| 24 Sep 2026
Beyond rainfall thresholds: hydrologic controls on debris-flow initiation in alpine channels
Abstract. Debris flows are a severe natural hazard in mountain regions. As a spatially and temporally highly irregular phenomenon, meaningful quantitative description and generalization of the underlying processes in contrasting environments as well as associated robust local prediction of debris flows is hindered by a lack of sufficiently detailed observations. As a consequence, debris flows are frequently predicted based on simple regional precipitation (or rainfall) intensity-duration thresholds that lump many channels over larger spatial domains, irrespective of potentially different underlying debris flow trigger processes due to hydro-climatic or geomorphic differences between the channels. Based on long-term debris flow records of up to 18 years, hydro-climatic observations and modelled hydrological variables across three contrasting active debris flow channels (ID1-ID3) in the Austrian Alps, this study (1) analyses differences between the local debris flow regimes in these channels, based on their strength of temporal coupling with hydro-climatic and hydrological extremes, (2) quantifies differences between regional and local debris flow initiation thresholds that emerge from these differences in their temporal coupling, and (3) identifies the pairs of threshold model variables that allow the most reliable debris flow predictions in each of the three study channels. We have found that the three study channels are characterized by considerable difference in seasonal debris flow timing. In channel ID1, debris flows events are largely confined to early summer, occurring, on average, several weeks before the wettest conditions in each year, thus being only weakly coupled to hydro-meteorological extremes. In the close-by channel ID2, debris flows occur throughout summer, on average a few days after the wettest conditions of any specific year. The high debris flow frequency in this channel in addition suggests multiple cycles of sediment exhaustion and re-supply to the channel over one season. Together, this is evidence for a stronger coupling and a highly dynamic balance between the availabilities of sufficient sediment and in-stream transport capacity, respectively. Due to the geographical proximity of and thus similar hydro-climatic conditions in ID1 and ID2, the differences in debris flow regimes can largely be attributed to differences in landscape characteristics and thus geomorphic predisposition. In contrast, in channel ID3, debris flows are largely confined to late summer and temporally coincide with the wettest annual conditions, suggesting a strongly coupled debris flow regime. The differences between ID3 and the other channels arise from the combined influence of hydro-climatic and geomorphic predisposition. These differences in debris flow regimes are reflected in the differences between the individual local thresholds across the three channels and the systematically superior performance of local thresholds to detect debris flows than regional thresholds across all channels. It was further found that traditional precipitation and rainfall thresholds were consistently outperformed by thresholds that explicitly account for the sum of rainfall and snowmelt. The overall strongest thresholds were found to be the ones based on channel discharge, providing a direct descriptor of in-stream transport capacity. Together the results provide evidence for the benefit of local thresholds and explicitly accounting for hydrological variables in threshold-based debris flow prediction models.
Competing interests: At least one of the (co-)authors is a member of the editorial board of Hydrology and Earth System Sciences.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.
In this work the authors investigate several hydrological quantities, rainfall intensity, discharge and so on for obtaining six couples of thresholds, including the classical ID (rainfall intensity versus duration) for three debris flow channels with different geomorphological behaviour ranging from supply-limited to transport-limited.
It results that some thresholds that consider the effect of snowmelt and saturation degree of the basin, or the discharge, have performance much better than the classical ID.
However, the results may not be fully generalizable because the rainfall records have an hourly resolution and therefore do not capture sub-hourly precipitation. For example, a 30-minute rainfall burst that triggers a debris flow would appear with half its true intensity when aggregated to hourly values. This effect has been documented in the literature: Marra (2019) showed that coarse temporal resolution leads to a systematic underestimation of rainfall thresholds, and Bernard and Gregoretti (2021) reported that many rainfall events triggering debris flows are sub-hourly. This temporal smoothing may partly explain the lower performance of the classical intensity-duration (ID) thresholds (see Sect. 5.4.2). The hourly time step is also unlikely to affect all threshold types equally: discharge Q and relative soil moisture SU are outputs of a storage-based model and are therefore already low-pass filtered, whereas P, PR and PL are directly degraded by the aggregation. The advantage of the discharge-based thresholds over the rainfall-based ones may thus be partly an artefact of the time step. A test with sub-hourly forcing (if possible), at least for one channel, would clarify this.
The paper is suitable for HESS. Below some specific comments.
Specific comments
Lines 44-45 "The detailed physical trigger mechanisms of debris flows are not yet fully understood and remain subject of ongoing debate." The writer does not agree. The triggering mechanisms are mainly the entrainment of debris into a water stream (runoff-generated debris flows) and the failure of a slope with possible additional contribution of runoff (landslide-induced debris flows).
Lines 46-53 "Theoretical and empirical evidence suggests that climate-change leads, amongst other factors, to increases of precipitation intensities and associated river flood peaks (e.g. Haslinger et al., 2025; Astagneau et al., 2026). As a consequence, it therefore needs to be expected that occurrence frequencies and magnitudes of debris flows will likewise be affected (e.g. Hirschberg et al., 2021; Kaitna et al., 2023a; 2025), emphasizing the need for reliable prediction models." Really both floods and debris flows are influenced by runoff volume. In the case of debris flows the potential sediment volume that a debris flow can transport depend on the runoff volume (Barbini et al., 2024).
Line 79 A regional model that performs particularly well is that of Staley et al. (2017). In this specific case, post-wildfire debris flows, the reduced permeability of the soil makes the hydrological response of the area more homogeneous, which in turn leads to a high reliability of the rainfall threshold during the three years following the wildfire. A description of the sediment source area is missing for channel ID1. Are ID2 and ID3 transport-limited catchments?
Line 212 please explain the distance of the rain gauge from the centroid of the basin with the outlet at the initiation area. Is it possible to show them on the maps of figure 1?
Line 217 Please explain the methods adopted for estimating the discharge; is it possible to show the point where discharge was measured in figures 1 or 2? The number of in-situ discharge measurements should also be reported.
Lines 390-397 The constraints on Q are limited: ENS,Q and R²WL are based on data from 2021 only (camera-based water-level classes and in-situ measurements for a few days), and the number of discharge measurements used is not reported. The Q of the third catchment (ID3c) is not constrained by any flow data: its parameters are sampled from the posteriors of ID1c and ID2c and calibrated against snow cover only, which does not constrain flow magnitudes, although ID3 yields some of the best discharge-based thresholds. The identifiability of the parameters (Table 2), in particular those controlling snowmelt, should be discussed.
Lines 400-405 The model calibrated on the whole catchments is applied to the initiation sub-catchments by changing only the HRU fractions and the elevation distributions, without recalibration or any verification at that scale. The hourly 1x1 km INCA precipitation (l. 210-211) is also applied to sub-catchments of the order of 1 km2 or smaller. The uncertainty intervals shown (e.g. notes to Figs. 9-11) reflect the parameter uncertainty of the hydrological model only: forcing errors and structural uncertainty are not considered, and the error bars of PR and PL are omitted because the uncertainties were "minor", which holds only for the parametric component. The authors should at least discuss how forcing errors and structural uncertainty may affect the ranking of the threshold types.
Line 417 perhaps it is better Berti et al. (2020) rather than Berti et al. (2012) that concerns shallow landslides
Line 636-638 "In contrast, the wettest conditions in a year, here defined as timesteps with PL,i and Q both exceeding lowest PL,i and Q at the moment of observed debris flows in that catchment, have throughout the study years been observed to occur between end of June and beginning of September, with a median date of August 10" the lowest PL,i and Q are the smallest value (measured and modelled??) among those corresponding to the 12 occurred debris flows? Why assuming the wettest year timesteps based on the debris-flow occurrences. In addition, it seems in contrast with what written at line 640: although ~60% of debris flows occurred at discharge magnitudes exceeding the 99th quantile. More generally, since the wettest conditions are defined from the PL,i and Q values at the observed debris flows, the time lag between debris flows and the wettest conditions depends on the levels at which debris flows happen to occur in each catchment. The coupling metric, and the regime classification based on it, are therefore partly dependent on the definition itself.
Lines 650-662 The median dates and the lag Δt = −9 d for ID2 are based on only three seasons (2020-2022), with no observations before the end of May (l. 651-653), whereas ID1 and ID3 cover 18 years. The comparison of the regimes, and in particular the moderate coupling attributed to ID2, should be made with more caution, and this limitation should be mentioned in Sect. 6.4.
Lines 984-988 "These results suggest that pronounced soil moisture deficits and thus also low pore pressures, that develop in early summer due to a lack of further snowmelt, in combination with the only occasional and irregular occurrence of high intensity rainstorms before late summer severely limit both, sediment supply to and transport capacity, i.e. Q, in the moderately steep channel of ID3 and lead to much later occurrence of debris flows." Too long and difficult sentence.
Lines 1048-1054 The authors rightly note that Q results from a non-linear low-pass filtering of the water input. However, the statement that Q is "a more direct and meaningful descriptor of in-stream transport capacity" is supported only by the higher performance of the Q-based thresholds, which does not demonstrate a physical meaning: since Q is a filtered function of the past liquid input, it also carries antecedent-wetness information, and it can detect events at PL = 0 simply through the time lag between input and discharge peaks (l. 1054-1056). The authors should benchmark Q against simple antecedent-wetness indices (e.g. an antecedent precipitation index, or rainfall plus snowmelt accumulated over 24-72 h) to show what Q adds beyond hydrological memory.
Figure 6. How can it represent the whole period if the x-axis runs from January to December? I can see at most two overlapping lines.
References
Barbini, M., Bernard, M., Boreggio M., Schiavo, M., & Gregoretti C. (2024) An alternative approach for the sediment control of in-channel stony debris flows with an application to the case study of the Ru Secco Creek (Venetian Dolomites, Northeast Italy). Front. Earth Sci. - Geohazards and Georisks, 12, doi: 10.3389/feart.2024.1340561
Bernard, M. Gregoretti, C. (2021) The use of rain gauge measurements and radar data for the model-based prediction of runoff-generated debris-flow occurrence in early warning systems. Water Resources Research, 57(3), doi: 10.1029/2020WR027893
Berti, M., Bernard, M., Gregoretti, C. & Simoni, A. (2020) Physical interpretation of rainfall thresholds for runoff-generated debris flows. Journal of Geophysical Research - Earth Surface, 125, e2019JF005513, doi: 10.1029/2019JF005513
Marra, F. (2019). Rainfall thresholds for landslide occurrence: systematic underestimation using coarse temporal resolution data. Natural Hazards, 95(3), 883-890, doi: 10.1007/s11069-018-3508-4
Staley, D. M., Negri, J. A., Kean, J. W., Laber, J. L., Tillery, A. C., & Youberg, A. M. (2017). Prediction of spatially explicit rainfall intensity–duration thresholds for post-fire debris-flow generation in the western United States. Geomorphology, 278, 149-162, doi: 10.1016/j.geomorph.2016.10.019