Impacts of wetland loss and fragmentation on the hydrology of Zimbabwe's highveld

Globally, wetlands are one of the most productive ecosystems that provide several functions and services (Weise et al. 2020; Kundu et al. 2021). The ecosystem services and functions of wetlands account for more than 40%, even though they only make up less than 8% of the Earth's surface (Mahdianpari et al. 2020; Berhanu et al. 2021). Wetlands provide number of hydrological and ecological functions that are essential for human survival, which include flood attenuation, water purification, base flow maintenance, groundwater recharge and recreation, among others (Gxokwe et al. 2020; Rojas et al. 2020). Concerns about the wetland fragmentation are growing as more and more wetlands are converted to agricultural land and built-up areas. The world's wetlands used to make up roughly 5–10% of the land surface, but over the past century, more than 70% of them have been destroyed or altered (Kingsford et al. 2016), with about 35% of those losses occurring since the 1970 s (Davidson et al. 2018). Wetland losses are thought to have reached more than 80% percent globally, with annual rates of losses accelerating from 0.68 percent in the 1970s to about 1.60 percent in the 21st century (Davidson 2016). Further analysis reveal that over the previous 20 years, approximately 6% of East and Southern Africa's wetlands and around 2% of West and Central Africa have been degraded (Rebelo & McCartney 2019). However, due to lack of historical documentation and monitoring, estimating the amount of degraded wetlands in some parts of Africa poses a significant difficulty. Majority of Africa's wetlands are found between 15 °N and 20 °S, with the main wetland systems constituting riverine systems, lakes and marshes (Kabii 2017). In Zimbabwe, wetlands cover approximately 1.8% of the total surface area, with about 60% of wetlands found in communal and resettlement areas (Marambanyika & Beckedahl 2016). The main wetlands in Zimbabwe include dambos, floodplain, riverine system and pans, with the majority of wetlands located on the highveld plateau (Dube & Chitiga 2011). Zimbabwean wetlands have been facing serious threats from both natural and anthropogenic forces which have resulted in about 50% of wetlands being lost and fragmented over the past 50 years (Musasa & Marambanyika 2020).

Wetland fragmentation is the impairment of wetland spatial configuration and extent, which results in alteration of its functions, due to conversion of wetland to non-wetland areas (Kadziya & Chikosha 2013). Fragmentation occurs when habitat is reduced in size or the distance between remaining habitat patches increases resulting in increased spatial isolation of the wetlands (Cui et al. 2021). The process of fragmentation occurs in four phases which include perforation, in which the wetland experiences initial small openings, dissection, in which there are bigger intrusions of change, often along with physical features, dissipation, which is the spread of alteration, shrinkage, which results in the reduction in patch size and eventually wetland loss, in which the wetland disappears (Flowers et al. 2020). Despite the patchy nature of wetlands, fragmentation results in decreased density and increased physical isolation (Cosentino & Schooley 2018).

In the literature, a number of approaches have been used to quantify wetland fragmentation. Satellite images with different spatial resolutions and various change analysis methods have proven to be the most effective techniques of quantifying changes in wetland area (Amler et al. 2015). Due to long record of continuous observation and moderate spatial resolution, LANDSAT series of satellite images have been widely used for forest degradation, urbanization and wetland loss and fragmentation (Lin 2018). Zhao et al. (2015) in the Heihe river basin in China used image classification for land cover change analysis and then aggregated the land uses into wetland and non-wetland areas, for which a track change model was used for analysis of wetland fragmentation. Ma et al. (2018) used the combination of moderate-resolution imaging spectroradiometer (MODIS) and Landsat using the Spatial and Temporal Adaptive Reflectance Fusion Model. This model obtained the high-resolution image in the time series and Modified Normalized Difference Water Index in Panjin City, which showed wetland dynamics through the changes in Normalized Difference Vegetation Index (NDVI), and the Normalized Difference Water Index (NDWI).

In addition to use of remote sensing to monitor wetland loss and fragmentation, models have also been integrated to analyze the impacts of wetland fragmentation on flow regimes. Blanchette et al. (2019) used the PHYSTEL/HYDROTEL hydrological model to simulate wetland hydrological processes and quantify hydrological services they provide and concluded that wetlands can exhibit different flood regulation behaviors within a catchment, with efficiency of wetland service varying depending on location and size of the wetlands. Tang et al. (2020) in Houston, Texas used HEC-HMS in simulating the impacts of wetland size on downstream flood area found that inland wetlands alleviate floods when used as extra storage in the watershed, with upstream wetlands providing more regulatory mechanism. However, most researches on wetland loss and fragmentation have been focussing on impacts on ecosystems (Rahimi et al. 2020; Kundu et al. 2021), amphibian's population (Cosentino & Schooley 2018; Mester et al. 2020) and also water quality (Wu et al. 2020; Chang et al. 2021), with impacts of wetland fragmentation on hydrology, especially on Zimbabwe highveld headwater catchments remaining an unresolved issue.

Wetland fragmentation results in alteration of various components of the hydrograph (Tang et al. 2020), however there is lack of consensus on hydrological role of wetlands on highveld headwater catchments, hence it is important to investigate hydrologic implication of wetland loss and fragmentation on hydrology. On a global level, there are seemingly contradictory conclusions on impacts of wetlands fragmentation on hydrology. Bullock & McCartney (1996) in exploring the relationship and river flow interactions on Zimbabwe highveld concluded that wetland densities are neutral in their impact on dry season flows. Mccartney & Gifford (2000) also disputed the findings by Howard-Williams & Thompson (1985), which concluded that highveld headwater wetlands act as flow regulators, storing water during the wet season and releasing it during the dry season, thereby maintaining dry season river flows. Despite the controversial conclusion by Mccartney & Gifford (2000) that highveld headwater wetlands have no significant contribution to low flow maintenance, no other researches in recent years have explored the link between wetlands and hydrology in the highveld catchments of Zimbabwe.

Most studies on the wetland hydrology of Zimbabwe's highveld were published between 1990 and 2000 (Bullock & McCartney 1996; McCartney et al. 1998; Mccartney & Gifford 2000). These studies made the connection on how the flow regimes respond to variations in water input as catchment land uses change. However, lack of coordination and financing for wetland research is likely to blame for the decline in multidisciplinary research on wetlands. Since 2005, the majority of studies have concentrated on employing use of remote sensing to identify changes in land use and cover. Chikodzi & Mapfaka (2018) in Murehwa District looked at wetland fragmentation, but never explored how wetlands affect flow regimes, overlooking the link between wetland fragmentation and hydrology. Moreover, Musasa & Marambanyika (2020) also looked at the causes of wetland fragmentation and sustainable utilization of Zimbabwean wetlands but there was a missing link of how wetland fragmentations alter the hydrological processes of a catchment. The question on the impact of wetland fragmentation on hydrology is still open and unanswered, therefore this research seeks to enhance our understanding on the relationship between wetland loss and fragmentation. Because of this unresolved issue on the lack of understanding and unavailability of information on fragmentation and flow regimes, policy on wetland utilisation has been a challenge. Therefore, this study analyzed the impacts of wetland fragmentation on flow regimes for the three catchments. The objectives of the study were to: (1) quantify wetland fragmentation from 1984 to 2021 (2) simulate streamflows through the TOPMODEL rainfall runoff model using historical observations and (3) assess the hydrological impacts of wetland fragmentation on Zimbabwe's highveld. Results from the study will be important to the Ministry of Environment and the Environmental Management Agency (EMA) in contributing to the national wetland master plan, which will be essential in identifying the most affected wetlands and those with greatest impacts on hydrology, hence prioritize them in restoration programs.

Macheke, Nyagui and Upper Manyame (Figure 1) are located on Zimbabwe's highveld plateau between 30.5 degrees and 32.8 degrees east and 17.3 degrees 19.2 degrees south. Upper Manyame catchment is the headwater catchment for the Manyame River, which is a tributary of the Zambezi River, while Nyagui catchment is the headwater catchment for Nyagui River, which rises on the central plateau and flows through a largely rural area before joining the Mazowe River, while Macheke catchment is the headwater catchment for Wenimbi and Ruzawi rivers, which drain into Save. All the three catchments start from Marondera Town which forms part of the watershed. Upper Manyame catchment area encompasses settlements such as Harare the capital city, Chitungwiza and Ruwa, with Nyagui catchment cover several districts that include Marondera, Goromonzi, Murehwa and Bindura, with Macheke covering areas like Macheke, headlands and Rusape.

The three catchments cover approximately 15,299 km², with Upper Manyame measuring about 3,786 km², Nyagui (4,900 km²) and Macheke (6,613 km²). The climate of the area is described as transitional semi-arid humid climate as it receives annual rainfall total that varies between 700 and 1,000 mm, but is highly seasonal and the rivers flow strongly during the rainy season (November–April), but diminish during the dry season (between September and October). The altitude of the area ranges from 900 to 1,642 m above sea level. The three catchments are endowed with a lot of water resources such as dambos, rivers and reservoirs. Upper Manyame, Macheke and Nyagui forms part of the highveld central plateau that has been experiencing rapid urbanisation and increased economic activities. Moreover, these catchments are endowed with abundant wetland resources, which however are under serious threats from human activities that are in these areas.

Cloud-free remote sensing Landsat Thematic Mapper (TM) for the years 1984, 1994, 2004 and Landsat OLI from 2014 and 2021 imagery (spatial resolution: 30 m & temporal resolution: 16 day) were used in the study. Landsat images were selected, because they have shown to be valid in analysis of landcover change (Xu & Chen 2019). Landsat images were downloaded from Earth Explorer United States Geological Survey (USGS) website (https://earthexplorer.usgs.gov/). The dry season months were selected, since there is a good probability of getting minimal cloud cover during that period and grass would have died and crops removed from the fields, hence providing for improved landcover classification. Furthermore, given the time of year of the satellite images acquisition (dry and summer season), the differentiation between wetlands and bare ground is evident, making the classification much easier.

To assess the wetland output map from classification, the classified map was compared to the output from national wetland delineation. The classified maps with the five classes were first reduced to two classes of wetland and non-wetland, for easy comparison with the national wetland output.

After image classification, Lecos plugin in Quantum GIS (QGIS) was used to calculate the class level metrics for the three catchments. QGIS is a free and open-source geographical information system (GIS) which can be used to create or manipulate geographic data. With the help of user-written plugins that can be downloaded from the desktop suite, QGIS makes it simple to extend the functionality of its main features. Lecos is one of the important plugins that can be used in computing landscape metrics within QGIS. The plugin provides detailed statistical information on various landscape levels. For this study, the class metrics derived using lecos in QGIS were used to determine how wetland structure and composition has changed. To characterize the spatial characteristics of wetland fragmentation in this study, two landscape metrics were utilized at the class level (number of patches (NP) and mean patch size (MPS). These landscape metrics were selected because they can illustrate changes in land-use activity. Moreover, Number of patches was used as a measure of wetland fragmentation as it indicates the loss of connectivity within a wetland.

Daily discharge data for the study was obtained from the database of Zimbabwe National Water Authority (ZINWA) for gauging stations in Upper Manyame, Nyagui and Macheke catchments. A graphical method was used to assess the quality of the discharge data, in which it was used to contrast the change in rainfall and runoff. Although the network of the runoff gauges is generally poor, the network may serve the objectives of this study as stations are well distributed across the study area, and screening of runoff data using Double Mass Curves indicated that observation time series are complete and of good quality.

Thiessen polygon method was used to estimate daily areal rainfall in Upper Manyame, Nyagui, and Macheke catchments. Thiessen weights were calculated for all the three stations in ILWIS.

An STRM DEM (30 m resolution) covering the study area was retrieved free of charge from the website (www.earthexplorer.usgs.gov). The digital elevation model (DEM) was used to extract flow direction, flow accumulation, create streams, delineate the catchments, and calculate sub-catchment parameters in Integrated Water and Land Information System (ILWIS) software. A DEM was also used to derive the topographic index.

The area distance map was computed from routing of overland flow by the use of a distance-related delay. A point outlet map generated in QGIS and imported into ILWIS was used for distance calculation in ILWIS. After producing the distance map, the distance map was sliced into segments for the routing of surface flows to the outlets for the three catchments. The tabular distance distribution was transformed to a text file that was used as a TOPMODEL input file.

The IDL TOPMODEL code was modified to allow application in a semi distributed fashion. Once operational, the code was then applied to Zimbabwe highveld catchments. The code is a conversion of Keith FORTRAN version of TOPMODEL where an infiltration excess overland flow component is considered to allow for landuse simulations (Gumindoga 2010). The selection of the IDL code was based on the fact that it can handle relatively large sizes of the topographic index histogram.

The model was calibrated and validated based on three parameters using knowledge obtained from literature of (Beven 1997; Peters et al. 2003; Gumindoga et al. 2011, 2014) to find out which parameters are the most sensitive. The parameters that were found to be more sensitive include the soil hydraulic conductivity decay parameter (m), the soil transmissivity at saturation (To) and the average root zone available water capacity (SRmax). A time step of 24 hours was selected for computations to calibrate the model. The value of parameter ‘m’ was varied, holding values of remaining parameters constant and the value of m which yielded the highest efficiency was determined. This method was repeated for all the parameters in order to obtain parameters which gave highest value of efficiency. Model calibration for the years 2006–2008 was done by changing TOPMODEL parameters manually to optimize model performance for each model run. After model calibration and validation, model performance was assessed using visual comparison and statistical performance indicators. Nash Sutcliffe efficiency (NSE) and coefficient of determination (R²) were the performance indicators which were used. After calibrating and running the model, model validation was carried out to test whether the model (using the same parameter set obtained by optimization, but with independent data sets) still performed well. Model validation was done using data for the period from 2008 to 2010.

To assess the impacts of wetland fragmentation on flow regimes, the model parameters were calibrated first. After model calibration, a model based theoretical approach in quantifying hydrological impacts of wetlands change was implemented. Figure 2 illustrates the approach that was used to assess the impacts of wetland fragmentation on flows. As a factor of wetlands change m was found to be the most sensitive parameter so was varied proportionally to wetland change based on results from image classification (Table 1). T_o and SR_max were also varied but their sensitivity was found to be very low so their variation was not significant. Rainfall for the decade 2004–2014 was applied for all the decades and the model was run to mimic the catchment characteristics. In using rainfall and evapotranspiration of a single decade for all other decades, climatic factors were held constant. Rainfall for the decade 2004–2014 was run for all the years so that the climatic variables could be held constant and only impacts of wetland change could be observed.

Moreover, after obtaining the modelled discharge, the flow duration curves (FDCs) (Figure 14), were used to assess the distribution of probabilities of discharge being greater than or equal to a specified magnitude, with the slope of the curve being used to observe the changes in streamflows. The flow duration curves for the three catchments were plotted to show the relationship between the impacts of wetland fragmentation on the streamflows for the highveld headwater catchments. Complimentary to the flow duration curves in describing the impacts of wetland fragmentation, some quantiles were derived from the streamflows in order to acquire information about the behaviour of the catchments. Quantiles were also calculated from the simulated streamflow data for the contrasting periods of wetland change.

After the image classification process and slicing of the classified map into wetlands and non-wetland classes, the sliced map output map was compared with the national wetland delineation output. Results from the map comparison show that the classified wetland map displayed wetland that closely matched well with that of the national wetland delineation output (Figure 9). The riverine wetland matched well across all the catchments, as illustrated on Figure 9, with some dambos not being captured by the classification process, despite being captured by the national wetland delineation output. This could be due to the criteria used to classify and define wetlands, since national wetland delineation output in used the vertical distance to channel network (VDTCN) process which uses elevation to identify wetlands, while the classification process considered water, soils, and vegetation. Moreover, national wetland delineation output produced a lot of wetlands, with some of the wetlands having already been fragmented and lost. As conceptualised by Bian et al. (2021), the inundation area of a wetland depend on the antecedent hydrological conditions, the morphological characteristics of the depression, and the utilization purpose of the wetlands, therefore changes in these factors can cause the inundation area of wetlands to change, even between the time span of the data acquisition.

Change in landuse class and landscape metrics over time was used as a proxy for the six phases of landscape fragmentation as conceptualized Flowers et al. (2020). The phases of fragmentation include perforation, incision, dissection, dissipation, shrinkage and attrition. In relation to the sizes and number of the patches for the wetlands across the study area, the mean patch wetland area decreased dramatically for all the three catchments. Upper Manyame catchment, in period from 2004 and 2014 experienced massive wetland fragmentation as it had the largest decrease in average size of the patches (2 hectares), as compared to the other periods, while the period from 2014 to 2021, it experienced a decline in wetland fragmentation (see Figure 10), as the average patch size of wetlands reduced by only 31% as compared to 66% in the previous decade. In Macheke catchment, the average patch sizes were 3.7 hectares in 1984 and decreased to 0.937 hectares in 2021, indicating massive wetland degradation that has taken place over the years. Nyagui catchment also experienced rapid decrease in mean patch area for wetland landscape, as it reduced by 253% (1.891 hectares) between 2004 and 2014, while the period from 1984 to 2004 experienced a steady decrease in the mean patch area. The pronounced wetland fragmentation experienced in the three catchments can be attributed to rapid urbanization, coupled with increased agricultural activities. Results have shown that there is a significant change in the number of patches between the years, with most fragmentation taking place since 1994. In 1984, 7,899 wetland patches were observed in the Upper Macheke catchment, with 17,853 in Nyagui and 16,388 in Macheke. However, the number of wetland patches significantly increased, with 47,873 observed in 2021 for Upper Manyame, marking a 600% increase in number of patches, with Nyagui experiencing a 535% increase in the number of patches, while Macheke experienced a 648% increase. Increase in wetland fragmentation could be due to rapid urbanisation, wetland drainage and increased agricultural activity in wetland areas. A limited number of researches have been done concerning fragmentation of wetlands in Zimbabwe. Chikodzi & Mufori (2018) looked at wetland fragmentation in Murehwa, however the researcher only used google earth imagery to assess the wetland changes. Kowe et al. (2020) also conducted study on fragmentation in Harare and discovered that landscape metrics showed a decrease in the core vegetation cover and an increase in isolation and number of small vegetation patches. The results of this study are consistent with that of Zhao et al. (2015), who found that wetlands in Heihe river have been fragmented, resulting in the core area reducing by 12,8% while wetlands patches increased by nearly 9%. The results from various studies show that wetlands continue to be degraded despite the fact that demand for goods and services they provide continue to increase.

Moreover, the results from Macheke FDCs show that high and low flows for catchment have significantly changed for the period from 1984 to 2021. Low flows (<3 m³/s) exceedance probability declined from 97% in the first decade to 75% in the last period (2014–2021). Comparing the flow duration curves for the four decades (1984–2021), there is evidence of change in flow regimes, driven by a decline in baseflow. This could be due to decreasing groundwater recharge and increasing length of the dry season. Increased peak flows are accompanied by low flows showing the diminishing hydrological function of highveld wetlands, reducing their sponge properties. Low recharge on the highveld areas, which mostly depend on rain-fed wetlands, results in low flows that are typically brought on by diminishing infiltration rates.

At the watershed scale, results for Nyagui catchment show that low flow support provided by wetlands has reduced and high flow attenuation of wetland also diminished as the wetlands fragmented. Low flows (<3 m³/s) probability of exceedance declined from 86 to 71% in the last decade. The median flows (6 m³/s–15 m³/s) have no variability for all the decades. On high flows, pronounced peaks can be observed, owing to increased streamflows which result from flood attenuation function loss.

These findings on changes in streamflow in response to wetland fragmentation are similar to the findings of Blanchette et al. (2019) which found that the loss of wetlands modify the surface routing of water, as the river's capacity to support low flows and attenuate high flows is diminished. Furthermore, wetland degradation is accompanied by a significant reduction in sponge hydrological function of the wetlands as water is no longer absorbed into the wetland, but rather flow over the wetland surface. As expected, wetland degradation and loss resulted in reduced low flows and increased peak flows. The results are also similar to those reported by other researchers (Ahmed 2014; Makungu & Hughes 2021). Moreover, Blanchette et al. (2019) also reached the same conclusion in Des Hurons, Jaune, Lorette and Du Berger catchments that a decrease in wetland areas is associated with a decrease in low flow support, whereas an increase in wetland areas is reflected by an increase in low flow support. However, these results contradicts the findings by Bullock & McCartney (1996), who concluded that dambos do not operate as sponges but they even reduce dry season river flows because dambo storage is depleted more by evapotranspiration than by baseflow. From the available literature, it can be observed that the distinct water storage capacity and their role in controlling hydrological processes can be different from catchment to catchment, depending on a number of factors. Tang et al. (2020) found that hydrological functions of wetlands specifically rely on the natural combined influence of wetland type, location, and landscape characteristics.

Complimentary to the FDCs in describing the impacts of wetland fragmentation, some quantiles were derived from the streamflows to acquire information about the behaviour of the catchments (see Table 3). Quantiles calculated from the streamflow data for the contrasting periods indicate an increase in lower quantile flows, while there is an increase in the higher quantiles, indicating an increase in high flows. Increase in high flows may have been caused by low permeability of the soils due to wetland degradation, which resulted in low infiltration and more runoff. Moreover, from the slope of the curve calculated, steep slope at the high flows indicates the flashy response to rainfall as a result of rapid changes associated wetland fragmentation, especially between the decades 2004–2014 and 2014–2021. Moreover, the changes in low flows may have been due to the streams that dried more frequently in the 2004–2014 and 2014–2021 decades.

Results from landcover change analysis of Landsat imagery for the years 1984, 1994, 2004, 2014 and 2021 indicated that wetland area has been decreasing, while built up and agriculture have been increasing for the three catchments. The built-up area increased by 474, 293 and 211% between the period from 1984 to 2021 for Upper Manyame, Macheke and Nyagui respectively, with bare land increasing by 11.7, 27.6 and 22% for Upper Manyame, Macheke and Nyagui. Increase in built up and agriculture was followed by 44.3, 57.6 and 39.7% decrease in wetland area for Upper Manyame, Macheke and Nyagui catchments. Complementary to landcover changes analysis, class level landscape metrics results shows that there has been massive fragmentation of wetlands, shown by increase in the number of patches, which increased from 7,899 in 1984 to 46,873 in 2021 (492%), from 17853 in 1984 to 95,513 in 2021 (535%) and from 16,388 in 1984 to 106,194 (648%), for Upper Manyame, Nyagui and Macheke. Furthermore, results streamflow simulation using TOPMODEL indicated that the model was able to successfully simulate streamflows in the catchments of Upper Manyame, Macheke and Nyagui, shown by NSEs of 0.78, 0.86 and 0.64 and R² of 0.74, 0.83, 0.79. The ability of the model to achieve the values above 50% have shown that the model is very useful in water resources management, as it can be used beyond the period under study. Moreover, model validation results have also shown a fair model performance, as NSEs of 0.69, 0.78 and 0.63, and R² values of 0.65, 0.75 and 0.63 being obtained for Upper Manyame, Macheke and Nyagui catchments, respectively.

To get an insight on hydrologic impacts of wetland fragmentation, flow duration curve analysis was done on streamflows. Flow duration curve analysis have shown that wetland fragmentation has resulted in increased peak flows, with probability of exceedance for 100 m³/s increasing from 1% in 1984 to 3% in 2021, in for Upper Manyame, 1% in 1984 to 4% in 2021 for Macheke and 0% in 1984 to 2% in 2021 for Nyagui catchment, while low flows reduced, with flow (3 m³/s) probability of exceedance declining from 96% in 1984 to 71% in 2021, in Macheke, 95% in 1984 to 73% in 2021 for Upper Manyame and 85% in 1984 to 64% in 2021 for Nyagui. A substantial increase in wetland fragmentation resulted in wetland vegetation loss, which in turn result in more runoff as flow resistance by vegetation is diminished, and hence the degree of flood attenuation by the wetlands reduced, causing increase in peak flows. Moreover, increased wetland fragmentation results in soils becoming hydrophobic, as reduced infiltration and storage capacity, potentially contributing more runoff. In addition, reduced infiltration and increased runoff due to reduced permeability of soils results in low recharge and therefore low baseflow. Furthermore, results from the slopes for the flow duration curve for the three catchments have shown that the slope have become steeper for all the catchments, with that of Upper Manyame increasing by 25.32%, that of Macheke by 17.22% and that of Nyagui increasing by 12.9%, showing that the catchments are becoming flashy, in relation to runoff response. For all the three catchments, wetland fragmentation was followed by diminishing hydrologic wetland functions of low flow maintenance and peak flow attenuation, as indicated by increase in peak flows and decline in low flows. Results from this study are relevant for both Zimbabwean catchments and African ones are experiencing high rates of wetland fragmentation. Future work on wetland modelling should also focus on seasonal variation of wetland fragmentation as wetlands vary depending on water availability, as influenced by changes in seasons.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.