Relative contributions to suspended sediment variability under extreme events (Gironde Estuary, France)

The frequency and intensity of extreme events associated with climate change are projected to increase continuously in the coming decades. Within these scenarios, the effects and ramifications of extreme events on coastal ecosystems are still poorly understood. In particular, the spatiotemporal footprint of extreme events is required to devise a strategy for better mitigation of impacts. Satellite data provide a unique spatial capability to address the effects of extreme events, for example, on Suspended Particulate Matter (SPM) in coastal waters. However, the low temporal resolution (e.g., associated with cloud disturbances) leads to small or insufficient samples to capture the dynamics of a given coastal system. On the other hand, although hydrodynamic sediment transport models provide continuous spatial-temporal estimates of SPM, refining their realistic flow of SPM importation or accurate sediment class distribution, especially capturing extreme events, remains challenging.

The new generation of statistical approaches comprising machine learning techniques is a valuable tool for comprehensive cube data time series of satellite remote sensing data with spatial and temporal gaps. Here, we propose a machine learning framework.  The framework allows not only filling spatial gaps in satellite imagery (compromised due to cloud disturbances) but also the estimation of the spatial estuarine domain affected by extreme events in river discharge and windbursts. Preliminary results also suggest that SPM dynamics is largely influenced by hydrodynamic forcings (river discharge, tides, winds), but depth can also play a significant role. Our study demonstrates that machine learning might be useful to synthesize coherent spatial and temporal distribution patterns of SPM variability, highlighting where extreme events most and least likely affect the estuarine system. The latter provides valuable insights for coastal management, such as prioritizing regions mosltly influence by extreme events for ecological monitoring and maintenance of critical habitats.