Why verification and validation of public databases matters: Comment on “Using key and critical biodiversity areas to identify gaps in the protected area network in the Limpopo Province, South Africa”

Broadly used in South African conservation studies, the South African Protected Areas Database is known to contain limitations that require restructuring before use. By replicating the methodology of a recent study in northern Limpopo Province, South Africa, we show how failing to account for such limitations leads to unintentional errors which substantially misrepresent the size of the protected area estate, resulting in a questionable assessment of its contribution to biodiversity conservation targets. Given the implications of employing inaccurate protected area data alongside proposed industrial development and mining that would compromise the ecological integrity of the study area, we stress the importance of validating the national dataset against updated information, provincial-level data, and local knowledge. Dalziel and Evans (2024) assessed protected areas (PAs) contribution to conserving Key Biodiversity Areas (KBAs; BirdLife International, 2024) and Critical Biodiversity Areas (CBAs; Limpopo Department of Economic Development, Environment and Tourism [LEDET], 2018) within the 25,596 km2 Vhembe District, noting considerable threats to ecological integrity posed by the Musina Makhado Special Economic Zone (MMSEZ) industrial development (Gudani Consulting, 2024). Using the South African Protected Areas Database spatial dataset updated quarterly by the Department of Forestry, Fisheries, and the Environment (DFFE, 2023), the authors based their analysis on an ostensible 68 PAs comprising 38% of the district. Spatial datasets of PAs may accumulate errors over time, and, without interrogation, usage may incorporate unintentional errors that misrepresent PA estate size and coverage, leading to erroneous conclusions of their contribution to national (South African National Protected Areas Expansion Strategy; Department of Environmental Affairs, 2016) and international (Target 3 of the Kunming-Montreal Global Biodiversity Framework; Convention on Biological Diversity, 2022) biodiversity protection targets. The DFFE dataset contains limitations including intersecting polygons (190 instances identified by De Vos et al., 2019) and other inconsistencies that require restructuring before use (Skowno et al., 2019; Statistics South Africa, 2021). Overlaps arising specifically from the inclusion of World Heritage Site (WHS) coverages are a recognized feature of the dataset (De Vos et al., 2019: Skowno et al., 2019), distinguishing these from other PA categories which should not overlap and where the most recent proclamation takes precedence (Statistics South Africa, 2021). World Heritage Site buffer zones, rather seen as other effective area-based conservation measures (OECMs; Department of Environmental Affairs, 2016; Statistics South Africa, 2021), are included in the DFFE dataset. Discrepancies between DFFE data and provincial PA data (e.g., LEDET) also exist (Skowno et al., 2019), and some studies combine sources, ranking their authority by prioritizing the former (De Vos et al., 2019) or the latter (LEDET, 2017a). Users of the DFFE data should also acknowledge that private PAs, many declared before the 2003 National Environmental Management: Protected Areas Act, are not validated and some may not function as PAs (Desmet et al., 2013; Skowno et al., 2019; Statistics South Africa, 2021). We replicated Dalziel and Evans (2024) workflow using DFFE quarter two 2024 data (spatially consistent with data used by the authors) in QGIS (version 3.38.3; QGIS Development Team, 2024), analyzing the resulting layer for residual geometry before quantifying PA intersections (see Supporting Information for details). We also compared DFFE data to other datasets of provincial PAs (Desmet et al., 2013; LEDET, 2017b, 2018) for Vhembe and assessed spatial discrepancies, also referring to relevant proclamations in government gazettes. The DFFE dataset contains the 2009 Mapungubwe Cultural Landscape WHS buffer zone of 2630 km2, incidentally replaced in 2014 with a revised buffer of 1048 km2 (Department of Environmental Affairs, 2014; South African National Parks, 2019; Figure 1), and the former is included by Dalziel and Evans (2024), although both buffers are rather considered potential OECMs, not part of the formal PA estate. Three PAs included by the authors only remain in the study area as geoprocessing remnants (0.02 km2 combined area; Dalziel & Evans, 2024, supplementary dataset). Another PA (Manombe Nature Reserve, 9.08 km2) is erroneously mapped in DFFE data and is in an adjacent district (LEDET, 2016) while Makuya Nature Reserve (141 km2, figure from this study), a long-established provincially managed reserve (LEDET, 2012, 2013), is included in neither DFFE nor provincial (LEDET, 2018) spatial data despite reflection in conservation planning reports (Desmet et al., 2013; LEDET, 2013, 2017a, 2017b). Sixteen PAs included by Dalziel and Evans (2024) are intersected with other PAs by 826 km2 (Table 1 and Figure 1). The (deprecated) WHS buffer zone used contains or overlaps 12 of these PAs, and these intersections are separately counted by the authors toward the PA total, as is the buffer zone itself (double counting). This is further complicated by a PA, contained in the buffer zone, itself intersecting two other PAs (triple counting). Four other PAs included are contained within more recent PAs in the DFFE dataset, including a “national park” proclaimed by an Apartheid “homeland” (Department of Agriculture and Forestry, 1980) and a provincial PA made up of two contiguous polygons each intersecting a third (Table 1). Comparing provincial PA extents in provincial datasets (Desmet et al., 2013; LEDET, 2018) with those in the DFFE dataset revealed discrepancies in size and coverage for six PAs (Table 2 and Figure 1). Provincial authorities noted such inconsistencies and considered the former correctly aligned (LEDET, 2017a, 2017b), confirmed when consulting government gazettes proclaiming the PAs. Excluding the 2009 WHS buffer zone, removing areas of PAs partly intersected, removing PAs fully intersected, using provincial PA data (LEDET, 2018) to replace inaccurate DFFE (2023, 2024) data for six PAs and adding Makuya omitted in both datasets (acknowledging that its status remains intention to declare as a PA; LEDET, 2012), we determined 59 spatially distinct PAs with a combined area of 7017 km2, or 27% of Vhembe District (see supporting information). This is considerably less than the 9811 km2 or 38% under protection within 68 PAs reported by Dalziel and Evans (2024). Using the DFFE dataset without acknowledging inherent limitations (including the questionable validity of private PAs declared before 2003), accounting for intersecting geometry and incorporating methods other studies employed to overcome these, has resulted in Dalziel and Evans (2024) misrepresenting the size and coverage of the Vhembe PA estate. Critically, they included the WHS buffer zone OECM of 2630 km2 and also did not cross-validate PAs with other source layers, despite using CBA layers from such a dataset (LEDET, 2018). Their subsequent assessment of PAs contribution to CBA and KBA conservation therefore needs revision. Widely used for South African conservation studies of varying geographical scale, the DFFE dataset is also contributed to the World Database on Protected Areas (WDPA; where it is also updated quarterly), an authoritative source used for indicators on global-scale biodiversity targets (UNEP-WCMC, 2019). The WDPA replicates the DFFE dataset errors of overlapping PAs (although it excludes the WHS buffer zone; UNEP-WCMC and IUCN, 2024) and we support calls for improved validation and review of PA databases by both national authorities and the WDPA, rather than relying on users or journal reviewers to do so (Elsen et al., 2018; You et al., 2018). Notwithstanding this, our comment on Dalziel and Evans's (2024) study underlines the importance of accounting for overlapping PAs by firstly “flattening” or “dissolving” the coverage (as recommended for the WDPA; UNEP-WCMC, 2019) to make the database analysis ready and of further verifying and validating such data at a local scale, ideally with local knowledge. Given the unsustainable development pressures faced by Vhembe District from the proposed MMSEZ development and associated large-scale coal mining, we have concerns about the implications of this flawed study for environmental assessments. Peter Taylor is funded by the University of the Free State. The authors thank an anonymous reviewer for suggestions that improved the manuscript. The authors declare no conflicts of interest. A shapefile (including sidecar files) and comma separated values file resulting from the GIS workflow are available at: https://figshare.com/s/2e68b61b2fd8655dbe4e. A GIS workflow description is available as supporting information. Data S1. Supporting Information. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.