Distributed specific sediment yield estimations in Japan attributed to extreme-rainfall-induced slope failures under a changing climate

Introduction Conclusions References

sedimentary rocks and granites; these were considered based on their likelihood in the 23 formation of slope failures. 24

Estimation of the hydraulic gradient 25
The hydraulic gradient is defined as the rate of hydraulic head change per unit distance in a 26 particular direction. Temporal changes in hydrological conditions (changes in soil moisture 27 content from unsaturated to saturated and vice versa) due to variations in extreme rainfall and 28 the resulting infiltration rate have an intensive impact on slope failure formations. The 29 unsteady nature of this parameter offers a unique opportunity to combine our assessments 30 with climate change studies. Nevertheless, it requires a large computational effort as 31 compared to other triggering parameters. To estimate the hydraulic gradient attributed to 1 extreme rainfall at a 1-km resolution, we followed the method previously developed by 2 Kawagoe et al. (2010). The two-dimensional form of Richard's equation was employed to 3 obtain the hydraulic gradient, which was numerically solved by considering soil data, the 4 slope angle and extreme rainfall as the independent input variables in each grid cell (more 5 details can be found in Kawagoe et al., 2010). To estimate extreme rainfall events, 24-hour 6 maximum rainfall data covering 21 years (1980-2000) from the Automated Meteorological 7 Data Acquisition System (AMeDAS) were employed with the Generalized Extreme Value 8 (GEV) probability distribution function. For 1024 AMeDAS meteorological stations 9 throughout Japan, GEV analysis indipendently generates 1024 extreme rainfall values to each 10 station. Considering the fact that the rainfall patterns in mountain areas are largely influenced 11 by irregular topography (Buytaert et al., 2006), to distribute the estimated extreme rainfall at a 12 resolution of 1-km, we used the "Mesh Climate Data 2000" rainfall database developed by the 13 Japanese Meteorological Business Support Center (2002). In this database, the rainfall 14 distribution over Japan was estimated by regression models constructed using independent 15 variables developed from geographical factors (Lookingbill and Urban, 2003;Ueyama, 2004). 16 The data set includes the monthly averaged rainfall over 30 years  assembled at a 17 1-km grid resolution. A relationship between the estimated extreme rainfall and maximum 18 monthly rainfall from the Mesh Climate Data 2000 was developed for distributing the 19 extreme rainfall to a 1-km grid resolution. To develop a statistically better relationship, the 20 AMeDAS stations were categorized into three seasonal classes, winter (December-February), 21 spring-summer (March-August) and autumn (September-November) based on the probability 22 of an extreme rainfall event. As an example, the mountain areas at the seaside receive their 23 maximum rainfall during the winter, while the south islands of Japan receive the most rainfall 24 in the spring-summer category (Kawagoe et al., 2010). Therefore, three separate regression 25 analyses were performed to obtain the relationship between the extreme rainfall and 26 maximum monthly rainfall from the Mesh Climate Data 2000, and later extreme rainfall 27 values were distributed with a 1-km grid resolution based on the maximum monthly rainfall 28 values from the Mesh Climate Data 2000 at each grid box. These extreme rainfall values were 29 then used as the main input in the infiltration analysis to find the hydraulic gradients. 30 2.2 Probability model for slope failure 1 By following the above procedures, the probability of slope failure occurrence was 2 determined by accounting for past events of slope failures at each grid cell. A stepwise 3 logistic regression method was then employed to find the relationship between the triggering 4 parameters and slope failure probability (Eq. 1). Instead of considering the geological type as 5 an independent variable with the appropriate weighting factor in the model, four different 6 models were developed for each geological type. 7 where P is the probability of slope failure occurrence, σ 0 is the intercept, σ h is the coefficient 9 of the hydraulic gradient, σ r is the coefficient of relief energy, hyd is the hydraulic gradient, 10 and relief is the relief energy. these relationships reflect the influence of triggering parameters, such as rainfall, and the 15 geomorphologic setting of the catchments. In this study, we developed a relationship between 16 the annual average specific sediment yield and the average probability of slope failure in the 17 representative catchment. Altogether, 59 dams were selected throughout Japan. For each of 18 these dams, the catchment areas are larger than 185 km 2 and more than 15 years of annual 19 sediment yield records are available. The same relationship was developed for various return 20 periods of extreme rainfall, and the goodness of the fit was evaluated against the coefficient of 21 determination to select the best-fit relationship for climate predictions. 22

GCMs for climate predictions 23
The Special Report on Emission Scenarios (SRES) along with the IPCC AR4 report has given 24 widely recognized GCMs for climate predictions. In this study, we used two climate scenarios 25 from two GCMs developed in Japan: the Meteorological Research Institute Regional Climate for MRI-RCM20-Ver.2), were selected to show the transition of the impact in the future. The 10 two selected GCMs have some advantages as compared to other models presented in IPCC 11 AR4. Firstly, the model outputs produced finer resolutions, which is particularly useful for 12 use with mountainous topography. For example, the HADCM3 model has a very coarse 13 resolution that is approximately equal to 90,465 km 2 of the grid boxes in Japan, while the 14 MRI-RCM20-Ver.2 model resolution is only 400 km 2 . Secondly, MIROC and MRI-RCM20-15 Ver.2 have been proven to be very effective in simulating the climate variables that eventually 16 produced the impacts for extreme cases over wider ranges (GERF S-4 project document, 17 2008). Therefore, they avoid the extensive downscaling efforts that are necessary for many 18 GCM scenarios for predicting the impacts in a reliable range. 19 In the first step of downscaling, the procedure explained in Sect. 2.1 was applied with the 20 GCM-produced daily rainfall to obtain the extreme rainfall distribution at the GCM grid scale 21 (hereafter referred to as ERF GCM ). Because the bias correction was to be performed at the 22 GCM grid scale, the extreme rainfall in the present climate (hereafter referred to as ERF PC ) as 23 derived in Sect. 2.1 at a 1-km grid resolution was aggregated with the grid scale of the 24 climate model (e.g., 20 km for MRI-RCM20-Ver.2). The ERF GCM data for each time period

Slope failure probability and sediment yield 2
The spatial distribution of sediment yield in Japan attributed to extreme-rainfall-induced slope 3 failure probability was estimated. In the first stage of the research, the spatial distribution of 4 the slope failure probability was estimated by considering the extreme-rainfall-induced 5 hydraulic gradient, relief energy and geological formation as the triggering parameters. The 6 results portrayed two distinct aspects of the slope failure probability ( Table 3 in Kawagoe et  7 al., 2010). Firstly, the calculated standardized partial regression coefficient produced 8 noticeably different values for the two triggering parameters. This coefficient explains the 9 change in slope failure hazard probability in the model when one triggering parameter 10 (hydraulic gradient or relief energy) is changed by one unit while the other parameter is held 11 constant. The standardized partial regression coefficient was higher for the hydraulic gradient 12 than for the relief energy for all four geological formations. This suggests that the hydraulic 13 gradient is more influential than the relief energy in terms of triggering slope failures. 14 Secondly, differences in magnitude of the coefficient of each parameter in different geological 15 settings indicate variations in their resistance to slope failures. The probability of slope 16 failure occurrence varied from the highest in colluviums to the lowest in granites. The 17 generally loose, non-consolidated nature of the colluviums has been proven to be more 18 significant in the occurrence of slope failures than hard compact formations such as granite 19 (Restrepo et al., 2006). 20 In the next step, a regression model for the slope failure hazard probability and subsequent 21 sediment yield was developed. Among the various return periods of extreme rainfall 22 considered, the five-year return period gave the best fit with a determinacy coefficient of 0.65 23 (Fig. 1a). The underlying reason for the best fit with respect to the five-year return period was 24 tested by examining the number of years of extreme sediment yield in each dam. Annual 25 sediment yield records covering 15-20 years at each dam were examined. Assuming that the 26 annual sediment yield averaged over the catchment is normally distributed throughout the 27 recording period (15-20 years), the lower bound of the extreme sediment yield (SY LB in 28 m 3 /km 2 /year) in each catchment is defined as in Eq. (2). 29 where SY Avg (in m 3 /km 2 /year ) is the annual average sediment yield in the catchment and SD is 1 the standard deviation of the annual sediment yield data series. Sediment yields exceeding the 2 threshold of SY LB are defined as extreme sediment yield events. The threshold of SY LB is then 3 used to separate the years with extreme sediment yield events at each dam site. Figure 1b  4 indicates the average recurrence interval (Chow et al., 1988) of extreme sediment yield events 5 at all selected dam sites throughout Japan. According to this, over 55% of the dams studied, 6 experience an extreme sediment yield event every 5-7 years, and over 80% of the dams 7 experience one every 5-10 years, on average. These figures clearly explain the reason for the 8 statistically better relationship obtained between the extreme rainfall and extreme sediment 9 yield over a five-year return period. 10 where SY is the annual average sediment yield (m 3 /km 2 /year) in a particular dam and P is the 15 spatially averaged probability of slope failure occurrence in a specific dam catchment. The 16 exponential shape of the relationship indicates that the sediment yield may substantially 17 increase with increasing probability of slope failure. 18 The validity of the developed relationship was tested prior to its use in climate impact 19 predictions. Another 43 dams, which were not considered in developing the original 20 regression relationship, were selected, covering all of Japan. Figure 3a shows the locations of 21 these dams in Japan based on their annual average sediment productions. Figure 3b  The sensitivity of the sediment yield model to the triggering parameters was also tested. 7 Figure 4 shows the variations in sediment yield with relief energy and hydraulic gradient for 8 four selected geological formations. Similar to the slope failure hazard probability, the 9 sediment yield potential is highest in colluviums and decreases in the order of Neogene 10 sedimentary rocks and Paleogene sedimentary rocks to the lowest potential in granites. As an 11 example, for a unit change in the hydraulic gradient, colluviums formations produce 12.2 × 12 10 3 m 3 /km 2 /year of sediment yield, which is 94% higher than the sediment yield production of 13 granites under the same conditions. Similarly, Neogene sedimentary rock and Paleogene 14 sedimentary rock produce 8.6 × 10 3 m 3 /km 2 /year and 6.5 × 10 3 m 3 /km 2 /year of sediment load 15 for a unit change in the hydraulic gradient, respectively. 16 Figure 4 also reveals an important aspect that would be critical under changing climate 17 conditions. The hydraulic gradient is a rainfall-sensitive parameter that can be significantly 18 elevated with an increase in the intensity and frequency of rainfall with climate change 19 effects. According to Fig. 4, the rate of change of the sediment yield (gradient of the curve) is 20 more sensitive to a small change in the hydraulic gradient, especially within the rising limb of 21 the curve (e.g., 12.2 × 10 3 m 3 /km 2 /year per unit change of hydraulic gradient for colluvium 22 formations). Therefore, areas that will cross the lower edge of the rising limb in the future 23 may have a critical impact on the sediment yield under changing climate conditions. 24

Spatial variability of sediment yield 25
By applying the developed sediment yield model with the distributed slope failure probability, 26 the spatial variability of the sediment yield can be estimated. Figure 5a shows the spatial 27 distribution of the sediment yield estimated at a 1-km grid resolution. Moreover, the model-28 predicted sediment yields were further aggregated to the major river basins of Japan. These 29 river basins were categorized based on the existence of first-order rivers in Japan. Figure 5b  30 depicts the average sediment yield based on the different basins. Areas with significantly 31 higher specific sediment yields (over 2000 m 3 /km 2 /year) are distributed throughout the 1 Tenryu, Ooi and Kiso river basins in the Hokuriku and Tokai regions and the Shimanto and 2 Naka river basins in the Shikoku region. The lithology and relief energy differences between 3 the various regions may play an important role in producing sediment yield (Fig. 4). As an 4 example, the greater yields corresponding to the Yoshino river basin in the Tokai region 5 consists of 40% colluviums and 27% Neogene sedimentary rock formations, whose soils have 6 a low resistance to the sediment yield, while granites with high resistance to the sediment 7 yield cover only 1% of the area. Moreover, the Tohoku and Hokuriku mountain seaside 8 regions with comparatively high relief energy have a significantly higher specific sediment 9 yield (spatial average of 600-800 m 3 /km 2 /year in Fig. 5b) as compared to areas with low relief 10 energy, such as the eastern side of the Kanto region. 11

Sediment yield distribution under changing climate conditions 12
Two climate change scenarios applied to two time periods in the future produced four sets of 13 results to demonstrate the transition of the sediment yield with climate change effects (Fig. 6). while the MRI-RCM20 future estimations gave an only 8% extreme rainfall increase. Out of 6 105 river basins that cover the whole area of Japan, the model predicted an approximately 7 constant or decreasing trend of sediment yield for only 9 river bains in future as compared to 8 the present estimations. The average percentage of the sediment yield reduction in these nine 9 river basins was less than 10%, suggesting that almost all of the river basins in Japan will 10 suffer from an increasing sediment yield risk in the future. For 15 river basins, the model 11 predicted a more than 50% sediment yield increment in the future (for at least three out of 12 four estimations in the future), and among them, 5 river basins will experience a more than 13 90% change as compared to the present sediment yield. 14 When looking at the spatially averaged sediment yield over the whole country, both model 15 scenarios predicted an increasing trend for the intermediate climate (Fig. 7), implying a 16 potential impact in the first half of the 21 st century. With respect to four future estimations for 17 105 river basins, the MRI-RCM20 future climate, however, predicted a higher sediment yield 18 than the other three estimations for only 8 river basins. Therefore, for the future climate, the 2009), it is quite difficult to incorporate them into a probabilistic model in regional-scale 32 analysis. Therefore, the inclusion of more detailed information on land use and sub-basin 1 watershed characteristics in site-specific approaches should provide more accurate 2 predictions. After identifying the hazard-prone basins as done in this study, such a detailed 3 analysis would be appropriate for designing infrastructure facilities for mitigating future 4 climate change impacts. 5 6

Conclusions 7
To facilitate the decision-making process by identifying hazard-prone areas under changing 8 climate conditions, this study developed a probabilistic model for the relationship between the 9 slope failure probability induced by extreme rainfall and sediment yield. There are three 10 triggering parameters; the hydraulic gradient, the relief energy and the geology type 11 representing the hydro-climate (hydrology and extreme rainfall), topography and geological 12 effects, respectively, were considered in developing the probabilistic model for slope failure. sedimentary rocks to the lowest potential in granites. Moreover, it is known that the hydraulic 26 gradient is more influential than the relief energy. 27 The results of the GCM scenarios predict that the sediment yield impact will increase in the 28 future. When the spatial average sediment yield for all of Japan is considered, both scenarios 29 produced an approximately 16-17% and 14-21% increase around the first half and second half 30 of the 21 st century, respectively as compared to the present climate. On the regional scale, 31 substantially higher sediment yield changes (over 1000 m 3 /km 2 /year) were estimated in the 32 1 experience a moderate sediment yield increase (250-500 m 3 /km 2 /year), while the Tohoku 2 region is predicted to have a 0-to 250-m 3 /km 2 /year increase in sediment yield. Due to 3 variations in extreme rainfall events, the sediment yield estimations at the basin scale 4 predicted changes of different magnitudes. Out of 105 basins in Japan, 96 showed an 5 increasing trend of sediment yield under changing climate conditions. Among them, five river 6 basins will experience a more than 90% change as compared to the present sediment yield.