Hyperconcentrated flows shape bedrock channels

The paper addresses how bedrock gorges form and the relative role of single extreme events versus steady fluvial incision. While some canyons are attributed to long-term moderate flows and others to catastrophic floods, the timing and mechanisms of inner gorge formation in the Central Alps remain debated. Turbulent-flow incision mechanisms (abrasion, plucking, cavitation) and channel geometry controls (sinuosity, slope, confinement) are well studied, but direct quantitative observations of bedrock erosion during extreme events are rare. Existing stream-power models typically omit debris and hyperconcentrated flows, despite suggestions that such events may dominate bedrock scouring. The study aims to directly observe and quantify bedrock erosion from a single hyperconcentrated flow and to test whether erosion patterns depend on turbulent-flow indicators. The authors hypothesize that: (i) hyperconcentrated flows cause massive bedrock erosion, (ii) erosion is not governed by turbulent-flow geometric indicators (e.g., sinuosity, channel shape), (iii) erosion results from mechanically induced breakout of rock fragments by suspended-load impacts exceeding rock strength, and (iv) repeated hyperconcentrated flows during phases of enhanced precipitation or sediment supply can drive rapid gorge formation.

Prior work distinguishes fluvial bedrock incision processes such as abrasion, plucking, and cavitation, modulated by the tools and cover effects of sediment supply and by channel geometry (sinuosity, curvature, width, slope). Highest fluvial abrasion rates often occur at moderate bedload where cover is limited. Lateral migration and bank erosion increase with bend curvature and slope in turbulent flows, with erosion concentrated near the bed and rapidly diminishing with height. Stream-power-based models capture broad patterns but often fail to represent process-specific erosion and neglect extreme events like debris and hyperconcentrated flows, which can have high transport capacity and may dominate long-term incision. Observations of debris-flow and flood erosion exist, but few direct quantifications in bedrock channels can attribute erosion to a single extreme event. The literature suggests that suspended loads in hyperconcentrated flows can enhance wear and that long-term incision rates commonly exceed short-term fluvial measurements, implying a role for infrequent, highly erosive events.

Study area is the Höllental gorge (Germany), a 900 m long, up to 180 m deep bedrock gorge in Triassic Wettersteinkalk limestone (UCS 91±27 MPa; tensile strength 7.2±1.9 MPa). The catchment (~10 km²) has a mean slope of 110%. On 13 June 2020, a locally intense rainfall (50–60 mm/h) produced a ~60,000 m³ hyperconcentrated flow lasting ~1 h. Pre- and post-event terrestrial LiDAR surveys (RIEGL VZ-400) were conducted on 28 May 2020 (84 scans) and 19/25 June 2020 (89 scans), respectively. Scans overlapped sequentially for robust co-registration (ICP in RiSCAN Pro 2.9), achieving RMSE <1 cm. Point density was normalized to 2.5 cm in XYZ. The channel bed below water had sparse coverage, so analysis focused on lateral wall erosion below the mapped erosive flow height (from color-change high-flow marks documented via 360° imagery). Change detection used M3C2 in CloudCompare with normal scale D=1 m, projection scale d=0.5 m, and max cylinder length 3 m; LOD95% combined intra-cylinder point-cloud variability and alignment uncertainty. Significant changes (LM3C2 > LOD95%) were manually validated with photographs; positive distances indicated deposition and negatives erosion. For each of 232 significant erosion patches, pre- and post-event point subsets were isolated, triangulated, projected onto a plane parallel to the wall, and volumes/areas were computed via Cut and Fill in RiSCAN Pro. Mean erosion depth per patch was the mean M3C2 distance within it. The gorge was segmented into eighty-nine 10 m sections along a manually defined baseline (lowest recorded points at 1 m intervals), and section-wise mean lateral erosion was computed by normalizing eroded volume by scanned pre-event wall area below the erosive height. Channel metrics per section included mean/min/max width (perpendicular to axis, measured at the erosive height), average slope, convergence/divergence percentage, curvature (deviation from adjacent segments), sinuosity (deviation from regional valley azimuth of 33.69°), and boulder density (planform boulder area from pre-event meshes rasterized at 0.1 m divided by estimated bed area). Statistical analyses of magnitude–frequency relations for eroded volume V, area A, and mean M3C2 distance D used kernel density PDFs and CDFs with power-law fitting via maximum likelihood estimation (estimating xmin, β, γ) and robust linear regressions for both CDF and PDF tails; goodness-of-fit p-values and bootstrap-based confidence intervals were computed. Erosion height dependence was assessed by binning erosion metrics in 0.1 m increments above the baseline.

• The hyperconcentrated flow caused extensive geomorphic change over the entire 900 m gorge within ~1 h, including mobilization, shifting, and disintegration of boulders. A ~20 m³ boulder moved 16 m downstream; a 3.5 m³ rock fragment detached from a wall with up to 1 m lateral erosion depth.
• Change detection identified 232 statistically significant lateral erosion patches totaling 20.60 m³ of bedrock removed from 137.01 m² of wall area; 2.4% of the 5708 m² scanned wall area below the erosive height experienced detectable erosion.
• Erosion was spatially heterogeneous: >33% of the total eroded volume occurred between 420–500 m downstream (10% of channel length), a section pre-charged with boulders. The largest single eroded fragment volume was 3.3047 m³ with a maximum mean erosion depth (per patch) of 60.11 cm; at that site (∼4 m width) the channel widened by ~15%.
• Section-wise mean lateral erosion ranged from 0.018 to 42.97 mm; the mean over all sections with detected erosion was 4.89 mm, and 3.41 mm when averaged over the entire gorge including sections without detected erosion (minimum estimates given method detection limits ~2 cm per patch).
• Erosion-height dependence: Highest mean erosion depths occurred between 1.0–1.4 m above the bed (13.4–14.1 mm per 1 h event), decreasing generally with elevation up to ~5 m; erosion also occurred higher where flow depth reached.
• Correlation tests (21 combinations) showed no detectable relationship between erosion metrics and local channel geometry proxies indicative of turbulent flow (sinuosity, curvature, width, slope, convergence/divergence) or boulder density at the 10 m scale.
• Magnitude–frequency distributions of V, A, and D exhibit rollovers with power-law tails; xmin values indicate 19.8% (V), 26.3% (A), and 31.9% (D) of patches in the tail account for 85.8% (V), 81.2% (A), and 60.7% (D) of totals. Power-law fits had p-values >0.1 and stable exponents under bootstrapping. The exponents are consistent with rockfall/rock slope failure inventories, implying mechanically controlled wall retreat.
• Relative efficiency: Compared to a study of 208 induced turbulent flushing events yielding ~0.4 mm/a lateral erosion, this single ~1 h hyperconcentrated flow produced 3.41 mm average lateral erosion across the gorge (8.5× larger annualized magnitude for this event), highlighting far greater erosive efficiency.

The results demonstrate that a single hyperconcentrated flow can laterally erode massive limestone walls by millimeters to centimeters over hours and widen narrow sections measurably, far exceeding typical short-term turbulent fluvial incision. The lack of correlation between erosion and turbulent-flow proxies (sinuosity, curvature, width, local slope, convergence/divergence) suggests that during hyperconcentrated flows, erosivity is not governed by near-bed bedload dynamics and secondary flow structures. Instead, elevated suspended sediment concentrations throughout the flow depth increase density and viscosity, enhancing particle-particle and particle-bedrock collisions and generating normal and shear stresses capable of fragment breakout. The vertical distribution of erosion, peaking near 1–1.5 m above the bed and declining upward, is consistent with a decrease in suspended coarse fraction and collisional stresses with height, potentially modulated locally by transient bed cover during peak flow. The power-law magnitude–frequency behavior aligned with rockfall datasets indicates that erosion proceeded via mechanically enhanced wall retreat controlled by rock mass predisposition (fracture spacing/orientation), not by local hydraulic geometry. Collectively, these findings support the hypothesis that repeated hyperconcentrated (and debris-flow-like) events, especially during periods of abundant water and sediment supply (e.g., paraglacial phases), can produce punctuated, highly efficient incision and widening of bedrock gorges, helping reconcile the mismatch between short-term fluvial measurements and higher long-term incision rates. Incorporating such event-driven processes into landscape evolution models is therefore crucial for realistic predictions of gorge development.

• This study presents, to the authors’ knowledge, the first gorge-scale quantitative change detection of natural hyperconcentrated-flow-induced bedrock erosion, resolving 232 erosion patches over 5708 m² of wall area.
• A single ~1 h hyperconcentrated flow produced average lateral wall retreat of 3.41 mm across the gorge, with section maxima up to 42.97 mm and local widening by ~15% in a ~4 m wide reach.
• Erosion decreased with height above the bed, peaking around 1.1–1.2 m at ~14.1 mm, and showed no dependence on turbulent-flow geometric indicators (sinuosity, curvature, width, slope, convergence/divergence) at the studied scale.
• Magnitude–frequency relations of eroded patches follow power-law tails similar to rockfall datasets, indicating mechanically dominated breakout controlled by rock mass predisposition rather than local hydraulic forcing.
• Hyperconcentrated flows have greater erosion efficiency, broader vertical influence, and higher transport capacity than typical turbulent flows, underscoring their major role in bedrock gorge incision and the need to include such events in landscape evolution models.
• Repeated hyperconcentrated/debris-flow events during periods of high sediment and water supply (e.g., Lateglacial) likely accelerated gorge formation. Future research directions: quantify vertical incision concurrently with lateral erosion during such events; acquire multi-event datasets for recurrence quantification; integrate event-based mechanics into catchment-scale incision models; assess lithology/fracture control generality across diverse bedrock types; and constrain thresholds for transitions between turbulent, hyperconcentrated, and debris-flow regimes in natural gorges.

• Data coverage limitations due to gorge geometry (overhangs, narrow sections) and water-covered bed led to occlusions and sparse bed data; analysis focused on lateral wall erosion, not bed incision.
• LiDAR detection threshold (~2 cm at rock faces without ground control) likely underrepresents thin erosion, making reported lateral erosion conservative minimum estimates; vertical incision may be up to three times lateral values but was not measured directly.
• Single-event, finite dataset: the hyperconcentrated flow lasted ~1 h and was bracketed by scans two weeks before and within ~1–2 weeks after; repeating such observations is impractical.
• Section-scale (10 m) correlations may miss sub-section-scale patterns below detection thresholds; local reach-scale slope influence cannot be fully excluded despite no local correlations.
• Bedload and transient sediment cover during peak flow could bias near-bed erosion estimates due to limited bed visibility post-event.