Methodology for Risk Assessment
Rashmin Gunasekera
______
Framework for a Methodology to Integrate Vulnerability toDevelop Natural Hazard Risk Profiles for Sri Lanka
Rashmin C. Gunasekera
()
1.0Executive Summary
Sri Lanka is vulnerable to a range of natural hazards, and as such, its population, infrastructure, buildings and (urban and agricultural) land are at risk of damage and/or destruction.Natural hazard risk assessment is crucial in enabling effective disaster risk reduction and increasing disaster resilience.As such, it is vital that there is a coherent method of evaluating these risks in order that policies can be evolved for risk mitigation and for the benefit of other stakeholders. Heretofore, the problem with creating such a methodology has been manifold; amongst its biggest stumbling blocks were the differences in spatial scales involved, the variable parameters, the many influencing factors, the gaps in data required, and the conceptual problem of designing a model which could adequately address all these issues.
This report outlines a methodology which could assess natural hazard risk at national and sub-national level, identifying areas of high exposure and vulnerability.It also evaluates the exposure and vulnerability of the population, agriculture, property and transportation (+ utility networks). This methodology has the advantage of being designed to take into account a wide range of mutually influencing factors in the form of an interactive map, which can identify the location, relative severity, and extent of risks, for individual and for multiple natural hazards. It is also a significant advantage that this methodology is able to utilise high resolution local data and translate it into a risk analysis at sub national level.
2.0Introduction
The term risk is a highly complex one, and been defined in number of ways in the academic literature of natural hazards, incorporating several sophisticated concepts within this single term. Kron’s (2002)succinct definitionidentifies three variables which determine “risk”: hazard, vulnerability and exposure. At its most basic, these are the ways in which each of these terms (which together comprise risk) are understood:
– Hazard: the threatening natural event,its extent and severity of occurrence.
– Exposure: the values and/or humans that are present at the location of interest.
– Vulnerability: represents the damageability. It is the lack of (or low) resistance to damaging/destructive forces.
As mentioned above, risk is a complex combination of the components of hazard, exposure and vulnerability. This particular concept of risk,widely employed in US and Europe, aims to quantify loss and could be regarded from a scenario-based or probabilistic view point. This is a mathematical multiplication of exposure and vulnerability components.
However, utilising mathematical operators requires strictly quantitative indicators. As a consequence, most risk assessments conducted using this derivation are calculated in economic terms. This makes it difficult to take less immediately tangible, less easily quantifiable factors such as environmental, social and indirect damage,and the inability to avoid or recover from a natural disaster, or lack of resilience into consideration.
The physical sciences attempt to quantify this definition of risk, measurable in absolute values, while the social sciences consider relative and perceived risk, not easily quantifiable and space and time dependent. For a comprehensive approach to risk profiling, these different disciplinary approaches have to be integrated for the purposes of disaster preparedness, mitigation and policy planning, and the methodology has to reflect this interdisciplinarity. The new approach to risk profiling proposed in this report, which integrates socio-economic aspects as defined by UNISDR (2003), involves evaluating the full range of traditional and modernapproaches to quantify natural hazards. This approach involves viewing the problems of natural hazards and disasters and long-term safety survivability within the context of sustainable development (Parker, 2000).
3.0 Proposed Methodology
This section presents a stepwise approach to deriving a risk index for Sri Lanka at national level fromsingle and/or multiple hazards.The flow chart below details the key steps considered. It consists of three components of Risk:Hazard, Exposure and Vulnerability. The integral of the three provides a relative risk for each or combined hazard.
The proscribed methodology advocatesthe inclusion of Landuse data in grid cells in addition to the Exposure and Vulnerability data already collected in Grama Niladhari (GN) divisions. The Integrated Risk Map being proposed has to incorporate all these diverse elements and provide a method of bringing them together to mutually inform the national picture of risk, while remaining sensitive to the local variation of risk. The methodology suggested has to be one which is flexible and accommodates gaps in data sets, the changing nature of parameters that shape risk, and allows for a certain margin of error.
The proposed methodology has incorporated original and other modelling techniques from various vulnerability assessment approaches. These include:
- Application of Remote Sensing and GIS for Flood Risk Analysis: A case study at Kalu‐Ganga River, Sri Lanka (GeoInfromaticsCenter, 2008) – ranking methodology
- Fine Scale Natural Hazard Risk and Vulnerability Identification Informed by Climate in Sri Lanka (Zubair et al., 2005) - Grid based approach
- An assessment of weather-related risks in Europe: maps of flood and drought risks(Genovese et al., 2007) - Scaling methodology
- Dissolution based on landuse data (this study)
- Relative Vulnerability of GN Divisions to Food Insecurity, (UNWFP, 2006) -
Use of Principle Component Analysis
- Geographically weighted Regression analysis (Fotheringham et al., 2002)
- Multi-risk assessment of Europe’s regions (Greiving, 2006) - Calculation of Integrated Risk Index
Figure 1: Flowchart of proposed methodology for a particular peril deducing risk.
3.1 Hazard Component
The significance of the risk posed by any given hazard would be in part informed by its relative weight to other hazards. This concept it at the core of the Hazard index qualifying relative risk.
Hazard datain the form of maps of spatial extent and relative severity produced at different national institutes and universities would be collated by the Disaster Management Centre (DMC)Sri Lanka. These data sets will include Landslide data from the National Building Research Organisation (NBRO), Tsunami (and sea level rise, erosion and storm surge) data from the Coastal Conservation Department and windstorm (cyclone) data from the Department of Meteorology. There will also be flood data and drought data from the University of Peradeniya and the Department of Irrigation respectively.However, this data and its parameters currently differsfrom hazard to hazard.(For example, cyclone data mayhave a 10 km spatial resolution,while Tsunami inundation data may have a 50 m spatial resolution.) It would be necessary to generate a consistent multi-hazard dataset based on an agreed standardised spatial resolution for national level risk analysis. A possible common spatial resolution would be 1 km grid cells.
Once a consistent hazard layer has been established, clarification would be required for what could be deemed high, medium and low risk levels across hazards. The classification and the commonalities for the different hazard types would also need clearer definition and consensus. This would be carried out in consultation with the partner departments and institutes mentioned above.
In order to determine the relative weight assigned to each hazard,two approaches are suggested. First, a hazard or loss database could be used to evaluate frequency of occurrence and severity. Examples of such a database would be DesInventar or Centre for Research on the Epidemiology of DisastersCRED. However, these databases are limited bythe paucity and quality of historical records. The second option would be to conduct an expert elicitation exercise using techniques such as the Delphi method.
3.2 Exposure Component
We will consider four key types of exposure:population, property, agriculture and transportation (+ utility) network. These exposures were selected based on the following criteria:
a)Wide spectrum of parameters that covers socio-economic impacts which would directly affects policy planning
b)Ability to complements other studies conducted without as well as within Sri Lanka at national, regional and local scales.
c)Availability of high resolution census data (at GN level).
However we do recognise that these 4 exposures alone will not adequately incorporate all influencing issues such as effect of poverty. Therefore, there is flexibility in the approach to extend to more than the four parameters proscribed.
For each hazard, the most suitable exposure parameters for population, property, agriculture and transportation (+ utility) network would be considered. For each of theseexposure types 3 parametersare considered. The parameters are limited to 3 in number to preserve computational efficiency in calculating the vulnerability matrices for each type of exposure. It should be noted that for the parameters of demography, residential property and agriculture, each parameter has a binary option.
Table 1 outlines the parameters considered and the proscribed ranking for flood hazard.For demography, the selected parameters include gender (male or female), age (20-59, or other), and economic status (active or non-active). Parameters of residential property (all properties were considered as single 1 storey buildings) are type of construction (brick-cabook-cement or other), type of toilet(water sealed or other) and access to water supply (yes or no). For agriculture, the parameters considered were crop type1 (home gardens or commercial), crop type 2 (paddy or non-paddy) and spatial extent (< 20 ha or other). For transport, infrastructure parametersconsidered are A-roads, B-roads and rail tracks.
As the spatial extent of the GN divisions are not uniform, both absolute numbers and percentage contribution of each factor in each parameter have to be retained. The significance of using percentage contributions or proportions in addition to absolute numbers is to ensure that the size of the GN divisions do not unduly skew the representation of data. This would provide a better platform from which to conduct statistical analysis and avoid spatial biases within the datasets.
3.3 Vulnerability Component
The vulnerability of demography, agriculture, transport and residential property was also assessed for each hazard. To assess the vulnerability of the chosen parameters of the four exposure classes, first a matrix of each exposure type was created, with 8 permutations for the 3 parameters and binary options provided (2x2x2). These 8 permutations were then ranked dependant on the hazard concerned with the ranking of 1 being the least vulnerable and the ranking of 8 being the most vulnerable. The ranking is dependant on subjective interpretation which requires expert elicitation.Therefore, equation in Figure 1 represents:
Vul stp1GN(i) = ∑ CExp% GN(i) RHzE GN(i)
Where Vul stp1GN(i) is the step I of the vulnerability assessment for a particular GN division i, CExp% GN(i) is the calculated exposure percentage for a particular GN division i and RHzE GN(i) is the ranking of the exposed permutation for the particular hazard for a particular GN division, i. Constriction to 8 permutations decreases the variability of the model but still maintains sufficient flexibility and computational efficiency (an example of which is shown in Table 1).
Age Group / Employment / Gender / Ranking20-59 / Econ. / Male / 1
20-59 / Econ. / Female / 2
20-59 / Non. Econ. / Male / 3
20-59 / Non. Econ. / Female / 4
Other / Econ. / Male / 5
Other / Econ. / Female / 6
Other / Non. Econ. / Male / 7
Other / Non. Econ. / Female / 8
Construction Type / Access to Safe Water / Toilet / Ranking
Brick, Cabook, Cement / Yes / Water seal / 1
Brick, Cabook, Cement / Yes / No Water seal / 2
Brick, Cabook, Cement / No / Water seal / 3
Brick, Cabook, Cement / No / No Water seal / 4
Other / Yes / Water seal / 5
Other / Yes / No Water seal / 6
Other / No / Water seal / 7
Other / No / No Water seal / 8
Crop Type1 / Crop Type2 / Extent / Ranking
Non Home garden / Other / >20 ha / 1
Non Home garden / Paddy / < 20 ha / 2
Non Home garden / Other / >20 ha / 3
Non Home garden / Paddy / < 20 ha / 4
Home garden / Other / >20 ha / 5
Home garden / Paddy / < 20 ha / 6
Home garden / Other / >20 ha / 7
Home garden / Paddy / < 20 ha / 8
Table 1: Example of Population (top), residential property (middle) and agriculture (bottom) exposure type and matrix permutations for flood hazard.
By multiplying the percentage exposure of each GN by the ranking used, we derive a weights matrix for each GN. By employing the landuse classification at 50 m cells (to comply with tsunami inundation hazard data at 50 m resolution), we distribute the weighted matrix over GN division based upon landuse type. This calculation would be on a GIS based multi-band raster. This step would also act to normalise the spatial variability of the different sizes of GN divisions.
However, each of the exposure datasets also show significant inherent spatial and attribute inter-dependence. To negate this bias and reduce spatial variance, a well known spatial statistical approach, Principal Components Analysis (PCA),is used (eg. Shaw, 2003). This technique is used to transform the data weighting key principal components whereby compression of data would eliminate redundancies. PCA does this by transforming from an input multivariate attribute space into a new multivariate attribute space whose axes are rotated with respect to the original space. Therefore, the resultant axes (attributes) in the new space are uncorrelated. The first principal component will have the greatest variance, the second will show the second most variance not described by the first, and so forth. For example, data redundancy is evident in a dataset comprising elevation, slope, and aspect (on a continuous scale). Since slope and aspect are usually derived from elevation, most of the variance within the study area can be explained by elevation. PCA conducts computations faster, while heightening the accuracy of the computation.
This process will create spatially uncorrelated eigen values that could be then used as the contributing weights for the parameters considered for vulnerability analysis. The methodology could be extended to include spatial regression, making it applicable on a national scale.
Geographically Weighted Regression (GWR) is a fairly recent contribution to modelling spatially heterogeneous processes (Brunsdon et al., 1996; Fotheringham et al., 2002). The underlying premise of GWR is that parameters may be estimated anywhere in the study area given a dependent variable and a set of one or more independent variables which have been measured at places whose location is known. Taking Tobler’s (1970) observation about nearness and similarity into account, we might expect that if we wish to estimate parameters for a model at some location, thenusing observations which are nearer that location should have a greater weight in the estimation than using observations which are further away.
The resultant matrix would be an integrated vulnerability index that could be used at national or sub-national level. However, a significant component of this methodology and key to its success would be the careful selection of vulnerability classifications described above.This should be done at least at GN level and we need to address the problem of applicability of the data for a particular hazard taking into accountthat the boundary marks for rivers could also be the boundary of the GN division.
Transport and utility networksshould be considered under the binary approach of areas intersecting with hazard area as totally vulnerable. There are well established methodologies to determine common cause failures and system overloading which are particularly applicable for vulnerability of utility and lifeline networks (Javanbarg et al., 2009; Berdica, 2002). However, these considerations need to further evaluated for their appropriateness within the context of Sri Lanka. Therefore, to assess vulnerability of transportation network, we propose a road length density weights matrix for each grid cell to be adopted. This same methodology could be extended to utility networks such as electricity and water.
3.4 Risk Ranking Component
Risk would be regarded as a combination of the degree vulnerability and hazard intensity. To progress from this single hazard risk matrix to multiple hazard, several of the matrices could be stacked together to provide a composite risk profile.
We could also then modify the resilience matrix to consider aspects such as seasonality, mitigation (evacuation of people from cyclone paths due to Disaster Risk Reduction measures being implemented).The resilience index that would conduct this mitigation calculation would also be subject to expert elicitation to determine common consensus among risk practitioners.
Degree of Vulnerability1 / 2 / 3 / 4 / 5
Hazard Intensity / 1 / 2 / 3 / 4 / 5 / 6
2 / 3 / 4 / 5 / 6 / 7
3 / 4 / 5 / 6 / 7 / 8
4 / 5 / 6 / 7 / 8 / 9
5 / 6 / 7 / 8 / 9 / 10
Table 2: Risk Ranking matrix for each hazard (Modified from Greiving, 2006)
4.0 Other Considerations
The risk ranking matrix merges local data (at GN level) withsub-national level data. The methodology for this composite risk matrix is concise, enabling policy makers to identify areas of concern quickly and easily. It also allows the user to identify which component has a greater impact on an area of interest (for example, hazard or vulnerability).
4.1 Limitations
This methodology is not without its limitations. Since the methodology proscribes an index, the risk is quantifiable only as a relative risk. Moreover, the methodology does not include any probability of the likelihood element of risk. It does not also include a measurement of uncertainty, either epistemic or alletory.
It should also be noted that there are problems concerning data quality. Some of the key data quality problems include:
i) the availability of data
ii) limits to measurability – some aspects being difficult to quantify
iii) the weighting problem – the Delphi method provides weighting results which would need regular and expert updating
4.2Future Work
The next stage would be to develop an inactive risk map within a GIS-based macro. This would allow the user to have some control over the hazard and vulnerability options being used, which would strengthen this methodology.
To address the issue of data quality, the next Sri Lankan census of 2011 could incorporate information elicitation which could go towards mitigating some key limitations. Further, better and higher resolution landuse mapping (currently being at a scale of 1:50,000 from 1992) should be used to differentiate between urban and non-urban; residential and non-residential areas.