Exploiting LSPIV to assess debris flow velocities in the field

The assessment of flow velocity has a central role in quantitative analysis of debris flows, both for the 7 characterization of the phenomenology of these processes, and for the assessment of related hazards. Large scale particle 8 image velocimetry (LSPIV) can contribute to the assessment of surface velocity of debris flows, provided that the specific 9 features of these processes (e.g. fast stage variations and particles up to boulder size on the flow surface) are taken into 10 account. Three debris flow events, each of them consisting of several surges featuring different sediment concentration, flow 11 stage and velocity, have been analyzed at the inlet of a sediment trap in a stream of the eastern Italian Alps (Gadria Creek). 12 Free softwares have been employed for preliminary treatment (ortho-rectification and format conversion) of video-recorded 13 images as well as for LSPIV application. Results show that LSPIV velocities are consistent with manual measurements on 14 the ortho-rectified imagery and with front velocity measured from the hydrographs in a channel reach approximately 70 m 15 upstream of the sediment trap. Horizontal turbulence, computed as the standard deviation of the flow directions at a given 16 cross-section for a given surge, proved to be correlated with surface velocity and with visually estimated sediment 17 concentration. The study demonstrates the effectiveness of LSPIV in the assessment of surface velocity of debris flows, and 18 permit to identify the most crucial aspects for improving the accuracy of debris flows velocity measurements. 19


61
To our knowledge, the application of LSPIV on debris flows from video images has not been deeply investigated whereas it 62 could provide direct measurement to quantify rheological behavior of debris flows. Our objective is to test the LSPIV method this work are to explore: 1) the spatial and temporal variation within one study reach of debris-flow surges occurred in the period 2013-2015, 2) a detailed analysis of an individual surge dynamic, 3) the quantification of a "horizontal turbulence   (Cavalli et al., 2017) and amounted to about 5200 m 3 km -2 yr -1 .

78
Instrumented monitoring of the Gadria catchment began in 2011, for detailed information of the study site and monitoring 79 setup, refer to Comiti et al. (2014).

80
Two cameras are alongside a sediment trap (retention basin) near the alluvial fan apex, one looking upstream (Cam1) and the 81 other looking down at a more perpendicular angle to the flow (Cam2). The third camera (Cam3) is in the next reach upstream 82 from the sediment trap at a closer proximity to the flow (Fig. 2). These three cameras are connected to a cabin equipped with 83 power supply and a server (8 Tb storage capacity) collecting all the monitoring data. The fourth camera is in an upstream 84 ravine and it is triggered by a rain gauge when there is at least one minute of rainfall. For this study, we focused on the 85 application of LSPIV using only one of the four MOBOTIX M12 video cameras, Cam 2.

86
We attempted to utilize the other cameras for LSPIV application, but Cam 1 and Cam 3 were too close with an upstream view 87 to measure the large scale of the debris flow. Within the area of high incidence angle of the images, the number of reference 88 points is restricted, there is little spatial coverage, and there was too much pooling of water in front of Cam 1 located at the 89 dam. Cam 2 was the best option because it was located higher on top of the levee (10 -52% incidence angle), 12 -46 m from 90 the LSPIV area, and had an orientation more perpendicular to the flow path. Cam 4 was problematic due to the unchannelized 91 nature of the recorded events, in combination with the relative long distance between the camera and the moving sediment.  The M12 Mobotix security camera that we used is an IP camera (resolution 1689x1345 pixels) with a fish eye lens, at night 106 spotlights are activated during rainfall. This camera has limiting features such as an automatic adjustment for shutter speed 107 with illumination, and therefore the frame per second cannot be fixed. This initially was a problem since our aim was to have 108 a constant 10 frame per seconds (fps). During recording of the flow events, the frequency reduced to 2 -3 fps because of the 109 low lighting of the storms. We needed a standard frame rate for LSPIV calculations, therefore we subsampled the images to 110 the minimum frame rate of each flow event (Table 1).

111
Also, since the camera had a fisheye lens, significant distortion correction was required. A checkerboard pattern image from 112 the camera was used in a free software Hugin (http://hugin.sourceforge.net) which has a tool for distortion correction. This

113
was then applied to all the video imagery and converted to an ASCII grey scale format using batch processing in the XNview 114 freeware (www.xnview.com). This used to be necessary for the Fudaa software, however the more current version can now 115 handle jpeg and tiff colored formats.

131
For orthorectifying the video images, targets and natural features were used as reference points for matching between the SfM 132 point cloud (both pre-event and post-event) and video imagery ( Fig. 2A, 2B). Corners of rocks next to the flow line were 133 typically used on each side of the channel, and sometimes exposed stable rocks within the channel. Alignment errors of the 134 reference points (Table 1)

145
To have a good spatial distribution of the flow with a manageable dataset, we selected a grid with an approximate spacing of

151
The spatial distribution of velocity vectors covering the reach provided an opportunity to examine their variation (direction 152 and velocity fluctuation) to characterize the turbulence of the various debris-flow surges (Costa 1984). Since our LSPIV 153 method is in the two dimensions, we define it as the horizontal turbulence index according to directional variation (Td) and 154 velocity variation (Tv). We measure the turbulence (Td and Tv) by taking the standard deviation of vector orientations (Td) and 7 the changing characteristics of the surges rather than the spatial distribution. Therefore, small Tv and Td should characterize 157 laminar flow conditions and higher values should be associated to more turbulent flows.

158
The LSPIV results were taken from cross-section XS (Fig. 1C) to have accurate comparisons of debris flow surges. This is the 159 most stable cross-section before the widening in the sediment trap. It is also the closest and most perpendicular view from the 160 camera resulting in the most accurate LSPIV calculations. The LSPIV study reach experienced important deposition and 161 remobilization during the debris flow surges, therefore we did not attempt to measure the discharge rates.

174
The 2013 event featured one important surge, very typical debris-flow formation with a boulder front and the slurry-like tail.

175
The singular surge provided a convenient detailed analysis of the front, intermediate stage (transition from front to tail), and 176 the tail (described later).

177
The 2014 event had a small preliminary surge (pre-surge) and four debris flow surges (S1-S4) passing through the study reach.

178
It should be noted that there was a discontinuous surge that stopped just upstream of the LSPIV measurements before the first 179 measured surge passed through the reach. The first two measured surges were large enough to distinguish the front (S1 and 180 S2) and tail (S1 tail S2 tail) and the latter two were too small and were kept undivided (S3 and S4). There seemed to be a 181 higher water content with longer sustained fronts (compared to 2013). The S4 was unusually fast which behaved more of a 182 wave passing through the filled-up sediment trap of highly saturated deposit.

183
The 2015 event was especially interesting because of the variable rheology of the surges. High-intensity rainfall covered the 184 entire catchment triggering many different source areas. The first surge (S1) had little sediment but carried a lot of large woody 185 debris. S2 was a slower muddier flow, however cobbles and boulders were also transported. S3 was a larger and even slower  * the first actual debris flow surge stopped between the LSPIV and the radar, it remobilized with S1.

236
The intermediate (transition from front to tail) shows a distinct decrease in velocity with a more homogeneous distribution.

237
Zero velocities correspond with the boulder front deposition. The low-velocity tail becomes more confined traveling around 238 the boulder front as a more laminar flow (Fig. 6C).

239
Three cross-sections were examined to compare the velocity-time profiles of the event (Fig. 7b). The peak velocity in the front  Figure 7 with the boulder front depositing on the higher convex features with the water surge 248 passing around in the lower confined channel.

249
The longitudinal profile of the average velocities combined with the video imagery and multi-date topography (Fig. 8) 250 distinctly show the boulder front depositing after the sudden decrease in local slope (down to a negative slope) and channel 294 however, more surges need to be measured to better define the function. The influence of spatial and temporal sampling 295 resolutions also needs to be better understood for further application.

304
Higher image resolution and camera speed might give further insight on boulder dynamics and log jamming.

(c) (d)
We have presented LSPIV-derived velocities for three debris flow events in the Gadria channel, for a total of 11 surges and 313 these velocities were compared with manual measurements on the ortho-rectified imagery (mean difference of -0.1 m s -1 ) and 314 upstream radar sensors (mean difference of -0.9 m s -1 ). LSPIV appears to be a reliable method for measuring velocities of such 315 flows, and to the best of our knowledge, this is one of the first studies on the topic. The variation of vectors from the LSPIV 316 was introduced as an index of horizontal turbulence according to directional variation (Td) and velocity variation (Tv).

317
Within the studied reach, debris flows varied in velocity and turbulence among different events, among individual surges

327
LSPIV application on debris flows has shown to be very effective but there still needs to be a better understanding of the spatial

340
Funding for this research came from the research project "Kinoflow" funded by the Autonomous Province of Bozen-Bolzano.

343
The debris-flow monitoring station of Gadria catchment is managed by the Civil Protection Agency of the Autonomous

344
Province of Bozen-Bolzano. A preliminary analysis of the debris-flow hydrograph conducted by V. D'Agostino and F. Bettella