Revealing 35 years of landcover dynamics in floodplains of trained lowland rivers using satellite data

Abstract

Lacking substantial erosive and sedimentation forces, regulated rivers allow their floodplains to become overgrown with forest, increasing the flood risk of the hinterland. In the Netherlands, floodplains have therefore been subjected to interventions, like clear cutting, lowering and creation of side channels, and management, consisting of grazing and mowing. However, the comprehension of how those activities influence landcover dynamics is lacking. The aim of this study is therefore to investigate long‐term landcover dynamics of a regulated river system through the lens of remote sensing. What transitions between landcover classes can be observed? And how (if) do management and interventions impact succession and retrogression of landcover classes? The study area comprised the upstream part of the Dutch Rhine River, its three branches and five adjacent floodplains. Satellite data (LandSat 5 and 8), encompassing a 35‐year period (1984–2018), were used for studying landcover dynamics. Landcover classification was based on seven classes: water, built‐up area, bare substrate, grass, herbaceous vegetation, shrubs and forest. Retrogression was highest for the landcover classes obstructing water flow (shrubs, forest and herbaceous vegetation), succession was most frequent on bare substrate, and water and grass were the most stable landcover classes. The regulated nature of the system became apparent from the spatial and temporal cacophony of landcover dynamics which differ from those of natural meandering rivers. This study showed that satellite data are useful for analyzing the impact of human activities within floodplains of regulated rivers and may assist in floodplain management aimed at combining water safety and nature policies.

Environmental drivers of space-time dynamics in floodplain vegetation: grasslands as habitat for megafauna in Bardia National Park (Nepal)

“…Uncertainties in our results could originate from the classification models used, their input data (field samples and satellite image composites), validation data (collected in the field and with Google Earth), the change detection algorithm, and the datasets of the environmental drivers. Although these factors pose 655 limitations to optimal classification accuracies, the accuracies are in line with similar studies (Hermosilla et al, 2018;Harezlak et al, 2020;Van Iersel, 2020). For a study in Canada a more extensive algorithm was used, including a Hidden Markov Model and logical transition rules.…”

“…This was caused by the LANDSAT 7 Scan-line error. The open access cloud platform software Google Earth Engine (Gorelick et al, 2017), earlier applied for mapping changes in floodplain ecosystems (Harezlak et al, 2020;Zurqani et al, 2018;Donchyts et al, 2016;Van Iersel, 2020), was used for selecting, processing and classifying LANDSAT imagery, whereas RStudio was used for post-classification analysis. 265…”

“…The Random Forest model (Breiman, 2001) was used for classification of the remote sensing images, considering its earlier successful use in mapping of vegetation in and near floodplains (Van Iersel et al, 2018;Belgiu and Drăgu, 2016;Harezlak et al, 2020) and most satisfactory results during initial tests. Two time series of land cover maps were created based on the 295 above-described classification schemes.…”

“…It is especially vital to understand the functioning of such a valuable ecosystem in relation to possible changes in drivers (such as the indicated redistribution of discharge in 2009) in order to provide insight for nature management 115 strategies aiming to conserve grasslands and associated fauna, such as the abundance of tiger's prey and the endangered tiger itself (Harihar et al, 2014), and also to expand on the knowledge for similar systems where grasslands sustain prey-predator relationships. A promising method for studying land cover dynamics and its drivers is the use of earth observation techniques, which enable 120 diachronic analysis contributing valuable data and insights on the development of land cover (Lallias-Tacon et al, 2017;Dufour et al, 2019;Harezlak et al, 2020;Solins et al, 2017;Basumatary et al, 2021;Louzada et al, 2020;Van Iersel, 2020;Corenblit et al, 2010;. With remotely sensed imagery from satellites, vegetation in and near floodplains can be mapped at the following scales and characteristics: vegetation types (Alaibakhsh et al, 2017), species composition (Rapinel et al, 2019;Plakman et al, 2020), physiological processes (Wagner-Lücker et al, 2013) and vegetation structure 125 (Straatsma and Baptist, 2008;Jalonen et al, 2015).…”

Initial biological development of a newly established side channel at the Lower Rhine

“…It is to be expected that the intense morphodynamic development of the side channel will decrease in the medium run and, therefore, pioneer habitats and species communities will decline for the benefit of more stable communities and proceeding vegetation succession (Geerling et al, 2008;Harezlak et al, 2020;Stoffers et al, 2020;van Geest et al, 2007van Geest et al, , 2005avan Geest et al, , 2005b.…”