1. EGU General Assembly 2014
May 2nd, 2014, Vienna
Using the TRIGRS Model to Predict Rainfall-Induced Shallow Landslides Over Large Areas
Gioia, E.1, Speranza, G.2, Ferretti, M.2, Marincioni, F. 1, Godt, J. W.3, and Baum, R. L.3
Department of Life and Environmental Sciences, Marche Polytechnic University, Via BrecceBianche, Ancona, 60100, Italy,
Regional Forecast Center, Marche Region, Via del ColleAmeno 5, Ancona, 60126, Italy;
U.S. Geological Survey, Box MS 966, Denver, CO
Good afternoon. My presentation is about the use of the TRIGRS model as a predictor for rainfall-induced shallow landslides over large areas.

2. 99% municipalities affected
Introduction
(pictures courtesy of the Marche Civil Protection)
99% municipalities affected
Rainfall-induced landslides are a widespread and recurrent natural hazard in Italy in general and in the Marche region in particular. In 2008, a report of the Ministry for the Environment, Land and Sea denounces that in this region 99% of the municipalities are exposed to high hydrogeological risk: 9% of the areas for floods, and 91% for landslides.
(Ministry for the Environment, Land and Sea - Report 2008)
USING THE TRIGRS MODEL TO PREDICT RAINFALL-INDUCED SHALLOW LANDSLIDES OVER LARGE AREAS

3. RECOVERY AND RESTORATION
Introduction
WARNING SYSTEM
FORECAST
EMERGENCY MANAGMENT
RECOVERY AND RESTORATION
PREVENTION
EVENT
For this reason, an appropriate management of the warning system for hydrogeological risk is required. This work is focused on the forecast activity for rainfall-induced shallow landslides.
USING THE TRIGRS MODEL TO PREDICT RAINFALL-INDUCED SHALLOW LANDSLIDES OVER LARGE AREAS

4. Introduction LANDSLIDES FORECAST Antecedent rainfall (A) ;
Rainfall intensity (I) ;
Rainfall duration (D) :
Probability
EMPIRICAL MODELS
(STATISTICAL)
Hydrological features;
Hydrogeological features;
Geomorphological features;
Geotechnical features;
Structural features.
PHYSICAL MODELS
(DETERMINISTIC)
Among the methods proposed in the literature, physics-based models determine the hydrologic response of soils to rainfall and the propensity for landslide initiation. These methods are accurate but require understanding of many spatially and temporally variable factors. For this reason, such models are commonly applied over small areas of few tens of km2. This presentation describes a preliminary approach for calibrating a deterministic model to predict shallow landslides over an area of 550 km2.
USING THE TRIGRS MODEL TO PREDICT RAINFALL-INDUCED SHALLOW LANDSLIDES OVER LARGE AREAS

5. Introduction
TRIGRS
(Transient Rainfall Infiltration and Grid-based Regional Slope Stability)
RANGE
c [kPa]
c ≥ 0
φ [deg]
φ > 0
ϒs [N m-3]
ϒs > 0
D0 [m2 s-1]
D0 > zmin
Ks [m s-1]
Ks > 0
θr [-]
θs > θr ≥ 0
θs [-]
θs > 0
α [m-1]
α > 0
We used the U.S. Geological Survey Transient Rainfall Infiltration and Grid-based Regional Slope-stability (TRIGRS) model to compute infiltration-driven changes in factor of safety based on an infinite-slope stability analysis and a one-dimensional analytical solution for vertical infiltration. As input, we used rainfall information from a gage network, a 20m DEM and a hydrogeological map of the study area, mechanical and hydrological soil properties from published studies.
𝑭𝒔= 𝒓𝒆𝒔𝒊𝒔𝒕𝒊𝒏𝒈 𝒇𝒐𝒓𝒄𝒆𝒔 𝒅𝒓𝒊𝒗𝒊𝒏𝒈 𝒇𝒐𝒓𝒄𝒆𝒔
USING THE TRIGRS MODEL TO PREDICT RAINFALL-INDUCED SHALLOW LANDSLIDES OVER LARGE AREAS

6. THE POST-OROGENIC COMPLEX OF THE ESINO RIVER BASIN
Study area
THE POST-OROGENIC COMPLEX OF THE ESINO RIVER BASIN
The study area is part of the Esino River basin, in the Marche Region (central Italy). This area is hilly and characterized by post-orogenic Quaternary sediments prone to rainfall-induced shallow landslides. The hydrogeology can be grouped in four main soil texture units: clay, sandy-clay, loam, and gravel.
USING THE TRIGRS MODEL TO PREDICT RAINFALL-INDUCED SHALLOW LANDSLIDES OVER LARGE AREAS

7. LANDSLIDES (82) OCCURRED DURING THE PERIOD 1990-2012
Data collection
LANDSLIDES (82) OCCURRED DURING THE PERIOD
82 landslides occurred during the period were mapped in the study area. For the TRIGRS model, we needed to know the total area affected by the instability. Thus, each point location was linked to an existing landslide polygon from the Italian landslides inventory (IFFI) (a), or to a buffer area when the connection with one specific polygon was uncertain (b).
USING THE TRIGRS MODEL TO PREDICT RAINFALL-INDUCED SHALLOW LANDSLIDES OVER LARGE AREAS

8. RAIN GAUGES USED FOR THE STUDY
Data collection
RAIN GAUGES USED FOR THE STUDY
To extract the list of the rainstorms that triggered these landslides we user the Mechanical (MR) and Telemetric (TR) rain gauges network of the Marche Civil Protection. Every landslide was linked to the closest functioning gauge.
USING THE TRIGRS MODEL TO PREDICT RAINFALL-INDUCED SHALLOW LANDSLIDES OVER LARGE AREAS

9. LANDSLIDES TEXTURE CLASSES
Data collection
LANDSLIDES TEXTURE CLASSES
As the histograms show, the majority of the landslides was triggered in clay and sandy-clay soil texture classes. Due to the small area, the low number of landslides and the lack of literature data for the material properties of gravel, this class was not considered within the study.
USING THE TRIGRS MODEL TO PREDICT RAINFALL-INDUCED SHALLOW LANDSLIDES OVER LARGE AREAS

10. BOX PLOT CHARTS OF CENTRAL TENDENCIES
Data collection
BOX PLOT CHARTS OF CENTRAL TENDENCIES
The values of the soil physical parameters, required as input for the TRIGRS program, have been gathered from literature review of works related to the study area or nearby regions. These parameters are cohesion (c), angle of internal friction (φ), unit weight of soil (ϒs), hydraulic diffusivity (D0), and saturated hydraulic conductivity (Ks). This graph shows the box plot charts of central tendencies, such as the smallest observation (sample minimum), lower quartile (q1), median, upper quartile (q3), and largest observation (sample maximum). Therefore, we defined a range of property values for each zone for the calibration of the model.
USING THE TRIGRS MODEL TO PREDICT RAINFALL-INDUCED SHALLOW LANDSLIDES OVER LARGE AREAS

11. Method SETTINGS STEP 1: TRIGRS calibration
Once selected the hydrological and mechanical parameters values to test, we calibrated TRIGRS for a single cell using representative input values. Based on field observations and reports, average landslide depth was assumed 1m at the median slope angle for landslides in each soil type. The storms chosen were those which triggered the largest number of landslides during the period ; October 1996 and November 1998 with 7 and 18 landslides respectively. We compared the modeled responses (pressure head and factor of safety) with observed landslides timing in clayey, sandy-clayey and loamy soils (here represented by the highlighted areas). Among the model runs, those with the best results were constrained for the subsequent analysis. The best results are identified as the runs in which the 1-D profile yield a change in pore-water pressure such that factors of safety (Fs) decreased to less than one in landslide locations.
USING THE TRIGRS MODEL TO PREDICT RAINFALL-INDUCED SHALLOW LANDSLIDES OVER LARGE AREAS

12. Method SETTINGS STEP 1: TRIGRS calibration
STEP 2: Outline test cases, choose the rainfall event, select the rain gauge and the rainfall intervals
Following the calibration on a single cell, we ran the model for the entire study area. We divided the storms into intervals (light blue lines) characterized by the mean rainfall of the period that they represent. The figure shows the rainfall intervals and intensities of the October 1996 storm as an example.
USING THE TRIGRS MODEL TO PREDICT RAINFALL-INDUCED SHALLOW LANDSLIDES OVER LARGE AREAS

13. Method SETTINGS STEP 1: TRIGRS calibration
STEP 2: Outline test cases, choose the rainfall event, select the rain gauge and the rainfall intervals
STEP 3: Evaluate an “Antecedent Water Index”
In order to estimate the initial water table depth, we computed an Antecedent Water Index (AWI) for every rainfall event. We considered the wilting point condition as the beginning of the meteorological season or the beginning of consistent rainfall. In the case studies the AWI was approximately equal to or greater than 0 (field capacity) on the days prior to the landslide event. Thus, we ran TRIGRS with saturated conditions and an initial water table depth at: 0.5m for clay and 1m below the ground surface for sandy-clay and loam. The graph shows an example of the AWI trend for the period August 1st – November 30th 1996.
USING THE TRIGRS MODEL TO PREDICT RAINFALL-INDUCED SHALLOW LANDSLIDES OVER LARGE AREAS

14. Method SETTINGS STEP 1: TRIGRS calibration
STEP 2: Outline test cases, choose the rainfall event, select the rain gauge and the rainfall intervals
STEP 3: Evaluate an “Antecedent Water Index”
STEP 4: Construct input grids of slope from a 20m DEM, property zones, rainfall input, soil depth and water table depth
Furthermore, we fitted a soil depth model assuming the correlation of the soil depth with the profile slope angle, so that the depth decreases with the increase of the slope. Based on the strength properties values (friction angle, cohesion and unit weight of soil) calibrated in the above-described tests we computed the critical soil depth for a factor of safety equal to 1. Using this approach, slope-dependent profiles of regolith depths (diamonds) have been developed for all the soil units.
USING THE TRIGRS MODEL TO PREDICT RAINFALL-INDUCED SHALLOW LANDSLIDES OVER LARGE AREAS

15. Results OCTOBER 1996 CASE STUDY
We visualized and compared in a GIS software the outputs. The figures show spatial and temporal distribution of Fs for the October 1996 event. The red cells represent a Fs less than 1 that suggests instability, the yellow cells a Fs between 1 and 1.3 that is intermediate and the green cells a Fs greater than 1.3 that denotes stability. The timing (0, 72 h) reflects the beginning and the end of the storm. Hence, the TRIGRS model minimize the unstable cells (red) at the beginning of the storm, but maximize those at the end. These results indicate that the model was well calibrated.
USING THE TRIGRS MODEL TO PREDICT RAINFALL-INDUCED SHALLOW LANDSLIDES OVER LARGE AREAS

16. RECEIVER OPERATING CHARACTERISTICS (ROC) ANALYSIS
Results
RECEIVER OPERATING CHARACTERISTICS (ROC) ANALYSIS
The Receiver Operating Characteristics (ROC) technique was used to assess the performance of TRIGRS as a landslide predictor for the study area. The True Positive Rate (TPR) represents the number of unstable cells also predicted unstable from the analysis. The False positive rate (FPR) represents the number of unstable cells predicted stable from the analysis. The upper left graph shows the results comparing the October 1996 event against the IFFI inventory and the whole inventory. The upper right graph shows the ROC analysis for different landslide scenarios. The graph indicates that for the events that triggered a single landslide TRIGRS tends to overpredict the extent of the landslides, but for scenarios with higher number of landslides, results are improved. the lower charts shows the ROC curves for the events of October 1996 and November 1998 plotted varying the range of Fs values considered to be unstable (True Positive). Acceptable predictions fall in the upper left quadrant of the graph (highlighted). Based on these results, considering also the intermediate Fs values (yellow cells of the previous figures) as unstable maximizes the agreement between known and predicted landslides for both events.
USING THE TRIGRS MODEL TO PREDICT RAINFALL-INDUCED SHALLOW LANDSLIDES OVER LARGE AREAS

17. Conclusion
We presented an innovative methodology for the application of a physical model (TRIGRS) for rainfall-induced shallow landslides forecasting over broad areas.
TRIGRS well predicts the instabilities, thus can be used to build an accurate susceptibility map;
Rainfall-induced landslides over large regions may be predicted even where geotechnical and hydraulic properties data as well as temporal changes in topography or subsurface conditions are not available.
In conclusion, we presented an innovative methodology for the application of a physical model (TRIGRS) for rainfall-induced shallow landslides forecasting over broad areas. Results showed that by evaluating the model’s output in a probabilistic framework, TRIGRS can be used map landslide susceptibility and may be suitable for the prediction of rainfall-induced landslides over large regions even where geotechnical and hydraulic property information are not available.
USING THE TRIGRS MODEL TO PREDICT RAINFALL-INDUCED SHALLOW LANDSLIDES OVER LARGE AREAS

18. Thank you for your attention.