In this study, we provided compelling evidence that the flow propagation generated by indiscernible experimental conditions indeed exhibits a substantial variability. First, we visualized this variability in form of total exposure probability maps, boxplots and bar charts with whiskers representing the total exposure’s standard deviation. Thereupon we analysed through a suitable statistical method (i.e. ANOVA) if and to what extent the considered experimental factors exerted a significant effect on the response in terms of the mean total exposure of the alluvial fan surface. Whereas we could corroborate the main effects due to both the solid fraction and the released total volume, no main effect due to stream power could be retracted. We could also statistically underpin that all interactions among the aforementioned factors explain a component of the overall variability of the response and we visualized them in detail through interaction plots.
These results have far-reaching implications for hazard and risk assessment on alluvial fans where sediment-laden flows may constitute the main formative processes:
First and foremost, the inherent variability of the flow propagation on the alluvial fan surface strongly suggests that components of randomness do play a relevant role in the distributary dynamics of extreme events on real-world alluvial fans as well (Fuchs et al. 2013). Sources of randomness can for instance reside in the effects of autogenic processes, in the inhomogeneity of grain size distributions and marginal changes during the hazard event as a suddenly mobilized boulder or disrupted log jam, minor fluctuations of model boundary conditions (i.e. unaccounted sediment-laden flow pulses), or any unexpected clogging of a flow section, which could induce singular behaviours so that patterns of deposition and exposure could change unpredictably. Although we concede that it was not possible to establish the causalities between the aforementioned sources of uncertainty and the variability of the observed patterns of exposure, we could experimentally prove the existence of this variability and quantify it accordingly for the considered alluvial fan layout.
Although the experimentally observed magnitude of variability is not representative of other experimental settings (i.e. different geometry and dimensions of the alluvial fan models, different types of flow, the scaling issue), recent experimental findings indicate that this randomness generally exists:
Santibañez et al. (2021), for example, conducted experiments on an analogous small-scale alluvial fan model (i.e. without the presence of the curved guiding channel) varying systematically the total discharge volume, the sediment fraction of the biphasic mixture and the applied stream power. In their experiments, a fixed proportion of the solid fraction was constituted by large wood. Their results confirm that the patterns of exposure associated with specific loading conditions exhibit remarkable randomness and that the fixed portion of large wood strongly also interfere with the distributary dynamics on the alluvial fan surface.
Blasi et al. (under review) accomplished experiments on two small scale alluvial fan layouts, one with a guiding channel equipped with a crossing bridge and one featuring the guiding channel without crossing structures. This study also confirmed that the patterns of exposure associated with specific loading conditions exhibit a remarkable variability.
Moser (2018) studied fluviatile hazard processes on a torrential fan (i.e. Schnannerbach alluvial fan, Tyrol, Austria), executing many complex experiments on a large Froude-scaled (1:30) model. The model accurately reproduced the natural conditions and covered a set of buildings on the torrential fan (for more details on the experimental set-up see Sturm et al. 2018a, b). The sediment deposition patterns of five experiments, each accomplished identically and with experimentally indiscernible boundary conditions, were analysed. By referring to one experiment as a reference case, the obtained spatial patterns of differences were remarkable. This highlights that the geomorphic work accomplished throughout the unfolding of each experimental run is essentially unique.
Based on this premise we contend that hazard assessment on alluvial fans based on a single deterministic flow simulation for specified initial and boundary conditions (i.e. inflow fluxes for a given return period) might be insufficient for reliably detecting the intervening autogenic processes and the abovementioned tipping process patterns that may change the exposed areas to a significant extent. In the authors view this problem might be serious on unconfined landforms, such as alluvial fans and fan-deltas, in particular if hazard mapping decisions are based on one single simulation with a given setup, without any sensitivity analysis and additional expert judgement.
Hazard maps elaborated according to the existing standards may be based on the aforementioned drawbacks, fail to adequately mirror hazard-prone areas or overemphasize the exposure of certain indicated sectors (i.e. by definition the probability of exposure is equal to one). This may contribute to jeopardize the effectiveness of public and private investments into supposedly hazard proof buildings and settlements.
In the introduction, we mentioned the Rio Blanco case in Chaitén (Lake district, Chile) as a paradigmatic example. Basso-Báez et al. (2019), for example, analysed the tipping point behaviour of the flow obstruction and diversion effects of different bridge clogging scenarios. Through tailored simulations of considered hydrologic scenarios, they could demonstrate that the cross-sectional blockage at the “Austral Road” bridge represents a pivotal factor in determining the extent of the inundated areas and associated impacts in the urban area of Chaitén which is partly located on a fan-delta.
In light of the experimentally detected process variability, a word of caution must be mentioned here also concerning the calibration of numerical models based on documented events. Forcing the simulation model to reproduce an observed event by calibrating the model parameters could be misleading, since different initial and boundary conditions contrived to identity in the simplified modelling environment could exhibit, upon occurrence, fundamentally different process patterns. A sensitivity analysis concerning the overlooked differences in both initial and boundary conditions would be precluded and hazard and risk assessments would potentially be inaccurate.
Again, concerning the Rio Blanco case in Chaitén, Basso-Báez et al. (2019) point out that complete disclosure of the different sources of uncertainty that may affect the predictability of hazardous phenomena (e.g. Pate-Cornell 1996; Merz et al. 2008) is essential for an informed decision-making process. As outlined by Fuchs et al. (2008) and Mazzorana and Fuchs (2010b), several sources of uncertainties should be considered concerning hydrological hazards and torrent processes including (i) uncertainties about the possible range of rheological behaviour and about the liquid-solid mixture concentration of the analysed flow phenomenon, (ii) uncertainties in system loading assumptions (e.g. duration-intensity related uncertainties, uncertainties related to sediment transport rates, uncertainties emerging from LW transport), (iii) uncertainties in system response mechanisms (e.g. localised obstructions that divert the flow patterns, influence of small-scale topological features), and (iv) uncertainties related to morphological changes in the response system. In this study, we specifically addressed the last source reported on this list and explored the effects on the generated exposure on the experimental alluvial fan surface.
To limit this source of uncertainty, we propose that the development of numerical models capable of mirroring more reliably the stochastic sediment-laden flow behaviour on alluvial fans should be conducted considering the ongoing experimental research on alluvial fans. Large-scale experimental alluvial fan models could be regarded as a valuable touchstone for model development. Specifically, we suggest, that the quality of a numerical model setup should be judged concerning its capability to reproduce not a single hazard event but sets of experimental responses associated with the same, experimentally indiscernible, initial and boundary conditions. So rather than minimizing the discrepancies with respect to one, deterministically obtained, model output, it should be attempted to minimize the probability of a mismatch between a set of simulations outputs and the associated probabilistic experimental results.
Adopting this probabilistic calibration concept, particular attention should be devoted also to enhancing the model-based detection of the autogenic phenomena occurring on the alluvial fan surface during process propagation. Again, regarding the flood hazard analysis in Chaitén (see Basso-Baez et al. 2019), such an enhanced modelling concept, integrating both experimental and numerical results, would have helped to better detect potential autogenic processes such as channel-backfilling and avulsions with profound repercussions on the resulting hazard propagation patterns. Of course, our “similitude by process” based experiments serve only as a gross analogy of the suite of processes that intervened during the unfolding of the inundation of Chaitén and extended benchmark research, with different model developers testing their products against large sets of experimental data obtained from a Froude-scaled experimental model would be warmly welcomed.
Our results suggest at least as a promising starting point, that allogenic factors such as stream power or discharge in the feeding channel may be rather ambiguous control factors. On our experimental alluvial fan, no main effect on the mean total exposure imputable to this factor could be retracted.
The elaborated exposure probability maps show that experiments conducted imposing half stream power generate extended areas with a non-zero probability of exposure. Moreover, the associated spatial patterns indicate a remarkable uncertainty and the areas assumed as safe (i.e. with a zero-valued probability of exposure) are patchier and, in several cases, less extended.
The performed statistical analysis (ANOVA) further indicated that the different interactions between the considered factors exert a significant effect on the mean total exposure. Overlooking the role of these interactions and basing investment decisions aiming at effective risk mitigation on a “design discharge” alone, as it is common practice in many mountain regions of the world, is a risky endeavour (Mazzorana et al. 2013).
Due to the profound implications for our ability as a society to proactively adapt and enhance our resilience to the impacts of flood hazards in mountainous and alpine environments, fostering tailored research aiming at experimentally and numerically unravelling the complexity and hidden variability of sediment-laden flows on unconfined landforms (i.e. alluvial fans) becomes essential.
The first fundamental research challenge consists in making this process complexity and hidden variability experimentally observable, since the available historical record, no matter how extended, remains largely incomplete (De Haas et al. 2018). Once dealt with this first challenge, a detailed detection of the underlying spatial and temporal variabilities is important for the generation of datasets with minimized imprecision and maximized accuracy. In a subsequent research step, these data sets should be analysed with suitable analytical and statistical methods. Thereupon, endowed with reliable quantifications of the emerged variability of the observed process propagations and the identification of all relevant sources of uncertainty, the attempt has to be made to devise computational modelling strategies capable of adequately mirroring process stochastics by nudging or coaxing the deterministic simulation models or by introducing elements of randomness. The gained insights and the generated knowledge would be essential requisites for further developing the societies’ ability of proactive adaptation.