115 years of sediment deposition in a reservoir in Central Europe: Topographic change detection

Reservoirs have become an important component in the worldwide river sediment flux. Reservoirs prevent downstream sediment transport and have become a major sediment sink. In this study, sediment deposition during the last 115 years in the Urft Reservoir in western Germany is reconstructed. The Urft Reservoir is the oldest reservoir in the Eifel Mountains and was almost completely drained in 2020. This enabled a detailed mapping of the lake bottom using an unmanned aerial system and the computation of a high‐resolution digital surface model. Topographic maps with a nominal resolution of 1:1000 from the time prior to the construction of the dam (around 1900) were used to construct a pre‐reservoir elevation model. A digital elevation model of difference (DoD) was calculated from these two datasets for the reservoir floor (0.72 km2). Based on the DoD, a net sediment accumulation of 1.16 × 106 m3 was calculated alongside a propagated volume error of 6.91 × 105 m3, resulting in a mean accumulation of 1.54 m. Conservative vertical error propagation results in an average level of detection (LoD) of 1.8 m. In contrast, the comparison of the DoD with 47 cores in the upper part of the reservoir showed a mean difference of −0.11 m, indicating a high, independently assessed accuracy of the DoD. Three depositional hotspots were identified in the reservoir. One is close to the Urft dam where very fine sediments are draped across the pre‐reservoir topography. Two areas are related to reservoir management. Sediment deposition in the Urft Reservoir has been comparably low in comparison to other regions globally, resulting in a 3.25% ± 1.93% loss of reservoir volume between 1905 and 2020. To analyse the effect of strong flooding events, a subset of the reservoir was analysed after an extreme event in July 2021, but accumulation did almost entirely not exceed the LoD.

The fluvial systems are particularly affected by these changes, influencing, for example, channel morphometry and sediment transport (Downs & Piégay, 2019;Lewin, 2013).Global climate change is resulting in changes in precipitation patterns as well as flood frequency and magnitude (Blöschl et al., 2017;Merz et al., 2021), which, in turn, is affecting soil erosion and sediment transport in rivers (Li & Fang, 2016;Mullan et al., 2019;Owens, 2020).River systems worldwide, but especially in Europe, have been directly and indirectly influenced by humans for several millennia (Brown et al., 2018;Dotterweich, 2008;Dreibrodt et al., 2010;Hoffmann et al., 2008;Macklin & Lewin, 2003).This influence has again increased significantly in recent centuries.Over the last 200 years, both the sediment flow in rivers and their catchments have changed drastically besides the geomorphological shape of rivers (Downs et al., 2013;Downs & Piégay, 2019;Goudie, 2020).
The construction of dams influences longitudinal sediment connectivity and in many rivers worldwide sediment transport downstream is reduced (Dethier et al., 2022;Fryirs, 2013;Kondolf, 1997;Vörösmarty et al., 2003).It is estimated that more than 45 000 (Dams, 2000) to nearly 60 000 dams (ICOLD, 2020) exist on earth.Other studies estimate around 2.8 million dams (Lehner et al., 2011); 95% of the total global reservoir capacity was constructed after 1950 (Syvitski et al., 2022).A recent study assumed, that nearly two-thirds of all major rivers with a length of more than 1000 km do not remain free-flowing on their total length, with especially high numbers in Europe and North America (Grill et al., 2019).Large reservoirs became one of the most important sediment traps, accumulating around 3200 Gt since 1950 (Syvitski et al., 2022).Accordingly, many rivers have a decreased sediment load, despite enhanced sediment mobilisation in numerous regions of the world (Kondolf et al., 2014;Li et al., 2020;Syvitski et al., 2005;Walling & Fang, 2003).Decreasing sediment load, in turn, often results in enhanced erosion downstream (Gierszewski et al., 2020;Schmidt & Wilcock, 2008;Singer, 2010;Syvitski et al., 2005).
More than 50% of the sediment transported in regulated rivers is potentially trapped in reservoirs (Vörösmarty et al., 2003).Sedimentation in reservoirs results in a loss of reservoir storage capacity, effectively reducing the efficiency of flood protection, energy production, drinking water supply (Wisser et al., 2013) and the construction's life time (Hargrove et al., 2010;Palmieri et al., 2001).At present, loss in storage capacity is higher than the capacity gain due to construction of new reservoirs (Oehy & Schleiss, 2007).Globally, 0.5% to 1% of the storage volume of reservoirs is lost due to sedimentation per year (Basson, 2009, cited in Schleiss et al., 2016;Lajczak, 1994).High loss rates have been observed especially in Southern Asia and in the Middle East, whereas low rates occur in North America and Europe (White, 2001).However, regional differences are large (White, 2001;Wisser et al., 2013).For Europe, several studies asses the sediment deposition in reservoirs (e.g.Vanmaercke et al., 2011;Verstraeten et al., 2006).An annual storage capacity loss of 0.26% is assumed on the basis of 352 reservoirs in Europe (Verstraeten et al., 2006).Much higher values were measured for reservoirs in Italy (Patro et al., 2022), which is in accordance with a high regional variability in sediment yield in Europe (Vanmaercke et al., 2011).
Despite an increasing number of research studies, there are still large uncertainties regarding the amount and spatial distribution of sediments in reservoirs (Fox et al., 2016;Vanmaercke et al., 2011;Verstraeten et al., 2006).Various methods have been used to reconstruct sediment distribution in reservoirs.These include bathymetric methods like sonar, sub-bottom profiler or ground penetrating radar (GPR) (Bennett et al., 2002;Bladé Castellet et al., 2019;Sedláček et al., 2022), direct surface measurements such as LiDAR or photogrammetry (Pacina et al., 2020;Randle et al., 2015;Schleiss et al., 2016).However, completely drained large reservoirs that allow for reservoir-wide surveys are rarely available.Thus, most direct measurements of reservoir surfaces related to sediment accumulation are only covering the upper part of a reservoir (Pacina et al., 2020) or accumulation behind check dams (Alfonso-Torreño et al., 2019;Huang et al., 2021;Rodrigues et al., 2021;Zeng et al., 2022).In addition, a 3D topographic change detection integrated over the total time span since reservoir construction requires a reconstruction of the prereservoir topography, which mainly relies on cores, GPR data or historical air photos and maps (Carley et al., 2012;Childs et al., 2003;Pacina et al., 2020).It has been frequently highlighted that knowledge of the topography prior to the sedimentation is crucial for a better understanding of the sedimentation process, distribution and volume (Childs et al., 2003;Odhiambo & Boss, 2004;Venteris & May, 2014).
In this study, the sediment deposition in the Urft Reservoir in the Eifel Mountains, western Germany, is analysed (Figure 1).It was the largest reservoir regarding storage capacity in Europe at the time of final construction in 1905.Due to maintenance work on its main outflow in 2020, the reservoir was nearly completely drained, enabling a detailed mapping of the present floor topography using an unmanned aerial system (UAS) and the photogrammetric construction of a high-resolution digital surface model (DSM).The objective of this study is to get detailed information on the amount of sediment deposition during the past 115 years and to evaluate the sediment distribution in a typical valley reservoir in central Europe.Furthermore, the loss of capacity, an important parameter for flood protection of the downstream area, will be calculated.To assess the valley topography before the filling of the Urft valley, a digital elevation model (DEM) was created based on topographic maps on a scale of 1:1000 from the years 1896 and 1897.In July 2021, the region was subject to a major flood event, resulting in large destruction in the catchment area upstream of the reservoir.To assess the impact of an individual extreme event on the sediment input, the reservoir was partly drained again in autumn 2021.This enabled the UAS-based photogrammetric creation of a second DSM from the upper segment of the reservoir.
Finally, the limitations and geomorphic implications of these datasets are discussed.

| STUDY AREA
The Urft Reservoir is located in the eastern Eifel Mountains, Germany (50 36' N, 6 26 0 E) (Figure 1).The reservoir has a length of 12 km and a maximum water volume of 45.51 Â 10 6 m 3 .Its catchment has a size of about 372 km 2 .The detailed study area was limited to the reservoir area between the Urft Dam and the reservoir kilometre (RKM) 9.2 in 2020 and extended to RKM 11.25 in 2021.RKM measures the distance in kilometres upstream from the dam along the thalweg.The Urft Dam, which is located in the lower part of the catchment, was constructed between 1900 and 1905.The Urft River has a length of 46.3 km and is the main inflow into the reservoir.Since the construction of the Olef dam in 1959, the same-named tributary and nearly 13% of the Urft catchment have been regulated.The mean annual inflow is 4.9 m 3 /s, equivalent to 1.54 Â 10 8 m 3 .The main outflow ($90%) is via the 2.7-km-long Kermeter Tunnel, an artificial outlet, which is providing water for a nearby electric power plant.
The reservoir is used for flood protection, hydropower and downstream water supply for industrial needs, causing varying water levels throughout the year.The median daily water level for the period 1960 to 2020 is 313.4 m (NHN, 'Normalhöhennull', the vertical datum in Germany) (Figure 2).A comparison of the water level and the inflow is presented in the Supplement Figure S1.The reservoir was partly drained in 1945, 1947, 1964, 1974, 1997 and 2020.No sediment removal from the reservoir is reported.Apart from the Urft, eight small and steep creeks with a maximum stream length of 3 km enter the lake.
The study area is under the influence of the Atlantic Westerlies with annual precipitation values of around 900 mm.In July 2021, the area was affected by a major flood event, which resulted in large devastations upstream of the reservoir.Regional precipitation exceeded 120 mm in 72 h (Dietze et al., 2022;Kreienkamp et al., 2021;Lehmkuhl & Stauch, 2023), resulting in the largest inflow in the Urft Reservoir in the past decades (Figure S1).
The area around the reservoir and in the southern part of the catchment mainly consists of Devonian siltstone and mudstone sequences, whereas in the central part of the catchment Devonian limestone predominates.Shallow Cambisols and Leptosols are developed in the area (Meyer, 2013;Walter, 2010).Large parts of the slopes and the upper parts of the Urft catchment are covered by broad-leaved and coniferous forests (Figure 3, Table S1) (Corine Landcover, 2018).The coniferous forest mainly consists of spruce which was introduced in the area in the 19th century (Montzka et al., 2021).Less inclined areas are used as pastures and for agriculture.Most of the villages are concentrated in the valleys of the Urft and Olef rivers.

| METHODOLOGY AND METHODS
To analyse sediment deposition during the last 115 years in the reservoir several DEMs were constructed.To evaluate the pre-reservoir topography historical topographical maps were used.UAS was applied to map the reservoir topography in 2020 and after the flood event in July 2021.To quantify topographic changes between these periods, DEMs of difference were calculated.A special focus was placed on a rigorous evaluation of the datasets.

| Digital elevation model 1897-Interpolation of elevation data from topographic maps
For the reconstruction of the pre-reservoir valley topography, a set of nine maps on a scale of 1:1000 was used.These maps were compiled by direct field surveys before the construction of the dam in the years 1896 and 1897 in order to assess the reservoir volume.The contour lines are spaced in intervals of 2 and 1 m on the hillslopes and flat valley bottom, respectively.The digitised map sheets were georeferenced using ArcGIS Pro 3.0 using 10 to 12 ground control points (GCPs) per map sheet.As the maps are missing a reference system, individual reference points had to be selected.The maps cover mainly areas that are not covered by the present-day national topographic maps as they are now regularly below the water level of the reservoir.Therefore, georeferencing was based on the contour lines of the DSM from the year 2020.Only contour lines in geomorphological presumably stable positions were selected (James et al., 2012).Thus, all GCPs are located on the slopes and none at the valley bottom.Due to the high accuracy of the maps, a first-order polynomial transformation could be used.The final reference system was ETRS89 UTM 32 N (EPSG: 25832).Each contour line was digitised manually on a working scale of 1:500.The total root mean square error (RMSE) for all nine maps amounts to 4.0 m.An error introduced by the digitisation estimated to range between a few to several tens of cm can be neglected.Topographic information from the contour lines was interpolated using the tool 'Topo to Raster' in ArcGIS Pro 3.0, which is based on the ANUDEM program (Hutchinson, 2008(Hutchinson, , 1989;;Hutchinson et al., 2011), and resulted in a hydrologically correct DEM with a spatial resolution of 1 m (DEM 1897 ) (Figure 4).The spatial resolution of 1 m was conservatively selected based on the distance of the contour lines and the scale of the nine maps (Henselowsky et al., 2021).
No ground control is available as no fix points can be reconstructed to assess the accuracy or precision of the DEM 1897 .The vertical error of topographic maps σ h is highly dependent on the slope inclination and can be estimated according to the adaptation of the formula of Koppe (1905) by Töpfer (1960): where M designates the scale number of the topographic map and α is the slope angle in degrees.We accordingly calculate estimates of the vertical error σ h,1897 for the DEM 1897 based on the slope raster derived from it.

| Digital surface models 2020 and 2021-Photogrammetric structure-from-motion (SfM) multiview-stereo (MVS) processing of UAS-based survey imagery
In October and November 2020, the entire reservoir downstream of RKM 9.2 was covered by 22 UAS flight plans surveyed in a double grid pattern using an FC6310S camera sensor with 70% frontal image overlap and 10 nominal camera angle (Table S3).Every plan was flown at 90 and 120 m (nominal) altitude above take-off in order to mitigate systematic topographic errors in steep terrain (Trajkovski et al., 2020), such as the widely described 'doming effect', which can arise from camera distortions (Eltner et al., 2016;James & Robson, 2014;Sanz-Ablanedo et al., 2018).All flights resulted in a total of 6,879 images.Prior to the flights, 154 markers for ground control were distributed across the valley floor (Figure 5) and surveyed with a Leica Zeno 20 GNSS receiver in real-time kinematic (RTK) and SAPOS connectivity and featuring a localisation accuracy of 1-2 cm horizontally and 2-3 cm vertically.The layout of the markers has been affected by limited accessibility as the lake sediments were still highly water-saturated; 139 of 154 markers were used for subsequent processing, 50% as GCPs and 50% as checkpoints (CPs), chosen randomly.The remaining 15 points were discarded due to very low accuracy.
In November 2021, the Urft Reservoir was drained partially, which enabled another UAS survey that covers the upper part of the reservoir from RKM 7.25 to RKM 11.25 (DSM 2021 ) (Figure 6) and has a significant overlap with the survey in 2020 between RKM 7.25 and RKM 9.2 in order to assess the effects of the July 2021 flood.In total, 12 flight plans yielded 1536 images for subsequent photogrammetric processing (Table S3).Ground control was assessed by surveying 240 locations in RTK mode using a Trimble R12 LT with a localisation accuracy of 1-2 cm horizontally and 2-3 cm vertically.Out of these, 171 points with total accuracy values ≤10 cm were used as GCP and CP.Consistent between both surveys in 2020 in 2021, photographic images were processed by SfM and MVS processing, which was carried out in the Agisoft software Metashape Professional, version 1.6.3.We followed the workflow described by USGS (2017) with adjustments by James et al. (2020) (cf.Walk et al., 2022).Photogrammetric processing results for the cameras, tie points and dense clouds are summarised in Table S4.
Resulting ground sampling distances (GSDs) for the surveys in 2020 and 2021 of 2.65 and 2.49 cm were yielded, respectively (Table S3), determining the spatial resolutions of the corresponding orthomosaics.Spatial resolution of the DSM amount to twice the GSD, thus 5.
The vertical uncertainty of the GNSS receiver used, σ z,dGNSS , was then propagated to assess the vertical errors σ z,2020 and σ z,2021 for 2020 and 2021, respectively: Although a constant σ z,dGNSS of 2 cm is reported for the Leica and the DoD between November 2020 and December 2021 as The DoD 2020-1897 was evaluated for the reservoir floor between  or DSM used as input elevation data (Brasington et al., 2003;Carley et al., 2012;Lane et al., 2003).On a 95% confidence level, LoD 0.95 is given as where σ z,1 and σ z,2 represent the vertical standard deviations of error for both points in time and given a random distribution of the vertical precision across the survey area (James et al., 2017(James et al., , 2020)).We calculated the LoD 0,95, 2020-1897 for the DoD 2020-1897 based on the spatial interpolation of σ z,1897 and σ z,2020 and the LoD 0,95, 2021-2020 for the DoD 2021-2020 using the vertical error estimates σ z,2020 and σ z,2021 .
Based on the DoDs, the total net topographic change volume ΔV in the reservoir can be integrated over the respective floor areas: with Δz i,DoD representing the vertical change in z stored in each raster cell i and x, the spatial resolution of the DoD raster.The error for the volume change σ V was then propagated based on the vertical errors σ z,1 and σ z,2 and integrated over all raster cells i: The volumetric errors σ V,2020-1897 and σ V,2020-2021 evolve from the corresponding vertical error estimations σ h,1897 , σ z,2020 and σ z,2021 .
The range, mean and RMSE of the vertical error σ z,2021 for the DSM 2021 amount to 0.02-0.56,0.13 and 0.09 m, respectively (Table S5, Figure S3).

| Surface characteristics
The

| Sediment deposition-Topographic change detection between 1897 and 2020
The water-covered areas in front of the Urft Dam and the slopes of the valley were excluded, resulting in a floor area of 0.753 km 2 (Figure 6).Throughout the reservoir, a declining trend in sediment accumulation can be observed, except at the lowermost part of the study area where accumulation increased again (Figure 8).Besides these general trends of accumulation, two zones of higher accumulation with deposits of more than 4 m outside of the former Urft channel were distinguished.These are between RKM 5.4 and RKM 5.9, and RKM 7 to RKM 8.5, and have sizes of 10.5 Â 10 3 m 2 and 13.5 Â 10 3 m 2 , respectively (Figures 6 and 8).They consist of several large stretches of sediments, which are located on the former floodplain.Maximum accumulation in these zones is up to 6 m.Both zones can be assigned to typical operation patterns in the transition zone.The lower one at RKM 5.4 to RKM 5.9 is close to the minimum water level in the reservoir during regular operations at around 300 m (NHN) (Figure 2).The second zone is related to a water level between 306 and 308 m (NHN), which is normally reached in autumn after the reservoir was prepared to accommodate the late winter and early spring floods.
According to the sediment distribution in the Urft valley, the main sediment input originates in the catchment upstream of the reservoir.
None of the small creeks entering the reservoir has accumulated a notable amount of sediment.Several landslides of unknown age are visible at the flanks of the valley, but they also did not result in a considerable sediment input.However, several small slumps with a length of several tens of meters have been observed at the bank of the Urft River.Consequently, sediment remobilisation also occurred especially during extreme low water periods like in November 2020.
Only six sites show considerable erosion according to the DoD 2020-1897 .All of these sites are related to human activity, especially quarrying (Figure 9).As the sediments were removed Consequently, results from the topographic change detection between 2021 and 2020 were not used for further analysis.

| DISCUSSION
We first discuss the quality of our data, before analysing the spatial pattern and sedimentation rates in comparison to other reservoirs.
Finally, the underlying landscape changes are discussed.

| Uncertainties of the topographic change detection
Comparing historical maps of the relief with other survey methods was successfully used previously for purposes of topographic change detection (Carley et al., 2012;Pacina et al., 2020); however, to the author's knowledge, this study is unprecedented in regard to a topographic change detection over a time span exceeding a century and relying on a UAS survey covering over 250 ha (2020).The DEMs constructed from historical maps have a higher spatial resolution than DEMs constructed with bathymetric methods, as these methods are partly limited by water depth and sediment thickness (Liu et al., 2020;Pacina et al., 2020).However, large-scale maps with high accuracy In comparison to, for instance, a multi-methodical geomorphic change detection over 6 to 7 decades conducted by Carley et al. (2012) along the lower Yuba River in the southwestern United States, the exclusion of 56% is-depending on the exclusion criteria-1.3-3.5 times higher, yet still allows to deduce meaningful interpretations despite the overall low sedimentation rate.As we avoided a refinement of the DoD 2020-1897 by subtraction of the LoD (Carley et al., 2012;Milan et al., 2011), our quantification of the error alongside the topographic change can be considered conservative.The independent vertical accuracy assessment of the DoD 2020-1897 by 47 cores is by a factor of $3 higher than the conservative LoD 0.95 and, consequently, can be deemed satisfactory.
In comparison to the 2020 SFM-MVS model, the mean CP accuracy of the 2021 model is by two to three orders of magnitude higher for both the horizontal and vertical orientation (Table S5).Although the altimetric relative precision is, with 4.4 (GSD units), still relatively low but acceptable, planimetric relative precision is one order of magnitude higher.Higher altimetric than planimetric uncertainties are values between RKM 9.2 and RKM 7.25 (Figure S2).Towards the SE, the distribution of ground control markers measured in 2020 is comparatively sparse.Especially for large UAS-based SfM-MVS models, significant systematic topographic errors have been regularly described and can have especially large effects in the marginal model areas with poor GCP coverage (e.g.Eltner et al., 2016;James et al., 2017;e.g. James & Robson, 2014;Sanz-Ablanedo et al., 2018).

| Spatial sediment pattern and sedimentation rates in the Urft Reservoir
The net volume change is 1.16 Â 10 6 m 3 since 1897.However, the conservative volume error amounts to 6.91 Â 10 5 m 3 , which is 50.1% of the total absolute volumetric change.Because of anthropogenic sediment removal between 1897 and the filling of the reservoir in 1905, a total sediment accumulation of 1.27 Â 10 6 m 3 is probably more realistic estimate for the sediment budget.Due to the reservoir size, the seldom and low outflow through the bottom outlets and the predominant outflow through the Kermeter Tunnel, where the water is directly used for electricity generation, it can be assumed that most of the sediment, which is transported into the Urft Reservoir remains there.However, no direct measurements at the bottom outlets/ outflow points are available.
Accumulation mainly occurred in two areas (RKM 5.4 to RKM 5.9, and RKM 7 to RKM 8.5) in the upper part of the reservoir and, to a lesser degree, close to the Urft Dam.Despite these accumulation hotspots, a general reduction of the thickness of the deposits can be observed towards the lower part of the reservoir.This gradient is typical for elongated reservoirs and has been frequently observed before (Dhivert et al., 2015;Sedláček et al., 2016).Typical delta deposits, as described in other studies of reservoir deposition (Fan & Morris, 1992), could not be observed-neither on the main inflow nor on the tributaries to the reservoir.Strong lake level fluctuations, either natural or human-induced are an important factor in sediment distribution (L opez et al., 2016;Shotbolt et al., 2005;Snyder et al., 2004) as they are causing different zones of a reduction in flow velocity (Schleiss et al., 2016).In the Urft Reservoir, the water level is controlled by anthropogenic and natural factors.A main task of the reservoir, besides the constant production of energy and water for industrial purposes further downstream, is flood protection.Therefore, the water level is regularly lowered in late autumn (Figure 2), providing enough storage capacity for winter and spring floods.Summer droughts are another important factor.Anthropogenically lowered water levels are generally 10-15 m lower than the summer water levels.Consequently, two main water levels are responsible for the sedimentary environment in the reservoir, one during spring and early summer at around 300 m (NHN) and another in late autumn and early winter at 306-308 m (NHN).In the lower part of the reservoir, low flow velocity is resulting in the accumulation of finer sediments (Shotbolt et al., 2005;Snyder et al., 2004;Tarela & Menéndez, 1999).As a result, the Urft Reservoir exhibits a spatially heterogeneous depositional structure.
Besides the main trends in sediment deposition, a clear influence of the pre-reservoir topography is evident.During most of the course, the position of the channel of the Urft River has been preserved since 1897.This persistence of the river channel is presumably caused by the ongoing water flow at the bottom of the reservoir (Loizeau & Dominik, 2000;Shotbolt et al., 2005).These river channels are the main sediment transport pathways in narrow reservoirs especially during flood events (Cesare et al., 2001;Oehy & Schleiss, 2007).However, the present-day channel is considerably narrower than the one mapped in 1897.Sediments were deposited at the sides of the channel as well as on the former floodplain, indicating a quick reduction in flow velocity (Mulder et al., 1998;Scheu et al., 2015).In contrast, considerably less sediments can be found on the higher terraces.Similar depositional patterns were observed in a study of several reservoirs in the Czech Republic (Sedláček et al., 2016).Because of the reservoir operation, the upper part is frequently falling dry, and lacustrine and fluvial processes are thus alternating (Sedláček et al., 2022;Snyder et al., 2004).As a result, frequent remobilisation of sediments occurs.
Even stronger sediment remobilisation may occur during reservoir drainage (Shotbolt et al., 2005).In 2020, especially in the middle and downstream segments of the reservoir, considerable zones of bank instability and slumping were observed along the banks of the Urft channel.Relocation of sediments is an important factor in reservoir sediment distribution (Bladé Castellet et al., 2019;Lambert & Giovanoli, 1988) and similar sedimentation patterns as in the Urft Reservoir were also observed in the Seč Reservoir (Sedláček et al., 2022).
As most of the sediment remains in the reservoir, the Rur River further downstream experiences a sediment deficit and resulting river changes (Schmidt & Wilcock, 2008;Syvitski et al., 2005) and coarsening of the river bed by clearwater erosion (Kondolf, 1997).
During the last 200 years, a reduction in sediment transport and changes in meander development was observed in the downstream area (Nilson, 2006;Wolf et al., 2021) (Sedláček et al., 2022).Similar rates of around 2.9 and 2.5 cm/year have been obtained for the Malter Reservoir in Germany and a reservoir at the Lot River in France, respectively (Audry et al., 2004;Müller et al., 2000).SAR in the Urft Reservoir are at the lower end of these examples, taking the average accumulation rate of 1.3 cm/year into account.During the last 115 years, the reservoir capacity was reduced by 3.25% ± 1.93%, resulting in 0.028 ± 0.017%/ year, based on the study area.This is much lower than the global average loss of storage capacity, which has been estimated to be around 0.5 to 1%/year (Basson, 2009, cf. Schleiss et al., 2016) and also a magnitude lower than the capacity loss in Europe, which amounts annually to 0.26% (Verstraeten et al., 2006).
The comparably low sedimentation in the Urft Reservoir is argued to be related to the relatively stable landscape upstream.Since the 19th century, considerable reduction in agriculture on the slopes of the Urft valley and substantial afforestation took place.The area of arable land was reduced by 80%, whereas the extent of forest and grassland roughly doubled (Nilson & Lehmkuhl, 2007), resulting in reduced sediment transport in the rivers of the Eifel mountains (Esser et al., 2020).
Furthermore, the construction of the Olef Dam in 1956 at the main tributary of the Urft might also have led to a reduced sediment flow from the Eifel Mountains into the Urft Reservoir.Sediment input into the reservoir by small creeks can be another important factor for sediment accumulation (Abraham et al., 1999).However, direct sediment transport to the lake from the surrounding slopes is quite limited, as no major deposits were detected.The forest around the lake has been reported to be stable for several centuries and has been converted to a National Park in 2004.Some of the beech trees have an age of over 200 years (Tebaldini et al., 2023).A protecting land cover, for example a stable forest, has a positive effect on preventing soil erosion and downstream sediment transport (Brown et al., 2018;Cebecauer & Hofierka, 2008;Cerdan et al., 2010;Dotterweich, 2008).

| The July 2021 flood event
The 2021 flooding was a severe event in the area caused by intensive rainfall from the 12th to the 15th of July 2021 in the northern Eifel Mountains (Kreienkamp et al., 2021).The event resulted in large devastations and prominent morphological changes (Dietze et al., 2022;Lehmkuhl & Stauch, 2023).Also, the Urft River valley was heavily affected by strong flooding and severe destructions in several villages upstream of the reservoir.Large amounts of debris as well as sediments were transported into the lake and deposited on the valley floor and the sides of the reservoir.However, the deposition in the study area was too low to exceed the LoD 0.95 .LoD 0.95 values in the order of some decimetres are in agreement with our observation during the mapping in 2021.Only at a few sites, an accumulation of more than 50 cm was observed.Therefore, a meaningful spatial evaluation of the consequences of the flood event on sediment deposition was unfeasible.
However, these study areas were considerably smaller than the Urft Here, fine sediments accumulate and form a homogenous cover on the pre-reservoir topography.
In a subset of the study area, the effect of a major flood event

F
I G U R E 2 Frequency distribution of the mean daily water levels, binned in 1 m intervals, from January 1960 to November 2022, showing the average yearly and quarterly management of the Urft Reservoir.The long-term median water level ($313 m NHN) and seasonal peaks in the water level distribution are highlighted.[Color figure can be viewed at wileyonlinelibrary.com] 3 and 4.97 cm for the DSM 2020 and DSM 2021 , respectively.Ground control accuracy and precision are summarised in Table S5.In accordance with the very low horizontal CP accuracy and especially precision, a significant planar offset could be observed for both the DSM 2020 as well as DSM 2021 in relation to the digital F I G U R E 3 Land cover in the Urft River catchment based on CLC 2018 (see Tables S1 and S2 for class size and class aggregation).[Color figure can be viewed at wileyonlinelibrary.com] orthophoto web map service (DOP NRW) provided by Geobasis NRW (2021).Therefore, both DSMs were additionally co-registered to the DOP NRW using spline transformation.For the DSM 2020 200 reference points were placed resulting in a total RMSE of 0.25 m, whereas 50 reference points were used to co-register the DSM 2021 yielding a total RMSE of 0.002 m.Subsequently, the vertical accuracy of the co-registered DSMs, Δz 2020 and Δz 2021 , was assessed by calculating the difference between the z-values of the ground control measurements z GC deemed sufficiently accurate (n = 139 for DSM 2020 , n = 240 for DSM 2021 ) and the z-values extracted from the final DSMs z DSM : Zeno 20 GNSS receiver used for the DSM 2020 , individual values of σ z,dGNSS , ranging from 1 to 5.5 cm, were recorded with the Trimble R12 LT GNSS receiver applied in 2021.Eventually, we conducted a simple spatial interpolation of the vertical errors σ z,2020 and σ z,2021 , quantified at the ground control markers, using the 'Geostatistical Analyst' extension of ArcGIS Pro 3.0.0and ArcGIS Desktop 10.7.1.11614.We applied inverse distance weighting (IDW) to an optimised power of 2.03 and 2.05 yielding RMSE for the interpolation of 0.23 m and 0.08 m for σ z,2020 and σ z,2021 , respectively.Spatial vertical errors were extrapolated to the extent of the respective DSMs and gridded with a spatial resolution of 1 m.DSM 2020 and DSM 2021 were resampled using cubic convolution to spatial resolutions of 0.25 as well as 1 m.The 1 m DSM 2020 and DSM 2021 have subsequently been used for topographic change detection.

3. 3 |
Topographic change detection Topographic change between 1897, 2020 and 2021 was quantified by calculating two DEMs of difference (DoD) restricted to the reservoir bottom.As the effect of vegetation and anthropogenic structures can be neglected within the reservoir, both SfM-MVS-based DSMs (DSM 2020 and DSM 2021 ) can be considered equivalent to DEMs and F I G U R E 4 Digital elevation model of the Urft Reservoir from the year 1897 (DEM 1897 ) based on topographic maps with a scale of 1:1000.An example of the maps from the central part of the reservoir is plotted in the upper right corner.[Color figure can be viewed at wileyonlinelibrary.com] thus served as input alongside the historical DEM 1897 .The DoD between 1897 and 2020 is given as RKM 1.35 and RKM 9.2, which was mapped based on the coregistered orthomosaic of 2020.Topographic change between the dam and RKM 1.35 was not considered due to residual water.Analysis of the DoD 2021-2020 was restricted to the floor area overlapping between the DSM 2020 and DSM 2021 extending from RKM 7.25 to RKM 9.2.In addition, the water-filled Urft channel was excluded from the further analysis.

F
I G U R E 5 (a) Digital surface model of the Urft Reservoir from the year 2020 (DSM 2020 ) and the orthomosaic (upper right) based on UAS flights.(b) Longitudinal analysis of the DSM 2020 (i) along the course of the Urft River (blue) showing the measured profile (continuous line) between RKM 9.2 and RKM 1.365 and inferred profile range (dashed lines) further downstream and (ii) along a 100 m wide, mid-valley-centred swath profile, binned in 256 (length: 27.3 m) cross-segments and displaying the median elevation (black continuous line), interquartile range (IQR; dark orange checked pattern) and minimum elevation (black dashed line); vertical exaggeration is approximately 50-fold.[Color figure can be viewed at wileyonlinelibrary.com]To evaluate the significance of any topographic change in the DoDs, vertical changes can be compared to a specified 'level of detection' (LoD), which depends on the vertical uncertainties of each DEM the DoD 2021-2020 features a consistent negative linear trend oriented from S-SSE to N-NNW, we applied, cropped to the overlapping floor area, a detrending by subtracting a first-order global polynomial interpolation raster (RMSE = 0.11 m) based on cubic resampling of the DoD 2021-2020 to 5 m.F I G U R E 6 DoD 2020-1987 of the Urft valley floor and the comparison with the cores (upper right, see also Figure S4).The black boxes A and B indicate areas with higher sediment deposition, and the black lines indicate the transects shown in Figure 7. Red circles with Roman numbers designate detailed figures with anthropogenic influence on sediment deposition (Figure 9).[Color figure can be viewed at wileyonlinelibrary.com] 3.4 | Evaluation of the DoD 2020-1897 based on sediment coresAt 47 sites in the upper part of the reservoir, between RKM 6.8 and RKM 8.9, the sediment thickness was estimated manually in 2020 parallel to the UAS flights (Figure7).At 39 locations, a metal stick was hammered in the soft sediments.At eight sites, cores were taken using a Russian peat corer, equipped with a 50-cm-long chamber.The peat corer was then repeatedly extended and used at the same location to obtain a continuous sediment sequence.The evaluation of the sediment thickness could only be done in the upper part of the reservoir due to safety concerns further downstream.Coring sites were selected according to the valley geomorphology and in order to cover several terraces along cross-transects.Coring was conducted until no further drilling was possible.However, due to sediment compaction and the diffuse contact zone between the lacustrine and older fluvial sediments, the obtained depths are estimates.Constituting an independent accuracy measure for the sedimentation between 1905 and 2020, the coring depth d c was then subtracted at each coring site from the respective value in the DoD 2020-1897 , Δz DoD,2020À1897 : δz DoD,2020À1897 ¼ Δz DoD,2020À1897 À d c : ð9Þ 4 | RESULTS The vertical accuracy of the different datasets is presented first before the reservoir topography is evaluated.The topographical changes are described and evaluated for the DoD 2020-1897 and the DoD 2021-2020 .4.1 | Vertical DEM/DSM accuracy Estimates of the vertical error σ h,1897 of the DEM 1897 reconstructed from maps depicting the pre-reservoir topography range for the mostly gently inclined to flat reservoir floor between 0.83 and 1.28 m, F I G U R E 7 Cross-sections through the Urft valley showing the elevation in 1897 (blue), 2020 (orange) and the accumulation based on the DoD 2020-1897 (grey).The scaling of the accumulation is equal in all panels.Note that the LoD is not considered in the cross sections.For the location of the sections, see Figure 6.[Color figure can be viewed at wileyonlinelibrary.com] featuring a mean and standard deviation of 0.87 and 0.04 m, respectively.Due to the direct dependence on the slope angle for the accuracy estimation, increased σ h,1897 is yielded where the inclination of the reservoir floor is relatively high.For the UAS-based SfM-MVS-generated and additionally coregistered DSM 2020 , vertical σ z,2020 lies between 0.02 and 1.55 m, with a mean of 0.09 m and a RMSE of 0.30 m (Table Urft Reservoir is a classical valley-shaped reservoir with a small width.In the upper part of the reservoir, the valley floor has a width of around 55 m, and it increases up to 65 m in the downstream part.A notable exception is the area south and west of the Krummenauel Island (RKM 7.2), featuring a valley floor width of around 110 m.The depositional area in the reservoir is characterised by a flat valley bottom with a relatively constant longitudinal slope of 0.3 (Figure 5b) and steep slopes on both sides.Directly beside the Urft River, a floodplain is developed throughout the whole reservoir.At locations with a wider river valley, a higher fluvial terrace one to two meters above the floodplain is preserved.An additional terrace up to 8 m above the floodplain is only recognisable around Krummenauel Island (e.g.RKM 7.2) (Figure 7, CS03).
Smaller water-covered areas in the basin were ignored, as these ponds were generally very shallow.Total sediment accumulation on the valley floor of the reservoir is estimated based on the DoD 2020-1897 to be 1.27 Â 10 6 m 3 .Negative volume changes by erosion were much smaller with À1.11 Â 10 5 m 3 , resulting in a net volume change of 1.16 Â 10 6 m 3 for the last 115 years.Error propagation yielded a volume error σ V,2020-1897 of 6.91 Â 10 5 m 3 , which is equivalent to 50.1% of the absolute topographic change volume.Individual values for the DoD 2020-1897 lie between À19.05 and 6.87 m with a mean of 1.54 m.Most of the values (96%) range from À1 to 5 m; 44% of the floor area surpasses the conservative LoD 0.95,2020-1897 , which ranges between 1.63 and 3.02 m (mean LoD 0.95,2020-1897 = 1.8 m).Based on the DoD 2020-1897 , the reservoir capacity has reduced by, in total, 3.25% ± 1.93%, or by 0.28‰ ± 0.17‰/year.In comparison, the 47 cores drilled to evaluate the DoD 2020-1897 result in an arithmetic mean and standard deviation of Δz DoD,2020-1987 of À0.11 and 0.49 m, respectively (FigureS4).An absolute difference of more than 1 m was observed in two cores between RKM 8.75 and RKM 9.At this location, the model assumes an accumulation of around 4 m, whereas coring depths of 255 and 265 cm were reached.It can be assumed, that these two cores do not reach the base of the sediments as the high water content close to the present-day river inhibited deeper coring.Although the general topography from the pre-reservoir times is preserved, the main geomorphic effect of the sedimentation is the considerable narrowing of the river channel.In 1897, the Urft River had a mean width of around 21 m, whereas the channel features a mean width of around 10 m in 2020, indicating substantial sediment accumulation in the former river bed.High accumulation values have also been observed on the former floodplain of the Urft River, especially in the upper part of the study area (Figure7, e.g.CS 01 and CS 02).Generally, less accumulation occurred on the different terrace levels.Sedimentation here is up to a few tens of centimetres on the highest (third) terrace and up to 2 m on the former floodplain (e.g.Figure7, CS 03).Higher values occurred in the former Urft River channel, either due to narrowing or to a relocation of the channel.A relocation of the channel occurred, for example at RKM 4.5 and RKM 5.5 (see also Figure7, CS05).At both sites, more than 4 m of sediment accumulation occurred since 1897.However, in most parts of the reservoir, the present-day channel is following the Urft channel as it was mapped in 1897.In the lower part of the reservoir, a lower but notably symmetrical accumulation occurred (Figure7, CS 06).
between the time of mapping in 1896/1897 and the infilling of the reservoir in 1905, it can be assumed that this material was used for the construction of the dam.Additional topographic changes can be observed close to the dam at the lower part of the study area.Here, bomb craters from World War II are preserved on the valley floor.However, these craters have a depth of up to 50 cm and are not detectable in the DoD 2020-1897 .The influence of the craters on the F I G U R E 8 Longitudinal analysis of the DoD 2020-1897 along a 100 m wide, mid-valley-centred swath profile, binned in 256 (length: 27.3 m) cross-segments, extending from RKM 9.2 (right) to RKM 1.365 (left).Displayed are (i) the median net change (green) for the total DoD and for topographic change exceeding the LoD 0.95 (continuous and dashed green lines, respectively), for the areas affected by (ii) sedimentation, the median (dark blue) and 95th percentile (P95, light blue) for the total positive DoD and for sedimentation areas exceeding the LoD 0.95 (bars and marker, respectively), and for areas affected by (iii) erosion, the median (dark red) and 5th percentile (P5, light red) for the total negative DoD and for erosion areas exceeding the LoD 0.95 (bars and marker, respectively).Cross-segments with affected areas <10 m 2 were ignored.[Color figure can be viewed at wileyonlinelibrary.com]F I G U R E 9 Anthropogenic influence on the topography.I: Large quarry and the old road through the Urft valley (background: DSM 2020 ).II: A removed terrace edge where 8 m of sediments were removed (background: DSM 1897 and DoD 2020-1987 , for the colour legend, see Figure 6.III: Bomb crater from World War II, the black line indicates the section (background: DSM 2020 ).Locations are shown in Figure 6.[Color figure can be viewed at wileyonlinelibrary.com] sediment budget can be ignored.Summarising the accumulation and erosion values, it can be stated that most of the erosion totalling À1.11 Â 10 5 m 3 is related to anthropogenic sediment removal prior to 1905, followed by fluvial sediment accumulation occurring mainly in the 115 years between 1905 and 2020.Consequently, the net volume change is biased by human influence prior to the construction of the dam and the total sediment accumulation on the valley floor of 1.27 Â 10 6 m 3 is a better assumption of the sediment budget in the Urft Reservoir since 1905.4.4 | Topographic effects caused by the July 2021 flood The topographic effects of the extreme flood event in July 2021 were assessed based on the detrended DoD 2021-2020 produced for the upper part of the reservoir between RKM 9.2 and RKM 7.However, despite the directly observed sediment accumulation which reached up to 30 cm in this part, topographic changes for almost all of the detrended DoD 2021-2020 (99% of the area) are insignificant at a 95% confidence level (Figure 10).Although, the LoD 0.95,2021-2020 , ranging between 0.11 and 2.22 m (mean LoD 0.95,2021-2020 = 0.58 m), indicates an uncertainty which is $2.5 to 3 times smaller than the one for the DoD 2020-1897 , sediment accumulation due to the flood event cannot be reliably reproduced by the reservoir-wide UAS-based and detrended DoD 2021-2020 .Accordingly, it yields a sedimentation volume of 4.83 Â 10 3 m 3 , which is almost exactly counterbalanced by a negative topographic change volume by erosion of À4.86 Â 10 3 m 3 and, hence, sums up to a minimal net volume change of 2.5 Â 10 1 m 3 .In addition, the propagated volume error σ V,2021-2020 of 4.5 Â 10 4 m 3 corresponds to 4.6 times the absolute topographic change volume.
from the time prior to the construction are seldom.Given the constant scale factor of M = 1000 for all historical topographic maps used, estimation of the vertical uncertainty of the DEM 1897 by the formula of Töpfer (1960) only depends on the slope angle and result in minimum estimates of σ h,1897 = 0.83 m for a horizontal surface, whereas it exceeds 1 m in steep slopes of at least 34 .These values are relatively large in comparison to the other studies (Pacina F I G U R E 1 0 Left: DSM of the upper Urft Reservoir between RKM 11.25 and RKM 7 from the year 2021 (DSM 2021 ) and the orthomosaic (upper right) based on UAS flights.The black box denotes the overlapping area with the DSM 2020 .Right: Detrended DoD 2021-2020 for the (overlapping) reservoir section between RKM 9.25 and RKM 7. [Color figure can be viewed at wileyonlinelibrary.com] et al., 2020).We thus argue that σ h,1897 needs to be considered a conservative estimate of the vertical error.DSM based on several thousands of low-cost UAS images and covering an area of several hundreds of hectares are very rarely constructed due to the long survey times and, in particular, time-consuming SfM-MVS processing.Sanz-Ablanedo et al. (2018) surveyed a mining area with a size of over 1200 ha using a similar number of GCPs (100+) but significantly fewer photos (2500+).In particular, the horizontal CP accuracy and precision of the 2020 SfM-MVS model are very low (x = 0.47 m ± 3.18 m, y = À5.91 m ± 17.84 m and z = À0.05m ± 0.73 m; TableS5).Accordingly, relative planimetric precision exceeds those achieved in the large-scale study bySanz-Ablanedo et al. (2018) by two orders of magnitude.We explain this by the missing satellite reception of the GNSS receiver over large stretches at many footslopes of the deeply incised Urft valley.In addition, accessibility to the drained reservoir floor was largely restricted downstream of RKM 5.5.Although the overall GCP density of 1.86 (h = 90 m) and 2.92 (h = 120 m) can be considered acceptable(Sanz-Ablanedo et al., 2018), the challenges resulted in an irregular control point network with segments of several hectares that could not be covered by ground control (Figure5).Moreover, we argue that the complex layout of the 2 Â 22 flight survey missions along the meandering valley provides a critical image constellation for the bundle adjustment in Agisoft Metashape Professional.Subsequent planar co-registration of the DSM 2020 to the DOP NRW(Geobasis NRW (2021)) resulted in a significant increase of the vertical precision from an RMSE of 0.73-0.30m (σ z,2020 ), which is substantially exceeded by the mean Δz 2020 (0.09 m; cf.James et al., 2020) and can be considered sufficient given the challenging circumstances(Sanz-Ablanedo et al., 2018;Tonkin & Midgley, 2016).LoD 0.95 for small areas of interest of single to few hectares and surveyed exclusively by UAS and processed by state-of-the-art SfM-MVS workflows range between 1.4 and 4.0 GSDs, equivalent to a few centimetres(James et al., 2020).The mean and range of the LoD 0.95,2020-1897 amount to two orders of magnitude larger (mean = 1.8 m, range: 1.63-3.02m), which is strongly determined by the propagation of the vertical error estimates for the DEM 1897 , featuring a mean almost thrice as large as those for the vertical precision for the DSM 2020 .The percentage of reservoir area that got excluded by applying the LoD at a 2σ confidence level is 56%.
unusual(Sanz-Ablanedo et al., 2018) and for those reasons, subsequent co-registration was deemed necessary also for the DSM 2021 .As to be expected, the DoD 2021-2020 , generated entirely based on UAS surveys, is characterised by a significantly, 2.5-to 3-fold smaller LoD 0.95 (mean = 0.58 m; range: 0.11-2.22m) than the DoD 2020-1897 , yet still high in relation to small-scale UAS surveys(Cook, 2017;James et al., 2020), despite using a relatively high GCP density >7 GCP per 100 images.The consistent linear trend observed in the original DoD 2021-2020 is apparently produced due to a systematic topographic distortion of the DSM 2020 in the overlapping area, visible in the σ z,2020 Reservoir.The surface of the almost completely drained Urft Reservoir was surveyed in 2020.The resulting DSM 2020 has a spatial resolution of $5 cm.Additionally, a DEM with a spatial resolution of 1 m was constructed from topographical maps with a scale of 1:1000 from the years 1896 and 1897 (DEM 1897 ).Using these datasets as input, a model of difference (DoD 2020-1897 ) was computed to quantify sediment accumulation across the reservoir floor.However, due to the conservative approach to estimating the vertical uncertainties of the two elevation models, especially the DEM 1897 , 54% of the topographic change at the reservoir floor did not exceed the LoD at 95% confidence (LoD 0.95 ).Considering accumulation and erosion since 1897, the net volume change in the reservoir amounts to 1.16 Â 10 6 m 3 ± 6.91 Â 10 5 m 3 since 1897.However, sediment removal mainly occurred between 1897 and 1905.Net sediment accumulation since the filling of the reservoir in 1905, the last 115 years, is thus considered to be close to the total sediment accumulation of 1.27 Â 10 6 m 3 .The mean accumulation in the reservoir is 1.54 m, resulting in an SAR of 1.3 cm/year.In comparison to other reservoirs in central Europe, this is a low accumulation rate.Consequently, the small relative loss in volume of 3.25% ± 1.93% since 1897 has no major impact on the lifetime of the Urft Reservoir.The relatively low accumulation rate can be attributed to LULC changes in the catchment upstream of the reservoir.During the past 200 years, arable land was converted into pastures and forests.Varying water levels due to reservoir operation resulted in two accumulation hotspots.These two areas have an accumulation of up to 6 m.A third area of high sediment accumulation has been observed close to the Urft dam.
was assessed by comparing the DSMs from 2020 and 2021.However, despite being the largest flood event for at least a century in the Urft area and massive destructions of infrastructure and river banks reported in the upstream area, sediment accumulation in the reservoir was almost entirely below the LoD 0.95 and, thus, mostly smaller than several decimetres.Field observations from the year 2021 confirmed the relatively low flood-related sediment accumulation.AUTHOR CONTRIBUTIONS Georg Stauch: Conceptualisation; funding acquisition; investigation; writing-initial draft; writing-reviewing and editing.Lukas Dörwald: Methodology; investigation; resources; writing-initial draft; writingreviewing and editing.Alexander Esch: Methodology; investigation; resources; writing-initial draft; writing-reviewing and editing.Janek Walk: Conceptualisation; methodology; investigation; writinginitial draft; writing-reviewing and editing.