the ERA5_land and TerraClimate rainfall dataset time series values are shown in Table 9. ERA5_land dataset showed a consistence abrupt change in the rainfall values for all the stations under consideration, as shown in the p-value that is less than 0.05. The TerraClimate dataset showed either constant or reduced long-term mean values for all sampled stations while the ERA5_Land dataset showed fluctuations in the long-term mean values. These fluctuations in the rainfall pattern across the catchment can be well correlated with the corresponding fluctuations being observed in river and stream levels.

A correlation analysis of the stream flow/water level and rainfall was done on the three stations that had a statistically significant trend: Zambezi River at Lukulu, Senanga and Victoria Falls. The Pearson correlation coefficient was used to determine the linear relationship between the two variables. The results showed that there was a positive correlation between rainfall and water levels at the Zambezi River Station at Lukulu and Senanga, while the Zambezi River at Victoria Falls showed a negative correlation, as shown in Table 10.

The amount of water available at any point in the stream, tributary or river is vital in supporting the social and economic development of a nation (Ndebele-Murisa et al. 2021). Its availability is usually affected by many factors such as land cover changes, amount of rainfall received and climate change. Hence these factors are crucial in determining the variation and establishment of trends in streamflow (Roudier et al. 2014). The results obtained from this analysis for average, minimum and maximum values for all the 14 gauging stations in the Kariba catchment indicate general declining water levels with a few exceptional increasing trends. The observed increasing trends are attributed to the length of records as established already on two stations that are fairly adjacent but indicating a different trend and also the variability of rainfall across the catchment has a significant influence on the phenomenon. The change points of these time series have been observed in the 1980s for the dataset with more than 50 years of records.

The change in mean average values for stations with a statistically significant change point (p ≤ 0.05) indicates that Zambezi River at Lukulu had its mean value reduced by 20%, at Senanga a reduction of 15% was observed. The Zambezi River at Victoria Falls had its mean value reduced by 23%. Ironically, Nana's Farm located some 33 km upstream of Victoria Falls (typically the same reach) had its mean value increase by 28%. The difference in trend indicators at Victoria Falls and Nana's Farm gauging station can be attributed to the length of records that were at the two stations. Zambezi River at Nana's Farm was opened in 1994 and the analysis is for 28 years of the record while Zambezi River at Victoria Falls has recorded as far back as 1907, with the analysis for this present study restricted from 1924 to 2022. Dixon et al. (2006) established the relationship between the length of record of analysis of streamflow data and the trend of the streamflow as a cause-and-effect kind of relationship. Furthermore, Ekolu et al. (2022) indicated that studies with a span of less than 30 years of streamflow analysis in southern Africa tend to show an increasing trend while those with more than 30 years of data analysis have a decreasing trend. This agrees with the findings of this present study where change points in the long-term mean were observed in the 1980s. The lower catchment of Kariba gauging stations had exceptional shifts in mean values as they have less than 30 years of dataset records. Kalomo River at Luezhi Ranch with 14 years of records had its mean value reduced by 55% and Ume River at Binga Road Bridge with 16 years of records had its mean value reduced by 46%. Coincidentally, the changepoint year for the two gauging stations in the lower catchment of Kariba happened in 2012.

The minimum average values had substantial changes observed in statistically significant change points (p ≤ 0.05). Zambezi River at Chavuma had its minimum average flow reduced by 34%, Lukulu by 45%, Senanga by 27%, Victoria Falls by 35% and the Luanginga River at Kalabo reduced as high as 59%. A rise in minimum average values was observed at Little Zambezi River at Matongo (11%) and Zambezi River at Nana's Farm recorded an increase of 28%. The lower catchment gauging stations, by nature or regime of the flow of being annual streams with a minimum value of zero were difficult to determine the change; however, they all showed a statistically decreasing trend.

In terms of the maximum average values, the entire catchment had a generally decreased trend too, with significant changes in maximum values being observed at Zambezi River at Lukulu (13%), at Senanga (13%) and Victoria Falls reduced by 30%. Exceptionally, the Gwayi River at Deka Road Bridge had its annual maximum average increase of 70%, with the year of change being 2019.

The results obtained are close to what Zimba et al. (2018) obtained when a Pettit homogeneity test was done on Lukulu and Senanga mean values where at both stations, changes were observed in 1981. Furthermore, a 19 and 25% reduction in mean values was observed at Senanga and Lukulu, respectively for the period of analysis, while for the present study, a 15 and 20% reduction has been observed for Senanga and Lukulu, respectively.

The trend and changes in long-term mean, minimum and maximum values can be attributed to changes in land cover and land use to a certain extent. However, climate change contribution cannot be ruled out as Beilfuss (2012) and Ndhlovu (2013) suggested that the Zambezi River basin whose total area is 47% Kariba catchment is at risk of the effects of climate change. Temperature rises and variations in precipitation have been noted in the Zambezi basin since 1980 (Kling et al. 2014), this has also been demonstrated in the results obtained from the analysis of ERA5_Land and TerraClimate precipitation dataset over the Lake Kariba catchment and point gauging stations. The fluctuations in the rainfall pattern being observed across the catchment can be well correlated with the corresponding fluctuations being observed in river and stream levels. For example, the statistically significant shift in the trend of water levels can be well correlated with the rainfall obtained at stations, such as Lukulu and Senanga, as shown in the positive Pearsons correlation coefficient, while Victoria Falls station showed an inverse correlation, which can be attributed to the length of the rainfall dataset that was used in determining the correlation. Further studies on the effects of climate change on the Zambezi basin indicated that a 20% decrease in precipitation and a corresponding increase of about 25% in evaporation coupled with the planned abstractions in the Zambezi basin would translate into reduced runoff and streamflow from as low as 5% to as higher as 40% (Chenje 2000; Desanker & Magadza 2001; Kling et al. 2014). This agrees with the findings of the present study as highlighted in the changes in the time series obtained on most of the gauging stations and those that Beilfuss (2012) obtained in the Zambezi basin.

The Kariba catchment is an important catchment to Zimbabwe, Zambia and the entire southern Africa region as it houses the two hydropower stations which are integral parts of the Southern African Power Pool (SAPP). The availability of water in the catchment for hydropower production has been of concern in the recent past. A trend analysis of stream time series at a 95% confidence level was conducted in the present study at all the 14 gauging stations to establish if there are any changes in the time series in the water level, discharges and precipitation over the Lake Kariba catchment, the trend and change points can be observed.

The results indicate a general decreasing trend in the water level and precipitation, with statistically significant trends and change points in time being observed at some of the gauging stations. Various factors that influence streamflow have been explored in the literature review such as land cover changes, rainfall and climate change, a thorough examination of long-term trends, change points and variations in streamflow at multiple gauging stations has been done, with a consideration to results and outcomes of similar previous studies conducted. The findings of the present study indicate that stream flow has been impacted by both land use and climate change.

The authors are immensely grateful for the support given by the National Research Foundation through the AJ–CORE Project and partners to this research project, Zambezi River Authority for the supply of streamflow and water level data used in this study. Partial support was also received from the tipping point explained by climate change (TiPPECC) supported by SASSCAL.

B.M. contributed to conceptualisation, data analysis, writing draft paper, and editing. K.B. contributed to conceptualisation, reviewing, editing, supervision, funding, and resource management. L.C. contributed to reviewing, supervision, funding and resource management. Y.U. contributed to reviewing, funding and resource management. I.N. contributed to conceptualisation, review, supervision, funding and resource management. The authors discussed the analysed data and the results of the final paper.

The authors declare that they have no conflict of interest.

Data cannot be made publicly available; readers should contact the corresponding author for details.