Uncertainty in measuring the role of climate change on debris-flow triggering on volcanoes - bulk-density, temperature and moisture analysis at Unzen Volcano (Japan)

As climate change creeps into the 21st century, the intensity of debris flows due to heavy and concentrated rainfall has increased in mountainous regions of Japan and East Asia. However, the relationship between climate change and an increase in debris flows is likely to be non-linear. Rainwater infiltrates more quickly into porous material, and the lack of vegetation cover increases evaporation and the temperature of surface sediments. In addition, periodic gully collapses bring fresh layers of porous material that increase the distance between the surface and the vadose zone. Therefore, to understand the relationship between volcanic debris flows (or lahars), parameters such as density, porosity, temperature, and moisture retention must be captured in detail in both time and space. The aim of this paper is to assess the role of loose, coarse, and porous sediments in lahar triggering. The present study was conducted at Unzen volcano in the Tansandani Guly between 31 May and 1 June 2023, 30 years after the last eruption. The dacite material is composed of a matrix of sand and coarse sand mixed with larger fractions and blocks, therefore traditional density measurement methods could not be applied, and a photogrammetric based method was used. In the field, sets of SfM-MVS photographs were taken before and after digging a hole in the sediments so that the measured mass could be compared to the volume of the hole in the sediments. After the sediments were dried, the dry and wet density, bulk density and porosity of the sample were calculated. When compared to the temperature data collected in the field, the following results were obtained: (1) The porosity of the volcanic material was highest in the lower reaches, followed by the upper reaches, and lowest in the middle reaches. This may be because fine sand washed out of the upstream area by rainfall is currently stored in the midstream area, which may facilitate debris flow generation. In addition, the downstream area has a high porosity, which may be due to the surrounding vegetation preventing the influx of new fine sand from the channel wall. (2) Because of the higher porosity and the lack of organic matter and vegetation cover, the increase in temperature acts more directly on the decrease in water content than in mountainous areas. Consequently, empirical equations for the potential for mudslides in volcanic areas with respect to surface temperature and soil moisture need to be developed for hazard and disaster risk management purposes.


Conceptual background
Recent developments in close-range photogrammetry such as SfM-MVS (Structure from Motion and Multiple-View Photogrammetry) and LiDAR (Light Detection and Ranging) has triggered an increase of topographic surface comparisons to measure erosion and deposition [1], at the coast [2], in landsliding areas [3], after post-fire geomorphological changes [4], and in volcanic areas such as Sakurajima Volcano [5].These changes analysis are based on surface changes, and do not include variability in bulk density, which is influencing the measure and calculation of erosion and deposition.There is thus a need to determine whether the density is constant over a deposit, and whether there are associated parameters that will change with it, especially in areas that are geomorphologically active, and where the accumulated sediments can become the source for hazards, such as debris-flows or lahars on volcanoes.

Research location and characteristics
Unzen Volcano is a stratovolcano that erupted last during the period 1990-1995 [6].The eruption was characterized by a series of dome-collapse block-and-ash flows, intercalated with volcanic debris-flows or lahars.The event took the lives of 44 and buried different hamlets and portions of Shimbara-City on Shimabara peninsula where the volcano is located (Figure 1).
Since the eruption, lahars have been diminishing in number and in amplitude, mostly due to erosion processes modifying the geometry of the watersheds over time [7], and due to change in the grain-size of the material with the finer fractions being preferentially transported away, leaving only coarser fractions, making the triggering of lahars more difficult [8].However, these watershed-scale parameters are certainly varying spatially within the watershed, and the present contribution assumption is that the different level of energy of the sediment transport events (wall collapse, rainfall rill erosion, debrisflows...) are known to organize grains in different patterns, and thus the sediment density must be differing as well.
The present contribution is therefore proposing to analyse variations in the density along the Tansandani gully (a tributary of the Mizunashigawa valley), because different densities will also influence the moisture content and the ability of the sediment to retain water after a rainfall event.Finally, different densities and different moisture contents will then modify the temperature at the soil surface, which in turn will drive evaporation and the residence time of water in the sediment interstices.3 Because sedimen1t bulk-density and moisture content will be essential to define in detail the number of antecedent rainfalls and rainfall intensity necessary to generate lahars and to define the sediment resistance to transport (for instance thinking through the lens of the Coulomb's law), the authors have investigated these parameters to refine the understanding of debris-flow tracks in the Tansandani Gully.This question and main objective, can thus be sub-divided into a set of sub-objectives: (1) Calculate the different bulk density values of the loose sediments.
(2) Measure the moisture content and define its relation to the bulk-density.
(3) Define spatial relations between these variables.( 4) Quantify the error of the density values based on site-specific data.

Methods
For this purpose, took 11 samples at the three locations (Figure 1).The authors generated 3D pointcloud models using close-range SfM-MVS (Structure from Motion -Multiple View Stereophotogrammetry) in the field, combined with temperature measurement, and with sediment sampling and laboratory characterization of porosity, water content (Figure 2).

3D model generation
The authors have used the method developed by Gomez [9], and Gomez et al. [10] It provides a tool to calculate the density of coarse unconsolidated material of heterometric size using two surfaces taken before and after sediment sampling.At the three different locations, a total of 11 samples were collected with photographs ranging from 60 to 80 photos.The photographs were taken using a Galaxy camera 2 (Samsung) and the built-in f/2.8-5.9 23-483mm equivalent lens (21X zoom), with a fix focal.The photographs were then loaded into the SfM-MVS Agissoft software to generate the pointcloud, using a well-known process.Then, by subtracting the two obtained point-cloud surfaces, the authors calculated the volume extracted (which then compared to the mass provides the wet-and dry-bulk density), as well as the error in Z, propagated twice (from the top and lower surface), using the open-source CloudCompare.(1)   ∶       ∶   ℎ   ℎ

Field Sediment sampling and analysis
At each location where the SfM-MVS data was collected, the were collected using a small shovel, taking care not to compress or disturb the material.Any material creeping in the hole was removed by hand.The sample was double-bagged, sealed, and brought back to the laboratory to measure the wet and dry bulk-mass.At the three different locations, the temperature of the surface was also measured using an Infrared Thermometer.Combining the 3D volume calculated from SfM-MVS and the mass of the samples, the density of the samples was derived (Figure 2).

Results
The average bulk density at the confluence was 1.41 g/cm³ (1.361 g/cm³ except for sample 5), with higher densities of 1.622 g/cm³ and 1.592 g/cm³ in the lower and upper midstream areas, respectively (Figure 3).Sample 5 at the confluence, where the density was 1.610 g/cm³, showed a large difference from the other samples 1 to 4. This is most likely because sample 5 was collected at a natural bank of the stream, where fine particles were deposited, and thus exhibited an outlier-like value.MAE values ranged from 0.018 cm to 0.134 cm (0.046 cm on average) for all samples, with very small density errors.The porosity of the samples ranged from 24.37% to 37.65% at the confluence of the valleys, with an average of 33.28%.Excluding sample 5 at the confluence of the valley, the mean at the confluence of the valley was 35.51%.Porosity in the lower middle reaches ranged from 20.72% to 28.15% with an average of 23.75%, and in the upper middle reaches ranged from 23.10% to 26.27% with an average of 24.46% (Figure 4).Moisture content ranged from 4.18% to 7.09% and was not characterized by location.
In relation to this moisture content value, the temperature of the material was found to range from 20.7°C to 45.1°C.Although temperatures increased from midday to late afternoon (samples 1-3 in the downstream and upper-midstream reaches), temperatures varied somewhat more complexly due to shading by trees and cliffs, especially in the narrow gullies (Figure 4).

Discussion
The value of MAE (Ave: 0.046 cm) is very small in this study, and we believe that the accuracy of density estimation using SfM-MVS is high.This is most likely due to the fact that the number of GCPs were spaced 10 cm apart, which resulted in a higher GCP density, as well as the fact that the positions of the GCPs were manually adjusted on a pixel-by-pixel basis.This method is useful for density estimation in volcanic areas where sediments are prone to collapse, but the manual adjustment of GCP positions is time-consuming.
Regarding the relationship between moisture content and temperature, a clear correlation between surface temperature and moisture content was observed in open areas without vegetation (valley confluence), but in the upper and lower middle reaches, the relationship between temperature and moisture content was complicated by shadows caused by surrounding cliffs and trees.Therefore, in order to predict water content from surface temperature, it is necessary to consider the position of the sun and the shadows caused by the sun's position at that location.The density and porosity results from this study suggest the condition (Figure 5).This condition consists of the following three components.(1) Fine-grained sediment supplied by weathering and erosion of the upstream and gully sidewalls is discharged downstream by small amounts of rainfall.(2) Because of the constriction between the lower midstream and the confluence of the gully, fine sediment does not flow out to the confluence of the gully, and sediment is deposited in the lower midstream.(3) A large amount of fine sediment accumulates in the lower middle reach.This fine sediment is easily transported by even a small amount of rainfall and may develop into suspended sediment or swept sediment that triggers a debris flow.This spatial division of the density, based on three locations where several samples were extracted show that in narrow and in broad gullies, the sedimentation process must be different.The narrow sections of the gully are arguably more like transport and transition areas, whereas the central part is broader, and a deposition zone.The processes involved are thus different and such division has been associated to grain-size variability notably [11], and internal architecture at Unzen Volcano [12].Furthermore, the analysis of the size and shape of the transversal profiles has revealed variable evolution rates, also emphasizing that processes at play must differ [7].

Conclusions
The relationship between transported moisture content and road surface temperature is relatively simpler in volcanic areas than in forested areas due to the absence of organic matter in the soil, trees, and tree roots, so a constant relationship between moisture content and temperature was found in open areas without shading by trees and gully.However, the transfer of thermal energy from water masses and air masses in sediment pore spaces must be fully considered.The present study revealed clear differences in local density and porosity at different locations in the Mizunashi River, suggesting that a small amount of rainfall transports sediments little by little.We believe that small changes in flow paths due to this transport are an important factor in predicting large changes in flow paths such as debris flows.
Finally, this study also shows that accurate measurement of erosion and deposition in any environment for geomorphological purposes using point-cloud technologies need to take into account density variability across a deposit.

Figure 1 .
Figure 1.Research location.(a) Location of Unzen Volcano in South Japan; (b) Zoom to the East of the summit of Unzen Fugendake, where the Tansandani gully is located; (c) sampling locations.

Figure 2 .
Figure 2. Methodological flow chart and photographs illustrating the onsite observations, the control points and the volume calculation in Cloud Compare.

Figure 3 .
Figure 3. Calculated densities and MAE for the 11 samples at the three locations (confluence, lower-midstream and upper-midstream).

Figure 4 .
Figure 4. Water Content, Porosity and Water content of the bulk samples collected at the three sampling sites (total 11 samples).

Figure 5 .
Figure 5. Spatial variation of the sediment density and temptative of interpretation.