Monitoring coastal estuarine habitats for biodiversity along the temperate bioregion of South Africa

Abstract

Coastal wetlands provide critical ecosystem services, including the enhancement of biodiversity, carbon sequestration, and flood protection. Although these ecosystems have been mapped for country-level biodiversity typing, improved extent mapping is necessary to account for estuarine dynamics and improved reporting to the Kunming-Montreal Global Biodiversity Framework (GBF) by 2030. We achieved an overall coastal wetland accuracy of 90.7% (95% confidence interval: 90.2%–91.4%) utilizing a dense time series of very high spatial resolution (3 m) PlanetScope satellite imagery to map coastal wetlands with a combination of Random Forest to develop training data, U-Net convolutional neural networks, and a final decision tree to determine discrete ecosystem extents. Across the 84 mapped estuaries totaling 67,452 ha and 2,135 images, we mapped 9,131.1 ± 1,596.9 ha (13.5% of total estuarine functional zone extent) of salt marsh & reed beds and 1,718.6 ± 234.3 ha (2.5%) of Submerged Aquatic Vegetation (SAV). In addition to our earth observation analysis, we calculated tidal amplitudes and water level trends for 20 water level gauges across the region. We found tidal amplitude was a significant driver of salt marsh extent, explaining 33.6% of the variation (F (1,19) = 9.62, p = 0.005). We demonstrate a repeatable methodology for improved mapping of ecosystem zonation and utilize water level data to explore potential drivers of ecosystem distribution. Our method could be incorporated into a robust earth observation approach for reporting progress toward the goals of the/reporting to the GBF and Sustainable Development Goals (SDGs).

PLAIN LANGUAGE SUMMARY Coastal wetlands provide many benefits to humans, including as habitat for a variety of species, accumulation of carbon in their soils, and protection from flooding and storm events. Global and regional maps of these ecosystems exist, but they lack precision in their identification of ecosystem zones. Improved maps could be used for improved reporting to international agreements and inform coastal management. We mapped three coastal wetland habitats to a high degree of accuracy (90.7%) utilizing a time series of commercial satellite data and machine learning algorithms. Across the 84 mapped estuaries totaling 67,452 ha, we mapped 9131.1 ± 1596.9 ha (13.5% of total estuarine functional zone extent) of salt marsh & reed beds and 1718.6 ± 234.3 ha (2.5%) of Submerged Aquatic Vegetation (SAV). We conducted additional analysis on how tidal amplitude, water level, and impervious surface influence the distribution of habitats in the region, finding that higher tidal amplitudes correlated with more salt marsh extent. Our methodology is repeatable and could improve the monitoring of these ecosystems in South Africa.

KEY POINTS • We mapped coastal wetland habitats (salt marsh, reeds and sedges, and submerged aquatic vegetation) with a U-Net approach at >90% accuracy • We identified tidal amplitude as a major driver of salt marsh habitat, explaining 33.6% of the variation (F (1,19) = 9.62, p = 0.005). • Salt marsh was only 19.4% of coastal wetland extent, 40% of this extent was found within Knysna Estuary and Langebaan Lagoon.