The monitoring of riverine sediments in Ötztal is part of Austria's national strategy to assess changes in sediment dynamics due to factors like deglaciation and land-use change for details, see.

Suspended sediment concentrations (SSC) have been monitored since 2006 at both stations by HD-Tirol using optical infrared turbidity sensors. SSCs are derived from the turbidity measurements, and calibrated with in-situ SSC samples manually taken at the gauge at a variety of flow conditions. At Tumpen, turbidity is continuously monitored throughout the year, while at Vent measurements are paused in winter (November–April) to protect the equipment from damage by ice. Sediment transport is considered negligible during this period (Fig. ).

For this study we used 15 min time series of streamflow, Qt, and suspended sediment concentration, SSCt. Time series of suspended sediment flux SSFt in tonnes per time step is calculated by multiplying Qt and SSCt. These data are available between 2006 and 2022, meaning that data on fluvial sediment transport is only available for 16 of the 21 years of the study period (2004–2024).

The main precipitation data used in this study are hourly precipitation grids at 1 km resolution which are supplemented by daily and hourly precipitation from 33 weather stations in and around Ötztal (Fig. b).

Gridded precipitation products in mountainous regions have limited accuracy e.g. due to the strong influence of topography on precipitation, an observation bias towards lower elevations, and challenging conditions for radar (e.g. beam shielding) . Even in the mountainous parts of the INCA domain (i.e. Austrian Alps), the high density of weather stations is somewhat biased towards elevations below 2000 ma.s.l. . Taking advantage of the higher density of rain gauges in Ötztal also at high elevations , we can estimate the uncertainty of hourly and daily INCA precipitation with our assembled rain gauge data in and around Ötztal using four metrics.

The mean error (ME, Eq. ) and the root-mean squared error (RMSE, Eq. ) are calculated for each station by comparing the observed precipitation of a station xiobs with the predicted INCA precipitation at the grid cell in which the station is located xiINCA 1ME=1N∑i=1N(xiINCA-xiobs)2RMSE=1N∑i=1N(xiINCA-xiobs)2 where N is the total number of time steps indexed by i with both valid INCA and observation values. The ME indicates whether INCA tends to over- or under-estimate precipitation at the station, whereas the RMSE quantifies the overall error magnitude of INCA .

To estimate the ability of INCA to capture the occurrence of precipitation we classified each time step of xiobs and xiINCA into a “precipitation” and “no-precipitation” category, using a threshold of 0.1 mmh-1 for the hourly data, and 1 mmd-1 for the daily data. The latter is the definition of a “wet day” commonly used in climate indices . Next, we computed contingency tables see listing the number of hits (true positives) a, misses (false negatives) b, false positives c, and true negatives d. From these four possible outcomes we estimated (1) the accuracy (Acc, Eq. ) and (2) the frequency bias (FB, Eq. ) . 3Acc=a+dN4FB=a+ca+b The Acc is the fraction time steps in which INCA correctly predicts the occurrence of precipitation. The FB quantifies whether INCA tends to over- or under-estimate the occurrence of precipitation.

To assemble a catalogue of heavy precipitation events, we device a multi-scale detection approach based on extreme value statistics that allows for detection of heavy precipitation peaks at multiple temporal and spatial scales. We use catchment-averaged, Pt, and grid-scale maximum precipitation, It, to represent the spatial scales of the catchment-wide and localised precipitation respectively (Fig. ). Events in Pt will tend to represent catchment-wide heavy precipitation such as frontal precipitation, while events in It will track localised heavy precipitation, generally convective cells. Using a duration-dependent generalised extreme value (d-GEV) distribution, we estimate intensity-duration-frequency (IDF) curves of Pt and It during May–October for Tumpen-Ötztal (Fig. S3, Supplement). With the IDF curves we set detection thresholds for each duration, extract peaks, isolate the associated precipitation event, and merge any duplicated or overlapping events. The procedure is described in detail below.

For each of the two time series, Pt and It, we fit a d-GEV distribution which allows us to calculate IDF curves from a single extreme value distribution see reducing the total number of parameters required compared to approaches which individually fits one GEV distribution for each duration. We use the the R-package IDF with the options allowing multi-scaling and curvature for small durations see.

The d-GEV distributions were fitted to annual May–October precipitation block maxima, M=(mdj), using the maximum likelihood method. We calculated M for each of the precipitation time series Pt and It. For each duration d, M were calculated by applying a d-moving average to the precipitation time series and extracting the maximum value m during May–October of each year j. We considered the durations 1, 2, 3, 6, 12, 24, 48, and 72 h. We restricted M to May–October precipitation, when most sediment is exported and rainfall is highest (Fig. ). Outside of this season precipitation predominantly falls and accumulates as snow and is therefore not relevant for our study. Furthermore, suspended sediment transport is negligible from November to April.

Detection thresholds, u, for each duration d and spatial scale (i.e. grid-scale It or catchment-scale Pt) were set at the 0.2 non-exceedence probability quantile of their respective IDF curves, which corresponds to a return period of 1.25 years. With this choice of u we should capture the major precipitation events each year for the spatial and temporal scales under consideration and thus ensure a sufficiently large catalogue of heavy precipitation events.

Next, we detected and isolated heavy precipitation peaks in the precipitation time series. From d-moving-averaged Pt and It we extracted peaks above u and their timing, tpeak (Fig. , explained in detail in figure caption). The peaks were identified using the get_extremes function from pyextremes , which applies a declustering procedure to ensure a minimum time window of 24 h between peaks. We detected event peaks for all durations in Pt but only for durations ≤ 24 h in It, under the assumption that localised heavy precipitation events detected in the grid-scale precipitation time series will generally be convective events that last less than 24 h.

For each heavy precipitation peak, we isolated the associated event by searching forward and backward in time from tpeak to identify when it started and stopped raining. The event start time, tstart, was defined as the first time step before tpeak that satisfied the criteria 5∑t=tstart-1tstart+1Pt<0.1mmh-1 or 6∑t=tstart-1tstart+1It<1mmh-1 depending on in which precipitation time series the heavy precipitation peak was detected. The event end time, tend, was defined as the first time step after tpeak that satisfied the criteria in Eqs. () and () (tstart substituted with tend).

Given the generous criteria for the detection of heavy precipitation peak, many heavy precipitation events were detected at multiple scales, meaning these events exceeded the detection thresholds for several durations, or were detected both in Pt and It. These events are either duplicates (i.e. same tstart and tend) or temporally overlapping, and were merged iteratively by passing over the whole collection of detected events several times. The events were selected for merging by different criteria in each merging pass:

duplicated or overlapping events with peaks for the same duration, detected in both Pt and It;

After the merging, each remaining precipitation event was tagged with the durations and spatial scales at which it was extreme. Finally, all events were manually checked by visually evaluating their accumulated precipitation maps and comparing their time series with station observations. Events with implausible values or precipitation patterns were checked thoroughly and removed if it was judged that the event was a data artefact or mistaken detection, such as precipitation only occurring in a single grid cell.

For each heavy precipitation event, we calculate a set of characteristics based on , which quantify rainfall and precipitation amounts and intensity (Table ). In addition, we calculated the average precipitation area Aprecip in order to estimate the catchment area affected by a precipitation event. We calculated Aprecip by determining the fraction of catchment area receiving > 0.1 mm precipitation for each time step, before averaging over the entire event duration. To give an indication of the moisture conditions in the catchment over the last seven days leading up to the event, we calculated the 7 d normalised antecedent precipitation index, NAPI7, after .

We further categorised each event according to the spatial and temporal scale of the set of heavy precipitation peaks detected. An event detected from Pt or both Pt and It was categorised as a catchment-wide event, while one detected only from It was categorised as a localised event. Sub-daily events only have heavy precipitation peaks above 1 to 12 h thresholds. Long-duration events contain at least one heavy precipitation peak above a 24, 48, or 72 h threshold. Hence, the temporal scale does not refer to the overall duration of the precipitation event, i.e. D in Table , but the set of heavy precipitation peaks detected witin the event (Fig. ).

To estimate the fluvial sediment response to precipitation events we first delineated hydrological events in Qt using local-minima hydrograph separation . This method separates the entire streamflow time series into pulses of river discharge, separated by the local minima which mark the end of recession after one hydrological event and the onset of the next. Following we use a 21 h search window, which is suitable for Ötztal's glacially-influenced hydrological regime , which ensures that two local minima are separated by at least 10.5 h.

The resulting hydrological event catalogue was compared to the detected heavy precipitation events, matching hydrological events to each heavy precipitation event: All hydrological events that overlap with a precipitation event, i.e. begin or end between tstart and tend, are assigned to that event. We discarded all cases, in which the first matched hydrological event began > 3 h before tstart, or where the last matched hydrological event began < 1 h before the end of the precipitation event.

The fluvial sediment response window of an heavy precipitation event was defined as the time window from tstart to the end of the last matched hydrological event thydro,end. For each heavy precipitation event we calculated the mass of suspended sediment exported, i.e. suspended sediment yield SSY in t: 7SSY=∑t=tstartthydro,endSSFt To take into account the varying event durations we also calculate the average suspended sediment flux SSFmean in th-1: 8SSFmean=mean{SSFt:tstart≤t≤thydro,end} During the melt season, and especially in July and August, the sediment load in the river will be elevated at the onset of an event due to other drivers such as high sub-glacial sediment discharge. To account for this, we also calculated the excess suspended sediment yield SSYex of an event in t. This metric only sums up the SSF in each time step that exceeds the initial SSF at the start of the event: 9SSYex=∑t=tstartthydro,endSSFt-SSFtstart,SSFt>SSFtstart,0,otherwise. Finally, we extracted the peak and averaged SSC in mgL-1 during the sediment response window of each heavy precipitation event: 10SSCmax=max{SSCt:tstart≤t≤thydro,end}11SSCmean=mean{SSCt:tstart≤t≤thydro,end}

To quantify the contribution of precipitation-driven sediment transport to annual SSY, we conducted an inverse analysis in which we classified all hydrological events based on the associated precipitation. We categorised hydrological events with influence of heavy precipitation if they matched with heavy precipitation events in Sect. . Of the remaining hydrological events, those with an average precipitation intensity of >0.1 mmh-1 were categorised as non-heavy precipitation and the rest as no precipitation.

For each hydrological event we calculated SSY, SSF, Ptot, and RFtot, substituting tstart and tend with the start and end times of the hydrological event (see Eqs. – and in Table ). Next, we calculated the contribution to annual SSY of each hydrological event under the influence of heavy, non-heavy, and no precipitation. Due to the high inter-annual variability of annual SSY we also calculated the fraction of annual SSY exported during each precipitation influence event class. Using a Mann-Kendall (MK) test with a 5 % significance level we detected significant annual trends, and estimated their magnitude with Theil-Sen slope .

Given the strong influence of melt-driven sediment transport in the study area, not all hydrological events with high SSC are linked to (heavy) precipitation. In another inverse analysis we extracted hydrological events with high peak SSC and classified them as influenced by heavy, non-heavy, or no precipitation as in Sect. . We defined these suspended sediment spikes as hydrological events with SSCmax above the 90th percentile of SSCt, SSC90 after . For both catchments we counted the number of such events affected by either heavy or non-heavy precipitation.

The uncertainty analysis of daily and hourly INCA precipitation shows that INCA tends to overestimate the precipitation amount with an average ME of 0.2 mmd-1 and 0.01 mmh-1 for daily and hourly precipitation respectively. The average RMSE of hourly precipitation is 0.5 mmh-1 and is unrelated to altitude (Fig. S4d, Supplement). The RMSE of daily precipitation is 1.0 to 2.5 mmd-1 at stations below 1750 ma.s.l. compared to 1.0 to 6.2 mmd-1 at higher elevations (Fig. S4c, Supplement). At both, the hourly and daily scale, the INCA data are highly accurate (Acc > 0.9) concerning precipitation up to about 2500 ma.s.l. (Fig. S4e–f, Supplement). At higher elevations, this accuracy drops to between 0.8 and 0.9 for precipitation hours and between 0.66 and 0.93 for precipitation days. Overall, the frequency of precipitation hours and days is overestimated by INCA with FB generally above 1 (Fig. S4g–h, Supplement). However, at three stations, Proviantdepot, Latschbloder, and Station Hintereis, located at high elevations (2737 to 3031 ma.s.l.) in the inner Rofental, INCA underestimates the occurrence of precipitation at both the daily and hourly scale.

A total of 169 heavy precipitation events were identified with the multi-scale detection approach. Three events were removed during the visual check, as the precipitation maps revealed that only one or two grid cells received very high amounts of precipitation while neighbouring cells did not receive any and no precipitation was recorded at the weather stations; these events were likely data artefacts (see Sect. S2.2.1 in the Supplement).

Of the 166 events compiled in the final catalogue, 62 were detected from It, 78 from Pt, and 26 from both time series. Only 41 events (25 %) were detected with a single threshold, while the majority contained heavy precipitation peaks at multiple temporal or spatial scales. The events ranged from intense short duration bursts to long-duration events with high precipitation totals (Fig. b). The average event duration was 24 h with the shortest event lasting 5 h and the longest 128 h (5.3 d). The highest Imax was 88 mmh-1 with an average of 19 mmh-1 (2019-g, Fig. b). The largest event in terms of precipitation amount recorded a Ptot of 106 mm (2023-g, Fig. b) compared to the highest RFtot of 97 mm. On average the total precipitation and rainfall amount was 24 and 16 mm respectively.

The 75 heavy precipitation events detected with sub-daily thresholds (≤ 12 h) tended to have more intense precipitation (Fig. f) consisting mostly of rainfall (95 % on average). They occurred mainly between June and August, which is when precipitation and rainfall amounts are highest (Fig. ). Two examples of sub-daily heavy precipitation events (Fig. a, d) highlight how localised precipitation was, with much the catchment receiving little to no precipitation. The 91 long-duration heavy precipitation events had higher precipitation totals (Fig. g) that affected larger areas (Fig. e), occurred throughout May–October without any particular seasonality, and with higher fractions of snowfall (45 % on average).

The 104 catchment-wide heavy precipitation events had low precipitation intensities (Fig. b, f) but affected a larger catchment area (Fig. e) with high precipitation totals (Fig. g). In contrast, the 62 localised heavy precipitation events affected a smaller catchment area (Fig. e), had high Imax (Fig. f), and precipitation totals of mostly < 20 mm (Fig. g).

The annual frequency of heavy precipitation events varied over the study period with events occurring more frequently towards the end (Fig. a), displaying a significant increasing trend (MK-test, 5 % significance level). For the first few years of the study period there were 3–5 events per year and in these years May–October precipitation was lower than the latter part of the study period. From 2010 onwards followed a few years with high inter-annual variability in event numbers, varying between 4 and 16 events per year. In the final part of the study period, from 2015 onwards, the annual occurrence of heavy precipitation events was at a higher level with a minimum of 8 events per year. In general, there were more events in years with higher May–October precipitation, with annual event counts and May–October precipitation totals being significantly correlated, r=0.65.

Average annual RMSE of daily precipitation was higher in the first half of the study period (median: 2.4 mmd-1) and dropped to a lower level (median: 1.4 mmd-1) between 2016 and 2024 (Fig. a). In both periods daily precipitation is overestimated, but with a higher median annual mean error of 0.24 mm in 2004–2016 compared to 0.14 mm from 2017 onwards. Heavy precipitation days with more than 10 mmd-1 had a tendency to be overpredicted, except in 2004 and 2015 where they were under-predicted. The years with the highest frequency bias are 2007, 2012, 2013, and 2014.

About two thirds of heavy precipitation events occurred during the summer months July and August with the lowest occurrences in May, September and October (Fig. b). Events with a high liquid fraction, i.e. mainly rainfall, were concentrated in the months June, July and August. Mid-season heavy precipitation events tend to have higher intensities and shorter durations, while events with longer duration and lower intensity occurred evenly throughout the year. Events with higher amounts of snowfall (liquid fraction generally below 0.5) mainly occurred in the colder months of May, June, September and October.

Of all precipitation event characteristics, the three rainfall characteristics show the strongest and consistently positive significant correlation with each of the sediment response variables (Table ), with the exception of peak SSC and total rainfall, which is not significantly correlated. The peak catchment averaged precipitation intensity, Pmax, shows significant but somewhat weaker positive correlation with the sediment response variables. The 7 d antecedent precipitation index, NAPI7, is not significantly correlated with the sediment response variables, with the exception of peak SSC which is weakly negatively correlated. Event duration, D, and average precipitation area, Aprecip, are negatively correlated with the sediment response variables. Both SWEmean, indicating the amount of snow in the catchment, and SWEloss, indicating snowmelt during the heavy precipitation event, are significantly negatively correlated with the sediment response variables. The correlation between SSYex and the precipitation event characteristics does not substantially differ from SSY.

The relationship between event rainfall and suspended sediment shows clear differences between sub-daily and long-duration heavy precipitation events (Fig. ). The sub-daily heavy precipitation events have an overall weaker positive relationship with rainfall intensity and amount. The difference is particularly pronounced for RFmax and SSY and SSFmean (Fig. b, f). Here, SSY and SSFmean markedly increase with rainfall intensity for long-duration heavy precipitation event, whereas for sub-daily heavy precipitation event, increases in rainfall intensity barely have an effect. Sub-daily heavy precipitation events overall have higher SSFmean compared to long-duration heavy precipitation events (Fig. h) and somewhat although not significantly higher SSCmax (confidence intervals of medians overlap, Fig. l). In terms of total exported suspended sediment mass there is little difference between the two categories of heavy precipitation events, except a larger spread for long-duration heavy precipitation events (Fig. d).

Between 2006 and 2022, both the total mass and the fraction of suspended sediment exported during heavy precipitation events increased significantly in both catchments (Fig. a–d). The fraction of annual SSY exported during heavy precipitation-driven hydrological events increased at about 1 % yr⁻¹ in both catchments (Fig. a, c). There are no significant annual trends in suspended sediment exported during hydrological events with non-heavy or no precipitation. Moreover, in both catchments the sediment mass exported outside of heavy precipitation events (Fig. b, d) follows a similar trend to the total annual SSY (Fig. a, c). In Tumpen-Ötztal, the sediment mass exported outside of heavy precipitation events was largely unchanged throughout the 17 years (Fig. b). In Vent-Rofental, the contribution of hydrological events associated with non-heavy or no precipitation decreased until 2021, though abruptly increased in 2022 (Fig. d).

There is a significant difference between the SSFmean, Ptot, and RFtot of hydrological events associated with heavy, non-heavy and no precipitation (Fig. e–g). Those with heavy precipitation have the highest magnitudes, followed by non-heavy and no precipitation. On average, 23 % of annual suspended sediments were exported in association with heavy precipitation, compared to 37 % during non-heavy precipitation and 40 % during hydrological events without precipitation. The discrepancy between the magnitude and annual contribution of each category is due to the differing event frequencies (Fig. e–g). Events influenced by heavy precipitation are rarer, thus despite their high SSFmean, their contribution to annual SSY is lower than the other two categories.

The results of the inverse analysis showed that only 10 % of the more than 1000 suspended sediment spike events, i.e. hydrological events with SSCmax>SSC90, were associated with heavy precipitation (Fig. ). About a third were associated with some, but not heavy, precipitation (Fig. ). More than half of the suspended sediment spikes in both catchments were not associated with any precipitation. The pattern was nearly identical for Tumpen-Ötztal and Vent-Rofental.

Over the study period, we observe an increase in the fraction of annual SSY exported during heavy precipitation events. Furthermore, the number of heavy precipitation events significantly increased. As discussed above, uncertainties in the INCA dataset do not appear to significantly affect the detected number of heavy precipitation events. Lower event numbers in the early study period likely reflect drier-than-average years, such as 2004 and 2006 , dominated by isolated convective storms. The significantly increasing trend in the number of heavy precipitation events is therefore more likely due to an overall increase in precipitation intensity leading more events to exceed the heavy precipitation thresholds. While we find no indications of precipitation or rainfall intensity of heavy precipitation events significantly increasing over the study period, this may be due to the high variability of precipitation and the relatively short period we are considering. May–October precipitation in Ötztal calculated from the SPARTACUS dataset , consisting of spatially interpolated station data see, has significantly increased with 2.6 mm yr⁻¹ since 1961 (MK test with 5 % significance level). Thus we can at least say that the total precipitation amount during the extended summer season increased during our study period and is part of a longer term trend. As heavy precipitation-driven transport events have significantly higher suspended sediment fluxes (Fig. e), a pattern common in mountainous areas e.g., the increasing number of heavy precipitation events means that the fraction of annual SSY also increases.

In years with a high frequency of heavy precipitation events, this effect is especially noticeable. Examples are 2010 and 2020, years in which around 40 % of annual SSY were associated with heavy precipitation (Fig. a). In 2020, heavy precipitation events contributed significantly to annual sediment export. For instance, during event 2020-k, persistent rainfall triggered a debris flow in the proglacial area of the Hintereisferner glacier (in Vent-Rofental). A thunderstorm in August (2020-j) and a cold front in October (2020-n) caused additional flooding and mass wasting. Together these three events contributed 124 tkm-2 or 25 % of the annual SSY of 2020. The remaining 13 events contributed 15 %. The year 2020 illustrates the cumulative effect of both moderate and severe storms on sediment yields . Notably, the years 2013 and 2019 had the same number of heavy precipitation events but did not contribute comparable amounts or fractions of annual SSY. This highlights that factors such as annual sediment production and availability play an important role in determining the fraction of heavy-precipitation-driven sediment exports, in addition to the annual frequency and intensity of heavy precipitation events.

Annual SSYs have remained stable over the study period in Tumpen-Ötztal but show a gradual decline in Vent-Rofental (Fig. ), excluding extreme melt years such as 2022 . In Vent-Rofental, peak sediment may already have been passed , with annual SSY now steadily decreasing as glacier retreat reduces subglacial sediment supply, a trend projected to persist over the long term . In Tumpen-Ötztal, sediment exported during non-heavy or precipitation-free events showed a non-significant decline (Fig. b), suggesting that increases in heavy-precipitation-driven sediment transport at larger scales may partially offset reductions in glacier-driven transport. These contrasting trends highlight the influence of different spatial scales of paraglacial adjustments, as the timing of peak sediment and the duration of sediment reworking vary with catchment size and glacier cover .

Despite the increasing prominence of heavy precipitation, only 10 % of extreme suspended sediment spikes in Ötztal can be directly attributed to these events, with nearly half occurring without precipitation (Fig. ). This is consistent with findings showing that high SSC peaks in Alpine catchments can be generated by rainfall, glacier melt, snowmelt, or combinations thereof . Melt processes, due to their frequency, remain the dominant driver of annual SSY , with the most extreme sediment discharge often arising from a combination of melt- and precipitation-driven processes . Extensive work in glaciated catchments has shown that increased water input to subglacial drainage systems, whether from melt or rainfall, may enhance suspended sediment export without necessarily increasing surface erosion. However, during heavy precipitation events increased cloud cover and reduced air temperatures may suppress melt rates and sub-glacial discharge. Consequently, rainfall- and melt-driven sediment production should not be viewed as entirely independent, and heavy-precipitation-driven sediment transport in this study likely includes a substantial glacial component, particularly during summer when subglacial sediment discharge is high.

While glacial influences on sediment transport are declining, projected increases in heavy precipitation, particularly at sub-daily scales, remains a critical factor. Conceptual models suggest that rainfall-driven sediment transport will dominate post-deglaciation sediment yield levels . Convective summer precipitation is expected to intensify in the Alps , with an ensemble of 1 km convection-permitting climate models projecting a 20 %–38 % increase in sub-daily (1–24 h) extremes in the Eastern Alps, including Tumpen-Ötztal . Our results demonstrate that sub-daily heavy precipitation events are particularly effective at generating high suspended sediment loads. This is amplified by the abundant unconsolidated sediments in high-elevation landscapes, a legacy of deglaciation, which will remain available for transport until these areas stabilise . While vegetation and soil development will eventually promote stabilisation , these processes are temporally variable and can be disrupted by geomorphic disturbances such as rainfall erosion and fluvial reworking of proglacial deposits . These dynamics suggest that future sediment regimes in high-elevation catchments like the Ötztal will become flashier, characterised by more frequent and intense rainfall-driven events, but with overall lower annual yields.