Deep Learning-based Derivatives for Shoreline Change Detection in Cape Town, South Africa

Abstract

Coastal environments are vital for sustaining key industries, including fisheries, mining, real estate, and tourism, through the provisioning of essential ecosystem services. However, the intricate interplay of bio-chemical and physical processes in these areas is susceptible to disruption by human-driven coastal developments. This disruption has gradually led to adverse consequences such as altered water movement patterns, land quality degradation, habitat loss, pollutant introduction, greenhouse gas emissions, and subsequent sea-level rise. With Cape Town's 240-kilometre coastline serving as a prime example of this complex relationship between coastal processes and livelihoods, this study employs innovative  remote sensing and deep learning techniques. We use Sentinel-2 satellite imagery and the Deep Learning-based CoastSat Python toolkit, specifically leveraging the Modified Normalized Difference Water Index (MNDWI) for sub-pixel level shoreline detection. Focusing on four critical Cape Town coastlines - Noordhoek-Kommetjie, Milnerton, Strandfontein-Monwabisi, and Strand, we assess and compare shoreline changes as detected by the CoastSat toolkit. Our findings reveal significant shoreline changes, with Strandfontein-Monwabisi experiencing the most substantial shift, approximately 284 metres. Noordhoek-Kommetjie exhibited a change of 110 metres, while Milnerton and Strand displayed similar changes, with 88 and 91 metres, respectively. This research not only offers valuable insights into Cape Town's dynamic coastlines, but also informs the strategic allocation of coastal management resources. Consistent data collection and analysis hold the potential to foster interactive coastal monitoring tools, enhancing the effectiveness and sustainability of coastal management practices.