Spatial and temporal variation in rainfall and streamflow – Dzalanyama Catchment, Malawi

Rainfall and streamflow in Africa display high levels of variability across a range of spatial and temporal scales, with important consequences for water resource system management (Conway et al. 2008). Throughout Africa, this variability brings significant implications for society and causes widespread, acute human suffering and economic damage (Conway et al. 2008). Rainfall is a renewable resource, highly variable in space and time, and subject to depletion or enhancement due to natural and/or anthropogenic causes and the rainfall trend is an essential aspect in its analysis (Onyenechere et al. 2011).

Rainfall is a dynamic parameter in the hydrologic cycle, generally varying spatially and temporally due to changes in atmospheric gas and particulate concentrations. Analysis of such variability is important for the sustainable use of water resources, and management of floods and droughts. Brunsell et al. (2009) and Bellu et al. (2016) both note that detailed knowledge of the spatio-temporal distribution of rainfall is crucial for accurate modeling of flood management. However, Terêncio et al. (2019, 2017) argued flood management using detention basins and/or surface water storage using for example rainwater harvesting systems. Kumar et al. (2010) reported that changes in rainfall distribution could influence both spatial and temporal distributions of runoff, soil moisture, and groundwater reserves, and change the frequency of floods and droughts. Therefore, it is important to detect trend variations at spatio-temporal scales around the world (Wang et al. 2013; Pingale et al. 2014; Cui et al. 2017).

Several studies have been conducted to detect spatio-temporal trends in hydrologic and meteorological time-series data. The methods used included parametric (simple linear regression) and non-parametric (Kendall rank correlation, Spearman's rho, Mann-Kendall, modified Mann-Kendall, Theil-Sen slope) tests (Conway et al. 2008; Onyenechere et al. 2011; Terêncio et al. 2019, 2017). Studies conducted in Nigeria by Odjugo (2009, 2004) showed declining rainfall there, resulting in water level reduction, or the total drying up of some rivers and lakes in Northern Nigeria. Climate change and anthropogenic factors can affect hydro-meteorological processes adversely and continuously, their impacts appearing as trends or sudden jumps (Şen 2012).

Rainfall patterns have also changed, due to climate change, in Malawi, where rainfall varies considerably both seasonally and from year to year. The country has one of the most erratic rainfall patterns in Africa, with frequent recurrence of droughts and floods (Gumma et al. 2019). Water availability depends largely on the season's rainfall. According to the Malawi Government (2012), experience has shown that the country is vulnerable to climate change and extreme weather events. On average, Malawi receives an annual rainfall of 800 to 1,000 mm (Kumbuyo et al. 2014). This is adequate for rain-fed crop production and recharging aquifers, but the rainfall's spatial and temporal distribution is very erratic and uneven. The country is said to be prone to hydrologic droughts because of climate change (Malawi Government 2012). Rainfall variation is influenced heavily by anthropogenic factors such as changes in land use and land cover, among other things. The main driver of much of the observed variability in streamflow is rainfall, particularly at the scale of large river basins (Conway et al. 2008).

An understanding of the temporal and spatial characteristics of rainfall and streamflow is central to water resources planning and management, and is important in agricultural planning, flood frequency analysis, flood hazard mapping, hydrologic modelling, water resource assessment, and climate change impact and other environmental assessments (Michaelides et al. 2009).

The information on rainfall trends in the Dzalanyama Catchment, Malawi, is limited and there has been no investigation of the variability and trends of rainfall and streamflow there. This study's main objective was to use historical rainfall and streamflow datasets to assess the spatial and temporal variation of rainfall and streamflow in the catchment.

This study was carried out in the Dzalanyama Catchment area, in Lilongwe District, south-west of Lilongwe City. The catchment is underlain by a range of steep hills that slope south-eastward and form the watershed between Malawi and Mozambique, Figure 1. The catchment covers 989,300 hectares, of which 619,100 form part of Lilongwe District, and the rest are in Mchinji and Dedza districts, bordering Mozambique.

Four main rivers – Lilongwe, Lingadzi, Diamphwe and Bua – drain Lilongwe District. The Lilongwe River flows from Dzalanyama Mountain across the district to merge with the Linthipe River, which forms the border with Dedza District at the north-eastern tip of Lilongwe District. The river rises in the Dzalanyama Range, on the Malawi-Mozambique border, around 1,650 m above sea level, flowing north-east across the Central Region Plateau through Lilongwe City to Lake Malawi. The Lingazi River runs from the north and merges with the Lilongwe in Lilongwe City, while the Diamphwe River rises in the Dzalanyama Mountains and runs into the Linthipe River, forming the border with Dedza District. The Bua River forms the border with Kasungu in the north-east of Lilongwe District (Gumma et al. 2019).

The study area has a warm tropical climate with a mean annual temperature of about 21 °C. The lowest temperatures are experienced in July, ranging between 3.5 and 12.5 °C, and the highest in October-November rising up to 39 °C. Generally, there are cool, dry, and rainy seasons. The cool season is from May to July; the dry season is from August to October; and the rainy season is usually from November to mid-April. Mean annual rainfall is 800 to 1,000 mm. Rainfall distribution is influenced heavily by aerographic effects so that windward slopes receive more than leeward ones, and areas with high elevation receive more rainfall than low-lying areas. The passage of the inter-tropical convergence zone, between December and June, also influences rainfall in the study area (Kumbuyo et al. 2014).

Lilongwe District is underlain by volcanic and metamorphic rocks. The most important are gneiss, granulites, and schist, with important pegmatite developments. Sandy clay loam, clay loam, and clay soils with ferruginous soil properties dominate the catchment.

Historical monthly rainfall data were collected for four stations – Dzalanyama Kanyungu, Sinyala, Chitedze, and Bunda College – near the Kamuzu dams from the Department of Climate Change and Meteorological Services of Malawi. Daily rainfall data were also obtained for Kamuzu Dam from Lilongwe Water Board. The obtained dataset were summed up to give monthly data and, subsequently, the annual rainfall for a particular year. The rainfall data were reported to three significant figures. Table 1 is a summary of the rainfall data collected.

Data were collected from several flow gauging stations in the Dzalanyama Catchment. The daily data came from the Department of Hydrology, Ministry of Agriculture, Irrigation and Water Development (MoAIWD) of Malawi. Table 2 is a summary of streamflow data collected.

The test statistic, S, is assumed for the series where the sample size, n > 10, is asymptotically normally distributed, with mean E(S) and variance var (S) – see Equations (3) and (4):

The presence of an increasing or decreasing trend is determined using Z_MK. A positive value indicates an upward trend and a negative one a downward trend. Z_MK has a normal distribution to test for either upward or downward trend at ‘α’ level of significance (usually 5% with Z _0.025 = 1.96). The null hypothesis (H_o) for this type of trend analysis considers the existence of a non-monotonic trend from a data series. Thus, H_o is rejected if the absolute value of Z_MK exceeds Z_1-α/2 (rejected H_o: |Z| > Z_1-α/2) where Z_1-α/2 is the standard normal deviation and α is the significant level for the test. The probability value (p) from the two-tailed test using the Z_MK value, can also be used to test the trend's significance. If p exceeds α, then the null hypothesis (H_o: no trend in the data series) cannot be rejected but, if p is less than α, the null hypothesis is rejected (Mann 1945; Kendall 1975).

In this study, the non-parametric MK test in Microsoft Excel Statistical Software (XLSTAT) was used. The test was carried out at a 5% level of significance. The major variables for test interpretation were Z and p.

The highest mean annual rainfall of 933 mm was recorded at Dzalanyama Kanyungu station and the lowest, 866 mm at Bunda College, Table 3. The highest standard deviation (SD) was the 269 mm recorded at Kamuzu Dam and the lowest 142 mm at Sinyala. 1,490 mm, the maximum registered, were reported at Dzalanyama, while Bunda College reported the minimum of 165 mm.

Table 4 shows the monthly and annual water balances for the Dzalanyama Catchment for the period 1960 to 2014. The long-term balance was determined as precipitation minus losses. The warm-wet season (November to March) has higher mean water balances, with excess runoff up to 185 mm/month. The cool-dry season (April to September) has lower mean values, the lowest being approximately zero. The SD is also higher in the warm-wet than the cool-dry season, reflecting the higher variability of the balance in the wetter months. The CV, which is the ratio of SD and mean (SD/mean) is comparatively higher in the dry than the wet months. For example, CV is 336% in July but only 39% in January. This means reliability is better during the wet months than the dry ones.

The monthly trend analyses for the five rainfall stations showed that at Dzalanyama Kanyungu, Sinyala, and Kamuzu Dam, there were increasing trends in most months, but not April, May, June, or October (Table 5). At Chitedze, there were decreasing trends for most months apart from January and October, but the trend was statistically significant at 5% for June only. The Bunda College showed decreasing trends in March and December, with increasing trends in all other months. None of the monthly trends for Bunda College were statistically significant at the 5% level. The decreasing trend for the Kamuzu Dam was significant (at 5%) for August only.

Both the increasing and decreasing trends from the analysis annual time-series in the catchment were not significant at the 5% level, Table 6. The stations that had decreasing trends in the catchment were Dzalanyama Kanyungu, Chitedze, and Bunda College, whereas Sinyala and Kamuzu Dam had increasing trends.

At the seasonal time-scale, the warm-wet season has an increasing trend while the cool-dry and hot-dry seasons showed decreasing trends. It was observed that all the seasonal trends from this analysis were not significant at the 5% level. All three seasons also showed decreasing trends but only that for the cool-dry season was significant at 5% (Table 7). On the seasonal time scale, Chitedze showed decreasing trends in all seasons, but the trend was only significant (at 5%) for the cool-dry season. Kamuzu Dam showed a decreasing trend in the hot-dry season, and increasing trends in the warm-wet and cool-dry seasons, none of which were significant at 5%.

Table 8 shows that warm-wet season has the highest mean, the peak being 16.95 mm/month in February. The cool-dry season has lower mean values, falling to 0.45 mm/month in September. The SD is also higher in the warm-wet season than in the cool-dry one, reflecting higher streamflow variability in wet months. The relatively higher CV in the wet months confirms that streamflow is more reliable in wet months.

Table 9 illustrates the trend analysis results of monthly streamflow for the stations in the catchment. At station 4.D.4 most months showed increasing trends, but those in February, April, and December were decreasing; only the trend in October was significant at 5%. Station 4.D.6 showed increasing trends in all months but only those in August, September, October, and November were significant at 5%. Decreasing trends were found in all months at 4.D.21, and those in July, September, and October were significant at 5%. Station 4.D.24 showed increasing trends in all months from July to January, inclusive, but decreasing trends for February, March, April, May, and June; none were significant at 5%. Station 4.B.4 showed decreasing trends that were significant at 5% for February, March, and April only.

Table 10 shows the annual streamflow analysis for the catchment. Station 4.D.6 showed a positive trend that was significant at 5%. 4.D.21 showed a decreasing trend at annual time scale that was significant at 5%. In general, the annual time-scale analyses showed increasing trends that were not significant at 5% while the decreasing trend was statistically significant.

Table 11 shows seasonal time-scale streamflow analysis. Seasonal analysis of streamflow showed increasing trends that were not statistically significant except for the cool-dry season that showed an insignificant decreasing trend. Station 4.D.4 showed decreasing trends that were statistically significant for the cool-dry season only. At station 4.D.6 an increasing trend in the hot-dry season was significant at 5%. The warm-wet season at station 4.B.4 showed a decreasing trend that was statistically significant. The other two seasons showed decreasing trends that were not statistically significant. The warm-wet season streamflow analysis at 4.B.4 showed decreasing trends that were not statistically significant at 5%.

Of the five rainfall stations, only Chitedze, Bunda College, and Kamuzu Dam showed statistically significant trends at 5% in rainfall at different time scales. Chitedze showed decreasing trends in June and in the cool-dry season. Bunda College showed a decreasing rainfall trend in the cool-dry season. Kamuzu Dam showed a decreasing trend in August.

At monthly time-scale, stations 4.D.4, 4.D.6, and 4.D.24 showed increasing trends that were significant at 5%, but 4.D.21 and 4.B.4 showed significant (5%) decreasing trends. Significant decreasing trends (5%) were observed in rainfall in the catchment, while significant increasing trends were observed in the streamflow. This suggests strongly that the changes in streamflow are not caused directly by changes in rainfall. The most likely alternative is changes in land use and land cover.

The results of rainfall and streamflow analyses are in line with the results of other studies in countries in southern Africa. A study in South Africa by Odiyo et al. (2015), who investigated the long-term changes and variability in rainfall and streamflow in the Luvuvhu River catchment, revealed decreasing rainfall trends in four stations while catchment streamflow showed increasing trends. They suggest that anthropogenic factors such as impoundments could be impacting streamflow. They also suggest that rainfall and streamflow variations are also affected by non-climatic factors such as increase in water use and consumption resulting from population increases, economic development, and changes in land use and land cover. Gwate et al. (2015) studied the dynamics of land cover and its impact on streamflow in the Modder River (South Africa) and showed that rainfall in the catchment remained essentially the same over the selected period while streamflow increased significantly. This suggests that the streamflow increases are largely unrelated to climate variability or change but caused by the substantial changes in land use and land cover during the period observed. Gumindoga et al. (2018) studied the impact of land cover changes on streamflow in the Middle Zambezi Catchment in Zimbabwe and showed that there were increasing but insignificant trends in rainfall while decreasing insignificant trends were observed in streamflow. This suggested that the streamflow changes were not driven directly by changes in rainfall but other drivers such as land use/cover changes.

Decreasing trends that are statistically significant were observed in rainfall while increasing significant trends were observed in streamflow. The increase in streamflow over time suggests that there is increased runoff, usually accompanied by increased rates of erosion and siltation. Therefore, the increase in streamflow is consistent with common knowledge on the prospective role of tropical forests: that reducing forest cover in forested catchments increases their water yield. This suggests strongly that the changes in streamflow do not arise directly from changes in rainfall, but from other factors in the catchment such as land use and land cover. The negative and positive trends' variability at various stations may indicate the effects of human activity.

The anonymous reviewers of this paper are acknowledged gratefully.

Authors have no competing/conflicting interests.

No funding was received for this research work.

Supplementary data consist of shapefiles of the Dzalanyama Catchment, study area boundary; location of rain gauges; Malawi and rivers. Additional data can be provided upon request.

The authors declare no conflict of interests. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, and in the decision to publish the results.

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