Monitoring channel response and recovery of ephemeral Mediterranean streams using diachronic orthophotography analysis with machine learning (Rambla de Cervera, Spain)

Since the 70s, Mediterranean ephemeral rivers in Spain have been subject to large-scale in-stream gravel mining. In the 90s, this activity ceased to a large extent, although nowadays new controlled in-stream gravel extraction pits have continued. The decreased on gravel mining activity, together with the longitudinal transport of sediment generated by flooding, has allowed these streams to begin to recover their alluvial landforms although their development and continuity is controlled by channel entrenchment and limited by sediment supply.

This study aims to quantify the spatial-temporal changes generated along the river in order to characterise the transmission of sediment longitudinally.  For this purpose, remote sensing methods offer efficient and powerful techniques. Particularly, a supervised classification with SVM (support vector machine) was carried out annually from 2018 to 2021 in the Rambla de Cervera, an ephemeral stream at the Castelló province in eastern Spain. Orthophotographs from the Institut Cartografic Valencia with a resolution of 0.25 m/pixel and RGBI bands were used for the classification. In this classification, three landforms have been differentiated: bedrock (exposed at riverbed), channel (channel gravels and unvegetated gravel bars) and vegetated gravel bars (gravel bars with vegetation cover). Subsequently, an automatic segmentation along the river corridor was performed every 100 m. Finally, the classification values for each segment were extracted and data analysis was performed.

Preliminary results show that river recovery is controlled longitudinally by geological and structural controls. On wide alluvial reaches lateral river supply from bank erosion contributes to forming lateral gravel bars within a narrower alluvial active belt. Conversely, on confined reaches with structural control, gravel bars in the channel bed are discontinuous and alternate with erosional stretches indicating supply-limited conditions. At the most depleted sediment transmission conditions, dense vegetation is stablished decreasing the alluvial longitudinal continuity. The diachronic analysis of the orthophotographs (2018-2021) indicates an incipient recovery of the alluvial landforms at the most downstream reaches which were subject to the most extensive in-channel gravel mining. In summary, the combined use of high-resolution orthophotography with machine learning algorithms provided an effective technique for monitoring spatial-temporal stream recovery and the identification of river sectors where management and restoration efforts are urgent.