Remote sensing lake level fluctuations in response to a changing climate

Water is the very essence of life (Nhedzi 2009). A continuous supply of water in the right quality and quantity is critical to meet the ever-increasing needs of agriculture, industry and domestic consumers the world over. This has prompted the incorporation of scientific enquiry into water resources planning, development and management, with the water balance approach being dominant (Drayton 1984; Ayenew 2002). Water resources in general are affected by a number of natural and anthropogenic factors. An understanding of the effect of these processes is an important step towards better management of these water resources. This is especially critical at a time when the effects of the changing climate are cascading to other sectors of the environment.

Man-made lakes and reservoirs are an important source of water for millions of the world's inhabitants. This is because together with groundwater, surface water reservoirs such as lakes are an important source of water in semi-arid regions. The same sources of water will play an important role in the mitigation of the impacts of climate change expected to affect these regions which are hypothesised to get even drier (Greve et al. 2014; Intergovernmental Panel on Climate Change (IPCC) 2014). Furthermore, reservoirs and lakes are also linked to key socio-economic development projects such as hydro-electricity power generation (Klunner 2012; Liu et al. 2013) and irrigation development. The levels of success and sustainability of these developments are hinged on the ability of the reservoirs or lakes to supply enough water in the medium to long term. Thus, the monitoring of reservoir levels and storage capacity is of socio-economic significance. Lake levels and their fluctuations are indicators of climate change and climate variability (Zhang et al. 2013). Lake level fluctuations have the potential to affect the biology of lakes by altering nutrient concentrations (Ozen et al. 2010; Zohary & Ostrovsky 2011). For example, the 1992 drought that affected the southern African region significantly lowered reservoir levels and resulted in countrywide economic losses in Zimbabwe (Chagutah 2010). Furthermore, the flooding of the low-lying areas from the rising levels of Lake Malawi was associated with social and economic problems for Malawians, Mozambicans and Tanzanians living on the lakeshores (Drayton 1984). Thus, the proper and scientific management of water resources is closely related to several development goals including poverty eradication, socio-economic progress and environmental protection.

Conventionally, water resources planning and management the world over, including in Zimbabwe, has often assumed a stationary mean climate with no significant changes or variations over a given time. However, the reality of the situation is that anthropogenic and natural factors affect the water resources of an area, including lake levels. Zimbabwe is one of the countries expected to receive decreased amounts of rainfall owing to the impacts of climate change (Greve et al. 2014). This has led to several scientific approaches being implemented to plan for, develop and manage water resources (Dube & Zaag 2003; Tererai 2005; Viriri & Musariri 2006; Nhedzi 2009), including the sensitivity of reservoirs to variations in climate parameters. Small reservoirs, for example, can carry over very little water from one season to another, especially in the face of a severe drought (Mugabe 2005). However, there is a paucity of information on the sensitivity of medium to large sized reservoirs in semi-arid regions to changes in climatic patterns, especially annual rainfall. An understanding of the sensitivity of the reservoir levels to changes in climate is an important step towards management of the demand for water by the various, and often, competing users (Dube & Zaag 2003). This is critical especially for reservoirs that support key and diverse socio-economic activities in the areas where they are located. It is thus important to analyse the sensitivity of these reservoirs and lakes to changes in catchment characteristics with a view to predicting their continued existence under different and changing climate regimes (Vallet-Coulomb et al. 2001). In this study, we demonstrate the utility of remote sensing in monitoring lake level fluctuations for water resources assessment and management. The aim of the study is to examine the nature of the relationship between the rainfall received in Lake Mutirikwi's catchment and the volume of water in the lake at the end of the rainfall season. The hypothesis here is that to be able to mitigate against rainfall variability and unreliability, the amount of water in a reservoir should be independent of the amount of rainfall recorded within that season. Therefore, we do not expect to find a significant relationship between the rainfall amount received in a particular season and the volume of water in the reservoir. In this study, the surface area of the lake (a property which can easily be determined from remote sensing) was used as a surrogate measure of the Lake's volume and is often functionally related to the volume of water in the lake (Vallet-Coulomb et al. 2001). This, we are confident, is an important first step in understanding the impacts of future changes in climate on the future of proposed socio-economic developments that are dependent on the lake (Klunner 2012; Liu et al. 2013).

Lake Mutirikwi (Figure 1) is the second largest inland water reservoir in Zimbabwe. It is located in the south-eastern low-veld of Zimbabwe and has a capacity of 1,378,082,000 m³ and a surface area of 9,300,000 m² (9,300 ha) at full capacity. However, the surface area of the lake fluctuates following the volume of water stored within the lake. The greater part of the lake's catchment is dominated by communal lands such as Gutu and Zimuto, with an average annual precipitation which ranges from 400 to 700 mm. The catchment area has three distinct seasons which are hot wet season (October to March), cool dry season (April to July) and hot dry season (August to October).

However, the starting and ending months of the seasons greatly vary annually. Vegetation in the catchment area of the lake is mainly miombo woodlands which show a dominance of brachystegia spiciformis and julbernadia globiflora in the relatively wet parts, to a mixture of miombo woodlands interspersed with acacia shrublands and open grasslands in the drier areas (Masocha 2010).

The study utilised Landsat Thematic Mapper 5 images, p169 r074 with a 30 m spatial resolution, downloaded from the NASA website (www.glovis.usgs.org). The satellite images used in this study and their dates of acquisition are detailed in Table 1. We used the images for 10 years because they were readily available from the website for the preferred months of May to August of each year since the launch of Landsat 4 in 1982. Only satellite data up to 2004 was used in correlation analysis because the rainfall data, measured at different rainfall stations for the post-2004 period, was not consistent and up to date. Images for the May-August period were preferred because they belong to the cool dry season where no significant losses to evaporation would have been recorded. Thus, these months were deemed appropriate for representing the volume of water available in the reservoir after each summer season.

To extract the surface area covered by water in the lake, the satellite images were first subset to show only the lake Mutirikwi area and its immediate surroundings. The resulting image was then classified into water and non-water areas. Using the IFF statement in ILWIS GIS, only the water areas were retained. The resulting raster files were converted to a segment map to allow for further editing of the classified maps based on visual inspection of the satellite images. After editing, the segment maps were vectorised, and later converted back to a raster format. The statistics function was then used to determine the area covered by water for each of the years. The area of Lake Mutirikwi (in square metres) was then converted to hectares by dividing by 10 000 to allow for easier calculations and interpretation of the equations. The percentage deviation of the lake area was also calculated by comparing the surface area at full capacity (given in the description of the study area) and the area of the lake for each year calculated from satellite images.

Monthly rainfall data for four rainfall stations which are close to and within the Lake Mutirikwi sub-catchment area were obtained from the Meteorological Services Department of Zimbabwe (www.msd.org.zw). The four rainfall stations are Mvuma, Zaka, Makoholi and Masvingo International Airport. These rainfall stations were used, as opposed to a single station close to the lake, because the amount of rainfall received in the catchment varies spatially. The rainfall data that was readily available covered the period between the 1923/24 and 2005/2006 rainfall seasons. Rainfall data beyond the 2005/2006 agricultural season was not consistent and up to date for all the ground stations, therefore it was not included in the analysis. The rainfall data was used to calculate rainfall totals for the following time intervals: (a) first half of the season starting in October and ending in December, (b) second half of the season starting from January and ending in March, and (c) full wet season from October to March. Average rainfall totals for the sub-catchment are shown in Table 2.

A full explanation and interpretation of the SPI including the nominal classes can be found in McKee et al. (1993).

The rainfall and SPI data were first imported into ILWIS GIS where point maps of the four rainfall stations were generated. Next, the spatial variations in both rainfall and SPI were determined for each year using the inverse distance weighting interpolation technique. The inverse distance weighting method was chosen because the values at unknown points are calculated using a weighted average of the values available at the known points (Wong 2016). Using the hydrology function in ILWIS GIS, and the outlet of Lake Mutirikwi, the Mutirikwi sub-catchment boundary was delimited from a digital elevation model of the area. The catchment map was later used to subset the interpolated maps so that only the variations in the rainfall and SPI for the lake's catchment were shown. This step was important in determining the average values for both SPI and rainfall within the catchment for each year. These values were later correlated with Lake Mutirikwi surface area values determined from satellite images. Lake level data corresponding to the month and day of satellite image acquisition was obtained from the Zimbabwe National Water Authority (ZINWA).

Correlation analysis was used to test for the nature and strength of the relationship between rainfall (first half total, second half total and seasonal total) and the surface area of the lake. The nature of the relationship between lake level data and lake surface area was also determined. Furthermore, the nature of the relationship between SPI and change in the surface area of the lake (as a percentage of the surface area when the lake is at full capacity) was also determined. Regression analysis was used to find the equation linking rainfall or SPI and lake surface area or change in surface area respectively, as well as to determine the significance of the equation in predicting lake surface area or the percentage deviation of the lake surface area from either rainfall or SPI.

Figure 2 shows that, for the period under consideration, the surface area of Lake Mutirikwi fluctuated over a very wide margin. The lake's surface area decreased by only 1% in the wettest season, while it decreased by as much as 90% in the driest season. However, for most of the time, the surface area ranged around 40% to 60% of its total area (the results are not shown here).

Results of this study (Figure 3) also show a significant linear relationship between lake surface area (determined from remote sensing) and lake level (field data). Thus, we conclude that lake surface area can be used as a surrogate measure of lake volume.

Table 3 shows that there is a significant positive relationship between (a) the total amount of rainfall received in the season, (b) the amount of rainfall received in the second half of the season and the surface area of Lake Mutirikwi.

The table also shows that the amount of rainfall received in the second half explains the surface area of Lake Mutirikwi better than the total amount of rainfall received in that season. The relationship between the amount of rainfall received in the first half of the season and surface area of the lake did not return significant relationships. The results showing a relationship between SPI and lake surface area are also shown in Table 2.

The surface area of Lake Mutirikwi shows a positive correlation with both the total amount of rainfall received, as well as SPI, for the second half of the season. Results from this study are in agreement with several studies done on the African continent and beyond, which used the more sophisticated hydrological balance approach to estimate lake surface area. While working with more detailed bathymetric data and several other variables, Ayenew (2002) concluded that precipitation trends played a significant role in explaining the observed lake levels and consequent fluctuations in the levels of Lake Abiyata in Ethiopia. After reviewing lake inflows and the method of calculating lake rainfall from lakeside gauges, Piper et al. (1986) concluded that the rise in Lake Victoria levels can largely be explained through rainfall totals. Zhu et al. (2010) also concluded that a significant portion (47%) of the increase in lake levels was accounted for by increased rainfall activities in China. Using the water balance model while working on the changes in the levels of Lake Malawi, Drayton (1984) concluded that variation in runoff rates (mainly because of changing precipitation patterns) was the single most significant factor accounting for the variations in lake levels while factors such as evapotranspiration, human abstraction rates and land-use/land-cover changes proved to have minimal explanatory power.

This research differs from previous studies in that we have used a simple remote sensing technique, a departure from the more data demanding empirical methods (Awange et al. 2008; Chu et al. 2008; Liao et al. 2013; Wang et al. 2013), to explain and monitor the temporal variations in lake levels. Feng et al. (2012) characterised the changes of inundation of Poyang Lake between 2000 and 2010. Using MODIS data, they determined that precipitation was the key driver of lake levels during the summer months, while other factors such as the flow of the Yangtze River explained variations in the lake levels during the dry season. However, this study related lake levels and surface area to rainfall patterns only.

The major highlight of this study is the high (75%) explanatory power of rainfall on the volume of water in Lake Mutirikwi for the same season. This is particularly worrying given that the annual rainfall totals for both the country in general and Runde catchment in particular are showing a downward trend (Chikodzi et al. 2013; Greve et al. 2014). This downward trend, however, has not been found to be statistically significant (Mazvimavi 2008). There is therefore an urgent need to review the future of the existing lakes and reservoirs as a safety net against the changing climate. Such a review is especially critical given that Lake Kariba, the largest man-made inland reservoir by volume, is reportedly being threatened by climate change (Decline of Lake Kariba 2016).

Furthermore, we have also shown that rainfall falling in the second half of the season significantly (p < 0.05) explains the lake surface area. Though this hypothesis has not been tested in this study, the weak relationship between other rainfall totals (seasonal and first half) and lake levels can possibly be explained by the fact that significant portions of the precipitation are lost to evapotranspiration during the mid-summer season dry spell (which often lasts between 2 and 6 weeks), contributing to soil moisture and some groundwater recharge which occurs at the onset of the wet season. While factors such as evapotranspiration and changes in land-use are important in determining the water balance of an area, these were not considered in this study. Our aim was to investigate the impact of rainfall variations on the amount of water stored in the reservoirs. The study was informed partly by the fact that reservoirs and lakes themselves are expected to cushion farmers against erratic rainfall patterns, and partly due to the fact that previous studies that used data intensive methods identified rainfall as a driver of lake levels (Drayton 1984; Zhu et al. 2010) and that human needs only account for a small percentage of the water uses from a lake (Zhang et al. 2011).

While (Dube et al. 2014) attributed the increase in the surface area of Lake Mutirikwi to stream bank cultivation and consequently siltation, results of this study have shown that fluctuations in rainfall amounts is the single most dominant factor explaining the surface area of Lake Mutirikwi. An assessment of the erosion hazard in Runde catchment by Mutowo & Chikodzi (2013) revealed that a significant portion of the catchment had negligible to low erosion hazard. These low erosion hazard areas make up most of the Lake Mutirikwi sub-catchment. Therefore, future studies should focus on a more detailed study combining several factors to improve on the factors explaining the lake level fluctuations.

We conclude that some lakes in arid and semi-arid areas are so sensitive to the amount of precipitation received in their catchments that they are not suitable for mitigating against the impacts of climate change. We have also shown the utility of remote sensing in monitoring lake level fluctuations within semi-arid landscapes. Such applications are important for water resources management where the reservoir(s) might be in an isolated/remote area or where resource constraints impact on the collection of consistent and objective field data. We recommend that to successfully mitigate against climate change, there is a need to consider the reservoirs' surface area-catchment area ratios. Such ratios point towards the level of fluctuations and variations of lake levels under changing rainfall patterns. This will cushion communities and aquatic ecosystems against challenges associated with falling reservoir levels.

The author is grateful for the free access granted to Landsat satellite imagery by NASA through their website www.glovis.usgs.gov, as well as the anonymous reviewers who improved the article through their useful comments.