The Round Robin Test on Landslide Hydrological Modeling at IWL2013

(1)

Procedia Earth and Planetary Science 9 ( 2014 ) 180 – 188

Selection and peer-review under responsibility of Dipartimento di Ingegneria Civile, Design, Edilizia e Ambiente, Seconda Università di Napoli. doi: 10.1016/j.proeps.2014.06.019

The Third Italian Workshop on Landslides

The Round Robin test on landslide hydrological modeling at

IWL2013

Thom A. Bogaard

b

_{, Roberto Greco}

a

_{*, Lucio Olivares}

a

_{, Luciano Picarelli}

a a_{Dipartimento di Ingegneria Civile, Design, Edilizia e Ambiente, Seconda Università di Napoli, 81031 Aversa, Italy} b_{Department of Water Management, Delft University of Technology, PO Box 5048, 2600 GA Delft, The Netherlands}

Abstract

During the Third Italian Workshop on Landslide, a special session was dedicated to a landslide hydrological modeling competition. The modeling exercise dealt with slopes covered with loose granular pyroclastic deposits typical of Campania (southern Italy). The data provided to the participants for the calibration of their models are described, consisting of: soil physical characterization experiments carried out over small soil specimens; controlled infiltration experiments in small scale slope physical models; field monitoring. After model calibration, the participants were asked to provide blind predictions of the following experiments: controlled infiltration in a physical model of a slope reconstituted in a laboratory flume, lasting until the failure of the slope; measured rainfall infiltrating in a monitored field site. The results obtained by the participants using very different models show that complex coupled physically-based models, requiring large sets of data for their calibration, allow to shed light upon the hydrological processes leading to landslide triggering, while simpler models, easier to calibrate, may be preferred when only the major macroscopic aspects of the phenomena, such as approximate time and location of the failure, are needed.

Selection and peer-review under responsibility of Dipartimento di Ingegneria Civile, Design, Edilizia e Ambiente, Seconda Università di Napoli.

Keywords: Hydrological modelling; Landslide modelling; Model intercomparison; Rainfall-induced landslide; Field monitoring; Lab experiments

1. Introduction

During the third edition of the Italian Workshop on Landslides (IWL2013), which took place on 23 and 24 October 2013 in Naples (Italy), a session was dedicated to a Round Robin test on landslide hydrological modeling. A Round Robin is an interlaboratory comparison test performed independently, centered around a competition among modelers with the ultimate aim to discuss and improve our modeling concepts for better prediction of landslide © 2014 Elsevier B.V. This is an open access article under the CC BY-NC-ND license

(http://creativecommons.org/licenses/by-nc-nd/3.0/).

Selection and peer-review under responsibility of Dipartimento di Ingegneria Civile, Design, Edilizia e Ambiente, Seconda Universit di Napoli.

Selection and peer-review under responsibility of Dipartimento di Ingegneria Civile, Design, Edilizia e Ambiente, Seconda Università di Napoli.

(2)

occurrence. In particular, the test dealt about how to ride over the problem of putting together information taken at different scales of observation (small laboratory samples, flumes experiments, lysimeter measurements, or field monitoring sites) to effectively model the hydrological triggering of a landslide, and about the pros and cons of different modeling approaches. To such aim, all the participants were provided with identical information from lab and field experiments carried out with one volcanic soil. Then the groups were asked to model this test, while calibrating their model with the provided information. Ultimately all groups were asked to model a field experiment and predict the soil behaviour and failure time. This was reported and discussed in a one day workshop.

More specifically, the participants were solicited to calibrate their models on the basis of the provided data, and then, a-priori of the workshop, they were challenged to simulate:

x an infiltration experiment carried out in a laboratory flume on a small scale slope model, lasting until slope failure;

x the response of the real slope of Cervinara to a measured precipitation event.

The aim of this paper is twofold. First to describe the data provided to the groups. Second, this paper aims to discuss the background of the data with respect to their importance, depending also on the chosen modeling approach and on the introduced simplifying hypotheses, for simulating and understanding the processes leading to the triggering of a landslide. Six modelling teams from several European universities participated. Their results, presented during the workshop, are summarized in six papers published in this issue1,2,3,4,5,6_{. The experimental data} which are given in the last section of this paper will allow to evaluate the (dis-)advantages of different approaches, concepts, and modeling strategies. Last but not least, with this selection of papers and with the publication of the data set we aim to stimulate research groups in the world, working on landslide modeling, to use the publications for their own benefit and model testing.

The paper is organized as follows: the first three sections are dedicated to the description of the experimental data - measured on small soil samples, on physical models and through field monitoring - provided for model calibration; the fourth section describes in detail the two experiments which the participants were asked to reproduce with their models; the last section briefly comments the various proposed models and the obtained results, which are thoroughly described in separate papers in this issue1,2,3,4,5,6_.

2. Geological and geotechnical characterization of the study site

The studied soil is a granular volcanic soil, typical of the mountains surrounding Naples, which is notorious for the disruptive flow-like sudden shallow landslides. In particular, the attention was focused on the soil of the slope of Cervinara (around 40 km north-east of Naples), about which the following data were made available to the participants: data from laboratory experiments over small undisturbed specimens7,8_{; data from flume infiltration tests}

over small scale model slopes9,10,11_{; data from a field monitoring station operating since 2009 at the slope of}

Cervinara12,13_.

The hydrological modeling contest was dedicated to the pyroclastic soil cover of the northeast slope of Mount Cornito, near the town of Cervinara, about 50km northwest of Naples, southern Italy, just besides the location where, in the night between 15th and 16th December 1999, a flowslide was triggered after an intense rain event lasting more than 24 hours. The soil cover consists of an alternation of loose volcanic ashes and pumices lying upon a fractured limestone bedrock. The data presented in this section were obtained by means of measurements carried out over either undisturbed or reconstituted soil samples, of prismatic or cylindrical shape, with a width ranging between 70mm and 100mm, and a height between 20mm and 150mm. All the samples refer to the ashes, which usually represent the thickest layer within the profile.

Table 1 gives the main physical properties of the ashes, as obtained from several undisturbed soil samples collected at various locations and depths along the slope. Some of the undisturbed samples were used for carrying out suction-controlled triaxial tests. Figure 1 shows the results of such triaxial tests, plotted as relationship between net normal stress and maximum deviatoric stress. The straight line plotted in the same figure has an inclination of 38°, assumed as being representative of the friction angle of the saturated material. The same figure gives also the

(3)

relationship between suction and the cohesion intercept of the unsaturated material, obtained by assuming the same friction angle8_.

Table 1. Main physical properties of the investigated volcanic ashes.

specific weight, Ȗs (kNm-3) 25-26

unit volume weight, Ȗ (kNm-3_{) 11-14}

porosity, n 0.67-0.75

saturated hydraulic conductivity, ksat (ms-1) 1.5×10-7 – 5.7×10-6

effective friction angle, I’ (°) 38

cohesion, c’ (kPa) 0 0 100 200 300 400 0 50 100 150 200 p -ua(kPa) q (kPa) 0 5 10 15 0 25 50 75 100 ua-uw(kPa) c' +( u a -u w )t gI b (k P a)

Fig. 1. Results of suction controlled triaxial tests. Left panel: stress plane. Right panel: cohesion vs. suction relationship.

In order to provide information about the deformability of the investigated soil, all the data collected during the triaxial tests were available for the participating modeling teams. In particular, such data included: the transient development of volumetric strain and of the change in water content, observed during the eqilibrium phase, prior the application of the deviatoric load; the axial, radial and volumetric strain, together with the change in water content, measured step by step during the application of the deviatoric load. Fig. 2a shows 5 examples of the observed relationships between deviatoric stress and axial deformation. Fig. 2b shows the relationship between the net pressure and the pore index observed during isotropic compression tests for the same samples.

0 50 100 150 200 250 300 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Ha q (kPa) 1.7 1.9 2.1 2.3 10 100 p -ua (kPa) 1000 e

Fig. 2. Measurement on undisturbed soil samples: a) Stress-strain relationship; b) results of isotropic compression tests.

The saturated hydraulic conductivity was estimated by means of constant head tests with confining pressure between 20kPa and 600kPa, and varied between 5.7×10-6_ms-1_{and 1.5×10}-7_ms-1_{, with a mean value of 1.4×10}-6_ms-1_.

(4)

Fig. 3 gives the measured relationship between soil suction and volumetric water content, as estimated from either undisturbed or reconstituted soil samples with various experimental techniques.

0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.0 20.0 40.0 60.0 80.0 100.0 ua-uw (kPa) T

Reconstituted sample (transient infiltration test) Reconstituted sample (transient evaporation test) Undisturbed sample (suction controlled triaxial test) Reconstituted sample (pressure plate)

Fig. 3. Water retention data observed over small samples of the investigated volcanic ashes.

3. Data from flume infiltration tests

The geotechnical laboratory of the Seconda Università di Napoli operates an instrumented flume for controlled infiltration/evaporation experiments on small scale slopes. The infiltration experiments can be continued until slope failure. The flume contains a number of devices allowing accurate monitoring of several hydrological and mechanical variables during the experiments, among which: spray nozzles for the application of artificial rainfall of prescribed intensity; minitensiometers to measure soil suction within a soil layer; pressure transducers to measure pore water pressure; TDR probes to determine soil volumetric water content; laser sensors to monitor slope surface settlements. A sketch of the flume is given in Fig. 4. More details about the flume and the installed devices can be found in Olivares et al.9_.

The complete record of two infiltration experiments in a slope covered with a layer of the ashes of Cervinara were provided to the participating modeling teams. In both cases the inclination of the slope was 40° and the width of the slope 0.50m. Both the infiltration experiments were carried out in the flume with impervious bottom and walls (made with Plexiglas). At the foot of the slope, a supporting element made with gravel wrapped by a geotextile was placed. It behaved as a capillary barrier, draining the layer only when the soil at the foot of the slope approached saturation. Both the tests were carried out with constant and spatially homogeneous rainfall intensity. The main characteristics of the two experiments are summarized in Table 2.

Table 2. Main characteristics of the flume infiltration tests.

Test nr. Soil Thickness _(cm) Slope Length _(cm) Initial porosity _n0 Rainfall intensity _(mm/h) _{suction (kPa)}Initial mean Duration of _{test (min)}

D3 10.0 100 0.75 55 17.5 36

D4 10.0 120 0.76 56 41.0 30

As an example, Fig. 5 reports the transient behavior of the soil capillary pressure and slope surface settlements observed during experiment D3. The failure of the slope occurred after nearly 30 minutes, as indicated in both the figures.

(5)

Fig. 6 (left panel) shows some of the water content profiles, orthogonal to the slope, measured by a TDR probe at various times during experiment D3. The soil moisture profiles have been estimated with the inverse methodology

described in Greco14_{. The positive pore pressure, measured at the bottom of the soil cover during the same}

experiment, is given in the right panel of Fig. 6, in which also the time is indicated when the failure occurred around 36 minutes after the beginning of the artificial rainfall infiltration.

Fig. 4. Sketch of the instrumented flume: LT = laser transducer; MT = minitensiometer; RG = rain gauge; TDR = TDR probe; PPT = pressure transducer. The locations and the number of the devices are sketched by way of example, as they were different for every experiment.

-60 -50 -40 -30 -20 -10 0 0 5 10 15 20 25 30 t (min) uw-ua (kPa) failure deep tensiometer superficial tensiometers -20 -15 -10 -5 0 5 10 0 5 10 15 20 25 30 t (min) settlement (mm) failure

Fig. 5. Experiment D4: capillary pressure measured by tensiometers (left); surface settlements measured by laser sensors (right).

4. Data from field monitoring

Since August 2009 an automatic monitoring station is operating at the slope of Cervinara, not far from the location of the catastrophic flowslide which occurred in 1999. Measurements of volumetric water content by Time Domain Reflectometry (TDR) and capillary tension by tensiometers are taken every two hours. In addition, a rain gauge for hourly automatic acquisition was installed, with sensitivity to rainfall increments of 0.2mm. The

(6)

monitoring station includes eight tensiometers and seven metallic probes for TDR measurements of soil water content. The TDR probes are mainly disposed in the immediate proximity of the ceramic tips of the tensiometers, so to allow coupling water content and capillary tension measured at the same depth. For the automatic acquisition and storage of the monitoring data, all the transducers were connected to a Campbell Scientific Inc. CR-1000 Data Logger. All the installed equipment receives its power by a 12V battery connected to a solar panel. Fig. 7 gives a sketch of the entire monitoring station. More details about the monitoring activities can be found in other papers by the Authors12,13_. 0.0 2.0 4.0 6.0 8.0 10.0 0.20 0.30 0.40 0.50 0.60 0.70 0.80 T z (cm) t=0'55" t=5'44" t=11'31" t=21'29" t=28'07" t=32'19" t=35'01" 0 0.5 1 1.5 2 0 5 10 15 20 25 30 35 40 time (min) uw (kPa) failure

Fig. 6. Experiment D3: volumetric water content profiles measured at various times by a TDR probe orthogonal to the slope (left); positive pore water pressure measured at various locations at the bottom of the soil cover (right).

Fig. 7. Sketch of the monitoring station operating at the slope of Cervinara.

The modeling teams participating in the Round Robin were provided with the following data, measured between 01.01.2011 and 27.07.2011: hourly rainfall; soil suction measured every two hours by five of the tensiometers at the monitoring station; soil volumetric water content measured every six hours by the seven TDR probes at the station (with some missing data); hourly data of air temperature. As an example, Fig. 8 shows the measured soil suction and water content together with the corresponding daily hyetograph.

5. Blind predictions

The participants in the Round Robin were solicited to carry out two blind predictions: one consisting of the simulation of the results of a controlled infiltration experiment in a small scale artificial slope reconstituted in the same laboratory flume, in which experiments D3 and D4 had been carried out; the other consisting of the prediction

(7)

of the hydrologic response of a weather forcing (rainfall and temperature) of the soil, observed in the monitoring station at the slope of Cervinara.

On the basis of the characteristics of the laboratory infiltration experiment given in Table 3, the participants were asked to simulate: time of slope failure, which actually occurred after nearly 35min; soil suction at some locations within the flume, which are plotted in the above panel of Fig. 9; the settlements at some locations along the slope surface, plotted in the below left panel of Fig. 9; the pore water pressure at various locations at the bottom of the soil cover, shown in the below right panel of Fig. 9.

0 20 40 60 01/01/2011 31/01/2011 02/03/2011 01/04/2011 01/05/2011 31/05/2011 30/06/2011 30/07/2011 h (mm) 0 0.1 0.2 0.3 0.4 T (m3 /m3 ) -0.30m -0.60m -1.00m -1.70m -60 -40 -20 0 uw-ua (kPa) -0.60m -1.00m -1.40m -1.70m 0 20 40 60 01/01/2011 31/01/2011 02/03/2011 01/04/2011 01/05/2011 31/05/2011 30/06/2011 30/07/2011 h (mm) 0 0.1 0.2 0.3 0.4 T (m3 /m3 ) -0.30m -0.60m -1.00m -1.70m -60 -40 -20 0 uw-ua (kPa) -0.60m -1.00m -1.40m -1.70m

Fig. 8. Field monitoring data observed between 01.01.2011 and 27.07.2011. From top to bottom: daily rainfall height; soil volumetric water content; soil suction.

Table 3. Main characteristics of the flume infiltration test to be blindly predicted during the Round Robin contest. Test nr. Soil Thickness _(cm) Slope Length _(cm) Initial porosity _n0 Rainfall intensity _(mm/h) _{suction (kPa)}Initial mean

C4 10.0 100 0.65 60 52

For carrying out the blind prediction of field conditions, the modeling teams received following input data: hourly temperature, measured between 1st_{September 2011 and 12}th_{February 2012 by two meteorological stations} close to the slope of Cervinara; hourly rainfall, measured during the same period by the rain gauge of the monitoring station. In order to initialize the simulations, also the following soil data were provided: soil suction at four depths, measured every two hours between 28.10.2011 at 00:00 and 29.10.2011 at 2.00PM; volumetric water content at three depths, measured every six hours during the same days.

The participants were asked to simulate the following data, measured at the monitoring station at the slope of Cervinara: soil suction at the depths of -0.50m, -1.00m and -1.40m, where tensiometers were installed, between 01.01.2012 and 12.02.2012 (plotted in the left graph of Fig. 10); soil volumetric water content at the depths of -0.30m, -0.60m and -1.00m, where TDR probes were installed, between 07.01.2012 and 12.02.2012 (shown in Fig. 16).

(8)

5. Discussion and concluding remarks

The models used by the various participating groups to carry out the simulations of flume and field data awere based on some quite different approaches. Therefore, they constitute a good example of how hydrological modeling of the behavior of a slope is far from being well-established. The proposed models span from coupled physically based models, accounting for numerous processes taking place in the soil besides water flow, such as energy exchanges with the atmosphere and root water uptake, and considering the soil as a deformable medium with a complex constitutive relationship3,6_{, to simpler uncoupled physically based models relying upon the simplifying}

assumption of rigid soil skeleton1,2_{, to commercial software conveniently adopted to the studied case}5_{, to a}

simplified, nearly conceptual, model taking into account macropore flow4_.

-60 -50 -40 -30 -20 -10 0 0 5 10 15 20 25 30 35 40 t (min) ua-uw(kPa) failure deep devices superficial devices -20 -15 -10 -5 0 5 10 0 5 10 15 20 25 30 35 40 t (min) settlement (mm) upslope transducers downslope transducers failure 0 0.25 0.5 0.75 1 20 25 30 35 40 time (min) uw(kPa) failure

Fig. 9. Experiment C4: soil suction (above); slope surface settlements (below, left); positive water pressure at the bottom of the soil cover (below, right).

Also the equations describing slope equilibrium, used to evaluate (un)stable slope conditions, are quite different: the models accounting for soil deformations identify the failure as the moment when an abrupt increase of deformation rate starts3,6_{; conversely, in the other cases, the hypothesis of infinite slope is introduced, and a safety} factor is evaluated with the limit equilibrium theory1,2,4,5_{. In the limit equilibrium equation, the stabilizing} contribution of soil suction is accounted with the help of the data of Fig. 1b, interpreted with different models.

The hydraulic characterization of soils is also another interesting issue, as the modelers had to cope with the problem of handling contrasting information about soil water retention curve and saturated hydraulic conductivity, estimated over samples of different dimensions and with different experimental techniques. In particular, most of the modelers adopted different retention curves and hydraulic conductivities to describe soil hydraulic behavior in the flume and in the field, justifying such an assumption as a consequence of the different soil density2,5,6_{or of the}

(9)

hydraulic hysteresis1_{. However, all the modeling teams chose to estimate the hydraulic properties of soils from the} back analysis of the provided experiments, rather than from the measurements carried out over small soil specimens.

0 5 10 15 20 25 01/01/12 08/01/12 15/01/12 22/01/12 29/01/12 05/02/12 12/02/12 ua-uw(kPa) -0.5m -1.0m -1.4m 0.2 0.25 0.3 0.35 0.4 01/01/12 08/01/12 15/01/12 22/01/12 29/01/12 05/02/12 12/02/12 T -0.3m -0.6m -1.0m

Fig. 10. Monitoring at Cervinara in the early 2012: soil suction (left) and volumetric water content (right) measured at various depths.

In conclusion, the obtained results clearly show that complex physically-based models allow a better description and deeper interpretation of the processes actually leading to the triggering of a landslide. However, this comes at the price that the calibration of these sophisticated physically-based models require large sets of high quality data, which are often not available. However, if only the accurate prediction of triggering time is required, also simpler models (also conceptual, if properly calibrated), may perform satisfactorily.

References

1. Ahmadi-Adli M, Toker NK, Huvaj N. Prediction of seepage and slope stability in a flume test and an experimental field case. Procedia Earth

Planet Sci 2014; xxx:yyy-zzz.

2. Airò Farulla C, Rosone M. Modeling Round Robin test: an uncoupled approach. Procedia Earth Planet Sci 2014; xxx:yyy-zzz.

3. Hoffmann C, Meler N, Pinyol NM, Alonso EE. Small scale slope failure benchmark test. Modelling and prediction. Procedia Earth Planet Sci 2014; xxx:yyy-zzz.

4. Montrasio L, Valentino R, Terrone A. Application of the SLIP model. Procedia Earth Planet Sci 2014; xxx:yyy-zzz.

5. Reder A, Rianna G, Pagano L. Prediction of suction evolution of silty pyroclastic covers in flume tests and field monitoring. Procedia Earth

Planet Sci 2014; xxx:yyy-zzz.

6. Villarraga C, Ruiz D, Vaunat J, Casini F. Modelling landslides induced by rainfall: a coupled approach. Procedia Earth Planet Sci 2014;

xxx:yyy-zzz.

7. Picarelli L, Evangelista A, Rolandi G, Paone A, Nicotera MV, Olivares L, Scotto di Santolo A, Lampitiello S, Rolandi M. Mechanical properties of pyroclastic soils in Campania Region. In: Proc. 2nd Int. Work. on Characterisation and Engineering Properties of Natural Soils. Singapore; 2006. p. 2331-2383.

8. Olivares L, Damiano E. Post-failure mechanics of landslides: laboratory investigation of flowslides in pyroclastic soils. J Geotech Geoenviron

Eng ASCE 2007;133(1):51–62.

9. Olivares L, Damiano E, Greco R, Zeni L, Picarelli L, Minardo A, Guida A, Bernini R. An instrumented flume for investigation of the mechanics of rainfall-induced landslides in unsaturated granular soils. ASTM Geotech Test J 2009;32(2):108-118.

10. Damiano E, Olivares L. The role of infiltration processes in steep slopes stability of pyroclastic granular soils: laboratory and numerical investigation. Nat Haz 2010;52(2):329-350.

11. Greco R, Guida A, Damiano E, Olivares L.Soil water content and suction monitoring in model slopes for shallow flowslides early warning

applications. Phys Chem Earth 2010;35:127-136.

12. Damiano E, Olivares L, Picarelli L. Steep-slope monitoring in unsaturated pyroclastic soils. Eng Geol 2012;137-138:1-12.

13. Greco R, Comegna L, Damiano E, Guida A, Olivares L, Picarelli L. Hydrological modelling of a slope covered with shallow pyroclastic deposits from field monitoring data. Hydrol Earth Syst Sci 2013;17:4001-4013.