Abstract. To date, knowledge on the effects of decadal-scale changes in climatic forcing on sediment export from glaciated high alpine areas is still limited. This is primarily due to the extreme scarcity of sufficiently long records of suspended sediment concentrations (SSC), which precludes robust explorations of longer-term developments. Aggravatingly, insights are not necessarily transferable from one catchment to another, as sediment export can heavily depend on local preconditions (such as geology or connectivity). However, gaining a better understanding of past sediment export is an essential step towards estimating future changes, which will affect downstream hydropower production, flood hazard, water quality and aquatic habitats. Here we test the feasibility of reconstructing decadal-scale sediment export from short-term records of SSC and long time series of the most important hydro-climatic predictors, discharge, precipitation and air temperature (QPT). Specifically, we test Quantile Regression Forest (QRF), a non-parametric, multivariate approach based on Random Forests. We train independent models for the two nested and partially glaciated catchments Vent (98 km2) and Vernagt (11.4 km2) in the Upper Ötztal in Tyrol, Austria (1891 to 3772 m a.s.l.), to gain a comprehensive overview of sediment dynamics. In Vent, daily QPT records are available since 1967, alongside 15 years of SSC measurements. At gauge Vernagt, QPT records started in 1975 in hourly resolution, which allows comparing model performances in hourly and daily resolution (Validation A). Challengingly, only four years of SSC measurements exits at gauge Vernagt, yet consisting of two 2-year datasets, that are almost 20 years apart, which provides an excellent opportunity for validating the model’s ability to reconstruct past sediment dynamics (Validation B). As a second objective, we aim to assess changes in sediment export by analyzing the reconstructed time series for trends (using Mann-Kendall test and Sen’s slope estimator) and step-like changes (using two complementary change point detection methods, the widely used Pettitt’s test and a Bayesian approach implemented in the R package ‘mcp’). Our findings demonstrate that QRF performs well in reconstructing past daily sediment export (Nash-Sutcliffe efficiency of 0.73) as well as the derived annual sediment yields (Validation B), despite the small training dataset. Further, our analyses indicate that the loss of model skill in daily as compared to hourly resolution is small (Validation A). We find significant positive trends in the reconstructed annual suspended sediment yields at both gauges, with distinct step-like increases around 1981. This coincides with a crucial point in glacier melt dynamics: we find co-occurring change points in annual and summer mass balances of the two largest glaciers in the Vent catchment. This is also reflected in a coinciding step-like increase in discharge at both gauges as well as a considerable increase in the accumulation area ratio of the Vernagtferner glacier. We identify exceptionally high July temperatures in 1982 and 1983 as a likely cause, as July is the most crucial month with respect to firn and ice melt. In contrast, we did not find coinciding change points in precipitation. This study demonstrates that the presented QRF approach is a promising tool with the ability to deepen our understanding of the response of high-alpine areas to decadal climate change. This in turn will aid estimating future changes and preparing management or adaptation strategies.

Abstract. Knowledge on the response of sediment export to recent climate change in glacierized areas in the European Alps is limited, primarily because long-term records of suspended sediment concentrations (SSCs) are scarce. Here we tested the estimation of sediment export of the past five decades using quantile regression forest (QRF), a nonparametric, multivariate regression based on random forest. The regression builds on short-term records of SSCs and long records of the most important hydroclimatic drivers (discharge, precipitation and air temperature – QPT). We trained independent models for two nested and partially glacier-covered catchments, Vent (98 km2) and Vernagt (11.4 km2), in the upper Ötztal in Tyrol, Austria (1891 to 3772 m a.s.l.), where available QPT records start in 1967 and 1975. To assess temporal extrapolation ability, we used two 2-year SSC datasets at gauge Vernagt, which are almost 20 years apart, for a validation. For Vent, we performed a five-fold cross-validation on the 15 years of SSC measurements. Further, we quantified the number of days where predictors exceeded the range represented in the training dataset, as the inability to extrapolate beyond this range is a known limitation of QRF. Finally, we compared QRF performance to sediment rating curves (SRCs). We analyzed the modeled sediment export time series, the predictors and glacier mass balance data for trends (Mann–Kendall test and Sen's slope estimator) and step-like changes (using the widely applied Pettitt test and a complementary Bayesian approach). Our validation at gauge Vernagt demonstrated that QRF performs well in estimating past daily sediment export (Nash–Sutcliffe efficiency (NSE) of 0.73) and satisfactorily for SSCs (NSE of 0.51), despite the small training dataset. The temporal extrapolation ability of QRF was superior to SRCs, especially in periods with high-SSC events, which demonstrated the ability of QRF to model threshold effects. Days with high SSCs tended to be underestimated, but the effect on annual yields was small. Days with predictor exceedances were rare, indicating a good representativity of the training dataset. Finally, the QRF reconstruction models outperformed SRCs by about 20 percent points of the explained variance. Significant positive trends in the reconstructed annual suspended sediment yields were found at both gauges, with distinct step-like increases around 1981. This was linked to increased glacier melt, which became apparent through step-like increases in discharge at both gauges as well as change points in mass balances of the two largest glaciers in the Vent catchment. We identified exceptionally high July temperatures in 1982 and 1983 as a likely cause. In contrast, we did not find coinciding change points in precipitation. Opposing trends at the two gauges after 1981 suggest different timings of “peak sediment”. We conclude that, given large-enough training datasets, the presented QRF approach is a promising tool with the ability to deepen our understanding of the response of high-alpine areas to decadal climate change.

<p>Suspended sediment export from partly glaciated high alpine catchments is not only relevant for ecosystems, but also for infrastructure and flood hazard alterations in downstream areas. In order to estimate future changes, it is important to assess long-term developments in past sediment yields. However, existing records of suspended sediment export are mostly too short to investigate these long-term changes. For example, for the two gauges “Vent Rofenache” and “Vernagtferner” in the high alpine and partly glaciated Upper Ötztal in Tyrol, Austria, only 15 and four years of turbidity measurements exist, respectively, precluding robust explorations of longer-term developments.</p><p>To compensate for this lack of measurement data, we use a Quantile Regression Forest approach, a non-parametrical, multivariate tool based on regression trees. It allows for reconstructing continuous sedigraphs based on short-term or point-like sediment concentration data and continuous predictor variables such as discharge (Q), precipitation (P) and air temperature (T).</p><p>At gauge “Vernagtferner”, turbidity-based sediment concentration data were available only for the years 2000, 2001, 2019 and 2020. To test the ability of our model to reconstruct past sediment concentrations, we trained our model using the 2019 and 2020 data and validated against the 2000 and 2001 measurements, which showed good agreement (Nash-Sutcliffe Efficiency of 0.73). At gauge “Vent Rofenache”, the hydrographic service of Tyrol, Austria, has recorded turbidity-based sediment concentration data since 2006. Our model showed to be well able to reconstruct sediment yields based on by these data (out-of-bag Nash-Sutcliffe efficiency of 0.66).</p><p>This validation enabled us to confidently use the long-term availability of the predictor variables (Q, P, T) to reconstruct sediment yields at gauge “Vernagtferner” since 1974 and at gauge “Vent Rofenache” since 1967.</p><p>The resulting dataset allows us to</p><ul><li>Analyze annual sediment yields with respect to trends and change points for time series of 47 and 54 years, respectively,</li> <li>Examine changes in the predictor variables,</li> <li>and connect developments in sediment yields to mass balances of the large glaciers within the catchment.</li> </ul><p>Current results point at an almost step-like increase in annual sediment yields at the beginning of the 1980s at both gauges. This coincides with a marked increase in discharge volumes that in turn correlate with a basic change in glacier mass balances.</p>

Abstract. Future changes in suspended sediment export from deglaciating high-alpine catchments affect downstream hydropower and reservoirs, flood hazard, ecosystems and water quality. Yet so far, quantitative projections of future sediment export have been hindered by the lack of physical models that can take into account all relevant processes within the complex systems determining sediment dynamics at the catchment scale. As a promising alternative, machine-learning (ML) approaches have recently been successfully applied to modeling suspended sediment yields (SSY). This study is the first to our knowledge exploring machine-learning approach to derive sediment export projections until the year 2100. We employ Quantile Regression Forest (QRF), which proved to be a powerful method to model past SSY in previous studies, at two nested high-alpine gauges in the Ötztal, Austria, i.e. gauge Vent (98.1 km² catchment area, 28 % glacier cover in 2015) and gauge Vernagt (11.4 km² catchment area, 64 % glacier cover). As predictors, we use temperature and precipitation projections (EURO-CORDEX) and discharge projections (AMUNDSEN physically-based hydroclimatological and snow model) for the two gauges. We address uncertainties associated with a known limitation of QRF, i.e. that underestimates can be expected if out-of-observation-range (OOOR) data points (i.e. values exceeding the range represented in the training data) occur in the application period. For this, we assess the frequency and extent of these exceedances and the sensitivity of the resulting mean annual suspended sediment concentration (SSC) estimates. We examine the resulting SSY projections for trends, the estimated timing of ‘peak sediment’ and changes in the seasonal distribution. Our results show that the uncertainties associated with the OOOR data points are small before 2070 (max. 3 % change in estimated mean annual SSC). Results after 2070 have to be treated more cautiously, as OOOR data points occur more frequently and as glaciers are projected to have (nearly) vanished by then in some projections, which likely substantially alters sediment dynamics in the area. The resulting projections suggest decreasing sediment export at both gauges in the coming decades, regardless of the emission scenario, which implies that ‘peak sediment’ has already passed or is underway. Nevertheless, high(er) annual yields can occur in response to heavy summer precipitation, and both developments would need to be considered in managing sediments as well as e.g. flood hazard. While we chose the predictors to act as proxies for sediment-relevant processes, future studies are encouraged to try and include geomorphological changes more explicitly, e.g. changes in connectivity, landsliding / rockfalls, or vegetation colonization, as these could improve the reliability of the projections.