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

Abstract

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 topoedaphic parameters. 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 2 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.

Optimal Bands Combination Selection for Extracting Garlic Planting Area with Multi-Temporal Sentinel-2 Imagery

“…Secondly, the RF classifier was used to classify mono-temporal data sets, images of each month were classified into two groups (garlic and others), and the garlic planting area was extracted from them. For 110 bands of multi-temporal data set, to rank the importance for each of the 110 bands, the Mean Decrease Gini [ 32 , 33 , 34 ] was computed. It is the RF’s meaningful metric about the importance of each variable and widely used in variable selection and importance evaluation in remote sensing [ 32 , 34 ].…”

“…This study aimed to find the optimal combination of the multi-temporal Sentinel-2 imagery for extracting garlic planting area accurately. Therefore, we used the Mean Decrease Gini [ 32 ] in the RF method to rank the importance of all bands in the whole growth cycle of garlic. According to the importance score of each band, we designed eight different multi-temporal schemes.…”

Assessing Riparian Areas of Greece—An Overview

“…The remote sensing methods (vegetation indices) and the model were applied in the Nestos River Delta (Figure 1G) [53]. Vegetation indices can showcase changes in vegetation through time [54] and have also been used for assessing riparian areas [55].…”

Do Protected Areas Improve Ecosystem Services? A Case Study of Hoh Xil Nature Reserve in Qinghai-Tibetan Plateau

“…This approach exhibited feasibility and practicality in various aspects that could be implemented in other PAs. First, with the use of remotely sensed imagery, which has wide observation range, fast acquisition speed, and short update period, a quick evaluation of ESV can be implemented on a large-scale, long-term basis, and at low costs [69], particularly in remote and data-scarce areas.…”

“…Previous studies developed various satellite-based vegetation indices applied in monitoring the vegetation [69,71]. Such indices (e.g., the Vegetation Condition Index (VCI), the Vegetation Health Index (VHI), the Perpendicular Vegetation Index (PVI), and the two-band version of the enhanced Vegetation Index (EVI2) et al) representing a range of spectral responses of vegetation conditions [69], would be appropriate to adjust the value transfer method. However, each vegetation index varies in performance due to environmental changes (e.g., weather) [71].…”