Project : Empirical Modeling

3 downloads 0 Views 2MB Size Report
James Varghese. As a part of M.Sc. Module ... STANDARDIZATION AND WEIGHTS IN SMCE. .... 2013_mmd Inventory was produced by Mott MacDonald.
Project : Empirical Modeling 2015

A National Scale Landslide Susceptibility Assessment for St. Lucia, Caribbean Sea

Submitted by James Varghese As a part of M.Sc. Module On Empirical Modeling of Hazard Processes

Project : Empirical Modeling 2015

TABLE OF CONTENTS INTRODUCTION .................................................................................................. 3 STUDY AREA & PROJECT MOTIVATION ...............................................................................................................................3

METHODOLOGY .................................................................................................. 4 LANDSLIDE INVENTORY CHARACTERISTICS ......................................................... 5 DISCREPANCIES IN LANDSLIDE INVENTORY MAPS...................................................................................................................6

ANALYSIS OF FACTOR MAPS ............................................................................... 7 EVALUATION OF FACTORS WITH ‘WOE’ TABLES ...................................................................................................................8

SPATIAL MULTI-CRITERIA EVALUATION (SMCE) DECISION TREE .......................... 9 STANDARDIZATION AND WEIGHTS IN SMCE......................................................................................................................10

LANDSLIDE SUSCEPTIBILITY MAP ...................................................................... 11 RESULT ANALYSIS.............................................................................................. 12 SUCCESS RATE ANALYSIS................................................................................................................................................13

CONCLUSIONS................................................................................................... 14 REFERENCES...................................................................................................... 14

Project : Empirical Modeling 2015

Introduction Study Area & Project Motivation St. Lucia is an Island country which is part of a chain of islands forming the eastern most boundary of Caribbean Sea. It is also one of the ‘Windward Islands’ (http://en.wikipedia.org/). Due to its vulnerable location, it is often defenseless while facing hurricane induced extreme winds and rainfall events. In addition, the topography and its position (Fig. 1, Table 1) along the edge of one of the tectonic plate boundaries heighten the effect of various disasters, especially landslides, floods and storm surges (Conference, 2014). In light of the greater objectives of CHARIM (Caribbean Handbook for Disaster Information Management) Project led by ITC, this miniature one week project was carried out as a part of Module 6 - ‘Empirical Modeling of Hazard Processes’. In particular, landslide susceptibility assessment was accomplished for St. Lucia utilizing both qualitative and quantitative methods. Notably, ‘Weights-of-Evidence’ (Bonham-Carter, 1995) & ‘Spatial Multi-Criteria Evaluation’ (Günther, Van Den Eeckhaut, Malet, Reichenbach, & Hervás, 2014) techniques which are representatives of ‘Data Driven’ and ‘Heuristic’ approaches respectively, were applied. Deliverable was a Susceptibility Map of St. Lucia.

Fig.1: Location of St. Lucia in the Caribbean

Table 1: Statistic of St. Lucia

Project : Empirical Modeling 2015

Methodology A straight forward methodology was employed in arriving at the Susceptibility Map for St. Lucia with three different classes. Quantitative data driven method of ‘Weights-ofEvidence’ (WOE) was chosen to decide the significance of each factor for shallow landslides initiation. Consequently, chosen factor maps were used in a Heuristic, Spatial-Multi-Criteria Evaluation (SMCE) model for the final product. All the processing was carried out in the Open Source ILWIS ver. 3.4 GIS software and the map was prepared in ArcGIS 10.3.

Fig.2: Methodology

Project : Empirical Modeling 2015

Landslide Inventory Characteristics

Four Landslide Inventories were provided for undertaking the susceptibility analysis which are specified in Table 2.

Sr. No. 1

2 3 4

Landslide Inventory

Period

Tropical Storm Debby Landslide_1995 Hurricane Tomas Christmas Eve Trough 2013_mmd 2013_dec

1994 1995 2010 2013 2013 2013 20102014

2010_2014_bgs Total

No. of Landslides 634

383 45 1243 2305

Landslide % Area (Sq. Total Km) Area Triggering Event Point Map Triggering Event Triggering Event 1.61 0.26% Point Map 5.62

Landslides/Sq. Km.

No. of Landslides used in Model

1

634

0.63 0.07

378 15

2

1233 2260

0.93%

Table 2: Landslide Inventories used in St. Lucia

Authors of Landslide Inventory



Landslide_1995 Inventory was produced by C. Rogers post Tropical Storm Debby



2013_mmd Inventory was produced by Mott MacDonald



2010_2014_bgs Inventories were produced by British Geological Survey

P.S. Landslide deposits & Rock-falls were not taken into consideration

Project : Empirical Modeling 2015

Discrepancies in Landslide Inventory maps



While analyzing different inventories it became evident that they were inconsistent and often subjective owing to the fact that multiple authors were involved in mapping the landslides.



There was an apparent shift in some landslide locations probably because of conflicting Projection Systems. Instead of the hill slopes they were marked near valley bottoms which seemed implausible.



Distinction between various landslide types were often very obscure and uncertain.



It was difficult to interpret symbols and codes used by different authors in describing the landslides as they did not contain full forms of the acronyms used.



Although the triggering events like Hurricanes, Tropical Storms, Strong Rainfall Events, man-made influences etc. were well documented still the landslide inventories themselves were not updated after each major triggering event constantly.



Landslide Inventories form the starting point of many other studies, including hazard and risk assessment. Therefore if the basis is problematic then the results are doubtful too.

In spite of the short comings in the Landslide Inventories, some attempt was made to filter those landslides which did not match the objectives of the study and some were discarded purely based on their descriptions which were ambiguous in nature. Eventually all landslides from different inventories were clubbed together to form a single binary map. It was then overlaid (crossed) with various factor maps for determination of weights with the application of ‘Weights-of-Evidence’, a bi-variate statistical technique.

Project : Empirical Modeling 2015

Analysis of Factor Maps The following table (Table 3.) provides an overview of the available and derived factor maps. Figures illustrate some of the factor maps derived from the Digital Elevation Model (DEM).

Fig.3: DEM (units in meters)

Even though the quality of DEM was not the best still some of the derivative factors were quite useful, if not all of them, in influencing landslide phenomena.

Fig.4: Slope in degrees°

Fig.5: Aspect along compass directions

Sr. No. 1 2 3 4 5 6

Provided Factor Maps Digital Elevation Model (DEM) Soils Roads Geology LandCover Elevation

Sr. No. 1 2 3 4 5 6 7 8 9

Derived Factor Maps Slope Aspect Road Cuts Local Relief Soils and Slope LandCover and Slope Topographic Position Index Topographic Roughness Index Slope and Flow Accumulation Table 3: List of Factor Maps

Fig.6: Elevation in meters

Fig.7: Local Relief in meters

Project : Empirical Modeling 2015 Evaluation of Factors with ‘WOE’ Tables After calculating the ‘Weights-of-Evidence’ for each factor map, they were evaluated based on their Contrast values and also their Final Weights. Two of the factors with their associated ‘WOE’ tables are depicted along with their corresponding discussion. 1. One of the most definitive factors which exhibited a straightforward correlation with the triggering event was Aspect. It can be clearly seen from Table 4, associated with the ‘WOE’ of Aspect that the hill slopes facing South-East, South, South-West and East display the highest Contrast factors which is in line with the Atlantic Hurricane and Trade wind directions (Fig. 9) which in turn influence the landslides.

Fig.8: Aspect

Table 4. ‘WOE’ table for Aspect

Fig.9: Hurricane & Wind Direction trends

2. Perhaps, the single most important criteria which impact Landslide is Slope. Here (Table 5.) the cross between Slope and Elevation reveals some interesting facts. The highest weights are found in moderately steep slope in conjunction with moderately elevated hills. Therefore, this explains to a certain extent shallow landslide locations since Rockfalls are generally concentrated at or near steeply sloping high cliffs, where the latter were anyway eliminated from the study.

Fig.10: Elevation

Table 5. ‘WOE’ table for Elevation And Slope

Fig.11: Slope

Project : Empirical Modeling 2015

Spatial Multi-Criteria Evaluation (SMCE) Decision Tree After all the factors were analyzed based on their weights derived from ‘Weights-ofEvidence’ technique, the most influential factor maps were introduced into an SMCE tool in ILWIS ver. 3.4. Since SMCE is a knowledge driven method it largely depends on the user who assigns weight based on their expert knowledge. Some factors were not taken into the SMCE Decision tree because they showed unreliable results. For example, ‘Water body’ class in ‘Land cover’ factor showed the highest weight for the potential location of a landslide. Unpredictable results were also found in case of factor map- ‘Geology’ where the most stable rocks showed the topmost chance for landslide locations. The factors were divided into groups on the basis of their source. Initially, eleven factors (Fig. 13) were considered for the Susceptibility Map creation which could then be used as a measure for comparing various other outputs from different versions of SMCE decision trees. After ten runs of SMCE, a model was selected which generated a modest success rate. .

Fig.12: SMCE Decision Tree (Selected Version)

Fig.13: SMCE Decision Tree (Trial Version)

Project : Empirical Modeling 2015 Standardization and Weights in SMCE

For standardization of factors, a combination of ‘Benefit’ & ‘Goal’ (Fig. 14) method were chosen. For all factors the lower limit was chosen to be zero and negative values were ignored.

Fig.14: Standardize Value (Dialog Box - ILWIS)

After standardizing the factor maps, the groups to which they belonged were weighted. The principle behind attaching weights was based on previous knowledge of the most probable factors which contribute to landslides. Therefore it was decided to grant 70% of the weight to ‘Major DEM Derivatives’ and 30% of the weight to ‘Minor DEM Derivatives’ (Fig. 12). Finally, a Susceptibility Map was generated from the selected SMCE Decision Tree.

Project : Empirical Modeling 2015

Landslide Susceptibility Map

Inset Map

Fig.15: Landslide Susceptibility Map of St. Lucia

Project : Empirical Modeling 2015

Result Analysis

Based on the Susceptibility Map, following table (Table 6) summarizes the number of landslides in the low, moderate and high susceptibility zones marked in green, yellow and red year-wise. It also contains the areal coverage of the three susceptibility classes in hectares. Additionally, the total number of buildings falling within the three zones are also taken into account.

Table 6: Susceptibility Map Results

Project : Empirical Modeling 2015 Success Rate Analysis

Success Rates were generated for all time periods corresponding to the Inventories. The inventory from 2013_MMD showed the worst success rates and the best results were obtained from the Inventory provided by the British Geological Survey. Overall, all inventories combined, showed an average success rate.

Fig.15: Success Rates for different Inventories

Project : Empirical Modeling 2015

Conclusions

It was realized from this short project that the methods for landslide susceptibility assessment are well established, depending on the flexibility of choice, driven either by data or knowledge characteristics. However, elements which influence to a large extent the outcome of the analysis depend substantially on the geometric and thematic quality of available data. Therefore, if the quality of data is poor, then no matter how hard one tries, one cannot surpass a certain limit of success rate achievable. There are ways to elevate the success rates of the result provided one endeavors to improve the quality of factor maps and the Inventory themselves, which seemingly requires more investment of time.

References 1. Saint Lucia. Retrieved February 14, 2015, from the Wikipedia: http://en.wikipedia.org/wiki/Saint_Lucia 2. Winward Islands. Retrieved February 14, 2015, from the Wikipedia: http://en.wikipedia.org/wiki/Windward_Islands 3. Conference, I. (2014). Analysis and Management of Changing Risks for Natural Hazards Analysis and Management of Changing Risks for Natural Hazards, (November), 1–10. 4. Günther, A., Van Den Eeckhaut, M., Malet, J.-P., Reichenbach, P., & Hervás, J. (2014). Climatephysiographically differentiated Pan-European landslide susceptibility assessment using spatial multi-criteria evaluation and transnational landslide information. Geomorphology, 224, 69–85. doi:10.1016/j.geomorph.2014.07.011 5. Bonham-Carter, Graeme F. (1995). Geographic information systems for geoscientists: Modelling with GIS. In Computers & Geosciences (Vol. 21, pp. 1110–1112). doi:10.1016/0098-3004(95)90019-5 6. Project Empirical Modeling, Lecture PowerPoint Slides,