4.1. 1997 Flood Reconstruction
From the data sources and using the abovementioned methodological approach, we have reconstructed the main 1997 flood characteristics, both the event peak flows and the floodable and hazard maps, and obtained partial conclusions.
The first attempt at hydraulic simulation was carried out from the flows obtained in the hydrological model that used the precipitations registered at the station C106U for the three days between 11 and 13 January 1997. It soon became clear that the Qpm
(peak flow value minimizing the RMSE) exceeded the Qr-r
(peak flow range value derived by rainfall-runoff transformation [28
] for this event). The RMSE value refers to the fit error between the heights of the scars (dendro-geomorphological evidence, FDEs) and the water surface height associated with flood modelling in those locations. To determine this peak flow value, a simulation was then run using the estimated 500-year return period discharge value as the second peak flow value. The results of the modelling continued to show a high RMSE, and the variable flow depth obtained was always below the variable scar height. From these results, simulation using different increasing peak flow values was developed until a lower limit value of the RMSE was obtained. This value was established (Figure 5
A) as a peak flow of 1235 m3
, and it does not appear to be driven by topographic and surface roughness variables, as the sensitivity analysis results show (Figure 5
The notable difference in the peak flow values leads to two possible approaches that will be discussed below: the first, that the rainfall record for the event analysed is not correct, either because of measurement errors or because the spatial heterogeneity of the precipitation meant that the amount recorded by the pluviograph was not representative of the rainfall in the upper part of the Caldera; and the second, the role of solid (sediment and wood) flow in the flash floods that occurred at the study site.
Another point to highlight in these results is that the hydrological model peak flow values [28
] are always higher for the “Cantos de Turugumay” gorge, even considering that the rainfall intensities are similar in the two gorges. However, best fit between hydraulic simulation and dendro-geomorphological data require the peak flow to be greater in the “Verduras de Alfonso” gorge. This would indicate that the rainfall intensities cannot be considered similar in these two basins, because the sizes of the precipitation convective cells are smaller than the dimensions of the basins.
Furthermore, the hydraulic model shows that the peak flows associated with the “Verduras de Alfonso” gorge need to be higher than those flowing along the “Cantos de Turugumay” ravine. For the latter, the necessary peak flow (400 m3
) to minimize the differences with respect to the dendro-geomorphological data is similar to that obtained by Garrote et al. [28
] for a 200-year return period rainfall (one-hour duration hyetograph). However, taking into account these flows in the “Cantos de Turugumay” ravine, the necessary peak flows in “Verduras de Alfonso” ravine (between 675 and 835 m3
) clearly exceed all those obtained from all hydrological models considered. This situation is similar to that previously noted by Ballesteros-Cánovas et al. [41
] in the Kullu district (Western Indian Himalayas), which could be related to the highly localized nature of extreme rainfall events in the headwater catchment of the Caldera de Taburiente N.P.
The limited flooded area linked to the 2D hydraulic modelling peak flows (derived from the rainfall-runoff model without rainfall-altitude gradient applied) is one of the most interesting items of the results obtained for the 1997 flash flood event (Figure 6
(A1)). This can be seen from a simple visual analysis of the flash flood prone areas. Thus, for the v2009 topographical model, only 6 of the 11 trees give a depth value, and this is a low value, so that the RMSE obtained is around 1 m compared with the height of the identified disturbances. In the 1997 model, the number and RMSE are the same as in the previous model, although it is not the same set of trees which provide the depth values.
In addition, these flow differences mean a distinct behavioural change in this gorge. As shown in Figure 6
(A1), it functions either as a braided channel with emerged bars (~50 m3
) or as a single channel occupying the whole river bed (~1200 m3
). The difference in channel behaviour is less evident (Figure 6
(A2)) if we compare both the peak flow obtained using dendro-geomorphological data height (1235 m3
) and that associated with the 500-year return period (510 m3
). In these cases, a single channel occupying the whole river bed is the main river shape. However, significant flow depth differences can be observed between the hydraulic models, which range from 0.25–1 m in the area of “Playa de Taburiente”. Water depth differences (Figure 6
B) for the entire area covered by the hydraulic model show an average value of 1.2 m. However, this number is clearly affected by the values calculated for the narrower sections of both gullies. For the flow velocity data, the differences obtained between the hydraulic models with 1235 m3
and 510 m3
peak flows have an average value of 2.6 m/s, although in this case, as with the water depth, the average value is affected by the flow velocities reached in those narrower sections of the gullies, with the average flow velocity in the area of “Playa de Taburiente” being around 2.2 m s−1
This water depth and flow velocity variability presents us with a situation in which the flow energy increases noticeably even though the area affected by the flood does not present significant topological differences, or not to the point of modifying the channel plan. Thus, it can be observed that the flow velocity in the 510 m3 s−1 peak flow hydraulic model has a clear distribution of high velocity zones in the channels on a braided riverbed, and low speed zones on the bars of that braided pattern, which are also the low water depth areas. This flow velocity distribution is much less clear in the 1235 m3 s−1 peak flow model. Here, although velocities are still higher in the channels, velocities on the bars are also high, showing smoother transitions in the velocity distribution.
The last point is of great importance in flood risk assessment and for generating the relevant mapping to determine the zones affected, or unaffected, by flash floods (Figure 6
(C1–C3) and Figure 7
). For example, taking into account the particular characteristics of the study area (a National Park with a large influx of hikers, over 60.000 people per year), the delimitation of risk areas where people could suffer loss of stability due to flow [39
] will be of great importance for Park management. The results from the hydraulic models (Figure 6
(C1–C3) and Figure 8
) show large differences in relation to the high-risk area. Thus, in the 50 m3
peak flow model, the high-risk areas in the “Playa de Taburiente” are small and are limited to the most developed river channels within a braided channel pattern. In the 510 m3
peak flow model, the high-risk zones are much larger, with all of the channels and the lower bars included in the high-risk zone. Finally, in the 1235 m3
peak flow model, the high-risk zones cover about 85% of the bottom valley area, defining a main central channel with considerable secondary channels on each side.
The variability in the results must be due to the different sources of information used. It has been shown that when only meteorological information is used, even when it is processed with the objective of maximizing both the intensity and the volume of precipitation in a theoretical storm event, the peak flows obtained are still far from those derived when dendro-geomorphological data is included in the analysis. Therefore, the use of indirect flood characterization data (in this case dendro-geomorphological data) has revealed that the magnitude of the 1997 flash-flood would have been underestimated if the assessment was only based on the limited meteorological information available, as previously noted [41
]. This is the case for both the hydrodynamic parameters (peak flow, water depth, and flow velocity) and the related flood hazard. Thus, an improvement in the spatial probability of the areas affected when flash floods occur in the zone can be established due to inclusion in the analysis of dendro-geomorphological data (with potentially major impacts on Disaster Risk Management—DRM) and to the design of mitigation and adaptation plans [41
], taking into account all the uncertainties inherent in the limited data available to compare and contrast the results.
4.2. Uncertainty Sources of Flood Hazard Results
Our data sources and analysis methods have identified several uncertainties from different origins and ranges that can be grouped into four sets: alluvial movable river bed, topography, spatial distribution of disturbed trees, and scar height and dating data.
The movable alluvial river bed, which is linked to processes other than solid load erosion, transport, and deposition during flash flood occurrence, can be considered the first source of error and uncertainty. This solid load modifies the flow hydrodynamics by raising the height of the water sheet and incrementing the density and fluid viscosity, which also increase the flow hydrodynamics’ transport capacity. Other processes can be related to the increment of solid load: forming pools, narrowing the effective section, concentrating the flow, and initiating the sudden release of pooled water. As regards the detritic solid load, the floating woody load promotes similar processes, leading to major hydrodynamic and turbulent changes that may be responsible for the generation and position of the dendro-geomorphological evidence itself [42
Therefore, channel erosion-sedimentation processes may lead to topographical variations, which may be significant in the results obtained from hydraulic modelling [44
], which is why most previous studies have been located over stable bed topography channel reaches [46
]. As the “Playa de Taburiente” river channel was found to have high sediment mobility, a theoretical v1997 DEM was defined with the aim of assessing the influence of this mobile riverbed on the results obtained. However, only slight variations were observed in flow depth values for the different hydraulic models. It was therefore concluded that the possible changes in the river bed topography associated with incision or deposition processes (linked to the liquid and solid flow volumes required to cause the tree disturbances observed) do not affect the results of the analysis.
Moreover, a geomorphological phenomenon that counteracts the uncertainty related to the movable river bed has been observed. It is as if the solid load transport in this reach were to some extent in dynamic equilibrium (there appears to be a compensatory phenomenon), so that the wet section area remains constant over time. This process implies that the position of the erosive and sedimentary forms changes in the cross section over time, but not the channel wet section area. Additionally, this effect is not limited to the cross-section scale, but also operated at the “Playa de Taburiente” scale. This means that the dendro-geomorphological evidence (such as tree scars), although they cannot be used in this case study for a precise flow estimate, give an idea of the order of magnitude of the flow volumes (water plus sediment), over time, for a given section.
Topographic data precision is another source of uncertainty to be considered, as it plays a key role in all hydraulic modelling and controls how representative the results actually are [48
]. If the DEM used in the hydraulic model is not able to represent the terrain morphology, the reliability of the results obtained is questionable [51
]. Our model is adapted to the conclusions by Casas et al. [50
], in the use of LIDAR data and 3D mesh element size for the topographic representation of the terrain. As with topographic data, the spatial resolution of the different variables used in hydrological and hydraulic modelling is another source of uncertainty [52
]. Although it has been impossible to eliminate the uncertainty related to this phenomenon, attempts have been made to limit it as far as possible by using the topographic information with the highest resolution available for the area (1 m pixel size), in both the hydraulic analyses and later in the topographic analysis.
Another source of uncertainty may be related to changes in surface roughness over time, which could be caused by the existence of a movable riverbed, although it can be observed from the aerial orthoimages used (dated in 2002 and 2013) that spatially and temporally most of the channel is occupied by boulder bars. Orthoimages taken at earlier dates, particularly in 2002, show small variations in the extent of the willow forests near the National Park services centre. However, the hydraulic models derived from the two different surface roughness maps do not show significant variations in flow depth values (Figure 5
A), so the uncertainty related to changes in the surface roughness due to movable bed may be considered negligible.
Another source of uncertainty which cannot be ignored refers to the spatial distribution and location of the disturbed trees, which were located through a topographic survey. However, due to the abrupt relief of the area, any minimal displacement or uncertainty in the real location of the disturbed trees may produce significant differences in its topographic elevation. Therefore, these minimal displacements may similarly produce significant differences in hydraulic results. To limit this uncertainty when analysing the match between the height of the disturbances and of the water sheet, a 3 × 3 m buffer was taken into account around the position of the disturbed tree (Figure 9
), and its 9 elevation values were used. The results obtained lead to the following conclusion: although the associated RMSE values (Figure 5
A) vary according to the topographical values used inside the buffer (minimum, maximum, and tree location values), these do not modify the peak flow value associated with the lower RMSE.
Finally, the uncertainty produced by the dendro-chronological methods must also be considered. Up to 63 different wounds were dated in the study area corresponding to 8 event years, but the dendro-chronological methods may possibly not have recorded all the flash floods that occurred [29
]. There is no guarantee that all flood events had been dendro-geomorphologically dated [24
] because, for example, the injuries may be masked by new tree tissues formed after the flash flood occurrence if the wound was old or not extensive [55
], or the damaged trees may have disappeared, swept away by flash floods after uprooting [29
]. With the aim of reducing the uncertainties linked to the scar height, Ballesteros-Cánovas et al. [46
] used a size-gradation classification of scars to estimate peak discharge in ungauged mountain streams. Due to the low number of reliable disturbed trees observed for each flash-flood event, the size-gradation classification is not available in our study area, so the sampling strategy followed was exclusively focused on tree scars in the flow direction of flash floods [56
The validity of the results obtained and their application are determined by these previously noted problems and uncertainties. These uncertainties are not exclusive to the case study of this reach in the Caldera de Taburiente National Park; on the contrary, they are common to all gorges and ravines in the Canary Islands (with limited or no gauging), to most of the Macaronesian archipelagos, and to many mountain basins in the Iberian Peninsula, especially ungauged reaches [27
]. It must be taken into account that the lack of gauge data or the possibility of obtaining quantiles from calibrated rainfall-runoff models are the basis for most of the uncertainty sources considered. For this reason, and in spite of its limited statistical reliability, the information on flood flow magnitude derived from dendro-geomorphological evidence is justified [41
] as the only evidence objectively available and therefore useful for flood risk and hazard assessment.