Cryptostegia grandiflora is a woody perennial climber in the dogbane family, endemic to Madagascar and introduced to South Africa as an ornamental. In its invasive range it behaves with alarming vigour: unsupported it forms a shrub one to two metres tall, but when it finds host trees it can climb as high as 40 metres, smothering canopies in dense, shade-casting thickets. The species thrives in watercourses and areas with a high water table, tolerating annual rainfall between 400 and 1400 millimetres, and it is intolerant of shade, confining it largely to open ground and forest margins. Its stems, leaves and unripe pods exude a milky sap that is toxic to humans and livestock. In Australia, where the vine is ranked among the worst invasive weeds, dense infestations of rangelands cost the livestock industry an estimated 24.8 million Australian dollars annually and ultimately prompted the introduction of a rust fungus as a biological control agent. Similar damage has been documented in Ethiopia, where the vine has reduced species diversity and forest cover, and in the oases of Baja California, Mexico, where it has altered vegetation structure, changed water flow, increased sedimentation and raised erosion and flood risks.

In South Africa the plant is regulated as a category 1b invader under the National Environmental Management: Biodiversity Act and its 2020 Alien and Invasive Species Regulations, which legally mandate its control. Yet control along riparian corridors is notoriously difficult. Dense infestations limit physical access, the plant resprouts rapidly after cutting, and in the Mogalakwena River Reserve there is the added hazard of crocodiles and hippos in the waterway. The vine was first detected on the reserve in 2005, and subsequent studies identified soil moisture as the primary driver of its spread, with densities rising close to water sources such as dams. Until now, however, no systematic monitoring had tracked how quickly the population was growing, and no one had tested whether drone imagery could substitute for laborious and sometimes dangerous ground surveys.

The new study, led by Vivien Kovacs of the University of Copenhagen with colleagues from the Mogalakwena Research Centre and MAP Scientific Services, replicated the methodology of earlier surveys on the reserve. Between May and July 2023 the team sampled a 4.5 kilometre stretch of the eastern bank and a 2.7 kilometre stretch of the western bank, dividing the river into 36 sections of 200 metres. Within each section a single 30 by 30 metre quadrat was centred on a randomly generated GPS point, and each quadrat was subdivided into four 15 by 15 metre sub-quadrats for counting. Because rubber vine grows clonally, the researchers counted distinct ramets, treating stems emerging more than 10 centimetres apart at the soil surface as separate individuals. To keep the workload tractable in heavily infested plots, counting was capped at 100 individuals per sub-quadrat, a threshold corresponding to roughly 400 plants per quadrat, more than double the density conventionally defined as a dense infestation. The authors acknowledge this cap means their figures are conservative minimum estimates in the worst-hit areas.

The results were stark. The 2023 survey recorded 4410 individuals, with counts per quadrat ranging from zero to 376 and the vine absent from only four of the 36 quadrats. The eastern bank was markedly worse than the western side, averaging 136 individuals per quadrat, with 40 percent of eastern quadrats classified as densely infested, compared with 100 individuals per quadrat and 25 percent dense quadrats on the west. Statistical tests confirmed a significant increase in density between 2013 and 2023 and between 2016 and 2023, but no significant difference between 2013 and 2016. The researchers attribute the 2016 lull to the severe El Nino drought of 2015 and 2016, one of the most intense on record, which drove the Mogalakwena River to historically low levels and likely suppressed germination and survival of a species that depends on consistent water availability. The rebound since then suggests that favourable conditions and ineffective control have allowed rapid expansion, with flooding disturbances opening resource-rich gaps along the riverbank for the vine to exploit.

The structural data add a further layer of concern. The predominant stem diameter was 2.1 centimetres and the predominant height 3.67 metres, and stem diameter correlated strongly with height, with a log-log regression model explaining 83 percent of the variance and an estimated scaling exponent of 1.09. Spatial analysis revealed significant spatial autocorrelation in plant height, and a spatial error model provided the best fit to the data. In practical terms, this means the invasion is not merely thickening but also growing in stature: taller, thicker-stemmed individuals can overtop host vegetation, access more light and suppress native flora more effectively. The team also confirmed that density declined significantly with distance from the nearest irrigation dam, reinforcing the role of water availability in shaping the invasion front.

The second half of the study tackled a question of growing importance for conservation technology: can a consumer drone replace boots on the ground? On June 17, 2023, the team flew a DJI Mini 2, an entry-level professional model with a 12-megapixel camera, at 80 metres above ground along a 60-hectare corridor covering part of the eastern bank, capturing 978 photographs. The timing was deliberate. Flown in the southern winter, when most native deciduous vegetation had shed its leaves, the vine’s bright green canopy stood out against the bare landscape, improving detectability. The images were stitched into high-resolution orthomosaics using ArcGIS Drone2Map, and the Segment Anything Model was applied to delineate candidate canopies, which a trained analyst then classified visually, distinguishing rubber vine from lookalike evergreen species such as Croton megalobotrys, Senegalia schweinfurthii and Philenoptera violacea by structural cues like whip-like trailing shoots and leaf texture. Suspected canopies were verified in the field and misclassifications removed.

The comparison between methods was statistically rigorous and revealing. Across 31 quadrats with usable imagery, drone-derived canopy cover ranged from zero to 91 percent and mirrored the field pattern, with greater invasion on the eastern bank. Spearman’s rank correlation showed a strong monotonic association between drone and field estimates, at 0.79 with a p-value below 0.001, and a linear regression explained about 53 percent of the variance. But the agreement was imperfect. The mean absolute error was 14.85 percent, the root mean square error 20.46 percent, and the mean bias of minus 9.97 percent revealed a systematic underestimation by the drone, which grew worse at higher infestation levels. Bland-Altman analysis placed the limits of agreement between minus 45.57 and plus 25.63 percent, and the Wilcoxon Signed-Rank Test confirmed the difference between methods was statistically significant at p equals 0.015. The underestimation is largely attributed to canopy occlusion: a top-down camera cannot see small or understory plants hidden beneath host trees. In a few lightly invaded quadrats the drone actually overestimated density, likely because vines climbing host trees just outside plot boundaries coincided with the quadrat edges. A linear correction model improved the error metrics, reducing the mean absolute error to 11.18 percent and the root mean square error to 14.13 percent, though the underlying bias remained.

The findings place this study alongside a small but growing literature. Previous work in Sabah, Malaysia, found strong correlations between drone-based and ground-based liana estimates, and research in the Republic of the Congo identified significant relationships between ground metrics and canopy liana leaf cover, but few studies have compared both methods for a single invasive climber in a savanna riparian system. The consistent message is that drones excel at broad spatial assessment, especially in terrain that is dangerous or inaccessible, while field surveys remain indispensable for accurate density measurement and ground truthing. The authors note that their entry-level drone lacked the capabilities of more sophisticated platforms, and they point to multispectral sensors and LiDAR as technologies that could overcome the occlusion problem by penetrating the canopy or distinguishing species spectrally.

For the Mogalakwena River Reserve, the practical conclusion is urgent. Rubber vine density has climbed significantly over the past decade, the eastern bank is the invasion hotspot, and proximity to irrigation dams predicts the worst infestations, giving managers a spatially explicit map of where to act first. More broadly, the study argues for hybrid monitoring strategies that pair drone surveys with targeted field validation, a scalable model for tracking invasive woody climbers worldwide as climate change intensifies drought cycles across southern Africa. Whether the technology can keep pace with a climber that can grow 40 metres into a tree canopy remains an open question, but the alternative, wading through crocodile-infested thickets with a tally counter, is clearly no longer the only option.

Subject of Research: Drone-based versus field-based monitoring of invasive rubber vine density along a South African river

Article Title: Comparing drone and field based methods for assessing Cryptostegia grandiflora density along the Mogalakwena river in South Africa

Article References: Kovacs, V., Kruger, C., van Rensburg, G. J., Kruger, F., & Schmidt, L. (2025). Comparing drone and field based methods for assessing Cryptostegia grandiflora density along the Mogalakwena river in South Africa. Discover Ecology, 1(1), Article 12. https://doi.org/10.1007/s44396-025-00013-w

Image Credits: AI Generated

DOI: 10.1007/s44396-025-00013-w

Keywords: Cryptostegia grandiflora, rubber vine, invasive species, drones, remote sensing, riparian ecology, South Africa, Limpopo, plant density assessment, biological invasions, UAV monitoring, conservation