Project Topics
Get Complete Project Material File(s) Now! »
Vulnerability in hierarchic systems and SES
The analysis of vulnerability in systems reveals high usefulness when it is able to identify i) the vulnerability of particular entities, ii) the vulnerability along nested levels in a scale, and finally, iii) the factors and mechanisms determining the latter (ii) in relation to the former (i) (Adger, 2006; Birkmann, 2005; Hiete and Merz, 2009; Turner et al., 2003a; Vogel and O’Brien, 2004).
Indeed, the fully interconnectedness of entities in hierarchic systems entails that « a disturbance introduced into an organization at any one level reverberates at all the levels it covers » (Feibleman, 1954, 6th law of Levels, p. 61). Vulnerability analysis in this kind of systems should therefore include the notion of the so-called domino effect (Michel-Kerjan, 2000; Turner et al., 2003a) 25: when a hazardous event —e.g. a flood— takes place, impacting any entity in a given system, the initial shock is expected to ripple through the system in a series of chained effects.
Inasmuch as that interconnectedness is driven by inter- and intra- level interactions in the given system, the spreading of the impact shall be considered bidirectional. First, intra-level interactions will spread impacts from the entities directly exposed to the hazard to the rest of the entities in the same level. Second, due to inter-level interactions, impacts in one level will reverberate through all levels in the system.
Furthermore, considering that the properties of entities at one given level emerge from lower levels, entities and subentities might be differently susceptible to the impacts of the hazard. As a consequence, some effects might not be observable but at certain specific levels, whereas the factors that explain them (or originate them) should be searched for in lower levels of the system.
Ultimately, the way the intra- and inter- level interactions either mitigate or amplify the magnitude of the initial shock in its spreading along the system depends on the system’s topology; i.e. the arrangement of system’s composing entities, pattern of interconnections between them and functional form adopted by those connections (Dekker, 2007).
Scales of analysis in the present work
Up until this point, the idea of scale and level has been somewhat abstract in this exposition. The scales in which we can distinguish the levels we have been referring to are multiple though. The selection of both scales and levels to be included in the study is an arbitrary decision driven by the research goals.
Hierarchic systems are dynamic. So is vulnerability according to our definition (see subsections 1.1 and 2). In effect, vulnerability is a property of the system that, emerging from the system’s bosom, co-evolves with it (Adger, 2006; Birkmann, 2005; Birkmann and von Teichman, 2010; Felbruegge and von Braun, 2002; Turner et al., 2003a; Vogel and O’Brien, 2004, among others). Time scale shall be thusly present in our analysis. Furthermore, the consequences of a concrete disruption (a flood) due to such vulnerability are not all observable in the same time span (Brémond et al., 2013; Merz et al., 2010). Hence, regarding levels in our time scale, we shall approach our study by using two different ones: immediate and belated time spans.
Observable consequences of floods are also going to be linked to the extent span of the geographical areas considered in the analysis. Namely, if the studied territory is limited to the flooded area, we would be focusing our analyses most likely on the consequences over the entities directly impacted. Larger territories would give the opportunity to also include the consequences over entities not directly impacted. Insofar in the study of flood impacts geographic distributions of entities play a fundamental role in the dynamics of observable impacts in the system, the present work also includes a spatial scale to measure the extent of the territory. As in the prior case, two main levels will be set: the territory corresponding to the flooded area (direct consequences) and the territory outside the flooded area (indirect consequences).
We have already stated that in a hierarchic system as the one we study i) the effect of the disturbance affects each entity according to its properties and state; and ii) the interaction of entities at a given level makes new different properties emerge at higher levels. Thus vulnerability analysis displays all its potential when capable of identifying vulnerability of individual entities and along nested levels. Therefore there is yet a third scale worth considering. We will refer to such a scale as aggregational scale.
Levels along this scale might be a little more complicated to establish though. Vul-nerability analysis available in the literature have been accomplished at different levels in what geographers call spatial resolution scales. In the context of SES, such a scale makes reference to the degree of detail with which the system is represented. Common levels along this scale are global, national, regional, local, even sublocal (e.g. communities). Levels alike are used by sociology and political economy when approaching the study of SES (compare the so-called « spatial levels of political jurisdiction » in Gibson et al., 2000, with Birkmann (2007); Birkmann and von Teichman (2010); Birkmann et al. (2013, 2014a)). Such correspondence is not surprising though. Those levels of jurisdiction can be associated with decisional levels, capable of influencing the evolution of the system by the encouragement/discouragement of adaptations.
Limitations to the vulnerability assessment in systems
To describe and measure how each factor and their potential combinations drive the vulnerability of every entity and the whole system, comprehensive vulnerability analyses should encompass the system in its totality. Nonetheless, in the type of system we are attempting to analyze —hierarchical complex systems coupling environmental and human realms—, such degree of comprehensiveness remains unrealistic 28 (Turner et al., 2003a).
Indeed, in hierarchic systems, studies are accomplished by setting clear boundaries that confine the dynamics of the particular processes at the specific levels we wish to study (McGinnis and Ostrom, 2014) 29. Such process though is rather arbitrary insofar it is based in research interests, hypotheses and assumptions that eventually results in a simplified landscape of the subjacent system. As a result, vulnerability analyses are performed on subsystems whose boundaries are artificially set according to the actors, interactions, outcomes and rules considered relevant a priori. Variables and levels out of bounds will be treated as constrains and noises that affect the system but are not affected by it.
Vulnerability assessments are therefore heavily case-dependent 30. They are also subject to a high degree of uncertainty, coming from three different sources: i) incompleteness, derived from the existence of arbitrary, yet necessary, boundaries to the study; ii) arbitrariness linked to the analyst choices and initial values; and iii) data quality and availability.
From vulnerability assessment to flood damage assessment
Insofar impact assessments are subjacent to vulnerability valuations and charac-terizations, the core of this work will rest upon flood damage assessments. Their implementation do not come absent of problems though.
Nature of damages included in this work
Floods impact economic systems in a wide range of ways, turning the assessment of the consequences a bit knotty (Hallegate and Przyluski, 2010; Przyluski and Hallegatte, 2011). For this reason, the existing literature discusses and establishes different typologies of flood impacts (see, for example, Brémond et al., 2013; Bubeck and Kreibich, 2011; Green et al., 2011; Hallegate and Przyluski, 2010; Merz et al., 2010; Meyer et al., 2013; Penning-Rowsell et al., 2013; Penning-Rowsell and Green, 2000, among others).
The first typology distinguishes between direct and indirect impacts. The former is commonly defined as the impact related to direct exposure to the flood (physical flooding). Regarding the latter, the literature is however less settled as we are going to see.
There are authors that understand indirect impacts as the impacts caused by the consequences of the disaster, not by the disaster itself (for example Hallegate and Przyluski, 2010; Przyluski and Hallegatte, 2011). Other authors though see the necessity of going further and establish subtypologies of indirect impacts. Thusly they distinguish between primary and secondary indirect impacts (Penning-Rowsell et al., 2013; Penning-Rowsell and Green, 2000). The first ones — primary indirect impacts— encompass both emergency costs and consequences over the economic activity of the elements directly flooded. Secondary indirect impacts, on the other hand, are understood as the impacts spreading further away from the flooded area, through economic linkages.
Further, some authors introduce the notion of business interruption costs (Green et al., 2011; Merz et al., 2010; Meyer et al., 2013). They are defined as impacts due to the interruption of economic processes in the flood aftermath in the areas directly affected by the flood. They can be presumed as a sort of alternative to primary indirect impacts, yet their proponents point out that they should not be understood as pure indirect impacts. Instead, business interruption costs are intended to serve as a bridge between direct and indirect impacts 31: they are caused by direct impacts in the aftermath of the flood, yet their scope is much more limited than the one considered for indirect impacts.
The practical reach of such abundance of typologies and classifications of indirect impacts is nonetheless debatable. Doubtlessly their existence enables analysts to accomplish thorough analyses of the consequences that a flood might bring over a system. However such richness in applied studies might be extremely difficult to reach, especially insofar it demands a high level of detail in the information needed.
In addition, all the prior definitions and classifications mix both spatial and temporal scales in a rather implicit way. In such regard, Brémond et al. (2013) propose to classify impacts discriminating explicitly along the two aforementioned scales. In such manner, they move along the spatial scale to distinguish between direct and indirect flood impacts. The latter are then defined as those which occur in an area that has not been exposed directly to flooding, whilst the former are defined in the same terms already stated at the beginning of the section. Similarly, they discriminate between instantaneous (or immediate) and induced (in the sense of belated) impacts based on a temporal scale. Instantaneous impacts are then defined as those ones which occurs during or immediately after the flood event. Induced impacts, in the other hand, denote those ones which occur later in time.
The advantages of the Agent-Based Modeling
Beginning with the most obvious one, ABMs provide utter flexibility to the modeler in relation to the model’s specification (Bonabeau, 2002; Richiardi, 2004). The plasticity in the definition of the entities —attributes and state variables— gives the modeler the capability to include, in a given model, tiers of variables representative of several dimensions of analysis. Thus, applied to vulnerability, ABMs offer the potential to accomplish multidimensional studies of systems.
In like manner, the potential for both topology definitions and their endogenous evolution, enables the modeler/analyst to perform analysis along several levels of aggregation of individuals within the same modeled system: individual agent, col-lectivity of agents, the whole system (Bonabeau, 2002). Thus allowing to study, to quote Schelling, « the micro-motives of the macro-behavior » observed in the system (Gilbert, 2008; Schelling, 1978); i.e. the behavior in different aggregation levels based on the incentives that govern the individual conduct. As it was mentioned in section 2, the analysis of vulnerability in systems turns more useful when it is able to identify nested vulnerability along different levels of aggregation of entities.
Not only levels of agents’ aggregation can be accounted for in ABMs. As it was stated, an ABM is a tool for the dynamic simulation of a system. Thus ABMs enable us to perform analysis according different terms in the time scale. Moreover, for our particular case, to be able to define spatially explicit environments is highly appealing. Insofar it facilitates the definition of flood prone areas, direct impacts of floods can be delimited. Making use of both the topology and the dynamic nature of the simulation, we should be capable of tracking down triggers, causes, and factors of impact spreading along the existing topology.
As well, ABMs may help us to tune alternative output trajectories (Loomis et al., 2008). As we stated, the output of a system in a bottom-up approach emerges from the interaction of the entities that compose the system. However the presence of characteristics such as:
1. Heterogeneity in the nature of entities.
2. Heterogeneity in entities’ attributes.
3. Heterogeneity in entities’ initial endowments.
4. Potential for topology evolution.
5. Flexibility in the initial topology construction.
The drawbacks of the Agent-Based Modeling
No simulation tool or technique is perfect, and ABMs are not an exception. Nonetheless, a few drawbacks commonly attributed to the technique can be traced back to the characteristics of the underlying kind of systems we represent with ABMs.
Inasmuch as what we eventually model is a subsystem of a much larger and complex hierarchic system (based on the property of near-decomposability), ABMs are always conceptualized with a concrete purpose. The model has to be designed at the right level(s) along the different scale(s) considered relevant to fulfill the purpose. Otherwise, the ABM can end up overfitting the submodel, i.e. fitting both main processes and noises in the system due to a too detailed design (Richiardi, 2004).
Depending on the available information and degree of comprehension of the sub-system that such information allows us, there might be several alternative designs to achieve the same purpose (so-called equifinality or multiple realizability). Each of those potential designs will require, to a greater or lesser extent, the formulation of different assumptions in relation to very different instances, such as unknown pro-cesses, unknown behaviors or initial endowments. This binomial design-assumptions sets the conditions to observe the emergence of particular sets of outcomes (Bonabeau, 2002; Norling et al., 2013; Richiardi, 2004; Sawyer, 2013). Given the complex na-ture of the systems we are studying, small deviations or modifications in the model specification (assumptions included) and/or initial conditions may result in great outcome variation (so-called butterfly effect). Such dependency of the outcomes on the model’s architecture, assumptions and initial conditions limits the capacity of ABMs to provide generalizable results.
Concerning assumptions in ABMs’ specification, Galán et al. (2013) introduce the notion of artefact. They first establish two overlapping categories of assumptions:
1. Core assumptions —which are those essential to attain the purpose of the model— and accessory assumptions —that encompass all of those not considered essential for the purpose of the model, but are needed to complete it.
2. Significant assumptions —those capable to influence the outcome of the model— and non-significant assumptions —which play a neutral role in relation to the outcome obtained.
Then, they define the artefact as the fail in identifying accessory assumptions as significantly influential in the outcome of the model. Such artefacts are especially problematic when the purpose of the model is to describe, characterize and/or explain causal processes in a given system 3. In these cases, the misunderstanding in the identification of relevant mechanisms or factors, namely to deem significant assumptions as non-significant, entails a problem of representativeness of the model in relation to the real system. As a consequence, both explanatory power of the model and its validity will be compromised.
Nonetheless, our ability to avoid the misidentification of significant hypothesis relies on the adequacy, quantity and quality of the information available. The amount of data required to build an ABM is related to its purpose and degree of detail. Higher degrees of detail and complexity in ABMs lead to the utilization and combination of several data sources and elicitation methods (Smajgl and Barreteau, 2014a). Nevertheless, sources of available information might not be representative of the same level of resolution, which will eventually force the establishment of additional hypotheses and the loss of detail. Moreover, the kind of systems we try to model can include what Bonabeau (2002) call soft factors. In business jargon, soft factors identify all factors not easily calculable, measurable and/or systematizable, often characterized by unpredictability and need of interpretation.
Table of contents :
Introduction
1 The notion of system in the present work
2 Vulnerability. Dissertation’s working definition
3 Vulnerability, systems and scales
4 Approach to vulnerability assessment
5 From vulnerability assessment to flood damage assessment
6 Modeling method
Research questions & dissertation outline
1 From the bottom up: Agent-Based modeling
1.1 The Agent-Based Model
1.2 The advantages of the Agent-Based Modeling
1.3 The drawbacks of the Agent-Based Modeling
1.4 Balancing the pros and cons of ABMs for the purpose of this dissertation
1.5 Implementation of an Agent-Based Model upon study cases
2 The COOPER model
The COOPER metamodel
2.1 Study case
2.2 The cooperative winemaking system (CWS)
2.3 The cooperative winemaking system (CWS): geographical aspects .
2.4 The cooperative winemaking system (CWS): analysis of the production
2.5 The cooperative winemaking system (CWS) in face of flood hazards
2.6 Overview of key hypotheses retained
The COOPER agent-based model
2.7 Purpose
2.8 Model conceptualization and design concepts
2.9 Entities, state variables, and scales
2.10 Process overview and scheduling
2.11 Submodels: flood impacts
2.12 Output
2.13 Model calibration
2.14 Initialization
The COOPER ABM. Financial analysis: extra hypotheses and model modifications
2.15 Overview of changes in relation to hypotheses
2.16 Process overview and scheduling
2.17 Submodels: flood impacts
2.18 Output
2.19 Initialization
3 Results I: system level
3.1 General discussion of results obtained
3.2 Article: Are interactions between economic entities determinant for the estimation of flood damage of complex productive systems? Insights from a micro modelling approach applied to wine cooperative system .
3.3 Homogeneous distribution, assuming every farm opts for the internal coping tactic
3.4 homogeneous distribution, comparison of internal and external coping tactics
3.5 Internal coping tactic. Comparison of configurations of interactions
3.6 External coping tactic. Comparison of configurations of interactions
4 Results II: from individual to system level
4.1 General discussion of results obtained
4.2 Article: Floods, interactions and financial distress: testing the financial viability of individual farms in complex productive systems and its implications for the performance of the system
4.3 Coping tactic
4.4 Variations in the configurations of links between elements within the system
5 Discussion, conclusion and perspectives
5.1 Summary and conclusions
5.2 Perspectives and further research
Appendices
A Resilience: Where does it fit?
B Case dependency in vulnerability and damage assessment. Analytical example
B.1 Case dependency of vulnerability assessments in hierarchical complex systems
B.2 Case dependency and influence of topology in flood impact assessment
C Wine classification
C.1 Before 2009
C.2 After 2009
D Complete list of stylized facts about the cooperative winemaking system
D.1 The Cooperative winemaking system (CWS)
D.2 The CWS in face of flood hazards
E Model implementation
E.1 Model implementation
E.2 Flowchart symbols cheat sheet
F Flood impacts. Technical cards
F.1 Brief explanation of the contents of the annex
F.2 Damages in soils
F.3 Damages in plants
F.4 Damages in harvest due to floods
F.5 Damages in harvest due to plant destruction
F.6 Damages in farm buildings and materials
F.7 Damages in harvest due to damages in farms
F.8 Impacts on vinegrowing cost
F.9 Damages in winery’s buildings and materials
F.10 Damages in harvest due to damages in winery
F.11 Impacts on winemaking cost
G Model data sources
H Resumé de la thèse en français Introduction
H.1 La notion de système dans ce travail
H.2 Vulnérabilité. Définition utilisé
H.3 Vulnérabilité, systèmes et échelles
H.4 Approche de l’évaluation de la vulnérabilité
H.5 De l’évaluation de la vulnérabilité à l’évaluation des dommages causés par les inondations
H.6 Méthode de modélisation
H.7 Plan de thèse
H.1 Aperçu général
H.2 Brève description du processus de production opérationnel
H.3 Brève description du processus d’investissement-financement
H.4 Brève description des autres variables intervenantes
H.5 Procédure de simulation, procédure d’inondation et impacts des inondations
H.6 Analyse financière
Resultats I
H.1 Discussion générale sur les résultats obtenus Resultats II
H.1 Discussion générale sur les résultats obtenus
Discussion et conclusion
H.1 Résumé et conclusions
H.2 Perspectives de recherche
Bibliography
