Plant-pollinator networks along an altitudinal gradient in the Drakensberg Mountain Centre – implications for climate change Victoria Roetger (2089081) A Dissertation submitted to the Faculty of Science, University of the Witwatersrand, in fulfilment of the requirements for the degree of Master of Science, Johannesburg, South Africa, 2025 Supervisors Prof. Glynis Goodman-Cron1, Prof. Sandy-Lynn Steenhuisen2 & Prof. Dave I. Thompson3 1School of Animal, Plant and Environmental Sciences, University of the Witwatersrand, 2Department of Plant Sciences and Afromontane Research Unit, University of the Free State, Qwaqwa, 3South African Environmental Observation Network, Ndlovu Node Declaration I, Victoria Roetger, declare that this dissertation is my own, unaided work. It is being submitted for the Degree of Master of Science at the University of the Witwatersrand, Johannesburg. It has not been submitted before for any degree or examination at any other University. _______________________________________ 10 th of March 2025 Wits University Faculty of Science post-graduate student AI declaration I understand that the use of generative AI tools (such as ChatGPT or similar) without explicitly declaring such use constitutes a form of plagiarism and is classified by Wits University as academic misconduct. I declare that in the course of conducting the research towards my degree or in the preparation of this thesis/dissertation/research report (select one by marking with an X): I did not make use of generative AI tools I did make use of generative AI tools for the following (tick all that apply): 1. Idea Generation (research problem/design, hypothesis) 2. Sourcing Related Work (summarising, identifying sources) 3. Methods and Experiment Design (experiment setup, model tuning) 4. Data Analysis (presentation, coding, interpretation) 5. Theoretical Development (theorem proving, conceptual analysis) 6. Code Development (generating algorithms, writing scripts) 7. Presentation (rendering graphics, formatting) 8. Editing (grammar, readability) 9. Writing (text generation, document structuring) 10. Citation Formatting (structuring, organising) If other uses were involved, please specify below: Generative AI tool used (list all) Used for? Student number: 2089081 Candidate signature: Date: 10/03/2024 x x Acknowledgements This project was made possible through funding from the University of the Witwatersrand (Postgraduate Merit Award), the National Research Foundation via the South African Environmental Observation Network, Ndlovu Node (Reference: PMDS22062226009), and the South African Association of Botanists (Masters Bursary). I am deeply grateful for this support which enabled me to pursue this research and contribute to the study of plant- pollinator interactions in the Drakensberg. A heartfelt thank you to Prof. Glynis Goodman-Cron for your invaluable mentorship throughout my Master’s degree. I am incredibly grateful to have had the opportunity to work with you for both my Honours and MSc projects and have learned so much under your guidance. A warm thank you to Prof. Sandy-Lynn Steenhuisen for providing guidance throughout this process and additional field resources and assistance. Thank you to Prof. Dave Thompson generously covering the cost of soil sample analyses. To all my supervisors, I sincerely appreciate your insightful feedback and support throughout this project. I extend my gratitude to Ezemvelo KZN Wildlife (Permit no.: OP 764/2023) for granting permission to conduct this study in the Royal Natal National Park. A special thank you to Ian Rushworth for assistance with permit processing and to Siphiwe Khoza, the ranger at Royal Natal, for organising our research accommodation. I am also grateful to the Alpine Research Unit (University of the Free State) for allowing us to stay at the Alpine Research Base during our high-altitude fieldwork and to Witsieshoek Mountain Lodge for accommodating us. Thank you to Dr Stephanie Payne, Dr Fairo Dzekashu Foryuy, Prof. Alice Classen and Prof. Michael Lattorff for your quick responses and assistance with statistics over the Christmas holiday period. I also appreciate Prof. Peter le Roux and Elsa van Ginkel for answering my questions and sharing your code. A special thanks to Prof. Chevonne Reynolds for chairing my proposal meeting and for your in-house statistical advice. Thank you to Dr Renee Reddy and Mr Sifiso Mnxati for your extensive help with plant identification. This project would not have been possible without assistance in the field. Thank you to my mum, Ingrid Roetger, for braving the chain ladders to help with observations, and to Liam and Nicholas Taylor for their invaluable support with data collection. Although not involved in the observational data aspect of the project, a dedicated team of students contributed immensely to the pollen imaging component of the project. A warm thanks to Brittany Pearson, Sydney Bagley, Naishe Chivandi, Mishon le Grange, Kate Edmunds, Gaby da Silva, Pascale Oberholzer, Joshua Marais, Ntandokazi Maqanda and Eve Cuyler for your invaluable efforts. I also extend my gratitude to the Evolutionary Studies Institute and Prof. Marion Bamford for granting access to your microscopes — without which the second part of this project would not have been possible. A special thank you to Dr Angela Effiom and Dr Abraham Dabengwa for showing me the ropes and making me feel welcome. A huge thank you to my dad, Joachim Roetger, for assisting with pollen-washing and slide preparation for over 1000 insects collected in this study. Finally, I am deeply grateful to my parents, my brother, Mathias Roetger, and my sister-in- law, Rebekka Roetger, for their unwavering support throughout this project, and to my dog, Lyra, who inadvertently became my steadfast writing companion and emotional support throughout this journey. Abstract Biotic pollination underpins ecosystem stability and biodiversity maintenance. However, pollinators and their interactions with plants are increasingly threatened by climate change, habitat loss and degradation. This is particularly important in montane ecosystems, which are characterised by high diversity and unique elevation-dependent species assemblages. Yet, local-scale assessments of pollination interactions in these highly diverse systems remain scarce and necessary to predict the future resilience of these systems. This study provides baseline data on plant-pollinator interactions in the northern Maloti- Drakensberg, South Africa’s highest-elevation region. Observations were conducted in three altitudinal zones — lower montane (~1350 m a.s.l.), upper montane (~2100 m a.s.l) and alpine (~3000 m a.s.l) during early, mid and late summer (2023/2024). I analysed interaction network properties (connectance, nestedness, dependence asymmetry, modularity and specialisation). Additionally, plant species (flowering) and insect functional group richness and diversity as well as insect order abundance across these three altitudes and over the summer. Plant diversity decreased with altitude in early summer but rose later in the season. Insect functional group diversity declined with altitude but increased over the summer. Networks shifted from bee- to fly-dominated with increasing altitude. Connectance peaked at low altitude, where invasive species were present and was lowest at mid-altitude where species richness was highest. High-altitude networks were most nested, reflecting adaptation to high environmental variability. Low dependence asymmetry at mid-altitude indicated frequent specialist-specialist interactions. At low altitude, pollinators relied more on specific plants and vice versa for high altitudes. Modularity and network specialisation did not vary across altitude or over the summer. Insect functional group specialisation peaked where the landscape was most heterogeneous. Plant species specialisation decreased with altitude, reflecting a typical response to high altitude conditions. The low species richness, habitat degradation and network property scores suggest that low-altitude grasslands are currently the most vulnerable to climatic changes. The mid-altitude networks are buffered by high species richness but may face long-term background species losses. While high-altitude networks currently appear to be the most resilient lack of upslope refugia and potential competition from lower-elevation species pose long-term risks to these networks and may eventually lead to ecological tipping points in the community. KEYWORDS: Altitudinal patterns; Climate Change; Maloti-Drakensberg; Network properties; Plant- pollinator interactions Table of Contents Declaration................................................................................................................................ ii Acknowledgements ................................................................................................................. iv Abstract .................................................................................................................................... vi Table of Contents ................................................................................................................. viii List of Figures ........................................................................................................................... x List of Tables ........................................................................................................................ xiii Introduction .............................................................................................................................. 1 Methods ..................................................................................................................................... 7 Study Area ............................................................................................................ 7 Sampling protocol ................................................................................................ 11 Plant identification ............................................................................................... 11 Pollinator observation protocol .............................................................................. 12 Insect sampling and identification ....................................................................... 13 Soil sampling ................................................................................................... 16 Weather data: Temperature and Precipitation ........................................................ 16 Data analyses ...................................................................................................... 17 Plant species richness ........................................................................................ 18 Shannon-Hill Diversity Index ............................................................................. 18 Relative abundance ........................................................................................... 18 Network indices ............................................................................................... 19 Weighted connectance (wC) ............................................................................... 19 Modularity (Q) ................................................................................................. 19 Nestedness (wNODF) ....................................................................................... 20 Dependence asymmetry ..................................................................................... 20 Specialisation ................................................................................................... 21 Network-level specialisation (H2’) ...................................................................... 21 Community mean of species level and functional group specialisation (d’) ................ 21 Statistical analyses ............................................................................................ 22 Results ..................................................................................................................................... 24 Temperature, precipitation and soil nutrients patterns during sampling period ............... 24 Daily Temperature Patterns ................................................................................ 24 Precipitation patterns......................................................................................... 25 Soil samples .................................................................................................... 27 Overall summary of the data .................................................................................. 28 Elevational and seasonal patterns in flowering .......................................................... 28 Observation pie charts ....................................................................................... 32 Insect functional groups ..................................................................................... 36 Observation Networks ....................................................................................... 41 Observation network properties .......................................................................... 56 Results summary diagram .................................................................................................... 59 Discussion................................................................................................................................ 60 Plant species patterns ............................................................................................ 60 Network Properties .............................................................................................. 71 Connectance .................................................................................................... 71 Nestedness ...................................................................................................... 72 Dependence asymmetry ..................................................................................... 73 Network-level specialisation .............................................................................. 74 Community-level specialisation .......................................................................... 74 Plant species .................................................................................................... 74 Insect functional groups ..................................................................................... 75 Implications under climate change conditions ........................................................... 75 Future work ........................................................................................................ 81 Discussion summary diagram ................................................................................ 82 Conclusion .............................................................................................................................. 83 Reference list .......................................................................................................................... 84 Appendix ............................................................................................................................... 118 Study plots ........................................................................................................ 118 Plant species list ................................................................................................ 119 Soil Fertility data ............................................................................................... 130 Temperature and precipitation patterns .................................................................. 131 GLM Coefficients .............................................................................................. 131 GAM Coefficients .............................................................................................. 132 Observational data ............................................................................................. 132 List of Figures Figure 1. A schematic profile through the Drakensberg Mountain Centre showing the major eco-thermal belts and associated plant communities ………………..……………..…………9 Figure 2. The study area in the Drakensberg Mountain Centre with 36 study plots in three altitudinal zones……...……………………………………………………………...………..10 Figure 3. Observation plot layout. Each 20 x 20 m2 plot was divided into three equal subplots………………………………………………………………………………...……..13 Figure 4. Minimum (dark blue) and maximum (red) daily ambient temperatures (℃) and daily precipitation (DP; mm) from the first of January 2023 to the first of February 2024 at (a) low, (b) mid- and (c) high altitudes.……….………………………………………………..……...26 Figure 5. Non-metric multidimensional scaling (NMDS) plot with a stress value of 0.088 constructed using Bray-Curtis dissimilarity…………….…………………………...………..27 Figure 6. (a) Species richness (number of plant species present and flowering) patterns along an altitudinal gradient and across the summer. (b) Altitudinal and across-summer patterns of Shannon-Hill Diversity of plant species………………………………………………………31 Figure 7. The interaction observation frequency of plant species and families at low altitude (~1350 m a.s.l.) during early summer (November 2023)………………….…………………32 Figure 8. The interaction observation frequency of plant species and families at mid-altitude (~ 2000 m a.s.l.) during early summer (November 2023)……………………..………………32 Figure 9. The interaction observation frequency of plant species and families at high altitude (alpine; ~ 3000 m a.s.l.) during early summer (November 2023)..…………………………..33 Figure 10. The interaction observation frequency of plant species and families at low altitude (~1350 m a.s.l.) during mid-summer (December 2023)……………………….…………......33 Figure 11. The interaction observation frequency of plant species and families at mid-altitude (~ 2000 m a.s.l.) during mid-summer (2023)..………………..………………………….……34 Figure 12. The interaction observation frequency of plant species and families flowering at high altitude (alpine; ~3000 m a.s.l.) during mid-summer (2023)……………....................…34 Figure 13. The interaction observation frequency of plant species and families flowering at low altitude (~1350 m a.s.l.) during late summer (January 2024)…………………………….35 Figure 14. The interaction observation frequency of plant species and families flowering at mid-altitude (~2000 m a.s.l.) during late summer (January 2024)………...………………….35 Figure 15. The interaction observation frequency of plant species and families flowering at high altitude (alpine; ~ 3000 m a.s.l.) during late summer (January 2024)……………………36 Figure 16. Altitudinal and across-summer patterns of Shannon-Hill Diversity of insect functional groups observed in interactions in sampling plots…………………………………….39 Figure 17. Number of insects observed interacting with flowering plants at low altitude (A), mid altitude (B) and high-altitude level (C) during sampling periods in early, mid and late summer 2023/24………………………………………………………………………...……40 Figure 18. Predicted relative abundance of main insect orders across three elevation bands based on beta regression models……………………….………………………...…………..41 Figure 19. Interaction frequency network plant-pollinator interactions observed at the low- altitude plots during early summer 2023 (~ 1350 m a.s.l.)..…….……………………………………..43 Figure 20. Interaction frequency network plant-pollinator interactions observed at the low- altitude plots during mid-summer 2023 (~ 1350 m a.s.l.).……………………………………………44 Figure 21. Interaction frequency network plant- visitor interactions observed at the low- altitude plots during late-summer 2024 (~ 1350 m a.s.l.).………………………………….…45 Figure 22. Interaction frequency network plant-pollinator interactions observed at the mid- altitude plots during early summer 2023 (~ 2100 m a.s.l.).……………………………………..……..48 Figure 23. Interaction frequency network plant- visitor interactions observed at the mid- altitude plots during mid-summer 2023 (~ 2100 m a.s.l.).………………………………….…49 Figure 24. Interaction frequency network plant- visitor interactions observed at the mid- altitude plots during late-summer 2024 (~2100m a.s.l.)..……………………………………50 Figure 25. Interaction frequency network plant-pollinator interactions observed at the high- altitude plots during early summer 2023 (~ 3000 m a.s.l). ………………………………....…53 Figure 26. Interaction frequency network plant- visitor interactions observed at the high- altitude plots during mid-summer 2023 (~ 3000 m a.s.l.).………………………….…………54 Figure 27. Interaction frequency network plant-pollinator interactions observed at the high- altitude plots during late-summer 2024 (~ 3000 m a.s.l.).…………………………….………55 Figure 28. Altitudinal and across-summer patterns of plant-pollinator visitor observation network indices.……………………………………………………………………………………..…57 Figure 29. (a) Network specialisation (H2’); ranges from 0 (no specialisation) to 1 (total specialisation). (b) & (c) Community mean of insect/plant species specialisation (d’) ….58 Figure 30. Summary diagram of the main study results………………………………….…59 Figure 31. Summary diagram of the potential implications of climate change on plant- pollinator interactions in the Northern Drakensberg as well as recommendations for future research and management……………………………………………………………………81 List of Tables Table 1. Functional grouping of insect floral visitors collected at study sites in the Royal Natal National Park and on the summit plateau within the Alpine belt near Mont-Aux-Source, South Africa..………………………………………………………………………………………..15 Appendix Table 1. Date, location, average slopes and aspect of each of the 36 plots………………………………………………………………………………………....117 Table 2. List of all plants observed in plots at low (L), mid (M) and high altitudes (H) across the summer flowering season 2023/24………………………………………………….118-128 Table 3. Soil fertility and texture analysis data from 20 samples taken at observation plots over the summer-sampling period……………………………………………………………………………..129 Table 4. Temperature and precipitation patterns at each study altitude (low, mid and high) between the 1st of January 2023 and the 2nd of February 2024…………………………………………….….130 Table 5. Beta regression coefficients for the effects of elevation, insect order and season on relative abundance. Significance code (R.4.4.1):0.01**, 0.001***…………………………..………………130 Table 6. Outputs of Generalised Additive Models, showing changes in network metrics along an altitudinal and summer season gradient in the Northern Drakensberg………………………………13 1 Introduction Plant-pollinator interactions are critical for biodiversity and ecosystem functioning, facilitating the reproduction of over 80% of wild flowering species and 70% of major food crops (Ollerton et al., 2011; IPBES, 2016; Chesshire et al., 2021; Siopa et al., 2024). However, wild invertebrate pollinators are in decline due to anthropogenic climate change, habitat loss, degradation and agricultural expansion and intensification (Artamendi et al., 2025). These declines may disrupt plant reproduction, with serious implications for food security, ecosystem stability and biodiversity (Ollerton et al., 2011; Burkle, Marlin and Knight, 2013; IPBES, 2016; Potts et al., 2016; Ramos-Jiliberto et al., 2020). Understanding the structure of plant-pollinator interactions —especially in biodiversity hotspots— is essential for their conservation (Bascompte and Scheffer, 2023). Montane ecosystems are highly diverse and support numerous endemic species (Rahbek et al., 2019; Hořák et al., 2023). Historically, these regions have served as climate refugia, fostering species isolation within small montane climate niches (Medail and Diadema, 2009; Bentley et al., 2019). Plant and insect richness and diversity exhibit distinct, often localised, patterns along altitudinal gradients, adding complexity to their interaction structures and influencing how these systems will respond to climate change (Carbutt et al., 2013; Vázquez et al., 2017; Classen et al., 2020; Malhi et al., 2020; Chesshire et al., 2021; McCabe and Cobb, 2021; IPCC, 2022). Given the variability of these patterns, locally specific research is needed to assess the structures and vulnerabilities of plant-pollinator systems (Thackeray et al., 2016; Ollerton, 2017). 2 A global review of 118 elevational studies (2001–2021) indicated that plant species richness most commonly peaked at mid-altitude (unimodal distribution), though other patterns— monotonic declines, increases, and inverse unimodal distributions—were also observed (Dani et al., 2023). Associated patterns of pollinator richness influence interaction network structure: Hymenoptera often dominate at lower elevations, while Diptera become more prevalent higher up (Pellissier et al., 2010; McCabe and Cobb, 2021). Understanding how these taxonomic and elevational changes affect network structure is crucial. To characterise these networks, several properties are commonly used including nestedness (the degree to which less-linked species form subsets of the interactions of more-linked species; (Renaud et al., 2020), modularity (clustering into distinct interaction groups), connectance (realised vs possible interactions; Renaud et al., 2020; Chesshire et al., 2021), dependence asymmetry (interaction imbalance; Bascompte et al., 2006; Vázquez et al., 2007; van der Kooi et al., 2021) and specialisation (niche partitioning) at both network (Olesen et al., 2007; Barker and Arceo-Gomez, 2021) and community level (Blüthgen et al., 2009). These metrics provide insight into the structural composition of networks and are key to evaluating the resilience and stability of plant-pollinator interactions and are frequently used to characterise plant-pollinator interactions along altitudinal gradients and across seasons (Delmas et al., 2019; Renaud et al., 2020). Such approaches have been used in elevational studies across the globe— the Andes, Alps and Mt. Kilimanjaro (Ramos-Jiliberto et al., 2010; Hoiss et al., 2012; Classen et al., 2020). In South Africa, work on Jonaskop (<1640 m a.s.l.; above sea level;) in the Cape Floristic Region has assessed pollination networks along altitudinal gradients (Adedoja, Kehinde, and Samways, 2020). However, in South Africa’s highest-elevation area, the Maloti-Drakensberg, most research has focused on individual taxa or specialised interactions (e.g. Johnson, 2005; Brown, Downs, and Johnson, 2009; Johnson et al., 2011; Barton, 2019; Cozien et al., 2019; 3 Kahnt et al., 2019; Springer, 2019; van der Niet et al., 2022). There remains a need for baseline data on overall network structures to assess current dynamics and potential future vulnerabilities. The Drakensberg Mountain Centre (DMC), located in the eastern section of the Great Escarpment of Southern Africa, is characterised by temperate montane and alpine grasslands and harbours 227 endemic angiosperm species across 90 genera (Clark et al., 2011; Carbutt, 2019). Ranging from 1318–3482 m a.s.l., the DMC spans three South African provinces (Eastern Cape, Free State and KwaZulu-Natal) and Lesotho, encompassing the country's highest elevations (Carbutt, 2019). The montane and alpine zones form two sub-centres of endemism: the Drakensberg Montane (1300–2800 m a.s.l.) and Maloti Alpine (>2800 m a.s.l.) (Carbutt, 2019). This region is an exceptionally biodiverse and vulnerable ecoregion (Bellard et al., 2012; Lamarque et al., 2014; Wilcox et al., 2017; Bentley et al., 2019). The montane and alpine regions of the DMC are separated into two sub-centres of endemism. The Drakensberg Montane sub-centre includes the lower-montane (1300–1900 m a.s.l.) and upper-montane (1900–2800 m a.s.l.) vegetation and eco-thermal belts below the Drakensberg escarpment (Carbutt, 2019), whereas the Maloti Alpine sub-centre is restricted to the alpine vegetation belt (2800–3482 m a.s.l.; Carbutt, 2019). The DMC forms part of a highly diverse and vulnerable biome and ecoregion (Bellard et al., 2012; Adamson et al., 2013; Lamarque et al., 2014; Wilcox et al., 2017; Bentley et al., 2019). Climate change is likely to intensify the system’s vulnerability in the future, especially in high-altitude areas (Bellard et al., 2012; Adamson et al., 2013; Soussana et al., 2013; Lamarque et al., 2014; Wilcox et al., 2017; Bentley et al., 2019). Although over 6% of the Drakensberg grasslands are formally protected, land degradation and productivity decline persist, driven by frequent burning, invasive species, overgrazing and inadequate funding (Carbutt et al., 2011; Singh, 2013; IUCN, 2020; DOPA, 2021; WDPA, 4 2023). These threats compromise biodiversity even within protected areas. Additionally, climatic changes threaten to disrupt fundamental ecosystem processes by affecting the phenology, morphology, abundance and distributions of plants, pollinators and their interactions (Burkle et al., 2013; Carbutt et al., 2013; Thackeray et al., 2016; Vázquez et al., 2017; Chmura et al., 2019; Gérard et al., 2020; Malhi et al., 2020). The rate and extent of these changes vary between taxa and are often influenced by local climatic and environmental factors (Thackeray et al., 2016). Climate projections for southern Africa suggest increasing temperatures—potentially 1.5–2 times the global average—along with more extreme rainfall events, particularly in the Maloti-Drakensberg (Davis et al., 2017; Engelbrecht, 2019; Scholes and Engelbrecht, 2021; IPCC, 2022). The rapid rate of change, rather than the magnitude alone, poses a major challenge to biodiversity (IPCC, 2022). High-elevation ecosystems are especially vulnerable due to elevation-dependent warming, where montane regions warm faster than lowlands (Pepin et al., 2015; Berauer et al., 2019). Species adapted to cold, nutrient-poor conditions with short growing seasons—such as those on summit plateaus—face competition from lower-elevation species moving upslope (Cotto et al., 2017; Lamprecht et al., 2018; Rosbakh and Poschlod, 2021; Möhl et al., 2022). Although available soil nutrients may increase with warming (Carbutt et al., 2013), this can reduce alpine plant diversity (Humbert et al., 2016) Distribution shifts (generally upslope) of plants and pollinators are predicted to occur in response to the projected climatic changes, creating novel assemblages (Gibson-Reinemer, Sheldon and Rahel, 2015; Pepin et al., 2015; Lamprecht et al., 2018; Kapuka and Hlásny, 2021; Scholes and Engelbrecht, 2021). Indeed, there is evidence that these shifts already occurring (Chen et al., 2011; Inouye, 2020; Zu et al., 2021). High-elevation areas such as the upper montane (1900–2800m a.s.l.) and alpine grasslands (>2800 m a.s.l.) of the Maloti-Drakensberg are likely to become important refugia (Bentley, 2018; Carbutt, 2019). However, these areas 5 will also experience pressures from invasive species and increased competition (Brown & Vellend, 2014; Inouye, 2020; Kapuka and Hlásny, 2021). Species with long lifespans may fail to track shifting climate niches, reducing their fitness and persistence (Couet et al., 2022). Studies from the Alps, Andes and Mt. Kilimanjaro indicate that upslope species shifts will result in significant changes to plant-pollinator networks, including increased generalisation and nestedness at higher elevations, less organized (connected) interaction networks and shifts in dominant pollinator orders (Ramos-Jiliberto et al., 2010; Hoiss et al., 2012; Classen et al., 2020; Gérard et al., 2020; Chesshire et al., 2021; Kapuka and Hlásny, 2021). Changes in temperature and precipitation are also reshaping phenology and morphology across taxa (Chambers et al., 2013), potentially leading to mismatches between plants and pollinators and reduced reproductive success (Memmott et al., 2007; Rafferty and Ives, 2011; Kudo and Ida, 2013; Ramos-Jiliberto et al., 2018). Not all species can adapt or migrate successfully (Brown & Vellend, 2014; Scholes and Engelbrecht, 2021; Kellner et al., 2023). Barriers such as dispersal limitations, soil type incompatibility, or competition may hinder shifts (Primack and Miao, 1992; Dahlhoff et al., 2019; Scherrer et al., 2020; Laiolo et al., 2023). These constraints threaten biodiversity and ecosystem functionality in mountain systems (Carbutt et al., 2013; Malhi et al., 2020; Kellner et al., 2023). As species assemblages homogenise, ecological stability declines (Wilcox et al., 2017). Pollination networks are valuable tools for assessing ecosystem health and monitoring change (Classen et al., 2020; Bascompte and Scheffer, 2023). Changes in network structure may signal broader shifts in biodiversity and function in montane regions such as the DMC. Monitoring these interactions will be key to detecting and responding to the impacts of climate change on biodiversity (Byers, 2017). This study will form the basis of future studies and thereby contribute towards serving this purpose. 6 Aim To use pollinator networks as a tool to compare plant species, insect functional groups and plant-pollinator interactions in three different altitudinal zones (and corresponding vegetation belts) in the northern region of the DMC, and to infer potential impacts of climate change on the interactions occurring within these zones. Objectives 1. To observe and record plant-pollinator interactions within the three altitudinal zones and over the summer to visualise pollinator networks in the DMC. 2. To compare species richness and diversity of plant species and insect functional groups across the three recognised altitudinal zones and across the summer in the northern region of the DMC. 3. To compare plant-pollinator interactions using network properties within these three altitudes over three periods in the summer of 2023–2024. 4. To infer the impact of climate change on current plant-pollinator interactions in the northern region of the DMC. 7 Methods Study Area The study was conducted in the northern region of the Drakensberg Mountain Centre, South Africa (Carbutt, 2019). Study sites were established at low, mid and high elevations to incorporate the three major vegetation types within the DMC. All study sites have a summer rainfall regime (Mucina and Rutherford, 2006). Observation and sampling of plants and visiting insects was conducted at three elevational zones, with the total altitudinal gradient spanning 1350–3065 m a.s.l. The low elevation sampling plots (1350–1500 m a.s.l.) were located in the Drakensberg Montane Sub-centre near the base of the Royal Natal National Park (28°41'8.87"S; 28°57'15.90" E) in Northern Drakensberg Highland Grassland (Fig. 1 & 2; Mucina and Rutherford, 2006; Carbutt, 2019). This grassland type mainly consists of short, sour grasses and is rich in forbs (Mucina and Rutherford, 2006). The landscape features broad valleys and steep slopes on mudstone and sandstone from the Clarens formation (Mucina and Rutherford, 2006). Summers are warm and frequently accompanied by mist, while winters are cold, often featuring severe frosts and occasional snowfall. Mean annual precipitation ranges from 720–1630 mm, and the mean annual temperature is approximately 13.4 ˚C (Mucina and Rutherford, 2006). The mid-elevation plots (2085–2205 m a.s.l.) were located near the upper reaches of the Royal Natal National Park, in proximity to the Witsieshoek Mountain Lodge (28°41'9.65"S; 28°54'1.94"E) within the Upper Montane eco-thermal belt (Fig. 1 & 2; Carbutt, 2019). The vegetation type at this altitude, uKhahlamba Basalt Grassland, is noted for its high species richness (Mucina and Rutherford, 2006; Carbutt, 2019). This landscape is defined by deep gullies as well as steep basalt rock faces interspersed with terraces, while gentler slopes contain nutrient-rich soils (Mucina and Rutherford, 2006). Summers in these grasslands are 8 hot with frequent mist, whereas winters are cold with frost and/or snow days occurring ~55 days annually (Mucina and Rutherford, 2006). Climate characteristics vary greatly depending on altitude and aspect. Mean annual precipitation ranges from 830–1820 mm, and the average annual temperature lies at approximately 11.6 ˚C, fluctuating greatly between seasons (Mucina and Rutherford, 2006). The high elevation plots (3000–3065 m a.s.l.) were situated on the summit plateau within the Alpine belt near Mont-Aux-Sources (28°45'17.90"S; 28°52'0.32"E) at around 3000 m a.s.l. This area comprises Drakensberg Alpine heathland characterised by short, patchy vegetation across the rolling basalt plateau (Fig. 1 & 2; Mucina and Rutherford, 2006; Carbutt, 2019). The colder temperatures and frequent frost days (~182 days annually) at this altitude limits soil microbe activities, resulting in functionally nitrogen-poor soils (Carbutt et al., 2013). Mean annual precipitation varies considerably with elevation, ranging from 634–1609 mm, and includes regular mist and occasional winter precipitation associated with cold fronts (Mucina and Rutherford, 2006). The mean annual temperature is approximately 4˚C (Mucina and Rutherford, 2006). 9 Figure 1. A schematic profile through the Drakensberg Mountain Centre showing the major eco- thermal belts and associated plant communities (Source: Carbutt, 2019; adapted from Killick, 1963). 10 Figure 2. The study area in the Drakensberg Mountain Centre with 36 study plots in three altitudinal zones. The low altitude zone in the valleys of the Royal Natal National Park (RNNP) at ~1350 m a.s.l.; the mid-altitude zone on the slopes of the north-eastern side of RNNP at ~2200 m a.s.l. (Free State & KwaZulu-Natal provinces); the high-altitude zone at ~3000 m a.s.l on the summit plateau (Free State province). The three sampling periods across the summer (2023-2024) are indicated as follows: Early summer (nine pink triangles; November), mid-summer (15 green circles; December) and late summer (12 purple diamonds; January). 11 Sampling protocol Sampling of plant and insect specimens and observations of plant-pollinator interactions took place over three periods in the flowering season of 2023–2024: viz., early summer (27 October–6 November 2023), mid-summer (14–23 December 2023) and late summer in 2024 (21 January–2 February 2024) to capture plant species composition and flower-insect interactions across the flowering season. For the purposes of this study, insects that were observed to make contact with the reproductive parts of the flower are referred to as pollinators. A total of 36 plots (20 x 20 m2) were sampled to encompass as much of the floral diversity at each elevation as possible: 11 low-altitude plots, 14 mid-altitude plots and 11 high-altitude plots (Fig. 2). Plot locations were chosen according to observed maximum floral abundance and diversity at sampling times and occurred across a variety of available grassland slopes and aspects (Appendix Table 1). Plots were located in recently burnt areas (the previous autumn/winter) because flowering was most prolific there. The aspect, average slope (using Google Earth Pro), and GPS location of each plot were noted (Appendix Table 1). Weather permitting, sampling began at the lowest site (Royal Natal National Park) and ended at the highest site (summit plateau) for each field period as the flowering season begins earliest at the lower elevations (Bucher and Römermann, 2020; pers. obs.). However, it was not always possible to sample sequentially up the mountain, and the order of sampling altitudes varied somewhat according to good weather windows (Appendix Table 1). Plant identification All plants flowering within each plot during the study period were identified using iNaturalist, a key (if available), and/or an appropriate field guide (Pooley, 2013, 2020; Uys 12 and Cron, 2013; Johnson and Bytebier, 2015; Swelankomo, Manning, and Magee, 2016) and confirmed using herbarium specimens at the C.E. Moss Herbarium (J). A list of all species flowering is to be found in the Appendix (Table 2). A voucher specimen of each flowering species within the study plots was collected at each altitude unless it was a ‘rare’ or threatened species. Rare species were recognised by their presence in low numbers and diagnostic pictures were taken of these plants instead of a voucher specimen. Photographs of all species are available on iNaturalist (https://www.inaturalist.org; locations of rare species set to private). No grasses, sedges or known wind-pollinated forbs were sampled. Pollinator observation protocol Pollinator observations took place in each plot over 90 minutes. The plots were divided into three sub-plots (Fig. 3). Each 90-minute session had three observers moving around their allocated sub-plot, observing flowering plants and recording any interactions where an insect made contact with the reproductive parts of the flower. If an insect was seen to visit multiple plant species during an observation session, each visit was recorded as a separate interaction. Two experienced observers consistently observed all plots over the three sampling periods, while the third observer varied between sampling periods. Observation sessions began once the sun had risen, and it had warmed up sufficiently (>20 ℃) for most diurnal insects to be active. However, sampling at low altitude on the 30th of October 2023 took place under colder conditions (~10 ℃) due to time constraints. At the low- and mid-altitude zones, observations and sampling took place between 9h00 and 14h00. At the high-altitude zones, sampling began at approximately 10h00 as it took longer for the air to warm up at the higher elevation. 13 Figure 3. Observation plot layout. Each 20 x 20 m2 plot was divided into three equal subplots. One observer was allocated per subplot (1, 2, 3) for the 90-minute observation period. The dotted line indicates the midpoint line of the plot where 3 soil samples were taken and combined into one sample. Insect sampling and identification Insects that visited flowering plants within a plot during pollinator observation sessions were collected using nets and/or vials (see observation protocol). Insects were fume killed in vials precharged with ethyl acetate and subsequently frozen after each observation session to make sure all specimens were dead and to preserve them. The caught insects were classified into one of 21 categories by insect order and then size and/or similar interaction types (Table 1). Each category of insect interacted slightly differently with the flowers in the plots. The insects were preserved using Phenol crystals and identified using the Field Guide to Insects of South Africa (Picker, Griffiths, and Weaving, 2019) and iNaturalist (www.inaturalist.org). The insects were pinned and added to a reference collection in the Life Sciences Museum at the University of the Witwatersrand, Johannesburg. Diptera were separated into categories to broadly compare how different groups of Diptera interact with the flowering plants. Syrphid flies, also known as hover flies, are recognised to be important pollinators, on par with Hymenoptera in many instances (Woodcock et al., 2014; 2 1 2 0 m 20 m 3 Midpoint line of plot 20 m Soil sampled here 14 Inouye et al., 2015). Non-syrphid flies have often been omitted from pollination studies under the assumption that they do not contribute significantly to pollination networks. However, work by Orford, Vaughan and Memmott (2015) suggests that non-syrphid flies have similar pollen- carrying capacities as Syrphid flies. To add nuance, the non-syrphid fly category was subsequently split into long-tongued flies (e.g. Bombilidae), short-tongued flies (e.g. Muscidae), mosquito-like flies (e.g. Empididae), predatory flies (e.g. robber and louse flies) and crane flies. These groups of flies appear to interact with flowering plants in varying ways as a result of their diverse morphology and physiology (Lefebvre et al., 2014; Woodcock et al., 2014; Inouye et al., 2015; Orford et al., 2015; Klecka et al., 2018; Toivonen et al., 2022). 15 Table 1. Functional grouping of insect floral visitors collected at study sites in the Royal Natal National Park and on the summit plateau within the Alpine belt near Mont-Aux-Source, South Africa. Insects were placed into 21 functional groups within the relevant insect order, based on size and how they interacted with the flowers they visited. 16 Soil sampling One combined soil sample was taken for each of the plots to account for the effect that different nutrient contents and soil properties may have on floral species diversity. Three soil samples were taken equidistant apart along the horizontal mid-line of each plot and combined into one plot sample (Fig. 3). The samples were taken at a depth of 0 – 10 cm to include the fraction of the soil used by plant roots. Samples were initially air-dried and then oven-dried at approximately 70 ˚C once returned from each field trip. A total of 36 soil samples were collected. Due to the proximity of some plots selected at different times during the summer, 20 soil samples were sent to the Soil Fertility and Analytical Services of the KwaZulu-Natal Department of Agriculture and Environmental Affairs (Cedara) to analyse soil fertility. A Principal Co-ordinates analysis using Bray-Curtis distance was performed using the software PAST (Hammer, Harper and Ryan, 2001) to visualise the differences among the 20 soil samples taken in observation plots at low, mid and high altitudes with respect to sample density (g/ml), Phosphorous (g/ml), Potassium (g/ml), Calcium (mg/L), Magnesium (mg/L), exchange acidity (cmol/L), total cations (cmol/L), acid saturation (%), pH (KCI), Zinc (mg/l), Manganese (mg/L), Copper (mg/L), Organic Carbon (%), Nitrogen (%) and Clay (%). Other studies have included these soil properties in their analyses (Carbutt et al., 2013; Carbutt and Edwards, 2016; Ratier Backes et al., 2021; Wang et al., 2024). All soil data are available in the Appendix (Table 3). Weather data: Temperature and Precipitation Average daily ambient temperatures (ADAT; ˚C) for low, mid and high altitudes were sourced from the Royal Natal Weather Station, the Alpine Research Unit (ARU) weather station at Witsieshoek and from three temperature TMI loggers at the ARU Alpine Research Base. 17 ADAT from the temperature loggers was averaged to get one ADAT for the high-altitude sites. Mid and high-altitude ADAT data spanned the first of January 2024 to the end of February 2024. At low altitudes, only data from the first of March 2023 to the end of February 2024 was available. Data analyses All analyses were performed using R Statistical Software (v 2024.12.0.467; Posit team, 2024). Using the R package “bipartite” (Dormann et al., 2024), bipartite networks were constructed for each altitudinal zone and summer period using quantitative data from observation-based plant-visitor interactions per plot (Blüthgen et al., 2009; Dormann et al., 2009; Dormann, 2011; Dzekashu et al., 2023). Plant species richness and Shannon-Hill Diversity index for plant species involved in plant- pollinator interactions were calculated. Additionally, the following network indices were calculated for each plot to describe the weighted networks (plant species–insect functional groups): weighted connectance (wC), modularity (Q), interaction strength (dependence) asymmetry, weighted nestedness (wNODF), network specialisation (H2’) and degree of specialisation of plant species, and insect functional groups (d’) (Blüthgen et al., 2007; Dormann et al., 2009; Classen et al., 2020). The effects of network size were not corrected for before determining network properties as network size is considered to be an important factor that depends on seasonal network composition (Schwarz et al., 2020; Dzekashu et al., 2023). However, using the hierarchical approach applied by Dzekashu et al. (2023), the effect of network size was accounted for by first calculating the seasonality pattern across elevations in the General Additive Model (GAM) analyses. 18 Plant species richness Plant species richness, the total number of plant species found present and flowering within a plot, was calculated per plot. Since species richness is strongly affected by sampling effort (Grey et al., 2018), the different numbers of plots per sampling period per altitude were accounted for as a weighted argument of the GAM function when analysing for seasonal or elevational trends in species richness. Looking at plant species richness within each plot, rarer plant species and plant species that were present but not observed to have been visited by potential pollinators were still recorded (Sgarbi et al., 2020). Shannon-Hill Diversity Index Shannon-Hill diversity indices for plant species per plot and insect functional groups were calculated for each of the 36 observation plots. This index is ideal for pollination networks because it does not overemphasise rare or common species and it is relatively robust to differences in sampling effort (Roswell, Dushoff, and Winfree, 2021). Shannon-Hill diversity was calculated using the package ‘vegan’ (Oksanen et al., 2024). Relative abundance The relative abundance of insect orders per plot was calculated for five main pollinator orders including Diptera, Hemiptera, Lepidoptera, Hymenoptera and Coleoptera. A Beta Regression was used to model the patterns of the relative abundance of insect orders across altitudes and the summer-period. This model is useful when analysing data between 0 and 1 and incorporates heteroskedasticity or skewness often found in proportions (Grün et al., 2012). The Beta regression was computed using the ‘betareg’ package (Grün et al., 2012). 19 Network indices Weighted connectance (wC) Weighted connectance was used to indicate network complexity and robustness to species loss as it uses the weighted realised proportion of all possible interactions in a network (Dunne, Williams, and Martinez, 2002; Tylianakis et al., 2010 Castro-Urgal et al., 2012; Tucker and Rehan, 2016; Renaud, Baudry, and Bessa-Gomes, 2020; Chesshire, McCabe, and Cobb, 2021). Weighted connectance produces a value between 0 and 1, where a value closer to 1 means that more of all possible interactions have been realised, i.e. most insect groups visited most of the flowers and the network should therefore be relatively robust to species loss. Conversely, a value closer to zero indicates that fewer interactions have been realised, i.e. flowers are only pollinated by a select few insects and the network may be more vulnerable to individual species loss (Tucker and Rehan, 2016). Weighted connectance (wC) was calculated using the networklevel function (index = “weighted connectance”; Dormann et al., 2024). Modularity (Q) Modular networks consist of sub-groups (modules) of plants and insect visitors that interact more frequently with each other than with other species outside of their module (Tylianakis et al., 2010; Carstensen, Sabatino, and Morellato, 2016; Petanidou et al., 2018). Values range from zero (no modules; random network) to 1 (complete modularity of network). Greater modularity may increase network stability by buffering the effects of disturbances between modules (Tylianakis et al., 2010; Carstensen et al., 2016; Petanidou et al., 2018). Modularity of the network was estimated for each plot using the computeModules function (Dormann et al., 2024). 20 Nestedness (wNODF) Weighted nestedness overlap and decreasing fills (wNODF) indicates the extent to which less- connected species (specialists) tend to interact with more-connected species (generalists; Almeida-Neto and Ulrich, 2011; Renaud et al., 2020; Dzekashu et al., 2023). Weighted nestedness considers both interaction patterns and interaction strength (Almeida-Neto and Ulrich, 2011; Castro-Urgal et al., 2012). Nestedness values range from 0 (fully nested network, i.e. generalist and specialist interactions fully overlap) to 100 (random network, i.e. no overlap between specialist and generalist species; Tucker and Rehan, 2016; Petanidou et al., 2018; Classen et al., 2020;). The higher the nestedness value, the greater the tendency of specialist species to interact with generalist species who in turn tend to interact with each other (Almeida- Neto and Ulrich, 2011; Petanidou et al., 2018; Classen et al., 2020). For example, since specialist species are often the first to go extinct in a network (Gallagher et al., 2015), a high overlap between specialist and generalist species within a network means that the remaining species will still have other species to interact with, thus buffering against secondary extinctions within the network (Bascompte, Jordano, and Olesen, 2006; Tylianakis et al., 2010). Weighted nestedness (wNODF) was calculated using the nest.smdm function (Dormann et al., 2024). Dependence asymmetry Dependence asymmetry was determined using the push-pull index function in R (R package “bipartite”; Vázquez et al., 2007; Dormann et al., 2024). Dependence (species interaction) asymmetry is a measure of the imbalance in the relationship between plants and pollinators (Bascompte, et al., 2006; Vázquez et al., 2007; van der Kooi et al., 2021). It produces values between ₋1 and 1, where positive values indicate higher dependence on the higher trophic level (Bascompte, et al., 2006). Mutualistic networks are characterised by few, highly asymmetric 21 relationships and a wide heterogeneity of relationships which contributes to biodiversity maintenance (Bascompte et al., 2003). Specialisation Network-level specialisation (H2’) Blüthgen’s H2’ index describes specialisation at the network level; it shows differences in niche partitioning among species (i.e., specialisation) between networks (Olesen et al., 2007; Barker and Arceo-Gomez, 2021). It produces values between 0 and 1, where higher values indicate greater average specialisation in the network. H2’ is robust against sampling intensity and network size, making it a suitable tool to compare networks across the three study sites (Blüthgen et al., 2006; Dormann et al., 2009). Network-level specialisation was calculated using the h2fun function (Dormann et al., 2024). Community mean of species level and functional group specialisation (d’) The degree of specialisation of insect functional groups and plant species was calculated. The species specialisation index (d’) indicates how much the observed interaction frequencies of insect functional groups and plant species deviate from the expected frequencies generated using random patterns of the total frequency of interactions available to the plant species or pollinator group (Blüthgen et al., 2009). The d’ index, which produces values between 0 (maximum generalisation) to 1 (maximum specialisation), is relatively robust to observation efforts and does not overestimate the specialisation of rarely observed species (Poisot et al., 2012; Classen et al., 2020). Specialisation was calculated using the dfun function, using an interaction-frequency-weighted average of dprime values per observation plot (Dormann et al., 2024). 22 Statistical analyses All statistical analyses were performed in the statistical software R (v 2024.12.0.467; Posit team, 2024) using the following packages: “MuMIn” (Bartoń, 2024), “mgcv” (Wood, 2023), “nlme” (Pinheiro et al., 2024), “ggplot2” (Wickham et al., 2024). Generalised additive models (GAMs) were used to analyse network patterns across altitudes and the summer period at plant species and insect functional group levels. The GAM is a non- parametric regression method which has less restrictive statistical assumptions than Generalised linear models (GLMs; Swartzman, Silverman, and Williamson, 1995; Guisan, Edwards, and Hastie, 2002). By using non-parametric smoothers, GAMs can handle both simple and complex linear and non-linear relationships between response and explanatory variables (Guisan et al., 2002; Wood, 2010; Peters et al., 2019). The GAMs were computed using the gam function in the “mgcv” package (family: Gaussian; link function: “identity”; Wood, 2023). To confirm that the GAM was a good fit for the data and that Gaussian was a suitable family to use, the distribution of residuals was checked. The basis dimension of the smoothing term was set to k=5 to minimize over-parameterisation (Peters et al., 2016; Dzekashu et al., 2023). Plant species richness, Shanonn-Hill diversity and network interaction indices (wC, Q, dependence asymmetry, wNODF and H2’) were individually used as response variables and elevation was used as a predictor variable. The seasonal period (early, mid and late summer) was included in the models as a factorial variable. The best model was selected using the hierarchical approach as used by Dzekashu et al. (2023), adding in weighted sampling effort to account for differences in total time sampled per altitude per seasonal period (e.g. mid-altitude plots were sampled for a total of 360, 450 and 450 minutes per observer during early, mid and late summer respectively). 23 1. Network ~ Elevation* Seasonal Period (interactive effect model) 2. Network ~ Elevation + Seasonal Period (additive effect model) 3. Network ~ Elevation (elevation-only model) First, the interactive effect model was run to check for the significance of the interactive effect between altitude and seasonal period. If the effect was not significant, the additive model was tested. If that effect was also insignificant, then the elevation-only model was used (Dzekashu et al., 2023). 24 Results Temperature, precipitation and soil nutrients patterns during sampling period Daily Temperature Patterns Temperature patterns between the three altitudes varied. Overall, temperatures were the lowest at high altitude and highest at low altitude. The low altitude experienced the greatest temperature range (−4.9–36.5 ℃), followed by mid altitude (−4.31–28.18 ℃) and high altitude (−9.54–22.15 ℃). At low altitude, the average maximum daily ambient temperature between 1 January 2023 and 29 February 2024 was 24.84 ± 5.34 °C. Data for the maximum and minimum temperatures on 16 days were excluded due to missing records from the weather station. The highest maximum temperature of 36.5 °C was recorded on 27 November 2023, while the lowest maximum, 5.8 °C, occurred on 30 October 2023 (Fig. 4a). The average minimum daily temperature over the period was 9.26 ± 5.24 °C, with the lowest minimum of −4.9 °C recorded on 11 July 2023, and the highest minimum of 26.1 °C on 11 January 2024 (Fig. 4a). At mid altitude, the average maximum daily temperature for the same period was 17.92 ± 5.11 °C, with one day excluded due to missing data. The highest maximum, 28.18 °C, was recorded on 5 October 2023, while the lowest maximum, −0.13 °C, occurred on 30 October 2023 during an unseasonal cold wave in early summer (Fig. 4b). Snowfall was observed at all three altitudes during this cold event (pers. obs.). The average minimum temperature was 8.22 ± 4.06 °C, with the lowest minimum of −4.31 °C recorded on 10 July 2023 and the highest minimum of 17.21 °C on 11 January 2024 (Fig. 4b). 25 At high altitude, the average maximum daily temperature between 1 January 2023 and 29 February 2024 was 12.15 ± 4.44 °C, with three days excluded due to missing records. The highest maximum of 22.15 °C was recorded on 5 October 2023, while the lowest maximum, −3.88 °C, occurred on 10 July 2023 (Fig. 4c). The average minimum daily temperature was 3.67 ± 4.12 °C. The lowest minimum, −9.54 °C, occurred on 10 July 2023, and the highest minimum of 11.66 °C on 11 November 2023 (Fig. 4c). Precipitation patterns Precipitation patterns differed significantly between altitudes. The mid-altitude site experienced the highest number of rainfall days (230 days, totalling 1510.2 mm), while the greatest total precipitation was recorded at low altitude (1917.4 mm over 187 days). In contrast, high altitude received substantially less rainfall, with only 306.2 mm recorded over 76 days. Low altitude sites experienced a mean of 10.25 ± 14.65 mm per rainfall day. The highest daily rainfall was 82.8 mm on 1 March 2023, while the minimum of 0.2 mm was recorded on 20 separate days (Fig. 4a). At mid altitude the mean daily rainfall was 6.57 ± 9.45 mm. The maximum single-day rainfall event was 74.8 mm on 29 December 2023, while the lowest (0.2 mm) was recorded on 26 days (Fig. 4b). At high altitude these was an average of 4.52 ± 6.46 mm per rainfall day. The highest precipitation, 37.4 mm, was recorded on 12 October 2023, and the lowest, 0.2 mm, on 11 days between 1 January 2023 and 29 February 2024 (Fig. 4c). An overview rainfall table is available in the Appendix (Table 4). 26 Figure 4. Minimum (dark blue) and maximum (red) daily ambient temperatures (℃) and daily precipitation (DP; mm) from the first of January 2023 to the first of February 2024 at (a) low, (b) mid- and (c) high altitudes. MMDAT and DP data were taken from (a) the Royal Natal National Park weather station, (b) the Alpine Research Unit (ARU) weather station at Witsieshoek and (c) the weather station at the ARU Alpine base. Breaks in the temperature lines indicate missing values. Sampling (i.e. observation) days are indicated in grey as vertical bars. 27 Soil samples The soil samples taken across the early, mid and late summer period at low, mid- and high altitudes did not form distinct groups in the non-metric multidimensional scaling (NMDS) plot (Fig. 5). Figure 5. Non-metric multidimensional scaling (NMDS) plot with a stress value of 0.088 constructed using Bray-Curtis dissimilarity to visualise the differences among the sample density (g/ml), Phosphorous (g/ml), Potassium (g/ml), Calcium (mg/L), Magnesium (mg/L), exchange acidity (cmol/L), total cations (cmol/L), acid saturation (%), pH (KCI), Zinc (mg/l), Manganese (mg/L), Copper (mg/L), organic Carbon (%), Nitrogen (%) and Clay % in the 20 soil samples taken in observation plots at low (black), mid (red) and high-altitudes (turquoise) during early (circle), mid (plus) and late (square) summer 2023/2024. 28 Overall summary of the data A total of 3 956 plant-pollinator interactions were observed over the summer study period of 2023 – 2024, with 612 in early summer, 1674 in mid-summer and 1670 in late summer. The 203 plant species across the sampling plots were identified to genus (15) and species level (188) where possible (See Appendix table 2 for a list of all species and their authorities). Plant species which were flowering during the summer sampling period will hereafter be referred to as flowering plants. Elevational and seasonal patterns in flowering Flowering plant species richness (number of plant species present and flowering) showed a unimodal response to altitude, with an interactive effect between early summer and elevation where overall plant species richness was lowest during early summer, especially at high elevations (n = 36, ED = 45.8%, Finteractive = 8.01, pinteractive= 0.001, Fig. 6a). During early summer, plant species richness peaked at low elevations (Fig. 6a), with no significant difference between mid- and high altitudes. In contrast, during mid and late summer, most plant species occurred at mid-altitude. Similarly, the Shannon-Hill Diversity Index for the observed flowering plant species changed significantly with elevation and showed a slight interactive effect between elevation and early summer (n = 36, ED = 31.2%, Finteractive = 3.44, pinteractive= 0.03, Fig. 6b). Plant diversity declined with altitude during early summer, while during mid- and late summer it peaked at mid-altitude and decreased slightly at high altitude (Fig. 6b). At low altitude, insects most frequently visited Helichrysum aureonitens (22.91%; Asteraceae) and Pentanisia prunelloides (18.75%; Rubiaceae; Fig. 7). Overall, most mid-altitude visits were recorded on Helichrysum krookii (Asteraceae; 14.97%), H. aureum (Asteraceae; 14.01%) and Watsonia lepida (Iridaceae; 14.01%; Fig. 8). At high altitude, Geum capense (Rosaceae; 29 31.91%) and the small shrub H. trilineatum (Asteraceae; 29.79%) were the most frequently visited species (Fig. 9). The geophyte Hesperantha schlepeana (Iridaceae) also received a notable share of visits during early summer, accounting for 19.15% of insect visits this elevation (Fig. 9). Across all altitudes and throughout the summer season Asteraceae species dominated plant-pollinator interactions. In early summer, they contributed 62.08% to observations at low altitude sites, 65.29% at mid-altitudes and 44.68% at the high-altitude level. Peak plant species richness occurred during mid-summer across the mid- and high-altitude sites (Fig. 6a). At low altitude during mid-summer, 15 Asteraceae species comprised 92.86% of flowering plants (Fig. 10). Among these, Senecio scitus (36.83%) and Helichrysum nudifolium (25.67%) received the highest number of visits (Fig. 10). The remaining 11 non-Asteraceae species represented 10 plant families and accounted for just 7.14% of insect visits. At mid- altitude, S. scitus was again the most visited plant species (27.69%; Fig. 11). Asteraceae (11 species) accounted for 81.00% of all recorded visits at mid-altitude, while the remaining 21 plant species from 16 plant families accounted for the remaining 19% of observed interactions (Fig. 11). At high altitude, 16 Asteraceae species accounted for 54.15% of observed interactions. Helichrysum marginatum (11.19%) and H. subfalcatum (7.22%) were the most frequently visited species among them (Fig. 12). The most visited species at the high altitude during mid-summer was Geranium incanum (Geraniaceae; 21.22%; Fig. 12). Other notable contributors included Scabiosa columbaria (Caprifoliaceae; 7.76%) and the geophyte, Dierama dracomontanum (Iridaceae; 7.40%). During late summer flowering was lowest at low altitude sites, with only 19 plant species in flower, while peaking at mid-altitude sites with 61 flowering species. At high altitude, the number of flowering species declined from 57 in mid-summer to 35 species in late summer (Fig. 6a). Asteraceae continued to dominate at low altitude site (52.82%), with Nidorella auriculata (31.17%) and N. pinnata (14.29%) accounting for most visits (Fig. 13). Non- 30 Asteraceae species such as Verbena bonariensis (Verbenaceae; 19.48%) and Hebenstretia oatesii (Scrophulariaceae; 15.58%) were also visited frequently by insect visitors. At mid- altitude, Asteraceae dominance decreased (44.90%), with the other observations split across 19 other plant families (55.10%; Fig. 14). Most interactions were recorded on Senecio isatideus (18.06%), Lotononis lotononoides (Fabaceae; 14.66%) and Cephalaria oblongifolia (Dipsaceae; 9.55%). At high altitude, Asteraceae species accounted for 85.33% observed plant- pollinator interactions. The most visited species were Helichrysum subfalcatum (23.56%), Berkheya macrocephela (22.07%) and H. albo-brunneum (12.00%; Fig. 15). The remaining 14.67%) of visits were distributed among 10 plant families and primarily on Erica frigida (Ericaceae; 3.41%) and Crassula setulosa (Crassulaceae; 2.67%). 31 Figure 6. (a) Species richness (number of plant species present and flowering) patterns along an altitudinal gradient and across the summer. (b) Altitudinal and across-summer patterns of Shannon-Hill Diversity of plant species flowering and observed in interactions in sampling plots. Generalised Additive Models (GAM) were used to analyse all seasonal and elevational trends (family = Gaussian, k = 5). Each sampled plot from early (pink), mid (light green) and late (dark blue) summer is represented. The number of plots per sampling period per altitude was used as a weighted link to account for varying sampling intensity. The p-value indicates the statistical significance of the differences among the low, mid and high altitudes across this summer period. Overlapping dashed lines indicate no significant differences between these summer periods. 32 Observation pie charts Early summer: Figure 7. The interaction observation frequency of plant species and families at low altitude (~1350 m a.s.l.) during early summer (November 2023). Plants are colour coded according to family. Figure 8. The interaction observation frequency of plant species and families at mid-altitude (~2000 m a.s.l.) during early summer (November 2023). Plants are colour coded according to family. Low altitude Early summer Mid-altitude Early Summer 33 Figure 9. The interaction observation frequency of plant species and families at high altitude (alpine; ~3000 m a.s.l.) during early summer (November 2023). Plants are colour coded according to the family. Mid-summer: Figure 10. The interaction observation frequency of plant species and families at low altitude (~1350 m a.s.l.) during mid-summer (December 2023). Plants are colour coded according to the family. High altitude Early summer Low altitude Mid-summer 34 Figure 11. The interaction observation frequency of plant species and families at mid-altitude (~2000 m a.s.l.) during mid-summer (2023). Plants are colour coded according to the family. Figure 12. The interaction observation frequency of plant species and families flowering at high altitude (alpine; ~3000 m a.s.l.) during mid-summer (2023). Plants are colour coded according to the family. Mid-altitude Mid-summer High altitude Mid-summer 35 Late summer: Figure 13. The interaction observation frequency of plant species and families flowering at low altitude (~1350 m a.s.l.) during late summer (January 2024). Plants are colour coded according to the family. Figure 14. The interaction observation frequency of plant species and families flowering at mid-altitude (~2000 m a.s.l.) during late summer (January 2024). Plants are colour coded according to the family. Low altitude Late summer Mid-altitude Late summer 36 Figure 15. The interaction observation frequency of plant species and families flowering at high altitude (alpine; ~3000 m a.s.l.) during late summer (January 2024). Plants are colour coded according to the family. Insect functional groups Insect functional group Shannon-Hill diversity decreased nearly linearly with altitude and increased over the summer sampling period (n = 36, ED = 51.7 %, Fadditive = 18.36 padditive= 0.001, Fig. 16). Diversity was highest at low altitude —especially during late summer— moderate at mid-altitude and was the lowest at high altitude throughout the sampling period (Fig. 16). The relative abundance of Hymenoptera decreased with altitude while Coleoptera abundance peaked at mid altitude (Fig. 18). Dipteran abundance was greatest at high altitudes (Fig. 18). Low Altitude At low altitude insect functional groups were the most diverse and wasps and true bugs were most frequently recorded at this altitude (Fig. 17 a,b,c). Early summer was dominated by medium and small solitary bees (14.46 %) as well as short-tongued flies (17.77 %; Fig. 17a). Insect visitation frequency peaked in mid-summer, with notable increases in long-tongued flies High altitude Late summer 37 (4.55 % to 13.80 %), fruit chafers (8.68 % to 11.54 %), monkey beetles (2.89 % to 6.79 %) and small beetles (7.02 % to 12.00 %). True bug visitation declined from 9.50 % to 2.49 % (Fig. 17a,b). Medium solitary bees, long-tongued and short-tongued flies were the most frequent insect visitors during mid-summer. By late summer, the most commonly observed insect groups were small solitary bees (14.72 %), wasps (12.99 %), and mosquito-like flies (14.29 %; Fig. 17c). Mid-Altitude Mid-altitude had the highest overall insect observation frequency with small beetles (Coleoptera) being especially prolific visitors at this altitude (Fig. 18). During early summer the most frequent groups observed included, small beetles (22.19 %), monkey beetles (19.61 %), and short-tongued flies (18.65 %; Fig. 17d). True bugs (9.00 %) and butterflies/moths (6.75 %) were also notable (Fig. 17d). During mid-summer there was an increase in insect activity, with small beetles dominant (40.08 %), alongside medium solitary bees (10.44 %), small solitary bees (7.72 %), short-tongued flies (11.69 %) and weevils (11.48 %; Fig. 16e). Observation frequencies peaked during late summer for honeybees (6.15 %), large solitary bees (8.77 %) and mosquito-like flies (4.84 %; Fig. 17f). High altitude The high-altitude (alpine) sites had the lowest insect diversity and activity. This altitude was characterized by the prevalence of Diptera (e.g. short-tongued and mosquito-like flies) as well as Coleoptera (e.g. small beetles and monkey beetles; Fig. 18). Many functional groups including true bugs, honeybees, large solitary bees, blister beetles, and fruit chafers being nearly or completely absent (Fig. 17g; 18). During early summer insect activity was limited. Short-tongued flies (38 %) were the most frequent visitors followed by small solitary bees with 23.40 % (Fig. 17g). Mosquito-like flies and small beetles were both recorded across 38 12.77 % of visits. True bug observation frequency peaked in early summer at 8.51 % (Fig. 17g). Insect activity peaked from mid- to late summer. During mid-summer, short-tongued flies (22.10 %) and mosquito-like flies (23.01 %) were the most prolific floral visitors (Fig. 17h). Monkey beetles (14.67 %), small beetles (14.40 %) were also observed relatively often, while small solitary bees were observed around half as often (12.92 %) as in early summer (Fig. 17g&h). Several insect groups appeared during mid-summer including medium solitary bees, tiny wasps, one cranefly, hoverflies, long-tongued flies, fruit chafers, monkey beetles, and weevils (Fig. 17h). By late summer interactions including short-tongued flies (28.30 %) and monkey beetles (25.93 %) were most frequent, while interactions involving mosquito-like flies (5.93 %) and small solitary bees (10.67 %) declined (Fig. 17i). Medium solitary bees (3.41 %), long-tongued flies (4.89 %) and small beetles (16.00 %) reached peak frequencies (Fig. 17i). While only one species of monkey beetle was collected at mid- and low altitudes, there were at least four species present at the high-altitude level. Notably, the few wasps (7) that were observed at this altitude were tiny, approximately 5 mm in size and only recorded at this high altitude. 39 Figure 16. Altitudinal and across-summer patterns of Shannon-Hill Diversity of insect functional groups observed in interactions in sampling plots. Generalised Additive Models (GAM) were used to analyse all seasonal and elevational trends (family = Gaussian, k = 5). Each sampled plot from early (pink), mid (light green) and late (dark blue) summer is represented. The number of plots per sampling period per altitude was used as a weighted link to account for varying sampling intensity. The p-value indicates the statistical significance of the differences among the low, mid and high altitudes across this summer period. Overlapping dashed lines indicate no significant differences between these summer periods. 40 Figure 17. Percentage of insects observed interacting with flowering plants at low altitude (a–c), mid altitude (d–f) and high-altitude (alpine) level (g–h) during sampling periods in early, mid and late summer 2023/24. Insect orders represented (from left to right) include Diptera, Hemiptera, Coleoptera, Lepidoptera, and Hymenoptera. 41 Figure 18. Predicted relative abundance of main insect orders across three elevation bands based on beta regression models. The main insect orders observed include: Diptera (blue), Hemiptera (yellow), Lepidoptera (pink), Hymenoptera (green) and Coleoptera (orange). Elevation bands refer to sites at low (~1350 m.a.s.l), mid (~2100 m.a.s.l) and high (~3065 m.a.s.l). Confidence intervals (95%) indicated by vertical lines around each average abundance point for each insect order per elevational band. Beta regression coefficients in Appendix, table 5. Observation Networks Low altitude At low altitudes, plant-pollinator interactions exhibited seasonal variation in both composition and network properties. During early summer, interactions were recorded among 16 insect groups and 22 plant species from eight families. This increased in mid-summer to 17 insect groups interacting with 26 plant species from 11 families. By late summer, interactions decreased to 14 insect groups and 11 plant species from six families (Fig. 19,20 & 21). Five plant species were present during both early and mid-summer sampling, while two were present during mid- and late-summer sampling periods (Fig. 19,20 & 21). The dominant plant-pollinator interactions also shifted across the sampling period. In early summer, the majority of interactions involved Helichrysum aureonitens, which was frequently visited by true bugs and short-tongued flies (Fig. 19). During mid-summer, the most prolific interactions involved Helichrysum nudifolium and fruit chafers as well as Senecio scitus and 42 short-tongued flies (Fig. 20). During late summer, there were many frequent interactions between specific groups of insects and plant species. For example, mosquito flies were frequently recorded to visit Nidorella auriculata and wasps were frequent visitors to Verbena bonariensis (Fig. 21). At low altitudes, interaction network properties varied across the summer sampling period. Weighted connectance (wC) during early summer was 0.18, dropping to 0.13 during mid- summer and increasing again to 0.20 during late summer (Fig. 28a). Modularity (Q) and weighted nestedness (wNODF) followed similar patterns. Modularity started at 0.32 in early summer, decreased to 0.29 in mid-summer and peaked during late summer at 0.35 (Fig. 28d). The weighted nestedness of the network during early summer was 30.35, then 25.30 during mid-summer and rose again to 40.00 in late summer (Fig. 28b). Network specialisation (H2) increased over the summer sampling period from 0.25 in early summer to 0.28 during mid- summer and 0.34 during late summer (Fig. 29a). The interaction strength asymmetry shifted from a negative value (-0.01) during early summer to a positive value (0.01) during mid- summer and back to a negative value (-0.08) in late summer (Fig. 28c). 43 Figure 19. Interaction frequency network plant-pollinator interactions observed at the low-altitude plots during early summer 2023 (~ 1350 m a.s.l.). Black blocks indicate the 16 insect-visitor groups (left). The blocks for 22 plant species (right) are colour coded according to their eight families. The height of the blocks indicates the observation frequency of each insect group (left) and plant species (right). The width of the grey lines between insect and plant blocks indicates the strength of the plant- pollinator interactions. Observed interactions are sorted from most frequent at the bottom to least frequent at the top. 44 Figure 20. Interaction frequency network plant-pollinator interactions observed at the low-altitude plots during mid-summer 2023 (~ 1350 m a.s.l.). Black blocks indicate 17 insect-visitor groups (left). The blocks for 26 plant species (right) are colour coded according to their 11 families. The height of the blocks indicates the observation frequency of each insect group (left) and plant species (right). The width of the grey lines between insect and plant blocks indicates the strength of the plant-pollinator interactions. Observed interactions are sorted from most frequent at the bottom to least frequent at the top. 45 Figure 21. Interaction frequency network plant-pollinator interactions observed at the low-altitude plots during late-summer 2024 (~ 1350 m a.s.l.). Black blocks indicate 14 insect-visitor groups (left). The blocks for 11 plant species (right) are colour coded according to their six families. The height of the blocks indicates the observation frequency of each insect group (left) and plant species (right). The width of the grey lines between insect and plant blocks indicates the strength of the plant-pollinator interactions. Observed interactions are sorted from most frequent at the bottom to least frequent at the top. 46 Mid altitude During early summer, plant-pollinator interactions between 14 insect groups and 22 plant species (flowering) were recorded (Fig. 22). This increased to 19 insect groups and 32 plant species in mid-summer and 15 insect groups and 41 plant species during late summer observations (Fig. 23 & 24). Of these plant species, 13 of 22 (59%) plant species during early summer, 16 of 32 (50%) species during mid-summer and 32 of 41 (78%) species during late summer sampling were unique to their respective sampling period. Notably, no plant species was present across all three sampling periods, although Helichrysum krookii, was recorded as present and flowering during both early and late summer. Eight plant species were present during early and mid-summer and nine were present from mid to late summer (Fig. 22, 23 & 24). Plant species present across two sampling periods were generally more abundant and frequently visited by insects during their earlier period of flowering. The composition of insect visitors also changed over the summer study period. For example, during early summer, Scabiosa columbaria was most frequently visited by butterflies and large solitary bees, whereas during mid-summer small beetles and medium solitary bees were its primary visitors (Fig. 22 & 23). Major interactions during early summer included monkey beetles with Watsonia lepida and short-tongued flies interacting with Helichrysum krookii. By mid-summer most interactions were recorded between small beetles and two flowering species, Senecio scitus and Haplocarpha scaposa (Fig. 22 & 23). In late summer, small beetles primarily visited Senecio isatideus (Fig. 24). Network-level indices at mid-altitudes indicate seasonal variation in plant-pollinator interactions. Weighted connectance (wC) peaked during early summer at 0.15 when more plant species were around, dropping to 0.12 during mid-summer and 0.11 during late summer (Fig. 28a). Modularity (Q) and network specialisation (H2’) were both high during early summer (0.42 and 0.38 respectively) and late summer (0.44 and 0.40) but declined during mid-summer 47 to 0.30 respectively (Fig 28d & 29a). The network was the least nested (wNODF) during mid- summer at 23.45 and almost the same during early and late summer at 30.20 and 30.06 respectively (Fig. 28b). In contrast, the heterogeneity of plant-pollinator relationships increased over the summer sampling period from 0.06 in early summer to 0.07 in mid-summer and 0.08 in late summer (Fig. 28c). 48 Figure 22. Interaction frequency network plant-pollinator interactions observed at the mid-altitude plots during early summer 2023 (~ 2100 m a.s.l.). Black blocks indicate 14 insect-visitor groups (left). The blocks for 22 plant species (right) are colour coded according to their eight families. The height of the blocks indicates the observation frequency of each insect group (left) and plant species (right). The width of the grey lines between insect and plant blocks indicates the strength of the plant-pollinator interactions. Observed interactions are sorted from most frequent at the bottom to least frequent at the top. 49 Figure 23. Interaction frequency network plant-pollinator interactions observed at the mid-altitude plots during mid-summer 2023 (~ 2100 m a.s.l.). Black blocks indicate 19 insect-visitor groups (left). The blocks for 32 plant species (right) are colour coded according to their 18 families. The height of the blocks indicates the observation frequency of each insect group (left) and plant species (right). The width of the grey lines between insect and plant blocks indicates the strength of the plant-pollinator interactions. Observed interactions are sorted from most frequent at the bottom to least frequent at the top. 50 Figure 24. Interaction frequency network plant-pollinator interactions observed at the mid-altitude plots during late-summer 2024 (~ 2100m a.s.l.). Black blocks indicate 15 insect-visitor groups (left). The blocks for 41 plant species (right) are colour coded according to their 20 families. The height of the blocks indicates the observation frequency of each insect group (left) and plant species (right). The width of the grey lines between insect and plant blocks indicates the strength of the plant-pollinator interactions. Observed interactions are sorted from most frequent at the bottom to least frequent at the top. 51 High altitude The flowering season at high altitude sites began later at high altitude than at the mid- and low altitude sites, with distinct shifts in species composition and interaction dynamics over the summer. During early summer, interactions between only seven plant species, from four plant families and seven insect groups were recorded (Fig. 25). Peak flowering occurred during mid- summer and corresponded with a diversity of insect-group interactions involving 15 insect groups and 37 plant species from 17 plant families (Fig. 26). In late summer, interactions between 16 insect groups and 32 plant species from 11 plant families were recorded (Fig. 27). Felicia rosulata was the only plant species recorded flowering across all three sampling periods and the only early-summer species that overlapped with the other two periods (Fig. 25, 26 & 27). Approximately half of the plant species (17) recorded during mid- and late-summer sampling occurred in both periods (Fig. 26 & 27). Dominant plant-pollinator interactions also shifted over the summer period. In early summer, the most prolific interactions were recorded between Geum capense and small solitary bees as well as Helichrysum trilineatum and short-tongued flies with a total of 48 interactions during this period (Fig. 25). This was substantially lower compared to the 743 interactions in mid- summer and 675 interactions in late summer (Fig. 26, 27). During mid-summer, Geranium incanum visited by monkey beetles and Scabiosa columbaria visited by mosquito-type flies accounted for the most frequent interactions (Fig. 26). In late summer, interactions were dominated by insects visiting Berkheya macrocephala and monkey beetles visiting a variety of plant species (Fig. 27). The high altitude (alpine) plant-pollinator interaction networks were most connected (wC) at 0.26 during early summer and dropped to 0.14 and 0.12 in mid and late summer respectively (Fig. 28a). In contrast, network modularity (Q) increased across the summer from 0.28 in early summer to 0.37 in mid-summer and 0.39 in late summer (Fig.28d).The networks also became 52 more specialised (H2’) over the summer period, from 0.21 in early summer to 0.34 during mid- summer and 0.33 during late summer (Fig. 29a). Weighted nestedness (wNODF) went from 37.02 in early summer to 28.93 during mid-summer and increased again slightly to 30.44 during late summer (Fig. 28b). Interaction strength asymmetry was lowest during early summer at 0.004 and peaked during mid-summer at 0.15 while decreasing to 0.08 at the end of the summer (Fig. 28c). 53 Figure 25. Interaction frequency network plant-pollinator interactions observed at the high-altitude plots during early summer 2023 (~ 3000 m a.s.l.). Black blocks indicate seven insect-visitor groups (left). The blocks for seven plant species (right) are colour coded according to their four families. The height of the blocks indicates the observation frequency of each insect group (left) and plant species (right). The width of the grey lines between insect and plant blocks indicates the strength of the plant- pollinator interactions. Observed interactions are sorted from most frequent at the bottom to least frequent at the top. 54 Figure 26. Interaction frequency network plant-pollinator interactions observed at the high-altitude plots during mid-summer 2023 (~ 3000 m a.s.l.). Black blocks indicate 15 insect-visitor groups (left). The blocks for 37 plant species (right) are colour coded according to their 17 families. The height of the blocks indicates the observation frequency of each insect group (left) and plant species (right). The width of the grey lines between insect and plant blocks indicates the strength of the plant-pollinator interactions. Observed interactions are sorted from most frequent at the bottom to least frequent at the top. 55 Figure 27. Interaction frequency network plant-pollinator interactions observed at the high-altitude plots during late summer 2024 (~ 3000 m a.s.l.). Black blocks indicate 16 insect-visitor groups (left). The blocks for 32 plant species (right) are colour coded according to their 11 families The height of the blocks indicates the observation frequency of each insect group (left) and plant species (right). The width of the grey lines between insect and plant blocks indicates the strength of the plant-pollinator interactions. Observed interactions are sorted from most frequent at the bottom to least frequent at the top. 56 Observation network properties Network properties differed by altitude and in some cases across the summer-flowering period too. Network connectivity was predominantly affected by elevation (n = 36, ED = 43.8%, Fadditive = 4.02, padditive = 0.01, Fig. 28a). Connectivity was lowest at mid-altitudes and highest at low altitudes. There was a slight seasonal effect where connectivity was generally lower during mid-summer and late summer relative to earlier in the season. Nestedness changed with elevation but did not differ across the summer period (n = 36, ED = 35.2%, Felevation = 4.55, pelevation = 0.01, Fig.28b). It was lowest at mid-altitude and highest at the high-altitude level. Interaction-strength (dependence) asymmetry did not differ across seasons; however, it was affected by elevation (n = 36, ED = 47.6%, Felevation = 8.24, pelevation < 0.001, Fig. 28c). Dependence asymmetry generally increased with altitude. Modularity was not affected by elevation, nor did it change across the summer (n = 36, ED = 8.89%, Felevation = 0.79, pelevation = 0.54, Fig. 28d). Network specialisation did not differ among altitudes nor across the summer season (n = 36, ED = 15.5%, Felevation = 0.21, pelevation = 0.84, Fig. 29a). Community mean insect species level specialisation oscillated between altitudes but did not vary across the summer sampling period (n = 36, ED = 40.8%, Felevation = 6.58, pelevation = 0.001, Fig. 29b). It was lowest at the low altitude, peaked at mid-altitude and then decreased again slightly at high altitude. Community mean plant species level specialisation (n = 36, ED = 15.8%, Felevation = 2.96, pelevation = 0.05, Fig. 29c) decreased in a near linear fashion with increasing altitude but was not significantly affected by summer sampling period. For a summary of results presented here, refer to Summary Diagram in Fig. 30. 57 Figure 28. Altitudinal and across-summer patterns of plant-pollinator visitor observation network indices. (a) Weighted Connectance (wC); ranges from 0 to 1, where a value closer to 1 indicates more of all possible interactions have been realised. (b) Weighted Nestedness Overlap and Decreasing Fills (wNODF); ranges from 0 (specialists and generalists fully overlap) to 100 (no overlap). (c) Dependence asymmetry; ranges from -1 to 1 and is measure of imbalance in the plant-pollinator relationship. For purposes of interpretation, plant species were the higher trophic level in the analysis. Heterogeneity of relationships contributes to sustaining biodiversity. (d) Modularity (Q); ranges from 0 (random network) to 1 (complete compartmentalisation of network). 58 Figure 29. (a) Network specialisation (H2’); ranges from 0 (no specialisation) to 1 (total specialisation). (b) & (c) Community mean of insect/plant species specialisation (d’); ranges from 0 (no specialisation) to 1 (total specialisation). Generalised Additive Models (GAMs) were used to analyse all seasonal and elevational trends (family = Gaussian, k=5). Each sampled plot from early (blue), mid (red)and late (grey) summer is represented. The number of plots per sampling period per altitude was used as a weighted link to account for varying sampling intensity. The p-value in each graph indicates the statistical significance of the differences across the summer and among the low, mid and high altitudes. Overlapping dashed lines indicate no significant differences across these summer periods. 59 Results summary diagram Figure 30. Summary diagram of the main results of the study. Shannon-Hill (S-H) diversity and species richness for plant species and insect functional groups as well as network properties from plant-pollinator interactions (i.e. weighted connectance (wC), nestedness (wNODF), dependence asymmetry and community level specialisation (d’) of insect functional groups and plant species) are summarised in the context of altitude and summer sampling period. 60 Discussion This study aimed to use pollinator networks as a tool to compare plant species, insect functional groups, and plant-pollinator interactions across the summer sampling period of 2023/2024 and three altitudinal zones, corresponding to distinct vegetation belts in the northern region of the Drakensberg Mountain Centre. It also aimed to infer the potential impacts of future climate change on these interactions. Patterns in plant species and insect group assemblages, richness and diversity varied along the altitudinal gradient and over the sampling period, largely aligning with the trends of other mountain systems, with a few notable exceptions. Plant-pollinator interaction properties primarily varied by altitude, with weighted connectance being the only metric to exhibit significant variation in early summer, whereas modularity and network-level specialisation showed no significant elevational or sampling-period trends. Finally, the potential impacts of anthropogenic-induced climate change, including rising ambient temperatures and changes in precipitation patterns, on plant and insect phenology, distribution and abundance are explored. The resilience of current networks to these changes was inferred, and potential strategies for maintaining or promoting network resilience in the future are discussed. See the summary diagram (Fig. 29) for an overview of findings and implications. Plant species patterns Plant species richness (based on plants present and flowering) displayed an unimodal distribution along the altitudinal gradient in this study. This is a common pattern that has been observed in various mountain systems. Similar patterns have been documented in the Bhabha Valley, Western Himalayas, India, where plant species richness peaked at mid-altitudes within temperate forests between approximately 2501 and 3000 m a.s.l. (Chawla et al., 2008). A mid-altitude peak in plant species richness was also recorded in the mixed conifer life zone of the San Francisco Peaks, USA, between 2550 and 2700 m a.s.l. (Chesshire et al., 2021). 61 Chesshire et al. (2021) attributed the mid-altitude peak in species richness to plant species from lower altitudes overlapping into the mid-altitude, which is consistent with the findings in this study. The grasslands of the Northern Limestone Alps in Germany and the Western Swiss Alps presented a similar species richness pattern, peaking at approximately 1200 m a.s.l. and 1800 m a.s.l. respectively (Pellissier et al., 2012; Hoiss, Krauss, and Steffan- Dewenter, 2015). Multiple factors likely contribute to the unimodal plant species richness in montane systems, including climatic influences, landscape heterogeneity, fire regimes, and the presence of invasive species (Stein, Gerstner, and Kreft, 2014). At high altitude (c. 3000 m a.s.l.), the flowering plant species richness was exceptionally low during early summer. The growing and flowering season at high altitudes is relatively short. It occurs later than at lower altitudes as climatic factors such as cooler temperatures and increased solar radiation as well as more frost days and variation in precipitation all, directly and indirectly, impose eco-physiological constraints on plant species occurring in these alpine or near alpine zones (Arroyo, Armesto, and Primack, 1985; Brand, Collins, and Du Preez, 2015; Lefebvre et al., 2018; Sekar et al., 2024; Ahmad, Rathee, and Krishnadas, 2025). While the high-altitude sites in this study occurred on basalt rock which usually forms nutrient-rich soils (corroborated by the soil sample results in this study), the lower temperatures during spring and early summer at high altitudes limit the activity of Nitrogen-mi