The complexity described above can only be understood through a scientific approach that combines several disciplines, conducts observations at a range of scales from single plants to ecosystems, and integrates land management activities. Using our two consecutive projects—Adaptive Resilience in Southern African Ecosystems (ARS AfricaE, 2014–2018) and Ecosystem Management Support for Climate Change in Southern Africa (EMSAfrica, 2018–2021)—as examples, we discuss the approach with a specific focus on the interlinkages between the different disciplines and scales.
The ARS AfricaE and EMSAfrica field site designs allow us to distinguish between land-use and climate-induced impacts on the structure and function of ecosystems. We established three focal areas along an aridity gradient in South Africa, with each area containing two contrasting observation sites based on different intensities of land use (and thus disturbance regimes), for instance, protected ecosystems compared with livestock grazing or peri-urban landscapes (Fig. 1). On these sites, we measure plant ecophysiological traits, monitor ecosystem-scale carbon fluxes, characterise the spatial dynamics of vegetation structure using remote-sensing, and conduct socio-economic surveys on human use of the ecosystems. The data are used to (i) create, calibrate, and test local ecosystem models, (ii) scale up the information to the biome level, and (iii) provide information adapted to the needs of land-use decision-makers by employing state-of-the-art multi-agent modelling and simulation techniques to integrate spatiotemporal ecological data with social-ecological data (Fig. 2).
Studying ecophysiological responses to climate change at the plant level
By using ecophysiological approaches, we can explore small-scale mechanisms at leaf or plant level, underpinning changes at higher organisational scales such as the community or ecosystem scale (Ainsworth et al. 2016). In our cross-level approach, ecophysiological field experiments are used to improve our understanding of how higher-level responses, such as canopy flux, are related to changes in short-term drivers such as humidity, temperature, and soil water. We collect data for key shrubland, savanna, and grassland species using manipulation experiments and in-field sampling. The traits examined include (1) vegetation structure; (2) water use and leaf-level gas exchange under a range of soil, water, and temperature conditions, and following herbivory; and (3) soil respiration at a range of soil, water, and temperature conditions.
In our cross-scale approach, the results are used to parameterise ecosystem models (the “Modelling ecosystem dynamics at various spatial and temporal scales” section) in order to predict the functioning of entire ecosystems under future climatic and management conditions. The soil respiration and photosynthesis measurements support the partitioning of net ecosystem carbon exchange (NEE) efforts from the nearby eddy covariance (EC) flux tower into its various components (the “Monitoring biosphere-atmosphere exchange of carbon dioxide and water vapor” section): gross primary production (GPP) and net primary production (NPP), and auto- and heterotrophic respiration. The results allow us to deduce responses under the varying environmental conditions brought about by climate change. Similar comparative experiments undertaken at the other project sites allow us to draw conclusions about how general or specific the responses to environmental factors are.
Monitoring biosphere-atmosphere exchange of carbon dioxide and water vapor
Understanding ecosystem carbon fluxes and stocks is important for predicting the responses to climate change and choosing climate change adaptation measures. We established EC flux towers at all six of our project sites (Fig. 1), providing continuous and long-term monitoring of biosphere-atmosphere exchange of CO2 and water vapor along the chosen aridity gradient as well as under different land-use management regimes.
Networks of EC flux towers with their associated meteorological measurements allow GPP and evapotranspiration to be quantified in a variety of climate zones and vegetation types (e.g. Brümmer et al. 2012). EC is the currently preferred method for continuously measuring exchanges of CO2, water vapor and sensible heat between ecosystems, and the atmosphere over time scales of hours to decades and at the landscape scale, thus enabling the evaluation of seasonal and interannual variability as well as the elucidation of their climatic controls (Baldocchi et al. 2001). As the productivity of water-limited ecosystems such as shrublands and savannas is highly dependent on rainfall, and interannual differences are typically significant (e.g. Veenendaal et al. 2004; Brümmer et al. 2008, 2009; Merbold et al. 2009), long-term measurements are essential to detect significant trends.
In our cross-scale approach, by linking flux data with on-site ecophysiological measurements and employing combined approaches from Earth Observation (EO) (the “Leveraging Earth Observation data to support ecosystem monitoring, modelling, and management” section) and vegetation modelling (the “Modelling ecosystem dynamics at various spatial and temporal scales” section), we are able to improve our interpretation and understanding of the carbon fluxes between the biosphere and atmosphere, and study the consequences of ecosystem change for processes such as NPP, which is the basis of many ecosystem services. Continuous measurements allow us to observe the impacts of short-term ecosystem perturbations, such as changes in management regime or weather anomalies, on the CO2 exchange of entire ecosystems. In the long term, the measurements will help to improve our understanding of the net carbon balance of Southern African ecosystems.
Leveraging Earth Observation data to support ecosystem monitoring, modelling, and management
EO data and products enable interdisciplinary studies at all scales of analysis, from the plant and household to the landscape and regional level (cf. Fig. 2). A comprehensive set of analysis-ready EO time series data, so-called space-time data cubes (cf. Baumann 2017) are collected on each of the project field sites (cf. Fig. 1) and for larger geographical areas (e.g. Kruger National Park). They consist of multi-temporal geospatial data from ground-based, air- and space-borne platforms at various sensing schemes. The pre-processed time series data comprise multispectral, thermal infrared, synthetic-aperture radar (SAR) as well as light detection and ranging (LiDAR) imagery and products that serve our diverse project applications and research topics. They are obtained at multiple spatial resolutions, with ground sampling distances (i.e. pixel sizes) ranging from a few centimetres to kilometres.
We further develop and test data fusion and analysis schemes by extracting image products, thematic maps, and spatial statistics from available EO data sets (e.g. Urbazaev et al. 2015; Odipo et al. 2016). Further emphasis is on computational approaches taking advantage of and adding value to publicly available satellite imagery such as data from NASA’s Landsat missions and ESA’s Sentinel (Copernicus) programme (e.g. Cremer et al. 2018; Urban et al. 2018). The resulting methods are used to derive land surface parameters related to the status and dynamics of South Africa’s terrestrial ecosystems (e.g. fuel biomass, woody cover, vegetation heights, land use) to implement environmental and socio-ecological mapping, monitoring, and management with direct societal benefits (e.g. Urbazaev et al. 2015; Odipo et al. 2016; Urban et al. 2018). An example application is the spatiotemporal characterisation of fuel biomass and fuel moisture content for improved fire management in the Kruger National Park.
In the cross-scale interdisciplinary approach, the EO data and products support the interpretation of EC fluxes (the “Monitoring biosphere-atmosphere exchange of carbon dioxide and water vapor” section), ecophysiological experiments (the “Studying ecophysiological responses to climate change at the plant level” section) and socio-economic surveys (the “Understanding human impact in ecosystem change” section). Moreover, they help to parameterise, calibrate, and validate our agent-based simulations, vegetation models (the “Monitoring biosphere-atmosphere exchange of carbon dioxide and water vapor” section) and biome shift predictions. Furthermore, they form an integral part of our anticipated system for data-driven and science-informed decision-making (the “Integrating models to support climate-relevant decision-making” section).
Understanding human impact in ecosystem change
The paired sites approach of the two projects aims at allowing comparisons between ecosystems under little human impact and those under high-intensity human management in all measurement scales. In addition, a case study on the local use of fuelwood is conducted around one of the observation sites, Agincourt village in Bushbuckridge. The socio-economic conditions in the Agincourt site are well researched due to the existence of the 27-year Agincourt Health and Socio-Demographic Surveillance System (HDSS). Surveys conducted on fuelwood collection and use and the monitoring of fuelwood removal in the areas surrounding the villages allow us to accurately quantify carbon removal from local ecosystems. Longitudinal survey data enable the study of interannual variation in the provision of ecosystem services to local communities, as well as the key socio-economic drivers of household dependence on these. In the interdisciplinary approach necessary for socio-ecological inquiry, the results are used in conjunction with vegetation models (the “Modelling ecosystem dynamics at various spatial and temporal scales” section) to construct a case study of resource use in local communities impacted by climate change. To investigate whether local resource extraction and carbon removal by human appropriation can be tracked from space, a further case study on the linkage between remotely sensed dynamics of woody vegetation and the Agincourt household survey data is envisaged.
Modelling ecosystem dynamics at various spatial and temporal scales
Dynamic Global Vegetation Models (DGVMs) integrate processes from the leaf level to the ecosystem and the biosphere level (Prentice et al. 2007; Smith et al. 2014). They simulate the distribution of competing plant functional types (PFTs) and different biome types, vegetation dynamics and structure, and the fluxes of carbon, water, and, increasingly, nutrients between the soil, vegetation, and the atmosphere. Disturbances, such as fire (Scheiter and Higgins 2009; Rabin et al. 2017) and grazing (e.g. Pachzelt et al. 2015) are included in some DGVMs. Site-scale or even farm-scale applications (e.g. within the ARS AfricaE project) make it necessary to adjust the models for local conditions (e.g. Hickler et al. 2012; Seiler et al. 2014).
Several DGVMs have limited applicability in savanna and shrubland ecosystems, because they do not allow for an accurate representation of the vulnerability of woody plants to fire (Scheiter and Higgins 2009) and the vulnerability of woody plants and grasses to herbivory (Scheiter and Higgins 2012). In addition, most DGVMs do not represent shrub growth forms adequately (Gaillard et al. 2018). In our projects, we use the adaptive Dynamic Global Vegetation Model (aDGVM), an individual-based model that was developed to simulate the response of tropical vegetation to impacts of climate change, fire (Scheiter and Higgins 2009; Higgins and Scheiter 2012), and human management, e.g. grazing and wood harvesting (Scheiter et al. 2019), as well as LPJ-GUESS (Lund-Potsdam-Jena General Ecosystem Simulator) (Smith et al. 2001), a global-scale model that will be adjusted for applications in southern Africa.
In our cross-scale approach, information obtained from ecophysiological (the “Studying ecophysiological responses to climate change at the plant level” section) and EC measurements (the “Monitoring biosphere-atmosphere exchange of carbon dioxide and water vapor” section) collected at the project sites is used to parameterise the models for local vegetation under varying grazing pressures. Remote-sensing data (the “Leveraging Earth Observation data to support ecosystem monitoring, modelling, and management” section) are used to test and benchmark model outputs. After successful model parameterisation and testing at various spatial scales, the aDGVM is used to estimate recent past, present, and future climate-driven changes in vegetation and ecosystem functioning.
Integrating models to support climate-relevant decision-making
A decision support system (DSS) is a platform for integrating, analysing, and displaying complex information to assist decision-making (Gibson et al. 2017). These systems typically consist of software components and complex algorithms, designed to support decision-making with the visualisation of different decision outcomes or scenarios.
In our interdisciplinary cross-scale approach, we use DSSs to integrate the various data and models in a way that allows exchange between the researchers and the local decision-makers. The DSSs cover the entire workflow from model development to result analysis and visualisation. Since agent-based modelling and simulation systems have a well-proven record to handle the complexity of coupled human-environmental systems (Le et al. 2012; Lenfers et al. 2018), the MARS (multi-agent research and simulation) framework (Dalski et al. 2017) is used in the projects described here. MARS is particularly well suited for the simulation of large-scale scenarios with a high number of individual agents (Hüning et al, 2016).
We implement decision support via specific case studies at core project research sites. The first study involves local land-use decision-makers at the Agincourt village, and the second focuses on livestock farming systems in the Karoo (Fig. 1). The related decision support systems are, from the outset, designed collaboratively by the MARS developers, the interdisciplinary team of researchers, and the local stakeholders. The active participation of such a diverse group requires the inclusion of a wide range of knowledge and values (Reed 2008; Hugé and Mukherjee 2018). We use stakeholder workshops and focus group discussions as the main way of facilitating exchange between researchers and local decision-makers at the early stages of planning (Nyumba et al. 2018). The information contained in the land-use planning tool as well as the format of the tool are designed together with the end users and adapted to local needs. This is a way to avoid one of the main limitations of many existing DSSs, i.e. their inability to represent available scientific knowledge in the most appropriate way for the intended local users (Dicks et al. 2014).
In addition to generating a practical, hands-on simulation tool for use by land users and/or decision-makers, this collaborative approach is also an important step towards scaling up information on the socio-ecological systems and their local impacts on climate into regional and global level assessments.