Estimating dry bed periods in non-perennial rivers using Sentinel-2 satellite data

Information on the number of dry bed reaches along non-perennial rivers is still lacking, as well as the duration of their non-flow periods. Measurements at conventional gauging stations are not exhaustive due to the high spatial variation of flow rate values and water presence along the non-perennial river network. The availability of moderate-resolution multispectral satellite data from the Sentinel-2 mission offers an unprecedented opportunity to monitor water presence on a broad scale. In this study, we developed a new, automatic approach to detect water, sediments and vegetation along non-perennial rivers by Sentinel-2 satellite imagery. Specifically, we implemented a classification method based on the minimum spectral distance between single pixel’s reflectance and reference spectral signatures, previously obtained from reference images. The classification results are, then, compared with very high-resolution images (resolution of 0.5 m or smaller) acquired by unmanned aerial vehicle and from Google Earth Pro. The performance (F1-score = 0.7) is significantly higher than the ones obtained with the classic algorithm based on the thresholding of Normalized Difference Water Index (F1-score = 0.5). Exploiting the proposed method, we estimated the duration of dry bed condition over two reaches of the Mingardo River (South Italy), from 2017 to 2022. The duration of the dry bed condition resulted to be significantly variable from year to year with the longest and the shortest dry periods respectively, in summer 2017 and in summer 2022. The study demonstrates the feasibility and robustness of using moderate-resolution multispectral images for large-scale monitoring of non-perennial rivers in a cost-effective way.