THE CHALLENGE OF MAPPING RIVER NETWORK DYNAMICS IN A SEASONALLY DRY CATCHMENT IN CENTRAL ITALY

Intermittent rivers and ephemeral streams, namely IRES, are those watercourses that periodically cease to flow. IRES have a global prevalence, covering the 60% of the world river network and they can be found in a variety of climate settings, from dryland to humid headwater catchments. One of the major issues that prevented the scientific community to get deeper insight on river network dynamics in IRES is represented by the difficulties in the monitoring of the temporal variability of the flowing streams. While in-person inspections through visual surveys represent the common method to map flow intermittency, this method is highly time-consuming. Here, motivated by the latest advancement in computer vision, a novel stage-camera system designed to estimate the space and time variations of the water level along the network was developed. This system comprised a consumer grade wildlife camera with near-Infrared (NIR) night vision capabilities and a white pole set in the thalweg as a reference in the collected images. The efficacy of the system was evaluated through a set of benchmark experiments in natural and complex settings. The maximum mean absolute errors between estimated water level and reference data were approximately equal to 2 cm. Moreover, further experimental tests were carried out to assess the optimal stage-cam setup in terms of bar colour and segmentation. The results indicated that the simplest stage-cam setup, represented by a white pole divided into 2 and 3 segments, is the optimal solutions to monitoring the water level fluctuations in most cases. The data gathered by a stage-camera network composed by 21 cameras distributed along the network was combined with the results of 40 field mapping of the active network to reconstruct the space-time network dynamics in a 3.7 km2 seasonally dry catchment of central Italy from October 2019 to August 2022. The two heterogeneous datasets (the data from the field survey and those derived from the stage camera network) were combined exploiting the hierarchical principle, which postulates that the nodes can be ordered in a Bayesian chain based on their local persistency. This chain dictates the activation/deactivation order of the nodes during the expansion/contraction cycles. The application of the hierarchical model allowed the reconstruction of the temporal evolution of the wet portion of the network during the study period, filling all the relevant data gaps in a robust and efficient manner. The combination of experimental data and modeling results highlighted the complexity of the network dynamics in the study area. The number of active nodes decreased during summer and increased during wet season, as imposed by the mediterranean climate, while their local persistency exhibited non-monotonic and highly heterogenous pattern along the network. As a result, some low order branches were found to be less persistent than some order branches. Despite the complexity of observed stream dynamics, the results shows that the hierarchical model well approximated the expansion/contraction cycle of the network with an accuracy exceeding the 99%. Importantly, the case study presented in this thesis emphasizes how the hierarchical principle allows the reconstruction of the entire active network just by monitoring a few nodes. Cameras and computer vision techniques can provide important quantitative information about the wet dynamics of some nodes in the network and can be used to support traditional monitoring methods, especially if they were combined with robust theoretical and modeling tools.