The classification and mapping of ecosystem types is foundational to their assessment and effective management (Borja et al., 2010; Galparsoro et al., 2012; Keith et al., 2022). Developing ecosystem classification frameworks and typologies can also provide an opportunity to ‘marshal’ biodiversity data and knowledge to inform efficient and appropriate action (Keith et al., 2020). As such, one of the key purposes of classifications is to simplify the complexity of biodiversity data into synthesized spatial information that is fundamental for reporting and management (Nikolopoulou et al., 2021). The main objectives for classifying ecosystems include providing: (i) surrogates to represent biodiversity patterns; (ii) ecologically relevant units to support management and planning; and (iii) ecosystem units that can support ecosystem service and accounting work (Dayaram et al., 2021). Applications of ecosystem classification and mapping outputs include those related to assessment, monitoring, spatial biodiversity planning, allocating environmental flows and natural capital accounting (Bland et al., 2017; Bogaart et al., 2019; Botts et al., 2020).

Globally, there have been multiple efforts to classify and map marine regions, habitats and ecosystem types (Gregr and Bodtker, 2007; Longhurst, 2007; Howell, 2010; Last et al., 2010; Gregr et al., 2012; Hu et al., 2021; Nikolopoulou et al., 2021). Approaches include those with a focus on biogeography (Spalding et al., 2007) or bio-physical habitats (McArthur et al., 2010; Sayre et al., 2017; Nikolopoulou et al., 2021), with more capacitated countries possessing substantial datasets increasingly applying statistical methods to support classification, mapping and dissemination (Costello, 2009; Dove et al., 2018; Gerovasileiou et al., 2019). In the context of this paper, ecosystems are considered as functionally connected complexes of living organisms and their non-living environment (often referred to as habitat), and it is recognized that ecosystems can be classified at multiple scales. It is also important to acknowledge the limitations of mapping continuous ecological patterns as distinct polygons that may be less discrete in reality (Keith et al., 2022). Despite the challenges and limitations, national maps of marine ecosystem types or marine ecosystem maps (MEMs) help scientists and managers to organize biodiversity information into spatially discernable units that can support spatial assessment and planning (Dove et al., 2018).

In 2020, the IUCN released a Global Ecosystem Typology (IUCN GET) “to support global, regional and national efforts to assess and manage risks to ecosystems” (Keith et al., 2020, 2022). The IUCN GET provides a globally consistent classification of ecosystems underpinned by ecosystem functioning that spans all realms and the entire biosphere. The typology is hierarchical and developed by top-down classification of biomes within each realm with functional ecosystem groups, but can also accommodate local-scale classifications, including those developed by bottom-up approaches that dovetail with the ecosystem functional groups (Keith et al., 2020). It includes indicative maps of the global distribution at the third hierarchical level (Ecosystem Functional Groups), and there are ongoing efforts to map selected ecosystem types at a biogeographic ecotype level (the fourth hierarchical level). For national applications, more detailed maps at hierarchical levels 4 (biogeographic ecotypes), 5 (global ecosystem types) or 6 (local ecosystem types) are required. The classification and mapping of ecosystems at a national or greater scale is challenging, because of the diversity of ecosystems involved, lack of data availability and coverage, limitations in understanding the drivers of ecosystem pattern, and discrepancies in knowledge among different types of ecosystems (Keith et al., 2022). These challenges are exacerbated in developing countries where there are often less financial and human resources to support strategic biodiversity data collection, analyses and translation of technical outputs into decision-support tools.

Marine ecosystem classification and mapping lags far behind the equivalent efforts on land. The collection of marine biodiversity and environmental data is particularly difficult, because the sea is less accessible (and most of it opaque to satellites) compared to the terrestrial and fresh-water realms (Roberson et al., 2017; Smit et al., 2022; Bell et al., 2022a). Ecosystem mapping is more complex in the marine realm because the ocean environment is three-dimensional, fluid with few prominent boundaries that limit connectivity, and can be highly variable and dynamic (Howell, 2010; Roberson et al., 2017; Sayre et al., 2017; Sink et al., 2019). The costs associated with deep-sea research increase exponentially with increasing depth, resulting in a rapid decrease in data at greater depths (Bell et al., 2022a,b). Furthermore, the technical skills needed to collect, process and disseminate relevant marine geological, oceanographic and biodiversity data are often limited or concentrated within isolated research groups. Different disciplines of marine science have often operated in silos, focused in different areas or conducted research at different scales, which has limited multi-disciplinary collaborative utility. Moreover, oceanographers or marine geoscientists rarely rely on biodiversity data, but ecosystem mappers rely on oceanographic, bathymetric and geological data that complement biodiversity datasets. These different types of data products rarely align in space and time.

South Africa presents a unique opportunity to report on the process and progress in advancing marine ecosystem mapping and is a relevant case study for three reasons. Firstly, we have an established record in harnessing biodiversity data to support systematic biodiversity assessment and conservation planning (e.g., Balmford et al., 2002; SANBI and UNEP-WCMC, 2016; Botts et al., 2019; Skowno et al., 2019; Holness et al., 2022), often leading to implementation success (Botts et al., 2019; Sink et al., 2019; Harris et al., 2022a; von Staden et al., 2022). Secondly, South Africa has three ocean systems: the Indian, Atlantic and Southern Oceans, with a high associated diversity of ecosystems and species (Gibbons, 1999; Costello et al., 2010; Griffiths et al., 2010) that exemplify the challenges in classifying and mapping many different ecosystem types. Thirdly, as a developing country, we can provide relevant lessons for other countries requiring evidence to support sustainable development of their ocean economies, but due to resource and capacity constraints, are faced with different challenges and opportunities to those of more developed nations (Smit et al., 2022). In addition, South Africa’s national MEM has been adjusted for improved alignment with global typologies where appropriate, and comparisons of classifications at multiple scales provide opportunities for joint learning and adjustments of classifications and typologies.

We aim to report on South Africa’s approach to produce and improve the national MEM as a foundation for assessment, planning and ecosystem-based management of the marine realm. The objectives are: (1) to document the process, results and application of the map; (2) to report key limitations, barriers and challenges; and (3) to identify enabling factors and opportunities for more transparent, data-driven and repeatable improvements. In doing so, we contribute an approach to collate, collect, share and integrate multi-disciplinary data to iteratively improve maps of marine ecosystem types in a developing-country context. Because this paper shares lessons from iterative development over the past two decades and also reports on future plans, it departs from the standard paper format and presents methods, results and reflections on several components before distilling key lessons. In so doing, we demonstrate a collaborative approach to iteratively improve ecosystem maps to feed into a national system that synthesizes biodiversity data and knowledge to support biodiversity assessment, spatial planning, decision-making and protection.

The challenges and size of the task of producing a national MEM, especially if it has not been done before, may paralyze efforts to get started. Collating the available and accessible knowledge into a first ‘best available’ version that could be iteratively improved upon, rather than focusing on perfection, was key to South Africa’s marine classification and mapping journey. Producing a first, even if rudimentary, version is better than none. South Africa’s national MEM has not only supported the assessment of marine biodiversity at a national scale (Lombard et al., 2005; Sink et al., 2012, 2019) but has also played a key role in spatial planning and prioritization (Botts et al., 2020; Kirkman et al., 2021; Harris et al., 2022a,b; Sink et al., 2023), and is being used to support national ecosystem accounts (Figure 5).

The ecosystem map is a foundational layer for the assessment of ecosystem threat status and protection level (SANBI and UNEP-WCMC, 2016; Keith et al., 2022) and even in its simplest form, provided an effective foundation to report on these headline indicators. One of the headline messages from South Africa’s first biodiversity assessment was that offshore ecosystems are the most poorly protected among realms (Driver et al., 2005). This message leveraged funding, research, stakeholder engagement and systematic conservation planning that culminated in the proclamation of 20 new Marine Protected Areas in 2019 representing a tenfold increase in area and a dramatic improvement (from 54 to 87%) in ecosystem representation in South Africa’s marine protected area network (Kirkman et al., 2021; Sink et al., 2023). South Africa has recently undertaken a review of its Ecologically or Biologically Significant Marine Areas (EBSAs), revising their delineation and descriptions, assessing their status, and providing management recommendations for each EBSA (Harris et al., 2022a). The MEM was a key dataset for the revised EBSA delineations, especially for assessing the EBSA criteria: biological diversity, and importance for threatened species and habitats, and in the EBSA status assessment. The MEM was also a key component of South Africa’s National Coastal and Marine Spatial Biodiversity Plan (Harris et al., 2022b) that in turn underpins the Biodiversity Sector Plan (DFFE, 2022), which is being developed as the biodiversity sector’s input to the national marine spatial planning process (Figure 6). Ecosystem characteristics, vulnerabilities, threat status and protection levels are considered in impact assessment and environmental authorization (Botts et al., 2020), which is important for ecosystem-based management. Analyses of the interaction between South Africa’s demersal hake trawl fishery and different ecosystem types have influenced eco-certification conditions (Andrews et al., 2022) with progress in ecosystem protection levels being a requirement for meeting eco-certification conditions (Sink et al., 2023). Although natural capital accounting has only recently been initiated in South Africa, the MEM is being used to create extent and condition accounts for terrestrial and aquatic ecosystems (Hein et al., 2022) with a view to ensuring that natural resources and assets, including ecosystem contributions to the economy, are considered, tracked, and maintained.

The limitations, challenges and enabling factors were derived through systematic reflection by members of the marine ecosystem network and by scientists using the classification and map for assessment and spatial planning. This built on the limitations of the MEM 2018 that were identified as part of the NBA 2018 process, because recognition and discussion of limitations is an explicit element used to identify research gaps and priorities for improvement in order to focus funders’ attention on nationally agreed priorities. The main impediments and most significant enabling factors for the classification and mapping process were identified in task team and committee meetings and reviewed in a dedicated workshop where opportunities for future improvements were jointly distilled.

Current limitations and areas for improvement of the marine ecosystem classification and map in South Africa relate broadly to the need for additional data coverage and quality, particularly in deep water (Supplementary Figure S1), with a recognized need to better document uncertainty in the resulting map (Jansen et al., 2022). Data management and provision platforms such as the Copernicus Marine Environment Monitoring Service (CMEMS) and the European Marine Obervation and Data Network (EmodNet) provide leading examples of digital platforms for data acquisition, sharing and production, with potential access to global datasets. Similar regional platforms for the Western Indian Ocean include the Ocean Data and Information Network for Africa (ODINAFRICA)¹ and the African Marine Atlas,² which could also provide data for future improvement of national maps. Identified gaps include improved mapping of Vulnerable Marine Ecosystems, reefs, rhodolith beds (Adams et al., 2020), rocky, mixed and shingle shores, and better understanding of benthic-pelagic coupling and pelagic biodiversity pattern. A limitation for species distribution modeling is the lack of a national bathymetry layer, which could be addressed by collating data from researchers, global databases, the navy and industry to produce one publicly available bathymetry map with the best available data at a national scale. Similarly, nationally collated datasets for oceanographic variables are of high priority, but are currently limited. Development of platforms such as the National Oceans and Coastal Information Management System (OCIMS) may be able to play this role at a national level in future. Linking in and collaborating with existing data platforms and networks, such as ODINAFRICA, the Ocean Info Hub and Indian Ocean Global Ocean Observing System (IO-GOOS), may improve access to data and streamlining data management workflows. An improved understanding of the role of freshwater flow in shelf environments and mapping of fluvial fans is also required. Those ecosystem types that posed challenges in aligning with international classification schemes such as the IUCN GET require further international collaboration to resolve. South Africa also needs to advance conceptual models of its ecosystem functional groups, drawing from, and providing feedback to the IUCN GET team.

South Africa could draw additional lessons from international efforts to collate and share relevant marine ecosystem data and indicators in support of marine management. This includes other international ecosystem classification programs such as the European Nature Information System (EUNIS; Davies et al., 2004), the marine habitat classification for Britain and Ireland (Connor et al., 2004; Parry et al., 2015), the Coastal and Marine Classification Standard (CMECS) (United States Federal Geographic Data Committee, 2012) and the Australian hierarchical seabed classification framework (Last et al., 2010). European platforms that also support the assessment and monitoring of environmental status include CMEMS and EModNet, which not only collate, curate and provide co-ordinated ocean observations but also ready-to-use data products and indicators of ocean health for application in ocean state reports and marine environmental decision making. Relevant improvements that South Africa must consider include the provision of robust, well curated open access datasets; advances in data architecture; application of new technologies to improve observations; modeling to address data gaps; development of forecasting approaches; improved workflows to integrate diverse data sources in a more transparent and reproducible framework; innovations in indicator development and steps to improve user uptake (Le Traon et al., 2019; von Schuckmann et al., 2022). Key challenges and barriers to data collation, curation and integration are shown in (Table 2) and include capacity, data and analytical challenges. Many scientists were concerned that data were inadequate, outdated or needed further time to be improved or completed before being incorporated into ecosystem mapping. Enabling factors and opportunities for improvement of data, analyses, maps and capacity are also outlined in (Table 2). Finally, the reliance on an expert-driven approach leads to potential bias and a lack of transparency in classification, which points to a need for more transparent, repeatable data-driven methods to improve classification and mapping at the national scale.

South Africa’s first lesson from the efforts to classify and map marine ecosystem types is the value of an iterative approach to develop and improve classification and mapping. Dayaram et al. (2019) describe how iterative versions of South Africa’s vegetation map have improved the mapping of terrestrial ecosystems and how this translated into improved foundation for biodiversity planning and management (Botts et al., 2020). The inclusion of a simple map of marine biodiversity pattern in 2004 was useful in ensuring that aquatic ecosystems were not omitted from assessments, and also highlighted discrepancies relative to the more advanced terrestrial effort. Increased participation by scientists was catalyzed by omissions or mismatches in the data used in marine ecosystem mapping, and also through evidence of the map being applied in decision support. There was an understanding that the process was iterative and that there would be opportunities to influence the map with contributed data, participation and knowledge, facilitated by scheduled updates to allow scientists to plan their research for future inclusion of their new data.

The second lesson that emerged was the value of a co-ordinated network of practitioners including early-career researchers to support this work, which is a lesson shared in other realms (Botts et al., 2019). Benefits include greater transparency and inclusiveness, sharing of experience and knowledge between multiple fields and institutes thus supporting learning and capacity development, and helping to leverage desirable data. This serves to enhance South Africa’s marine ecosystem network (Figure 3), which spans numerous disciplines including geology, geoscience, oceanography, biodiversity and ecology and draws from experience from different habitats and functional ecosystem groups. Inclusion of a dedicated emerging researcher task team within the marine ecosystem network and the participation of these young scientists in the Marine Ecosystem Committee meetings provides them with opportunities for learning, developing leadership skills and gaining insight into the science-policy interface, while also fostering institutional memories that are important for inter-generational continuity. The network and task teams and the emerging researcher group in particular have allowed for a greater diversity of researchers to participate in South Africa’s marine ecosystem work, which is important in transitioning to more inclusive and equitable efforts in marine science and capacity development (Amon et al., 2022; Harden-Davies et al., 2022). The established governance structures with multiple teams and committees provide opportunities for discussions that facilitate agreed priorities, highlight key gaps and knowledge shortfalls, and identify potential expertise, datasets, approaches or research opportunities to address gaps. Our experience accords with the expectations shared by Keith et al. (2022) in the context of the IUCN GET, where highlighting areas of limited knowledge promote research to fill significant gaps. In our experience, a formally appointed champion (individual or research team within an appropriately mandated institute) that can drive this process and integrate results as they become available is an important element in coordination, fostering research relationships and facilitating progress.

The third lesson relates to the importance of building and curating key foundational biodiversity datasets and research to understand drivers of biodiversity patterns. The majority of in situ marine biological or ecological data collection that is undertaken is focused on relatively narrow geographic extents and optimized to address specific hypotheses, rather than to map biodiversity. The development of national-scale maps of ecosystem types, however, requires expansive foundational data layers that inform biodiversity and environmental patterns. To produce these layers, substantial effort is often required to collate, ‘clean’ and prepare datasets from many sources into a single, standardized format. In developed countries this is typically a function of national data centers, but such infrastructure and services are often lacking in developing countries. The importance of curation, documentation, expansion and iterative improvement of these datasets is a key requirement to build improved biodiversity knowledge and maps over time (SANBI and UNEP-WCMC, 2016) and helps identify gaps that can guide future research priorities.

The benefit of identifying gaps and research priorities is that it can help funders focus their funding calls to strategically address key gaps. In South Africa, the Foundational Biodiversity Information Programme (FBIP)³ has played an enabling role in funding research to collect, collate and prepare foundational biodiversity data used in South Africa’s Marine Ecosystem classification and map. Similarly, the Department of Science and Innovation/National Research Foundation funded African Coelacanth Ecosystem Programme (ACEP) provides the research community with competitive access to funding and marine research platforms and infrastructure (such as remote imagery and mapping platforms) provided through the South African Institute for Aquatic Biodiversity. The ACEP Open Call draws research priorities from among other strategy documents, the National Biodiversity Assessment which communicates research priorities identified by the Marine Ecosystem Committee, to ensure that the research supported is relevant and addresses gaps in ecosystem understanding. Research supported by ACEP has contributed to ecosystem classification and mapping through the provision of data layers and dedicated research outputs in the form of several student theses and peer-reviewed publications.

The fourth lesson is the importance of developing capacity and transferring skills and tools that enhance data quality, analytical workflows and reproducible outputs. Developing transparent and repeatable workflows requires meticulous organization and documentation of datasets, data preparation, analyses and all decisions along the way (Wilson et al., 2017). Accomplishing this for a wide range of input data from different fields and many different institutions, with a small, constrained team, is a daunting undertaking. Fortunately, data science has adopted and developed many highly effective tools, most of them open source and freely available, that enable improved documentation, management and sharing of data and analyses. Developing such data science and data management skills within the team is critical to working ‘smarter’ and building repeatable workflows. This philosophy and its benefits to iterative programs of work are well demonstrated by Lowndes et al. (2017) in their annual Ocean Health Index assessments. The code, systems and approaches developed by better-capacitated international teams can be shared and adopted to greatly benefit resource-limited teams in developing countries.

Lastly, the fifth lesson is the importance of innovation and adopting modern solutions and emerging statistical tools to advance ecosystem classification and mapping. Rapid advances in the areas of genomics (van Oppen and Coleman, 2022), visual survey methodologies (Mallet and Pelletier, 2014) and machine learning (Beyan and Browman, 2020) have unlocked big data streams for inclusion in the mapping process. Improved environmental and biological data access and extent worldwide has led to increasing focus on data-driven approaches to marine ecosystem classification and mapping (Howell, 2010; Shumchenia and King, 2010; Hill et al., 2020; Woolley et al., 2020). These approaches are robust, transparent, repeatable and may include measures of uncertainty (Jansen et al., 2022), making use of the most complete available data for classifying and predicting biodiversity patterns. The existence of earlier versions of ecosystem maps may necessitate some level of continuity between successive iterations, which can be achieved through expert-based decisions or decision trees. This is feasible due to recent growth in the field of statistical ecology in Africa (Minoarivelo et al., 2021). Three aspects to consider when pursuing data-driven statistical approaches, include: (i) the types of data that are available; (ii) surrogacy (Mellin et al., 2011; Flannery and Przeslawski, 2015), prioritization or integration of datasets (Zipkin et al., 2021) at a relevant scale; and (iii) the statistical method that best captures or predicts biodiversity patterns at this scale (e.g., Verfaillie et al., 2009; Murillo et al., 2018; Hill et al., 2020; McQuaid et al., 2020). Due to the scale mismatch between localized project-based biodiversity research and the standardized, national-scale biodiversity datasets required for ecosystem mapping (Sink et al., 2019), there is typically a need for prioritization or integration of multiple datasets covering different regions and assemblages, or consideration of surrogacy in some cases. Though statistical data integration methods exist, these require parameters or species in common (Zipkin et al., 2021), which is not always the case for datasets collected using different methods such as eDNA, trawls, dredges, grabs, and visual surveys (Lange et al., 2014; Flannery and Przeslawski, 2015). Emerging statistical tools for marine ecosystem classification rely on multivariate biodiversity data, either environmental or biological (or both), and include a technique of clustering the data and predicting the clusters across space (e.g., Verfaillie et al., 2009; Howell, 2010; Ovaskainen et al., 2017; Murillo et al., 2018; Hill et al., 2020). Currently, there is no accepted standard methodology and multiple emerging methods can be explored to classify and predict biodiversity patterns from growing datasets.

A co-ordinated and inclusive process enabled the collation and improvement of multi-disciplinary data to advance ecosystem mapping in a developing country. The national MEM has supported spatial biodiversity assessment, a tenfold increase in MPA estate, the identification of biodiversity priority areas to inform marine spatial planning, and has been applied in biodiversity management. Reflecting on progress, limitations and failures in South Africa reveals important lessons for others working to synthesize multiple datasets for robust decision support and improved biodiversity management. Key lessons include the value of (1) a clear commitment to iterative improvement even if the ideal frameworks or data are lacking; (2) developing relationships and skills among a network of current and emerging practitioners; (3) strategically prioritizing and leveraging resources to build and curate key foundational biodiversity datasets; (4) enhancing data quality and analytical workflows; and (5) the application of new technology and emerging statistical tools to improve the classification and prediction of biodiversity patterns. South Africa’s marine ecosystem mapping process provides insights that are relevant for developing countries but are also applicable in global efforts to unlock, share and use multi-disciplinary data for better biodiversity decisions.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found at: http://bgis.sanbi.org/SpatialDataset/Detail/2681.

KS, NK, LH, AS, JC, LJA, and LAA contributed to the conception and planned content of this manuscript. KS, LAA, M-LF, JC, LH, SP, NK, AD, AS, ML-V, LJA, LD, SK, and RP wrote sections of the manuscript. AD, LAA, M-LF, MP, and KS produced the figures. All authors contributed data and analyses to the 2018 and 2022 versions of the ecosystem map and reviewed the draft manuscript. Most authors contributed to the reflections and lessons emanating from this manuscript.

The development and application of national maps of ecosystem types is part of the mandate of the South African National Biodiversity Institute. This manuscript was feasible due to funding received from the European Union’s Horizon 2020 research and innovation program under Grant Agreement No. 862428 (MISSION ATLANTIC). KS and NK acknowledge the ACEP Deep Connections, Agulhas Bank Connections and Deep Forests projects funded through the National Research Foundation (NRF Grants 129216, 129213, and 110765). LJA acknowledges support from the SeaMap project funded through the NRF Grant number 138572. LH and SH were supported by the Benguela Current Marine Spatial Management and Governance (MARISMA) Project. The MARISMA Project is funded by the German Federal Ministry for the Environment, Nature Conservation, Nuclear Safety and Consumer Protection (BMUV) through its International Climate Initiative (ICI), with in-kind contributions by the Benguela Current Commission (BCC) and its contracting parties (South Africa, Namibia and Angola). It is implemented by GIZ (Deutsche Gesellschaft für Internationale Zusammenarbeit GmbH; German Development Cooperation) in partnership with the BCC and its contracting parties. LH, LVN, SH, KS, MP, LD, and SK are also supported by CoastWise, a project of the MeerWissen initiative, which is funded by the German Federal Ministry for Economic Cooperation and Development, and implemented by GIZ.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.