Use of remotely sensed derived metrics to assess wetland vegetation responses to climate variability induced drought at the Soetendalsvlei wetland system in the Western Cape province of South Africa


 Wetland areas are the most vital ecosystems and they provide important functions towards stabilizing the environment. Hydrological processes in these wetland systems directly affects the productivity of plants. Therefore, assessing vegetation response to climate variability induced drought is vital in wetlands. In this paper, the subtle changes in vegetation distribution were used as a proxy to examine and quantify the extent of drought impacts on wetland ecosystems within the Heuningnes catchment, South Africa. First, vegetation health information was extracted by calculating the normalized difference vegetation index (NDVI) during the wet and dry seasons for the period between 2014 and 2018. The derived NDVI results were further statistical linked to the corresponding rainfall and evapotranspiration (ET) observed during the study period. An analysis of NDVI results revealed that gradual vegetation health change occurred across the study area. The highest derived NDVI (0.5) for wetland vegetation was observed during the year 2014 but progressively declined over the years. Change in vegetation health indicated a significant (α = 0.05) and positive correlation to the amount of rainfall received over the same period. The results of this study showed that healthy vegetation deteriorated between the study periods due to the 2015–2017 Western Cape drought.


Introduction
Wetlands are amongst the Earth's most productive ecosystems. Although they merely occupy 6.2 to 7.6% of land surface, wetlands are a valuable natural resource of considerable scienti c value because they are associated with high biological diversity (Ndirima 2007;Sghair and Goma 2013;Kuria et al. 2014). Wetlands within the Heuningnes catchment are important as natural ecosystem remnants facilitating nutrient cycling, cleaning and the puri cation process of water, as well as provide scenic attractions for tourists and wildlife habitats (Melendez-Pastor et al. 2010;Chen et al. 2014). Long-term threats to these wetlands include agricultural development, droughts, urban development, climate change and variability as well as other impacts associated with it, such as alien invasion species (Orimoloye et al. 2019;Rebelo et al. 2019). Wetlands are vulnerable and particularly sensitive to fluctuations in the quantity of water supply. In this respect, changes in precipitation due to climate change also pose great challenges to wetland conservation (Erwin 2009).
Inadequate rainfall can induce significant declines in overall plant productivity and even lead to high rates of plant mortality (Touchette et al. 2007;Yu et al.2019). Plants are excellent indicators of wetland condition for many reasons including their relatively high levels of species richness, rapid growth rates, and direct response to environmental change (Cronk and Fennessy 2009;Chatanga and Sieben 2019).
Many alterations to the environment that act to degrade wetland ecosystems cause shifts in plant community composition that can be quanti ed easily (Ehrenfeld 2000). Insu cient water supply may lead to the depletion of soil moisture (Bordi and Sutera 2007), which will further have adverse effects on the growth and health of plants. Increases in temperature also affect wetland systems by accelerating the rate of evaporation and transpiration (Abtew and Melesse 2013). Therefore, the ability to map and assess wetland vegetation productivity in detail, especially in response to climate change, will always be an objective in the management of wetland ecosystems.
Monitoring the response of vegetation to drought is important for the sustainable conservation of wetland ecosystems as it is related to the condition of water supply. However, continuous observation and investigation based on physical methods, remains restricted to small geographic coverage, for a speci c period of time and it focuses mainly on individual species (Hooper et al. 2005;Guo et al. 2017).
In addition, research done physical can be resource intensive and problematic when the study area is remote and hazardous (Daryadel and Talaei 2014). Similarly, developing models for monitoring wetland vegetation at individual levels remains impractical, especially in the light of the global effects of climate change (Xie 2008). Drought indices such as Palmer Drought Severity Index (PDSI) and Standardized Precipitation Index (SPI) become unreliable because their dependence on accuracy of ground observed meteorological inputs that provide sparsely (Zhao et al. 2017) and possess poor spatial resolution at regional scale, especially, in areas` where a few of ground observations are available.
Recent advancements in satellite remote sensing, as powerful means of Earth's surface assessment, have provide efficient, reliable, and affordable monitoring tools for identifying, describing, and mapping the distribution of wetland vegetation with various spatial, temporal, and spectral resolutions at wide scales from local to global (Jones et al. 2009;Kaplan et al. 2019). In particular, the normalized difference vegetation index (NDVI), precipitation and evapotranspiration (ET) products may provide valuable information to understand the wetland ecosystems response to drought because meteorological data obtained from ground observation stations often have poor spatial resolutions (Wan et al. 2004). The NDVI picks up the frequency that the plant leaf releases in order to measure its vigour of the plant's health (Xue and Su 2017;Onyia et al. 2018). Sensors typically capture some combination of visible and nearinfrared light using narrow lters to increase the sensitivity and speci city of the measurements (Lapray et al. 2014). When a plant becomes dehydrated or stressed, the spongy layer of the plant collapses and its leaves re ect less NIR light, yet they still re ect the same amount of light in the visible range (Jacquemoud and Ustin 2019). Thus, vegetation health is one of the most crucial factors to look at when studying the response of wetland ecosystems to a drought.
Investigating the relationship between NDVI and ET or precipitation can infer water stress from different plants. This because sufficient water promotes e cient transpiration and cool plant, while water de ciency promotes closing plant stomata and intense transpiration rate, thus, lower ET represents the stronger evaporative cooling for pixels with the same NDVI (Petropoulos et al. 2009;Yu et al. 2019). As an approach towards assessing wetland vegetation response to climate variability induced drought at the Soetendalsvlei in the Heuningnes Catchment, South Africa, this study mapped and assessed changes in vegetation health and distribution between the years 2014 to 2018, and also examined the relationship between wetland vegetation productivity and rainfall variability.

Description of the study area
The study focused on the Soetendalsvlei wetland system found in the Heuningnes Catchment, which occurs in the southernmost region of South Africa (Figure 1). The catchment covers an area of about 1 401 km 2 (Hoekstra and Waller 2014) and lies within the Mediterranean climatic zone. The area receives most of its rainfall during the winter season (mid-May to late August). The temperatures in the area vary signi cantly throughout the year, with an average range of 10ºC in winter and 28ºC in summer and a mean annual rainfall of 500mm (Hanekom et al. 2009;Herdien et al. 2010). The study site, is a natural fresh water lake which is about 8 km long and a width of up to 3 km, it occurs along the Nuwejaars River, between Elim and Soetendalsvlei. It is one of the major lakes in the catchment (~20 km 2 ) and South Africa's second largest freshwater lake after Lake Chrissie (Hoekstra & Waller 2014).
The area is considered a biodiversity hotspot because of the unique animals, ora and landscapes found in the region. It is a home to a highly threatened lowland fynbos type of vegetation and a prominent area for twitches (Gordon et al. 2012). The indigenous fauna and ora of the region form the basis of the shing and tourism sectors of the economy (Gordon et al. 2011). Marine resources such as line sh, rock lobster, and abalone as well as the bait species contribute a huge amount to the Western Cape economy, with the industry worth over R1.3 billion per year (Turpie et al. 2003). Both the lm industry and tourism are dependent on natural resources with an estimated 24% of foreign visitors to the region being attracted by its scenic beauty. Direct revenue is also generated from the fynbos through harvesting and cultivation of indigenous rooibos tea, wild owers like proteas, buchu for its aromatic oils, reeds for thatching, and various traditional and commercially marketed medicinal plants (Braschler et al. 2010).
In this study, time series of Landsat images were used to acquire more information about the extent and distribution of vegetation in the site. Landsat 8 (L8) Operational Land Imagery (OLI) Level 1 data acquired for the period of January 2014 to December 2018 were used, freely available from https://earthexplorer.usgs.gov/. The data are available every 16 days with a spatial resolution of 30m, different bands of the sensor and its speci cations are available in Table 1. Cloud-free images and images with less than 10% cloud cover were selected. Two images representing wet and dry season for each year were obtained and details of these data are provided in Dube and Mutanga (2014). Band 4 (Red) and 5 (NIR) were used for the estimation of NDVI for the wet and dry season of each selected year (Tucker 1979). The L8 images were atmospherically corrected using FLAASH atmospheric correction method. The selection of drought monitoring period was informed by the documented literature and information on the onset of drought (Botai et al. 2017;Leslie and Richman 2018;Otto et al. 2018).
Evapotranspiration (ET), and Precipitation data were acquired from https://wapor.apps.fao.org/catalog/1. The ET data was delivered on a dekad (10-days basis) and is mainly the sum of soil evaporation, canopy transpiration, and evaporation from rainfall intercepted by leaves. The value of each pixel represents the average daily ET in a given dekad (Sazib et al. 2018).

Precipitation dataset was obtained from CHIRPS (Climate Hazards Group InfraRed Precipitation with
Station), a quasi-global rainfall dataset, starting from 1981 up to near present. For CHIRPS, the value of each pixel represents the average of daily precipitation in the dekad expressed in mm (Funk et al. 2015). NDVI thresholds were de ned and set for each class and these thresholds were somehow informed by literature (Wilson and Norman 2018;Wang et al. 2018). In the study, thus thresholds were set as following non-vegetated (NDVI range between -0.21 and 0.19), vegetated (NDVI ≥ 0.2) and water (NDVI≤-02). We then conducted accuracy assessment for the derived classes by computing the user, producer and the overall accuracies, validation was done using ground control points, and Google Earth digitized sample points. Further, the derived results were compared to climate data for the areas to determine trends and relationships between derived vegetation metric and climate data. Speci cally, correlation analysis was used assess the response of wetland vegetation to drought by evaluating the relationship between NDVI results and rainfall variability. The Pearson product-moment correlation coe cient, better known as the as r was performed to derive the statistical analysis results. The coe cient was calculated for the 12 months data for each year from May to April. The correlation coe cient was computed as: Where Y is the precipitation or ET and NDVI is the normalized difference vegetation index and average monthly total precipitation or ET for the years 2014, 2015, 2016, 2017 and 2018 adopted in this study.
Possible values of r range from -1 to +1, with values close to 0 signifying little relationship between the two variables. When r is above 0.5, there is a positive relationship between two variables but there is no signi cant association. Value ranges from 0.8 to 1 represent a positive signi cant relationship between the two variables. A detailed description of the methodology is summarized in gure 2.

Remotely sensed mapping of wetland vegetation
The results of the study demonstrated that wetland vegetation was greatly affected by drought between the year 2014 and 2018 (Figure 3 and 4). Derived classi cation results showed that wetland vegetation can be mapped with very high accuracies.
High classi cation accuracies in terms of producer, user and overall accuracies were observed (Table 2). For all the remotely sensed derived wetland mapping results, all the accuracy assessment methods were ± 80%, demonstrating a commendable classi cation model performance.

Relationships between derived NDVI and climate data
The results indicated that wetland vegetation productivity was largely controlled by rainfall availability and evapotranspiration rates. The results from table 3 showed high correlations between wetland vegetation derived NDVI and rainfall as well as evapotranspiration. For example, for all the years NDVI and rainfall correlations coe cients were high and positive, on average above 0.80 whereas for NDVI and evapotranspiration the relationships were signi cantly but above -0.50. Figure 6 further details the observed monthly NDVI, precipitation and evapotranspiration trends for the entire study period. It can be observed that evapotranspiration and precipitation controlled or had a bearing on NDVI or wetland vegetation productivity.  Wetlands comprise of notable attributes of species diversity, richness, abundance and succession, and they are therefore considered to be the most dominant and important ecosystems, globally (Mitsch et al. 2015). This study examined changes in wetland cover to determine the ecosystem's response to drought by using remote sensing techniques. Work done in this study has relevance to the maintenance of ecological processes and quanti cation of natural disasters impacts because it explores: 1) spatial, temporal and seasonal variations of wetland cover; 2) seasonal variability of wetland vegetation health; 3) the link between wetland vegetation growth dynamics and rainfall variability to assess the response of wetland ecosystems to drought.
An analysis of classi ed maps revealed that gradual ecosystem change occurred across the study area.
Other studies such as that by Middleton and Kleinebecker (2012) done to assess the effects of climate change induced drought on freshwater wetlands, and that of Belle et al. (2018) in the eastern Free State, South Africa, con rm that vital wetland productivity processes that sustain biodiversity in the ecosystem may be critically affected by the occurrence of a drought. Climate change induced drought, especially in arid regions, drives change in hydrology and vegetation health, thus affecting ecological processes within the wetland ecosystem.
This study suggests that the decline in vegetation extent and water, and increase non-vegetated area in the wetland was a result of rainfall variability. Furthermore, climate change is predicted to increase drought, the number of high heat days, and the frequency of severe storms, all of which affected wetland ecosystems. Results for wetland transition shown in this study are comparable to Ridol et al. (2006), who observed that wetland ecosystems are vulnerable to disturbances such as a severe drought and may respond to biomass losses with highly irreversible catastrophic shifts to unvegetated conditions. Similarly, Nhamo et al. (2017), using the Landsat satellite data to delineate wetland extent and assess seasonal variations in South Africa during the period of 2000 to 2015, found a continues decline in wetland area and the minimum value was observed in 2015 which coincided with an El Nino associated drought in the study area (Rembold et al. 2016;FAO 2016).

Impact of meteorological data trends on wetland vegetation productivity
Based on long-term (5 years) data, this study examined the in uence of rainfall variability on the productivity of wetland vegetation in the Soetendalsvlei wetland system. The relationship between wetland vegetation health and quantity as well as the temporal patterns of rainfall variability were assessed and yielded two key results. Firstly, over the past 5 years, NDVI (Vegetation health) signi cantly and positively correlated with precipitation; and secondly, the NDVI and ET showed an opposite trend, ET exceeds the amount of precipitation during the period of this study.
The results of this study highlight the importance of rainfall variability on wetland vegetation productivity.
One explanation is that rain events provide su cient soil moisture and maintain high water availability (Merolla 2012). In arid and semiarid ecosystems, water is typically a limiting factor for plant health, and available moisture generally increases plant biomass (Twisa and Buchroithner 2019). Photosynthesis of plants depends on water availability, therefore, insu cient water availability can minimize the assimilation of carbon, thereby decreasing wetland vegetation productivity (Pinheiro and Chaves 2011).
For instance, the results of the study by Barros and Albernaza (2014) found that an elevation in water availability leads to a reduction in wetland vegetation growth rates or the reproductive success of many species. Wetland vegetation have highly developed root systems that holds the soil in place and lter pollutants, naturally improving water quality (Finlayson et al. 2015). Therefore, a drought will likely cause the loss of, or reduction in wetlands and will challenge the adaptability, composition and distribution of wetland plants. Moreover, if wetland vegetation productivity is challenged, pollutants could become more concentrated in wetlands and this will affect water quality.
Remote sensing spatial and seasonal variations of wetland vegetation Similar to other arid lands, vegetation, precipitation and ET in the study area is both spatial and temporally heterogeneous, making ground-based measurements invaluable. However, the study area is remote and the use of in-situ methods can be resource intensive and problematic when the study area is remote and hazardous (Adam et al. 2010). Remote sensing therefore provides invaluable means of monitoring vegetation to assess environmental conditions in wetland ecosystems (Amler et al. 2015).
The tool has been popular for collecting meteorological data, and offers spatially explicit data, as well as repeated observations and covers large geographic locations (Boisvenue and White. 2019).
Remote sensing images are key data sources for earth monitoring programs considering the great advantages that they have (Makapela et al. 2015). For instance, it is more easily obtainable to produce and update vegetation inventories over large regions if aided by satellite imagery and appropriate imagery analysis. A growing number of studies have examined the response of wetland vegetation productivity to drought by using remote sensed data (Santos et al. 2019;Easterday et al. 2019;Adamu et al. 2018;Wilson and Norman 2018;Nhamo et al. 2017 Vegetation index NDVI was used to detect any signi cant differences in vegetation cover between the years 2014 to 2018. The results indicated that NDVI was able to discriminate wetland vegetation from other classes within the study area. Results required for drought impacts assessment showed the change in landcover distribution and vegetation productivity between 2014, 2015, 2016, 2017 and 2018. These results clearly revealed that the wetland was negatively affected by the long-term drought. Temporal remotely sensed data enabled the assessment of wetland vegetation health condition as far as back as 2014, therefore remote sensing provided an effective tool in analysing and determining vegetation changes in wetlands under different management regimes. Frequent wetland monitoring is important for timely intervention in the case of an identi ed negative change. Remote sensing has shown its strength in wetland mapping and for monitoring wetland dynamics over time and is thus an important tool for wetland management.

Conclusion
Temporal and spatial distribution of wetland cover classes and vegetation cover was assessed using NDVI to examine the impact of rainfall variability (drought) on wetland vegetation. Results showed that a signi cant variation in the wetland surface area from 2014 to 2018. Speci cally vegetation and water decreased signi cantly over the monitoring period, while the extent of bare surface increased rapidly.
Wetland extent mapping was achieved with an average overall accuracies (85-90%) in this study. Further, Vegetation productivity signi cantly and positively correlated with precipitation over the past ve years, while ET showed a negative signi cant relationship, ET exceeds the amount of precipitation during the period of this study. From the observation of the whole study period, healthy vegetation has deteriorated due to drought that occurred in the study area between the monitoring periods. The amount of rainfall entering into an ecosystem is typically a limiting factor for plant health, the results of this study highlight the importance of rainfall variability on wetland vegetation productivity.

Declarations Competing Interests
There are no relevant nancial or non-nancial competing interests to report.

Funding Info
We thank the South African National Space Agency (SANSA) for funding this work.

Author contribution
Ndlala NC conceived of the presented idea, developed the theory and performed the computations. Prof Dube T veri ed the analytical methods and supervised the ndings of this work.

Data Availability
Data were derived from the following resources available in the public domain: https://earthexplorer.usgs.gov/ and https://wapor.apps.fao.org/catalog/1 Detailed statistics on the areal extents and observed changes in wetland vegetation between the wet and dry season for the entire monitoring period Figure 5