1. While the ecological impact of environmental change drivers, such as alien plant invasions, is relatively well-described, quantitative social-ecological studies detailing how these changes impact multiple ecosystem services, and subsequently stakeholders, are lacking.

2. We used a social-ecological approach to assess how Prosopis juliflora (Prosopis henceforth)¸ an invasive tree, affects the provision of multiple ecosystem services to different stakeholder groups in a degraded East African dryland. We combined plot-based ecological data on the impacts of the tree on indicators of ecosystem service supply with questionnaire survey data describing ecosystem service priorities from eight different stakeholder groups. This data was then used to quantify how tree invasion impacted individual ecosystem services, and the overall supply of services, relative to the priorities of each stakeholder group, using an ecosystem service multifunctionality metric.

3. In the study area, we found that Prosopis significantly increased the supply of shade, wood production and honey production, but reduced the supply of water availability, tourism potential and biodiversity protection.

4. Priorities for specific services differed between stakeholder groups. Although most groups assigned a high priority to provisioning services, such as water and crop production, it was either provisioning or cultural services which were a primary source of income, that were deemed most important.

5. Combining supply and priority data, showed that most stakeholder groups saw a net decrease in ecosystem service multifunctionality with increasing Prosopis invasion, or no significant change overall. Increasing Prosopis cover increased multifunctionality for only two stakeholder groups, charcoal producers and NGO’s involved in regional development.

6. Synthesis & applications: Our research highlights the need to account for stakeholder priorities in studies of how global change impacts ecosystem multifunctionality. We found that there are conflicting patterns of ecosystem service priority between different stakeholder groups, resulting in large variation in how different groups were impacted by the invasive tree. Our approach also highlights possible synergies and conflicts when managing this tree invasion. More broadly, we recommend that future studies of ecosystem multifunctionality explicitly consider differing stakeholder priorities, as these strongly influence the perceived impact of environmental and land management change.

Ecological data

Ecological data was used to estimate levels of ecosystem service supply and the extent of Prosopis invasion. Data on the ecological impact of Prosopis on ecosystem properties were collected along a Prosopis cover gradient ranging from 0-80% cover in 15 x 15 m plots. A total of 67 plots were sampled in the rainy seasons of 2016 and 2017. 5-8 plots were selected in each of 10 sub-locations, the smallest administrative unit in Kenya. Within each sub-location, plots were selected along a Prosopis cover gradient and plots within one sub-location all had a similar land use history, though land use history could differ between sub-locations; see Linders et al. (2019) for more details.

Total Prosopis woody biomass was used as an indicator for wood production and was measured by extrapolating basal diameter of each individual Prosopis stem; for details see Linders et al. (2020). Plant species richness was used as an indicator for biodiversity conservation, as it is known to correlate with the diversity of many other taxa (e.g. Manning et al., 2015; Sauberer et al., 2004) and measures of larger and more mobile taxa are more difficult to obtain reliably at the plot scale. For livestock production we used a combination of herbaceous fodder and Prosopis pods as supply indicators to generate a measure of total fodder production (g/ plot). There was a linear relationship between basal diameter and number of pods (number of pods = 5.781*basal diameter (mm) – 64.197; based on 28 trees: r²=0.51; P<0.0001). Prosopis pods weigh on average three grams (Sharma, Burman, Tewari, Bohra, & Harsh, 1994) and so we summed estimated Prosopis pod weight for each plot. Prosopis pods are only a healthy component of a livestock diet when they make up 20% or less of it (Mahgoub et al., 2004). Therefore, Prosopis pods only contributed to total fodder production when they were ≤ 20% total fodder weight. For details on the sampling of herbaceous biomass and plant species richness see Linders et al. (2019). Two indicators were used for honey production: abundance of Prosopis inflorescences and abundance of flowers in the herbaceous layer. Approximately, 1.48% of Prosopis flowers become pods (Oliveira & Pires, 1988) and Prosopis has, on average, 344 flowers per inflorescence (Zaitoun, Al-ghzawi, Samarah, & Mullen, 2009). Based on pod number we could thus estimate the total number of Prosopis inflorescences. The abundance of each herbaceous flowering species was also estimated from ground cover estimates, see (Linders et al., 2019) for details. Each species, including Prosopis, was then given a honeybee attractiveness score between zero and three, with three the highest, based on flower size and palatability (when known). For each species, attractiveness scores were multiplied with abundance to get an indication of flower availability. This method is likely to overestimate total herbaceous flower abundance as the abundance data included seedlings and non-flowering plants, which are often eaten before flowering. Therefore, the Prosopis flowers were given double weight. Tree cover was used as an indicator for shade by estimating visually both Prosopis and native woody plant cover (typically less than 5% of total cover) from 0-100%, which was done by two people independently. Most tourism activities in the region involve wildlife watching. Therefore, tourism potential was estimated by assessing habitat suitability for the four large charismatic mammal and bird species that could potentially occur in the plot and be seen by tourists: Ostrich (Struthio camelus), Günther’s Dik-dik (Madoqua guentherii), Plains Zebra (Equus quagga) and Vervet Monkey (Chlorocebus pygerthrus). For each plot it was assessed whether these species could occur here based on their habitat requirements, based on scientific literature (Cooper et al., 2009; Mittermeier, Rylands, & Wilson, 2013; Wilson & Mittermeier, 2011) and personal observations of the authors. For Ostrich, all plots with <50% tree cover were deemed suitable. Günther’s Dik-dik are browsers, with a habitat preference for woody cover but as Prosopis leaves are unpalatable we only deemed plots with >40% native tree cover as suitable. Plains Zebra are grazers, only occurring in open areas, <50% tree cover and at least medium grass cover, >25%. Vervet Monkeys need at least some trees, defined as >30%, but were never observed in Prosopis monocultures, so plots were only deemed suitable if there was at least some native tree cover. Each plot was given a score from zero to four depending on how many of these species could occur there. As we only scored services that the plots currently provide (i.e. current rather than potential supply), we did not measure how Prosopis affects crop farming. Although the plots could be converted to crop farming, this would mean the plots cannot supply the services listed above. We therefore gave each plot a value of zero for crop farming.

Water consumption of Prosopis was based on measurements taken in the Ethiopian Rift Valley, a region with similar climate (9.160 to 9.210N and 40.080 to 40.120E at 740 m.a.s.l.) as part of a wider project. The amount of water used, determined by sap flow volume of individual Prosopis trees was determined using the heat ratio method of the heat pulse velocity technique (Burgess et al., 2001; Dzikiti et al., 2017). In total, four sap flow stations were established, each with three trees of varying diameter. The individual tree sap flow volume in litres per hour were converted to stand level transpiration (in mm per hour) using the approach described by Dzikiti et al. (2017). To study the dynamics of total actual evapotranspiration from Prosopis stands, and to triangulate the estimate obtained from the sap flow set up, an open path eddy covariance system was installed in one of the sites (Shiferaw et al., 2021).

Stakeholder data

Data on the ecosystem service priorities of different stakeholders were gathered via individual face-to-face interviews, conducted in August 2019. To ensure that our research was conducted ethically we took the following steps: 1) All interviewees were informed about the purposes of the study and gave their informed consent orally before the start of each interview, as a proportion of the interviewees was illiterate, this was not done in written form. Interviewees were also given the option to halt the interview at any time. 2) Data were treated anonymously and are presented in the article in a way in which individual participants cannot be identified. 3) When uploading the data upon acceptance, individual level data will not be made available, thus ensuring anonymity of the interviewees. The Senckenberg BIK-F Institute, by whose employees the research was led, does not have an ethics committee for social science work. In total, 234 questionnaires were successfully completed, spread across eight stakeholder groups, each with their own interests in land management and income source. Relevant stakeholder groups were pre-defined based on discussions with local inhabitants and scientists familiar with the region, and represented the main land users and managers of the region. These groups were: conservationists, tourism industry, crop farmers, pastoralists, government officials, developmental NGO employees, charcoal producers and teachers, see Table 1 for affiliations and distinguishing characteristics of these groups. Stakeholder sampling was done in two different ways: by visiting villagers in their homes and by visiting institutions. Villages were selected throughout the study region, to cover all accessible areas (limited by road conditions and tribal conflicts) and included all areas where ecological sampling was done. In total, 60 villages were visited and in each village 2-10 interviews were conducted based on presence of stakeholders in their homes. Additionally, hotels, tourism guide associations, conservation organisations, government offices and NGO offices were visited throughout the study area. Within each institution 3-10 people were interviewed separately, depending on the size of the institution. In institutions the interviewees were selected by a supervisor to cover different hierarchical layers (e.g., from desk clerks to general managers) and interviewees were selected on availability. All interviewees were adults at the time of the study. Interviews were performed by a team of ten trained enumerators, who had previous experience in administering questionnaires and who were fluent in the local languages, and the questionnaire was pre-tested on roughly 15 villagers and natural resource management specialists.

Each stakeholder was asked which of the pre-defined stakeholder groups they identified as belonging to most closely. To measure the relative priority of multiple ecosystem services each stakeholder was then asked to divide twenty points, represented by dried beans, across fifteen potentially relevant ecosystem services (see Table 2). The meaning of all ecosystem services was explained by the enumerators. In addition, ecosystem services were represented by pictures depicting examples of this service when possible (see supplementary information 1). Ecosystem services were selected a priori by researchers with experience in the region, and a group of local people, based on the Common International Classification of Ecosystem Services (CICES) classification (Potschin & Haines-Young, 2016). We included only services with a direct link to final benefits (as defined in the cascade model, Potschin-Young et al., 2018), to minimise the misallocation of points between final benefits and the regulating services which underpin these. Interviewees were also given the option to name additional services which they deemed important and allocate points to them. The number of points to be distributed was limited to encourage stakeholders to prioritize, as otherwise stakeholders could assign high or low scores to all services (e.g. Washbourne et al., 2020). This method, which we term ecosystem service priority assessment (ESPA) was adapted from previous measures of ecosystem service priority (Washbourne et al., 2020) but tailored to generate values that can be used in the ecosystem service multifunctionality approach described above. This approach also allows multiple values of a single service (e.g. relational and material; Chan, Satterfield, & Goldstein, 2012) to be integrated without double counting. Discussion with stakeholders while conducting the interviews indicated that stakeholders were aware of interdependencies between services and allocated their points accordingly. E.g. those from the ecotourism sector gave points to both tourism and conservation, and farmers and pastoralists gave points to water. Nevertheless, overlap between certain services, e.g. (eco)tourism and biodiversity, and water and crop production could not be avoided altogether, meaning some points may be misallocated, and, the demand for regulating services that underpin these final benefits may be underrepresented. The priority some services that are currently at high supply may also be ‘taken for granted’ and so their priority underestimated.

The raw (anonymized data) is accompanied by readme files, which describes what is seen in each column.

Data is complete, there are no missing values.