Wetland water balance assessment: a case study of the Sosa wetland, Maseru, Lesotho

Wetlands are areas with marsh, fen, bog or water, either natural or man-made, temporary or permanent, with static or flowing water that is fresh, brackish or salty, involving areas of seawater with a maximum depth of 6 m at low tide (Ramsar 2016). Lesotho's wetlands are classified into three basic categories, namely palustrine, lacustrine and riverine (Olaleye 2019). The dominant form of the wetland is palustrine, which includes mires (bogs and fens), the majority of which are found at a high elevation (Olutayo 2012). Sosa wetland is a palustrine wetland.

In Lesotho, wetlands are under severe anthropogenic pressures due to overgrazing and the construction of housing (settlements) owing to the vastly growing population (Chatanga 2019). Wetland protection and restoration are critical for the planet's future sustainability, acting as safety nets for growing challenges such as climate change, food supply for an expanding global population, disturbance moderation, clean water and overall societal well-being (Zhang et al. 2021). Thus, comprehending land use change (LUC) and its spatiotemporal consequences for wetland ecosystems have become fundamental concerns in geography, ecology and sustainable science, all of which are critical for human well-being as well for regional sustainable development (Zhang et al. 2020). Wetlands supply important environmental services to the community and are surrounded by intensive farming that uses agro-chemicals and fertilizers in the management process (Moreira et al. 2021). Nowadays, drones are used to capture the extent of wetlands due to their high resolution (Bhatnagar et al. 2021). This method was found to improve the classification accuracy (87%) of the vegetation community compared to the unsupervised method (62%), which is the only option in the absence of ground truth (Bhatnagar et al. 2021).

The daily evapotranspiration capacity of the water area often exceeded that of the artificial wetland. Therefore, by comparing the water area and land use types (beach land, marsh and artificial wetland) in evapotranspiration, it was then concluded that water surface area is one of the driving factors influencing evapotranspiration (Wang et al. 2021). The wetland evapotranspiration partitioning and its relationship with abnormally low water levels were analyzed and results showed that the water evaporation rate was larger than the land evapotranspiration (Zhao & Liu 2016). Furthermore, it was found that the partition ratio increases up to 1 when the water level decreases showing that wetland evapotranspiration comes from surface land evapotranspiration than from water evaporation (Zhao & Liu 2016).

Wetland sensitivity varies with wetlands, size or productivity, which is caused by the variation in biophysical variables such as rainfall and evapotranspiration. Wetland condition describes the activities which are taking place within a wetland area. For most wetlands, the source of water is rainfall, hence their sensitivity depends mostly on the temperature and precipitation of the catchment. Wetlands are valuable ecosystems that provide many valuable services, yet many of these important ecosystems are at risk because of current land use and land cover (LULC) changes (Ouyang et al. 2014). In one of the studies based on regional climate sensitivity, a 1% increase in rainfall resulted in an increase of probable wetland area of 16.2%. While a 1% increase in temperature caused a negative impact on wetlands occurrence, reducing the probable wetland area by 37.9% (Nyandwi et al. 2016). Some studies conducted on wetland conditions showed that direct cultivation on wetlands, uncontrolled grazing or overgrazing, and aquaculture are common in areas extremely reliant on natural resources, particularly in Sub-Saharan Africa (Sakané et al. 2013).

Estimation of the water balance of wetlands is critical for understanding how they function and the way they respond to factors affecting them. Water balance analyses serve as an important foundation for decision-making at the watershed level regarding agricultural and forest plantation investments, water resource management and planning, water and carbon accounting and impact evaluations (Chemura et al. 2020). Accurate water balance management is critical for both short and the long-term water resource management. Obtaining an exact wetland water balance is challenging, as it is uncommon for all variables to be correctly monitored in the field concurrently. Therefore, hydrological models are normally used to estimate the water balance of wetlands. According to Belete et al. (2018), the spatial distribution of the annual water yield of the Nile River indicated a higher amount at the upstream whereas the contribution of the downstream was insignificant. Water balance varies from negative to positive. The positive value implies that inflows (rainfall) are greater than outflows (discharge and evapotranspiration). Therefore, the catchment or wetland is storing enough water for its hydrological and biological functioning.

Prolonged periods of low rainfall together with severe temperatures, and different land use types leading to land cover changes are hypothesized to contribute to the seasonal drying of wetlands, impairing their hydrological functioning, all over the world. Lesotho, like other countries, is not an exception to this problem. However, no study based on water balance assessment has been undertaken in Lesotho wetlands, and this study will focus on assessing the water balance of the Sosa wetland in the Jordan catchment. Specifically, this research sought to (i) estimate LULC changes in the upstream catchment of the Sosa wetland, (ii) compare site-specific evapotranspiration estimates using different methods, (iii) assess the sensitivity and conditions of the wetlands in the Jordan catchment and (iv) estimate the water balance of the Sosa wetland.

Landsat images with less than 20% cloud cover were used to detect changes in land cover and were downloaded from the Landsat online data repository (http://glovis.usgs.gov/). All images were downloaded between August and September, which is Lesotho's dry season. LandsatMSS, ETM and TM from 1975 to 2020 were then imported into Integrated Land and Water Information System (ILWIS) via geo-gateway for digital image classification. For Landsat 5 TM and Landsat 7, this study used multispectral pseudo-natural color band combinations of 5, 4 and 3, and for Landsat 8, it used a combination of 6, 5 and 4. The pseudo-natural color scheme is excellent for detecting land cover activity (Dube et al. 2014). The wavelength ranges and applications of the bands are listed in Table 1. The Landsat images were classified into themed maps using the maximum likelihood classifier, the algorithm that serves as the parameter in the classification process. Six (6) classes were classified, namely bare-land, water and marsh, cultivation, shrubs, grassland, and settlements. Table 2 has a detailed description of each class. Following the discovery of the themed map, the classified map was resampled using the study area geo-reference and then cross-referenced with the study area raster map in order to produce the classified map of the study area's size.

The accuracy of classification was assessed using 300 ground control points which were collected between January and April 2021 using a hand-held Global Positioning System (GPS) from various LULC classifications in order to validate the 2020 imagery. There were several classes, including bare-land, grassland, forest and shrubs, cultivation, settlements, and water and marsh areas. Additionally, coordinates were extracted from a Google Earth photograph of water bodies. The confusion matrix was utilized to quantify the error between the exact or real measurement on the ground, as well as to analyze the accuracy of the classification and identify areas of misclassification (Gumindoga et al. 2018). Additionally, the classification results were compared to ground truth data and the ones derived from Google Earth images. Each class received 50 control points in this trial. The areas were chosen based on their accessibility and concentration of the class of concern.

Since wetlands are hydro-morphological features, all wetlands of the Jordan catchment were described based on their position in the hydrological and terrain landscape. Then wetland sensitivity was determined using the twelve (12) biophysical variables, namely soil water content, bulk density, soil colour, soil texture, stream order, rainfall, soil pH, distance from dams, sediment transport index, evapotranspiration, topographic wetness index and flow accumulation according to Nyandwi et al. (2016). All the biophysical variables were resampled to a common 30 m spatial resolution using the nearest neighbor resampling technique in ILWIS-GIS. For each biophysical variable, weight was assigned based on their influence on wetlands formation. Furthermore, wetland condition, a feature that shows the status of the wetland in terms of anthropogenic activities within the wetland as well as at the catchment scale, was determined from the LULC map of 2020. It was determined by categorizing activities which were taking place in the Jordan catchment wetlands.

The water balance of the Sosa wetland was assessed using the Hydrologiska Byrans Vattenavdelning (HBV) model, which is named after the Hydrologiska Byrans Vattenavdelning unit at the Swedish Meteorological and Hydrological Institute (SMHI), which is where its development started in the 1970s. The HBV model has become widely used and exists in several versions. The version HBV light was developed using Microsoft Visual Basic and has become widely used in education at several universities. Daily streamflow, temperature and precipitation are combined to form a precipitation, temperature and discharge file, which together with evapotranspiration were used as inputs into the model. Model warming-up was done for the period of 2000–2001 while manual calibration was done for the period 2001–2010. Calibration for all parameters was continued until reaching the acceptable efficiency, with the mean difference closer to zero. Subsequently, model validation was done using data of the period 2011–2020 whereby the model parameters from the calibration were used on different data without any parameter being changed to see whether the calibrated model can simulate discharge properly for catchments with no observed discharge.

Streamflow data, which were provided by the LHDA had been collected from the Marakabei River weir located at a latitude of −29.55 and a longitude of 28.16 (Figure 1), at a daily time step. The data had 6% gaps, which were filled using basic arithmetic averaging of flows on the same day of the same month in the preceding and succeeding 2 years of the year with missing data. Filling gaps were critical to ensure a longer time of continuous data. The data were further cleaned to reduce errors that might be caused by mistakes done in data collection.

Furthermore, daily maximum and minimum temperature that were collected from the Mohale hydrological station for the period of 1975–2020, were obtained from LHDA. The daily temperature was then calculated as the average of the maximum and minimum temperatures. The average monthly evapotranspiration estimated from the Penman–Monteith method was used. Finally, hydrologic signatures, namely runoff ratio (RR), evapotranspiration ratio (ETR), the slope of the flow duration curve, baseflow index (BFI) and elasticity of streamflow (ESF) were also estimated for each subcatchment. The formulae used to calculate the respective hydrologic signatures were as follows (Sawicz et al. 2011):

• (i)
  RR

  

  (8)

  where Q is the total average discharge; P is the total average precipitation.
• (ii)
  ETR

  

  (9)

  where ETa is the total evapotranspiration; P is the total average precipitation.
• (iii)
  Slope of flow duration curve (SFDC)

  

  (10)

  where Q₃₃ is the 33rd quantile streamflow value; Q₆₆ is the 66th quantile streamflow value.
• (iv)
  BFI

  

  (11)

  where Q_90% is the 90th percentile streamflow value; Q_50% is the 50th percentile streamflow value.
• (v)
  ESF

  

  (12)

  where dQ is the difference in the streamflow between the previous and current years; dP is the difference in precipitation between the previous and current years; P is the mean annual precipitation; Q is the mean annual streamflow; median is a robust value (measure), used to eliminate outliers that have a substantial effect on the mean (Sawicz et al. 2011).

Table 4 shows the confusion matrix for validation of the classified map. The accuracy was calculated for the year 2020 for all six classes and the results revealed that the accuracy, in percentage, for all the six LULC classes was in the order of 42:60:64:64:68:80 for cultivation, bare-land, grassland, shrubs/forest, water and marsh and settlement, respectively. In this study, the classified 2020 map for the Jordan catchment and the sum of 150 ground truth points were used to calculate a confusion matrix that was used for accuracy assessment. The average reliability was found to be 64% and the overall accuracy of 61%, which is an acceptable accuracy.

The conductance of the stomata of any plant determines the maximum water loss during transpiration and this is considered an important factor in worldwide wetlands (Buckley 2019). Plants such as Neptunia natans and Eichhornia crasipes are floating macrophytes, having leaves that are located close to water surfaces, this keeps plants in a more saturated condition reducing water loss through evaporation, also these floating macrophytes reduces water loss through the shadow effect (Sikorska et al. 2017). Mulching of wetlands is also one of the best practices as it provides a creation of area to provide wetland vegetation, organic matter and soil organisms resulting in the reduction of water loss and an increase in wetland productivity (Denzler 2017).

According to Ouyang et al. (2014) the linear model was developed based on the current and historical climate data and this was used to determine the wetland sensitivity due to change in climate. The Palmer drought severity index (PDSI) of the current year and 2 previous years explains the annual wetland variation hence suggesting high wetland sensitivity to change in climate.

In the assessment of wetland sensitivity using remote sensing data, the area under vegetation was found to decrease in the wetland from 0.13 to 0.07 km², whereas the area under water declined by 0.85 km². On top of that, non-vegetated areas were also found to be increased by about 97% for the duration of the study (Ndlala & Dube 2021). The various wetlands and micro-catchments were transformed from naturally vegetative areas to agricultural land for the production of crops and grazing.

The water balance of the Jordan catchment was estimated using products of the HBV model and precipitation from the stations. Table 7 shows the annual water balance components simulated in the Jordan catchment. Annual water balance closure error, which is the change in storage for the Sosa wetland and Jordan catchment ranged between −3.13 for the 2020 water year and −3.51 for the 2004 water year. The annual mean water balance closure error is −3.34 compared to 0.15 km³/year, which was obtained in a water balance for a tropical lake in volcanic highlands (Alemu et al. 2020). This shows that the Sosa wetland and Jordan catchment receive less water than they release as discharge and evapotranspiration. Moreover, a decrease in storage change is due to vegetation which draws water from the stored water during the dry periods and uses it for photosynthesis and to meet transpiration demand.

Digital classification confirmed that the Jordan catchment is dominated by bare-land which subjects the catchment to loss of soil through erosion. Settlements are increasing at a lower rate, on the other hand, agriculture is not constant due to climate variability as well as soil erosion while forests and shrubs are decreasing at a high rate. Water and marsh, settlements, cultivation and bare-land increased by 2.04, 5.82, 4.1 and 28.71%, respectively, while forest and shrubs and grassland decreased by 38.83 and 1.76%, respectively, from 1975 to 2020.

There is a significant difference in evapotranspiration obtained from Penman–Monteith and Thornthwaite for the period of 1984–2020. Evapotranspiration from Penman–Monteith is by far the most reliable method that can be used in the absence of pan evaporation. This was proved by a test of difference between two methods which showed the sig(twotailed) value of 0.001, which is less than α of 0.05.

Rainfall and evapotranspiration affect the sensitivity of the wetland. Topographic variables such as Strahler order, flow accumulation, slope, topographic wetness index, sediment transport index and distance to water bodies as well as biophysical properties of the catchment being soil water content, soil porosity, bulk density and soil pH also affect the sensitivity of the wetland. The most sensitive wetlands are found at the headwaters and at the middle reach of the catchment.

The HBV light model shows effectiveness in simulating streamflow in the Jordan catchment based on model efficiency, mean difference and coefficient of determination performance indicators. The knowledge of catchment characteristics is very important in the process of calibration and also the quality of measurable data is of great importance in modeling. There is a negative water balance between components of Sosa wetland and this was proven by input (rainfall) being less than output (actual evapotranspiration and discharge), with water balance closure error ranging from −3.13 to −3.51. It is, therefore, not surprising that water bodies in the Jordan catchment including Sosa wetlands itself are at risk of drying up during prolonged drought spells due to very high water loss.

For the Sosa wetland revolution, the Green Revolution is advocated for. The Revolution where it has been practiced globally, led to high productivity within wetland systems through the addition of nutrients that encouraged the growth of phytoplankton and aquatic plant growth within these wetland systems. The above may reverse the risk of drying and the high risk of water loss.

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

The authors declare there is no conflict.