Assessing the impact of dams on riparian and deltaic vegetation using remotely-sensed vegetation indices and Random Forests modelling

Riparian and deltaic areas exhibit a high biodiversity and offer a number of ecosystem services but are often degraded by human activities. Dams, for example, alter the hydrologic and sediment regimes of rivers and can negatively affect riparian areas and deltas. In order to sustainably manage these ecosystems, it is, therefore, essential to assess and monitor the impacts of dams. To this end, site-assessments and in-situ measurements have commonly been used in the past, but these can be laborious, resource demanding and time consuming. Here, we investigated the impact of three dams on the riparian forest of the Nestos River Delta in Greece by employing multi-temporal satellite data. We assessed the evolution in the values of eight vegetation indices over 27 years, derived from 14 dates of Landsat data. We also employed a modelling approach, using a machine learning Random Forests model, to investigate potential linkages between the observed changes in the indices and a host of climatic and terrestrial predictor variables. Our results show that low density vegetation (0-25%) is more affected by the construction of the dams due to its proximity to anthropogenic influences and the effects of hydrologic regime alteration. In contrast, higher density vegetation cover (50-75%) appears to be largely unaffected, or even improving, due to its proximity to the river, while vegetation with intermediate coverage (25-49%)exhibits no clear trend in the Landsat-derived indices. The Random Forests model found that the most important parameters for the riparian vegetation (based on the Mean Decrease Gini and the Mean Decrease Accuracy) were the distance to the dams, the sea and the river. Our results suggest that management plans of riparian and deltaic areas need to incorporate and take into consideration new innovative management practices and monitoring studies that employ multi-temporal satellite data archives.