As most ecohydrological processes have been found to have a direct influence on the quality and quantity of ESs, a number of models designed to perform such assessments have been developed. Some ecohydrological models have, for instance, developed probabilistic model frameworks to predict the impact of climate and soil type variations on conditions of water stress of vegetation (Porporato et al. 2002; Rodriguez-Iturbe et al. 2001). Other hydrological models have also tried to predict the hydrological consequences on ESs. The vegetation dynamic model has been used to predict the hydrological consequences on ES variables. Such studies have predicted that ES variables are significantly influenced by what Han et al. (2015) have referred to as ‘environment variables’ such as soil water content and groundwater depth.

Despite the bulk of such assessments, very few studies have attempted to explicitly incorporate the element of risk or threat to ESs. Understanding the risk of extinction of certain ESs or collapse of ecosystem attributes will certainly require approaches that go beyond traditional assessment practices that seek to quantify only the amount of ESs available in a particular landscape (Maron et al. 2017). According to Maron et al. (2017), assessment efforts need to be directed at developing a standard set of criteria for pinpointing when and to what degree adequate provision of an ES in a given landscape is at risk and shedding light on how immediate the risk of complete loss of the service is. Absence of such a standardised set of criteria has seen us lacking a consistent basis for prioritising management interventions in overcoming threats to ESs or at least supporting their recovery.

The need to categorise threats to ESs has been underscored in the preceding literature. How to categorise is a daunting task as we currently lack an assessment framework that will help pinpoint when and to what degree adequate provision of ESs in a given environment is at risk, or how immediate the risk of complete loss of the service is (Maron et al. 2017). A plausible entry point to developing one is to define what constitutes ‘threat’. Within the context of ESs, threat should be conceptualised in two main ways. Firstly, we make reference to what we refer to as a ‘loss perspective’ and define threat as loss of service provision to a group of beneficiaries situated in a defined geographical location that is characterised by certain loss bearing ecohydrological characteristics. Such a perspective allows us to go beyond the traditional conceptualisation of threat that defines threat as a global loss of an ES. A second perspective that has been characterised in other literatures (see Maron et al. 2017) as a ‘supply versus demand’ perspective requires us to define threat as a failure of the existing supply mechanisms to meet the demand. In other words, ESs beneficiaries suffer in cases where the existing ecohydrological conditions are not sufficient enough to meet existing demands. Assessing threats to ESs therefore requires a framework that incorporates a human dimension – particularly by looking at the consequences of ecohydrological conditions characterising a particular landscape to benefits that accrue to humans.

We argue that net gains to humans are maximised when the supply of natural capital is able to meet the demand for that service by people. Maximising net gains would also mean that the resulting ecohydrological processes are able to ensure a steady supply of ESs for human consumption. Where such conditions are not permitting and where supply conditions of an ecosystem are not guaranteeing that demand of ESs will be met, risks to the well-being of people occur. Taking the cue from the risk register approach proposed by Mace (2015) and the Red List proposed by IUCN as well as the threat assessment approach developed by Maron et al. (2017), we developed a threat assessment framework (Figure 1).

Drylands are a critical terrestrial system of the Earth that is characterised by low water availability (Pravalie 2016). The quality and quantity of ESs available to communities occupying dryland or semi-arid landscapes are shaped by the hydrology of such landscapes. Like in many other ecosystems, semi-arid and arid landscapes are water controlled in that water availability is the single most important driver of the structure and organisation of (water related) ESs (Lehmann et al. 2014; D’Onofrio et al. 2015). The state of ecohydrological systems of arid and semi-arid landscapes has often been understood within the context of environmental variables such as conditions of water stress for vegetation (Porporato et al. 2002; Rodriguez-Iturbe et al. 2001), soil water content and groundwater depth (Shi et al. 2012; Zhao et al. 2005).

Certainly understanding dryland ecosystem responses to temporal and spatial changes in ecohydrological conditions is critical, as these ecosystems cover nearly 40% of the global land surface (Buntinga, Munsona & Villarrealba 2017). Such ecosystems also play an important role in supporting development for large human populations and providing ESs (Buntinga et al. 2017). A number of critical ecohydrology parameters and the important connections that exist between them and the subsequent effect they have on the availability of ESs have been extensively reviewed.

Commonly discussed ESs impacting parameters include, among others:

The hypothesised relationship between such variables and ESs provisioning is presented in the methodology section.

Ecosystems in semi-arid landscapes provide production, regulation, support and cultural entertainment service functions, and these functions in turn create a series of natural environmental conditions and socio-economic benefits for human survival and development (De Groot et al. 2002). Measuring such benefits, particularly the indirect ones, is a big challenge. In this study, we focused on benefits associated with ESs provision. We used focal ESs such as provision of food, raw materials and medicinal resources – similar to those used in the study by Ramirez-Gomez et al. (2015). We identified nine provisioning services that were deemed important by the local people through a series of participatory group discussions. The provisioning services that were considered in this analysis and that are also consistent with what Colophospermum mopane landscape can offer to rural households are supply of timber (for construction of houses and livestock kraals), thatch (grass for roofs), resins (tree exudate used as glue or sealant), wild fruits, bush meat (animals hunted for meat) – including mopane worms, fish (caught for commercialisation), natural medicines, materials for making crafts and traditional tools (fibres, stems and leaves) and ornamental resources (fibres, trees and tree bark used for making clothes for traditional dances and celebrations).

In developing the ES constructs, we employed the concept of a service provisioning area (SPA) referring to the source of ESs (Syrbe & Walz 2012). We only recorded the extent of household involvement in the collection of nine provisioning ESs within the confines of their village. Resource extraction behaviours of households were used as a proxy measure of ES values – largely depicting the amount of ESs derived from a given ecohydrological landscape using the following formula: ESVij=eij[Eqn 1] where ESV_ij is the value attached to jth ES obtained by the household from ecosystem i. e_ij is the amount of the jth ES obtained by the household from ecosystem i. The ESV_ij value was computed for the wet resource state (WRS) and dry resource state (DRS) as explained below.

Although our study focused more on spatial changes, we could not overlook temporal variation. We attempted to incorporate the temporal dimension, by employing a proxy measure that we found to be ideal in depicting the influence of variations in precipitation on availability of ESs. We borrowed insights from ‘pulsing hydrology’. Pulsing hydrology informs us that the spatial and temporal dynamics of an ecosystem are often influenced by resource availability and timing (Muvengwi et al. 2016). As argued by Parsons and Thoms (2013), there is a need to examine ESs associated with vegetation trends in flood, rain and dry states. Related studies have shown that vegetation and associated ESs increase as one moves from a dry to a flood state (Thapa, Thoms & Parsons 2016). To capture such temporal shifts, Thapa et al. (2016) distinguished between two important states in their research on semi-arid topic. These include the DRS (essentially the dry state) and the WRS (essentially the rain and flood state). The researchers therefore expected the level of resource extraction by communities to be relatively high in the WRS than in the DRS.

A number of ecohydrology parameters that would pose a threat to the supply of ESs to rural households were identified based on a critical review of the literature. The variable selection process was based not only on relationships that were clearly apparent on a variety of data sets but also on an ecological vulnerability assessment indicator system that we constructed. Following examples from Zang et al. (2017), we constructed an ecological vulnerability assessment indicator system using the ‘exposure–climate sensitivity–adaptive capacity’ framework according to the theory of ecological vulnerability (Beroya-Eitner 2016 in Zang et al. 2017). Our threat assessment approach was not misplaced, as empirical evidence from elsewhere has shown that ecological vulnerability is often assessed by combining the characteristics of study subjects and the objectives of the study based on the ‘exposure–sensitivity–adaptability’ framework (Zang et al. 2017). The final selection of ecohydrological variables was however severely constrained by data scarcity. We did not view this as a methodological deficiency as some studies have acknowledged that dryland regions are poorly gauged and therefore lack a detailed understanding of their ecohydrology (Jarihani et al. 2015). Variables that were selected for final analysis, their measurement metrics and hypothesised relationships are summarised in Table 1.

Prior to resolving the indicators, raw data were processed for ‘homogenisation’ and ‘non-dimensionality’, which is a standard requirement as there can then be questions of examining the homogeneity across the sites of the distribution of the scaled values (Hall 2003). Study constructs were first tested for normality. Ecohydrological indicator variables were further explored for reliability and validity through Exploratory Factor Analysis (EFA) using the Principal Component Analysis (PCA) method with varimax rotation. Important measures of reliability and validity were computed, including Cronbach’s alpha, composite reliability and the Average Variance Extracted (AVE). Both statistical measures sought to estimate internal consistency associated with the scores derived from the data scales (Hair et al. 2006). To evaluate discriminant validity, the AVE was calculated (Voorhees et al. 2015). The AVE was calculated as follows: AVE=∑λi2∑λi2+∑ivar(εi)[Eqn 2] where λ is the factor loading of item i and var(ε) is the variance of the error of item.

Within the EFA framework, the Kaiser–Meyer–Olkin (KMO) statistics and Bartlett’s test of sphericity were performed to examine the data suitability for PCA.

Most ecohydrological studies that seek to draw patterns from large data sets usually employ some form of clustering (Canedo-Arguelles et al. 2016; Chang, Han & Zhong 2009). Cluster analysis (CA) is a group of multivariate techniques whose primary purpose is to assemble objects based on their common characteristics (Kim et al. 2007). Central to all CA studies, hierarchical agglomerative clustering is the most common approach that is used to generate intuitive similarity relationships between any given data sets.

Hierarchical agglomerative CA was performed on the normalised data set. Conceptually, the analysis denotes ecohydrological threat outcomes associated with variable i in ecosystem and landscape j as Y^ij. This outcome is represented in Equation 1 as a function of the individual ecohydrological characteristics, Xqij, and a model error, rij (Bryk & Raudenbush 1992). Yij=β0j+β1jX1ij+β2jX2ij+…+βnjXnij+rij[Eqn 3] where r_ij~N(0,σ²).

Cluster analysis reduces the large number of ESs users into a small number of homogeneous groups, classified according to common ecohydrological attributes that depicted different resource thresholds in landscapes where resources are obtained. The various resource thresholds created were then used to draw important connections between associated ecohydrological conditions and the risk of loss of valuable livelihood source of the resource users.

The use of regression to estimate the influence of ecohydrological parameters is not new (Canedo-Arguelles et al. 2016; Han et al. 2015). We performed multiple regression analysis to determine which of the cluster-defining ecohydrological attributes were significantly shaping the resultant resource thresholds and their associated threats to the availability of ESs in both the WRS and the DRS. We used ES values as the dependent variable. Because of inherent normality problems associated with our dependent variables, we used the Box-Cox Transformation formula to stabilise the variance, make data more normal and subsequently improve the validity of measures of association depicted by the regression model.

To predict the model best fit, the analysis used the R² measure and performed an ANOVA test. Collinearity was diagnosed using the tolerance and the Variance Inflation Factor (VIF) as guided by O’Brien (2007). We used linear residual plots to detect model inadequacies in regression diagnosis. Specifically, we used such plots to assess nonlinearity and heteroscedasticity in regression diagnostics. Following the cue from Weisberg (1985) and Montgomery and Peck (1992), we assumed that a null linear residual plot shows that there are no obvious defects in the model and that a curved plot indicates nonlinearity. We also concluded that a fan-shaped or double-bow pattern would indicate non-constant variance.

In carrying out the study as well as in disseminating the research findings, the authors declare that all ethical issues in research have been addressed and that there has been no conflict of interest.