Climate plays a dominant role in erosional and sediment transfer processes in Alpine catchments (e.g. Huggel et al., 2012; Micheletti and Lane, 2016; Palazón and Navas, 2016). In such environments, three main hydroclimatic forcings drive the processes that contribute to suspended sediment concentration (SSC) along channels: erosive rainfall, glacial melt, and snowmelt. Erosive rainfall (ER), defined here as liquid precipitation over snow-free surfaces, is responsible for soil detachment and erosion along hillslopes (Wischmeier, 1959; Wischmeier and Smith, 1978), triggering mass wasting events such as debris flows and landslides (e.g. Caine, 1980; Dhakal and Sidle, 2004; Guzzetti et al., 2008; Leonarduzzi et al., 2017), which can mobilize large amounts of fine sediment (e.g. Korup et al., 2004; Bennett et al., 2012) and result in very high suspended sediment concentrations in the receiving streams. Together with erosional processes along hillslopes, which are strongly related to rainfall intensity (e.g. Van Dijk et al., 2002), precipitation events may also enhance channel and bank erosion through increased discharge. Ice melt (IM) is responsible for high concentrations of fine sediment produced with a variety of glacial erosion processes (Boulton, 1974). Ice melt may substantially increase suspended sediment concentration in glacially fed streams by entraining and transporting fine sediment previously stored in subglacial networks and paraglacial environments (Aas and Bogen, 1988; Gurnell et al., 1996; Lawler and Dolan, 1992). Snowmelt-driven overland flow (SM) generates hillslope erosion and potentially affects channel and bank erosion by contributing to streamflow. This hydroclimatic forcing is important in Alpine environments where snowmelt can produce high hillslope runoff and be a major contributor to channel discharge (e.g. Grønsten and Lundekvam, 2006; Ollesch et al., 2006; Konz et al., 2012). Due to the diversity of the erosion and transport processes (e.g. erosion driven by overland flow, mass wasting events) and the variety of sediment sources involved (e.g. hillslopes, channels, glaciers), sediment fluxes generated by these three hydroclimatic variables are expected to contribute to suspended sediment dynamics in a complementary way, both in terms of magnitude and timing.

In addition to natural hydroclimatic forcings, human activities potentially contribute to altering sediment dynamics, e.g. by changes in land use (e.g. Foster et al., 2003) and sediment storage in reservoirs (e.g. Syvitski et al., 2005). In Alpine environments, it is water impoundment and flow regulation due to hydropower production especially which may substantially influence the suspended sediment regime (e.g. Anselmetti et al., 2007). The impacts of hydropower operations on suspended sediment dynamics may vary substantially between catchments, depending on the specific features of the hydropower system (e.g. reservoir trapping efficiency, hydropower operations), and on the catchment characteristics (e.g. amount and grain size distribution of the eroded sediment, seasonal pattern of sediment production). Here, we focus on the two main effects of hydropower operations: sediment trapping in reservoirs and temporary sediment storage behind water diversion infrastructures (intakes), which may substantially reduce the amount of sediment delivered to downstream reaches and/or significantly alter the timing of sediment release to the river network (e.g. Vörösmarty et al., 2003; Finger et al., 2006; Gabbud and Lane, 2016; Bakker et al., 2018). Despite sediment trapping, water released from hydropower (HP) reservoirs may carry suspended sediment either previously stored in the reservoirs or entrained along the downstream channels.

In the context of environmental change, it is important to understand how the sediment regime has changed and what the relative role of different hydroclimatic forcings may have been. There are examples of studies which demonstrated alterations in suspended sediment yields driven by changes in land use, climate, or by disturbances such as wildfires, earthquakes, and flow impoundments (e.g. Loizeau and Dominik, 2000; Foster et al., 2003; Dadson et al., 2004; Yang et al., 2007; Horowitz, 2010; Costa et al., 2018). These changes are normally addressed by calibrating different sediment rating curve models, which express suspended sediment concentration as a power function of discharge, for different sediment supply regimes and by making the parameters of the rating curves time-dependent (e.g. Syvitski et al., 2000; Yang, 2007; Hu et al., 2011; Huang and Montgomery, 2013; Warrick, 2015). However, these approaches do not explicitly address the sources of sediment and their activation by different hydroclimatic forcings and are limited to using discharge as a predictor. As a result the hydroclimatic causality of changes in suspended sediment concentration in such analyses remains elusive. The approach proposed in this paper accounts explicitly for the hydroclimatic and hydropower activation and deactivation of different sediment sources, with the aim to identify their predictive power in estimating suspended concentration even without using discharge.

Our main objectives are (1) to explore the role played by the hydroclimatic variables erosive rainfall, ice melt, snow-melt, and the hydropower release, in controlling suspended sediment concentration of an Alpine catchment, and (2) to analyse long-term, climate-driven changes in suspended sediment concentration on the basis of a conceptual, data-driven approach accounting separately for the contribution of erosive rainfall, ice melt, snowmelt, and hydropower releases.

The upper Rhône Basin in southern Switzerland is used as the study catchment. The upper Rhône River contributes more than 65 % of the total input of particulate matter into Lake Geneva, the largest lake in the Alps (Loizeau et al., 1997), substantially influencing the morphology and ecology of the river delta and the lake (Loizeau and Dominik, 2000; Loizeau et al., 1997). The catchment is heavily regulated by hydropower infrastructure. Several large hydropower reservoirs have been in operation since 1960s, leading to a total retention capacity equal to roughly 20 % of the total annual discharge of the catchment (Loizeau and Dominik, 2000; Fatichi et al., 2015). In addition to reservoirs, a complex network of water intakes and diversions extracts water from headwater streams and delivers it either to the major reservoirs or directly to the hydropower plants. From a detailed map of the hydropower scheme, including reservoirs and water diversions (Fatichi et al., 2015), it is estimated that roughly 25 % of the catchment is affected by hydropower: 8 % flows directly into the reservoirs, and 17 % is diverted through tunnels and pumping stations. Sediment fingerprinting conducted in the catchment in a recent study from Stutenbecker et al. (2017) indicates that sediment originating in the lithological unit more affected by hydropower is under-represented at the outlet of the catchment, suggesting the impact of water impoundment on the sediment budget of the basin. In addition, alterations of suspended sediment concentration entering Lake Geneva have been observed in the recent past and attributed to human impacts (Loizeau and Dominik, 2000; Loizeau et al., 1997) and changes in climatic conditions (Costa et al., 2018).

The paper is organized as follows: Sect. 2 describes the data pre-processing, the hydrological modelling procedure to obtain the hydroclimatic variables (ER, IM, SM), the approach to obtain the hydropower releases (HP), and the analysis performed to infer their link to suspended sediment concentration; Sect. 3 presents the upper Rhône Basin and the data used in our analysis; Sect. 4 reports the main results which are discussed in Sect. 5; and Sect. 6 concludes the paper by summarizing the main findings.

To analyse the role of hydroclimate on the suspended sediment regime of a catchment regulated by hydropower reservoirs, we first divide the catchment into two distinct areas: (1) the area which contributes to the runoff accumulated in hydropower reservoirs (regulated area), including the fraction of the catchment draining directly into the reservoirs and the fraction connected to the reservoirs through tunnels and pumping stations, and (2) the remaining area, which naturally flows to the river network (unregulated area). We assume that the sediment fluxes originated in the unregulated area contribute directly to SSC at the outlet of the catchment, while sediment fluxes generated in the regulated area are diverted into the reservoirs and later totally or partially released according to hydropower operations. Finally, we estimate the contribution to SSC at the outlet of the catchment of sediment fluxes originated in the unregulated area by ER, IM, and SM, and of sediment fluxes carried by water released from the reservoirs during hydropower operations.

Our methodology consists of four main steps:

We derive mean daily SSCt, ERt, SMt, IMt, and HPtdatasets. Mean daily SSCt at the outlet of the catchment is derived from continuous measurements of turbidity (Sect. 2.1), the hydroclimatic input variables mean daily ERt, SMt, and IMt are derived from spatially distributed snowmelt and ice melt models, and the mean daily water releases from hydropower reservoirs HPt are derived by a conceptual approach based on a unique virtual reservoir, which is intended to model the cumulative effect of multiple reservoirs, when present in the catchment, and a target volume function (Sect. 2.2).

We apply our approach to the upper Rhône Basin in the Swiss Alps (Fig. 1). The total drainage area of the catchment is equal to 5338 km2 and about 10 % of the surface is covered by glaciers. The topography of the basin which has been heavily preconditioned by uplift and glaciations (Stutenbecker et al., 2016) is characterized by a wide elevation range (from 372 to 4634 m a.s.l.). The Rhône River originates at the Rhône Glacier and flows for roughly 170 km before entering Lake Geneva. The hydrological regime of the catchment is dominated by snowmelt and ice melt with peak flows in summer and low flows in winter. Mean discharge is equal to about 320 m3 s-1 in summer and 120 m3 s-1 in winter, while the mean annual discharge is around 180 m3 s-1. Basin-wide mean annual precipitation is about 1400 mm yr-1 and mean annual temperature is about 1.4 ∘C, estimated at basin mean elevation.

Porte du Scex is the measurement station at the outlet of the Rhône River into Lake Geneva (Fig. 1), where the Swiss Federal Office of the Environment (FOEN) collects discharge, SSC, and turbidity data. Mean daily discharge has been available since 1905, while SSC has been measured twice per week since October 1964. Quality-checked continuous measurements of NTU hve been available since May 2013 (Grasso et al., 2012). SSC at the outlet is characterized by a seasonal pattern typical of Alpine catchments (Fig. 2a). During winter (December–March) sediment sources are limited because a large fraction of the catchment is covered by snow and precipitation occurs in solid form. Streamflow is mainly determined by baseflow and hydropower releases (Loizeau and Dominik, 2000; Fatichi et al., 2015), and SSC assumes its minimum values. In spring, SSC increases when snowmelt-driven runoff mobilizes sediments along hillslopes and in channels. Simultaneously, snow cover decreases and rainfall events over gradually increasing snow-free surfaces erode and transport sediment downstream, resulting in SSC peaks. In July, SSC reaches its highest values in conjunction with streamflow (Fig. 2a). In late summer (August and September), when ice melt dominates, sediment-rich fluxes coming from proglacial areas maintain high values of SSC although discharge is decreasing (Fig. 2a). In terms of suspended sediment yield, low SSC conditions do not play a relevant role compared to moderate and high SSC conditions: more than 66 % of the total suspended sediment load entering Lake Geneva during the 4-year period May 2013–April 2017 is estimated to be due to SSC values greater than the 90th percentile (Fig. 2b).

The linear relationship between the logarithm of NTU and SSC for the overlapping period of measurement is statistically significant, with a coefficient of determination R2= 0.94 (Fig. 3). After applying the correction factor for back-transforming from logarithmic to linear scale, the calibrated parameters of the relation in Eq. (1) are a0= 0.56 and b0= 1.25. This relation was used to convert NTU observations to mean daily SSC. We are aware that the relation between SSC and turbidity (1) is site-specific, (2) may vary seasonally as function of discharge and transported grain sizes, and (3) depends on sediment sources, because the size, the shape, and the composition of suspended material may influence values of turbidity (Gippel, 1995). For this reason, in this analysis (1) we apply a site-specific SSC–NTU relation, (2) we calibrate the relation over a wide range of NTUs and discharge conditions to account for the seasonal variability in grain sizes transported by the flow, and (3) we derive the SSC–NTU relation based on a relatively short period of time, in which there is no evidence of changes in sediment sources. In addition, by allowing a nonlinear relation between SSC and NTU, we partially take into account the variability of turbidity with grain size. Higher suspended sediment concentrations are expected to transport proportionally larger grains, and the exponent in the SSC–NTU relation was expected to be greater than 1.

We estimate the hydroclimatic variables for the 40-year period 1975–2015 with the spatially distributed degree-day model of snowmelt and ice melt. The model is implemented using a DEM with a spatial resolution of 250 m × 250 m (Federal Office of Topography – Swisstopo). For the climatic dataset, we use gridded mean daily precipitation and mean, maximum, and minimum daily air temperature at ∼ 2 km × 2 km resolution provided by the Swiss Federal Office of Meteorology and Climatology (MeteoSwiss). These datasets are produced by spatial interpolation of quality-checked measurements collected at meteorological stations (Frei et al., 2006; Frei, 2014). Snow cover maps used for the calibration of the snowmelt rate were derived for the period 2000–2008 in a previous study (Fatichi et al., 2015) from the 8-day snow cover product MOD10A2 retrieved from the (MODIS) (Dedieu et al., 2010). We consider the GLIMS Glacier Database of 1991 to define the initial configuration of the ice covered cells. To calibrate the ice melt rate, we use mean daily discharge data measured at the outlet of two highly glacierized tributary catchments: the Massa and the Lonza (Costa et al., 2018).

To separate sediment fluxes originated in regulated and unregulated areas of the catchment (Fig. 1), we used a detailed map of the main hydropower reservoirs and water uptakes and diversions, available from previous work of Fatichi et al. (2015) and based on information included in the product “Restwasserkarte”, available from the Swiss Federal Office for the Environment (BAFU).

When applying the IIS algorithm (Sect. 2.3), we consider mean daily ERt-l, SMt-l, IMt-l, and HPt-l at time lags l from 0 day to 7 days. This choice is driven by the size of the basin and the expected flow concentration times in the basin. We calibrate the HMRC on data from the period 1 May 2013–30 April 2015 (730 days) and validate it over the period 1 May 2015–30 April 2017 (731 days). For the sake of comparison, calibration and validation periods are also the same when considering the RC.

The robustness of the hydroclimatic predictors of SSC in this work depends on the hypothesis that the hydroclimatic variables are independent drivers which activate different sediment sources in Alpine catchments. Indeed the high fraction of the daily SSCt variance explained by the first three hydroclimatic variables selected by the IIS algorithm, ERt-1, IMt-1, and SMt-1, is in accordance with the physical processes underlying the erosion and sediment transport dynamics in such environments. The higher intensity that characterizes rainfall events in comparison to the melting components is more likely to generate peaks of SSC during heavy rainfall and floods. In accordance, ER is responsible for a large fraction of the process variability (75 %). Indeed, intense rainfall events can detach and mobilize large amounts of sediment (Wischmeier, 1959; Wischmeier and Smith, 1978; Meusburger et al., 2012). The sharp rise in streamflow, which typically follows a precipitation event, results in an increase in sediment transport capacity that may further entrain sediment previously stored along channels. Precipitation is also one of the main triggering factors of mass wasting events, like landslides and debris flow (e.g. Caine, 1980; Dhakal and Sidle, 2004; Guzzetti et al., 2008; Leonarduzzi et al., 2017), in which large quantities of sediment may be instantly released to the river network (e.g. Korup et al., 2004; Bennet et al., 2012). Conversely, the physical processes of ice melt-driven erosion and sediment transport are more gradual and continuous. Similarly, the slow and continuous effect of snowmelt-driven runoff on hillslope and channel erosion contributes to the seasonal pattern of SSC and plays a secondary role in explaining its daily variability and peaks in SSC. Interestingly, hydropower releases do not influence significantly the variance of SSC at the daily scale, despite the fact that the Rhône Basin is heavily regulated by hydropower reservoirs. This is most likely related to the fact that water fluxes downstream of Alpine hydropower dams have lower concentrations of suspended sediment compared to fluxes entering the reservoirs, due to sediment trapping in the reservoirs (Loizeau and Dominik, 2000; Anselmetti et al., 2007). This is in agreement with results of the sediment fingerprinting analysis recently performed in the catchment, which suggests the under-representation of sediments originated in the most highly regulated lithological unit (Stutenbecker et al., 2017). This is also indicated by the negative coefficient of HP, which suggests that hydropower releases dilute suspended sediment in the HMRC model and therefore leads to a reduction of SSCt compared to natural flow. It should also be noted that the effect of hydropower reservoirs on sediment storage is grain size dependent (e.g. Anselmetti et al., 2007) and may be substantially different for coarser grains transported as bedload. Moreover, it is necessary to consider that this analysis focuses on the effects of hydropower on daily suspended sediment dynamics at the basin scale, and neglects potential effects at subdaily scale and localized in tributary catchments. For example, the instantaneous and intermittent flushing of sediment downstream of water diversions infrastructures, located at the most upstream headwater streams, may have substantial effects locally.

We calibrate and validate a rating curve based on the hydroclimatic variables selected by the IIS algorithm and hydropower releases and a traditional rating curve based on discharge only. While both the HMRC and the traditional RC show similar performance in calibration, the HMRC, by taking into account the physical processes which govern SSC in a more direct way, performs better in validation and simulates more accurately the seasonal pattern of SSC, especially in summer when melting of snow and ice are active and a large fraction of the catchment is snow-free and subject to erosion by rainfall. The traditional RC overestimates SSC in winter because it relies on streamflow only and does not account for the low concentration of sediment coming from hydropower reservoirs.

On the basis of the HMRC parameters, we find that although ER is responsible for the peaks of SSC and therefore contributes the most to the variance of SSC, IM fluxes generate the highest SSC per unit volume of water. This is in agreement with the fact that meltwater originated in glaciated areas is characterized by very high sediment concentrations (Gurnell et al., 1996; Lawler et al., 1992). For a catchment significantly glacierized such as the upper Rhône Basin (roughly 10 % of the surface is covered by glaciers) this implies also that among the hydroclimatic variables, IM represents the greatest contribution to SSC and suspended sediment yield from this Alpine catchment (as shown in Fig. 6). This supports the findings of Costa et al. (2018) in which the authors show that the increase in SSC observed at the outlet of the Rhône Basin in mid-1980s is most likely due to a significant rise in ice melt fluxes due to the enhanced glacier retreat associated with warmer temperatures. In concurrence with increasing ice melt, the mean annual SSC at the outlet of the catchment generated by IM, as simulated by the HMRC, increases after the mid-1980s (Fig. 8). This explains why the HMRC is capable of simulating the observed shift in SSC, although the simulation resembles more of a gradual increase than a sudden jump. The results show that a more process-based rating curve accounting for the different hydroclimatic forcing can not only separate the relative effects of the different forcings on SSC, but also explain climate-driven changes in suspended sediment dynamics, which is not possible by adopting a traditional rating curve based on discharge alone.

In this paper, we analyse how hydroclimatic factors influence suspended sediment concentration in Alpine catchments by differentiating among the potential contributions of erosional and transport processes typical of Alpine environments, driven by (1) erosive rainfall defined as liquid precipitation over snow-free surfaces, (2) ice melt, and (3) snowmelt. For regulated catchments, we include the potential effect of hydropower by considering the contribution to SSC of fluxes released from reservoirs due to hydropower operations. We obtained the hydroclimatic variables ER, SM, and IM by using a conceptual spatially distributed model of snow accumulation, snowmelt, and ice melt driven by precipitation and temperature at a daily resolution and we computed HP via a unique virtual reservoir which was operated on the basis of a target volume function, which is aimed at reproducing the cumulated effect of the historical operations of the several hydropower facilities. We then used the Iterative Input Selection algorithm to select the variables that play a significant role in predicting SSC and to quantify their relative importance and predictive power in simulating observed changes in SSC in the Rhône Basin over a period of 40 years. We tested our approach on the upper Rhône Basin in Switzerland. Our main findings can be summarized as follows.

The three hydroclimatic processes ER, IM, and SM are significant predictors of mean daily SSC at the outlet of the upper Rhône Basin, explaining respectively 75, 12, and 4 % of the total observed variance; hydropower releases do not play a significant role in defining the variance of SSC, most likely because fluxes released from reservoirs are poor in sediment due to sediment trapping. The characteristic time lag of 1 day for the ER, IM, and SM fluxes, representing the time necessary to produce sufficient runoff and to entrain and transport sediment from a given location in the catchment to the outlet, are in agreement with typical concentration times of the catchment; conversely for HP the time lag is lower than 1 day.

In summary, our approach provides an insight into how hydroclimatic variables control SSC dynamics in Alpine catchments, and the results suggest that a more process- and data-informed approach in predicting suspended sediment concentrations, which accounts for sediment sources and transport processes driven by erosive rainfall, snowmelt, and ice melt, instead of only discharge, allows climate-induced changes in sediment dynamics to be analysed. Although these results are specific for the upper Rhône Basin only, the approach is general and may be employed in other Alpine catchments with pluvio-glacio-nival hydrological regimes where sufficient data are available.