Memoire Online - A GIS-based modeling of environmental health risks in populated areas of Port-au-prince, Haiti

This thesis has been submitted in partial fulfillment of requirements for an advanced degree at The University of Arizona and is deposited in the University Library to be made available to borrowers under rules of the Library.

Brief quotations from this thesis are allowable without special permission, provided that accurate acknowledgment of source is made. Requests for permission for extended quotation from or reproduction of this manuscript in whole or in part may be granted by the head of the major department or the Dean of the Graduate College when in his or her judgment the proposed use of the material is in the interests of scholarship. In all other instances, however, permission must be obtained from the author.

Without any doubt the success of a study like this relies on a well-thought methodology, an excellent design, a good theoretical background, reliable data and tools, and a sound analysis. However, without the support of people who are expert in specific domains, this study would be more challenging and might not be made possible. A popular and wise biblical verse says: «People die for lacking knowledge». I would not physically die without access to some precious information released by many of those who helped, but I would be dying slowly with impatience, discouragement, sense of defeat, lack of inspiration, and frustration.

I want to take advantage of this occasion to thanks D. Phil Guertin who has been my advisor not only for the thesis but during the complete course of my studies at the School of Natural Resources. His support has gone beyond academics, and covered a large range of assistance that is not possible to list without missing some. I am more than certain Dr Guertin will continue to guide me even after the completion of my master's study. I want to express my gratitude to Dr Christopherson who opened the CASA Lab for me during the tedious digitization process and had provided me profitable guidance for the generation of Port-au-Prince's DEM. My appreciation goes to Wissler for sporadic but precious intervention when I was struggling with some ArcMap processes. Mickey Reed was irreplaceable for specific advice and access to fine-point tools and processes. Thanks to all the Advanced Resource Technology's staff for unconditional support and flexibility. I am grateful to Kareen Thermil, who granted me access to some precious information and data, and Juvenel Joseph, my brother who did any necessary arrangement to facilitate acquisition of much of the data needed and available from Haiti. I want to thank either the group of Haitian professionals and students who accepted to participate in the EOW survey. Finally, the best for last, I want to thank my wife who accepted heartedly to sacrifice our time and invest it in the accomplishment of the thesis. Her devotion and support were priceless for the completion of my study.

This thesis is dedicated to my mother who would not have the opportunity to witness the fulfillment of my dream; to my wife whose unconditional support helped me to be at the same time a father, a spouse, and a fulltime master's student; finally to my country which unfortunately is the inspiration of this topic.

UTSIG/CNIGS: Unite de Teledetection et de Systeme d'Information Geographique/ Centre National de l`Information Geo-Spatiale

ABSTRACT

In Port-au-Prince, Haiti's capital, the increasing occurrence and casualties from landslides and floods during the last few years has focused interest toward these natural disasters. The high pressure of human settlements associated with urban migration constitutes the main trigger of these deadly events by increasing the sensitivity of the environment as well as people's vulnerability. Long term impacts of environmental degradation on health related to human settlements have not received as much attention as natural disasters. The inconspicuous nature of environmental health hazards and their related consequences may have diverted stakeholders and people's attention from them. Health hazards derived from the environment are believed to be of a greater spatial extent, cause more losses than any other hazards, and concern more than two-third of the population within the study area. The objective of this study was to identify areas where such health hazards exist and assess neighborhoods' vulnerability to these hazards using a GIS modeling approach that offers the capability of superimposing multiple parameters. Nine factors were combined with different weighting schemes including an Expert Opinion survey. Moreover, several classification techniques were tested and compared in the final process of determining the four risk levels. Finally, a sensitivity analysis was performed to assess the responsiveness of the model to changes induced in the model's parameters.

Though this study was conducted in a context of poor data availability, the results suggest that about 41% of the entire area was subjected to high risk. Pollution originated from water bodies, traffic and waste were found as the most critical, while housing density, which is simultaneously a risk and a vulnerability factor represented the main trigger of many risks encountered. This study called for a deeper investigation of the state of pollution in Port-au-Prince by taking direct field measurement in order to validate the findings. In addition, it reveals the needs for a synergistic effort of governmental and non-governmental institutions to produce and make available spatial data at fine scale and resolution in a cost-efficient manner.

1. INTRODUCTION

Throughout the world, natural disasters have occurred over the last decades with increasing frequency and have resulted in significant mortality, morbidity, and disability among people affected, causing the destruction of physical assets and damaging social resources (UNDP 2004). They also increase the vulnerability of people, communities, and areas impacted by weakening or disabling local infrastructure, livelihoods, businesses, and regional economy (UNDP 2004). If on a global scale global warming, climatic change, and decadal variations are considered as external (but distant consequences of anthropogenic influences) triggers of the escalating of number and intensity of natural disasters, at local scale other human-induced factors such as population growth, unplanned urbanization, alteration of the natural environment, under-standard dwellings, augmentation of impervious areas, and land cultivation increases people exposure to hazards (Sorensen et al. 2006, Smucker et al. 2007). An increasingly larger proportion of the world's population is being exposed to locations at high risk (Huppert and Sparks 2006). This phenomenon is mostly observed in mega cities. Indeed, the world's population is being progressively concentrated in urban areas rather than in rural areas. It is predicted that in 2007, for the first time, more people will live in urban centers than in the countryside (Huppert and Sparks 2006).

Throughout the last three decades, the intensification of migration to urban centers, particularly in Port-au-Prince, Haiti, has resulted in the proliferation of anarchic and precarious habitats, the degradation of the resources of the environment, the deficiency of urban services, and the rupture of ecological equilibrium (IHSI 2003). Water supply and basic sanitation services are still very deficient. The 1998 edition «Health in the Americas» report of PAHO/WHO (1998) describes the situation in Port-au-Prince as follows:

«Solid waste management is a serious problem; bad excreta disposal practices are polluting almost all 18 water sources supplying Port-au-Prince. Drainage systems are inadequate and any major storm produces serious flooding. The growing number of motor vehicles and their inadequate maintenance have created a serious air pollution problem in Port-au-Prince.»

Some illustrations of the effect of environmental degradation on health are the high infestation of dengue in urban areas and the outbreak reported in 1994 in Port-au-Prince (PAHO 2001); acute respiratory infections (ARIs) accounted for 25% of deaths among children under 5 years of age, and suffered by 20% of children of this age group (EMMUS II 1994); and cardiovascular diseases that caused of the admission of 40% of patients at the State University Hospital in 1996 (PAHO 2001).

The United Nations Development Program (UNDP) (1994) considers an event a disaster when the number of human deaths is greater than 10. However, casualties resulting from environmental health hazards can be easily overlooked or even disregarded because, unlike floods or landslides that are spatially and temporally punctual, and with evident physical materialization, environmental health effects are discreet and continuous over space and time. Often the effects are not actually associated with their causes yet the casualties are countless and exceed victims of natural disasters.

On one hand it is recognized that a healthier physical environment is among the factors associated with a decline in morbidity and mortality in the past century (Corvalan et al. 1999). On the other hand high population density is considered one of the vectors responsible for health degradation and aggravation in poor countries (Campbell-Lendrum and Corvalan 2007). According to some estimates, one third of the global urban population and over 70% of people in urban developing countries live in slum-like conditions characterized by poor housing and meager basic services, with ineffective regulation of pollution (Campbell-Lendrum and Corvalan 2007). However, the impact of population densification on living conditions and health is not sufficiently well perceived, at least at the community spatial level. While qualitative information depicts the severe conditions in which the population is living, it is difficult, if not impossible, to find detailed quantitative data about the spatial incidence of a specific health hazard in Port-au-Prince. The spatial study unit used in surveys on living conditions is very coarse and doesn't allow acknowledging the reality at finer spatial scale. In Port-au-Prince, in addition to the intense degradation of the environment aforementioned, the average density found from the SDE^1(*) is more than 61,300 inhabitants per km². In some places, the density caps to more than one million per km². There's a proliferation of diseases among the population particularly in neighborhoods exposed to degraded environmental conditions such as water stagnation, waste accumulation, poorly cleaned channels, and high density housing. This situation puts in evidence the circumstances of vulnerability to which the disadvantaged urban neighborhoods are subjected.

Though Geographic Information Science GISc is relatively a new discipline in Haiti, whose usage started in the mid-1990s, no studies had attempted to link environmental health risk to causal factors using spatial analysis techniques either for Port-au-Prince or the entire country. Most studies have used traditional statistical approaches to address health issues. While the emphasis is being put on the establishment of projects and activities which aim to contribute to the realization of the Millennium Development Goals (Erenberg and Ault 2005), suitable GIS environmental models may represent an important component of such programs by attempting to link key determinants of environmental health risk to their spatial context. The ultimate output results in the localization of places where urgent interventions are to be initiated. The present study embraces this goal, and its framework comprises an assessment of the spatial susceptibility and the related vulnerability to environmental health hazards of populated places in Port-au-Prince using a geographic perspective.

2. LITERATURE REVIEW

Often the terms risks, hazards, and vulnerability are not well understood and consequently are repeatedly misused or used with different meanings (Schmidt-Thomé et al. 2006). In that vein, it is worthy to elucidate these concepts, which will further clarify the phenomenon we intend to assess.

2.1 Risk

A risk is perceived as: the losses derived from a specific hazard to a defined element at risk, over a certain time period (UNDRO 1979); the chance that a particular hazard will actually occur or the probability of experiencing loss from a hazard (Smith 1996); or simply the product of the vulnerability of a community or people to the effects of a specific event, and the potential for the occurrence of that event (Ferrier and Haque 2003). From these approaches, it is possible to express risk either as an average expected number of deaths, economic loss, or physical damage to property, or as the probability of the occurrence of an event. This probability is then dependant on social, physical, economic and environmental factors or processes, which increase the likelihood for people or communities to be harmed.

2.2 Hazards

Common definitions offered in the literature describe a hazard as a physical event, natural or man-made, that may cause damages to human life, property, assets and generate social and economic disruption or environmental degradation. It is also perceived as conditions that increase the probability of losses (UNISDR 2004, UNDP 1994, Smith 1996, Corvalan et al. 1999, City of Long Beach 1998). These definitions imply that natural hazards are normal phenomena that do not set nature necessarily to risk (Schmidt-Thomé 2006a). Risks and hazards are linked through vulnerability.

2.3 Vulnerability

The social, physical, economic, and environmental parameters that we referred to earlier and which boost the chance for the occurrence of a disaster embody the characteristics of the elements at risk, i.e. their vulnerability. The Stockholm Environment Institute (SEI) (2005) characterizes vulnerability as a lack of security from environmental threats and as the result of a mixture of processes that profile the exposure to a hazard, susceptibility to its impacts, and ability to recover in the face of those effects. As Schmidt-Thomé (2006a) noticed, vulnerability must be seen in a human perspective, since human beings put themselves at risk by their exposure to hazardous areas. Other definitions adhere to the concept of exposure to hazard but go one step further by adding the coping ability of people to adjust and reduce the negative impacts. Shortly, it is the potential for a geographic area and its belonging to experience losses from events (Hossain and Singh 2002, UNDP 1994, Chambers 1989, Cutter 1996, Clark et al. 1998, Liverman 2001). Though three categories of vulnerability are suggested (Cutter 2003, Weichselgartner 2001), namely the risk/hazard exposure, the social response, and the vulnerability of places, our perspective in this study is limited to the first and third approach. Yet the social response determined by the characteristics of the population studied can not be dissociated from the other elements (Cutter et al. 2003).

The concepts above suggest that hazards and vulnerability represent the two components of risk. How risk is assessed is another methodological aspect that we want to bring forward.

2.4 Risk Assessment

A risk assessment is formally an estimation of the types and the degrees of danger posed by a hazard. It comprises three elements, which are: 1) hazard identification, 2) risk and vulnerability estimation, and 3) evaluation of the social consequences (Ferrier and Haque 2003). The general formula that arises from this conceptual approach is R = pxV, where R represents risk, p the probability of occurrence, and V, vulnerability to loss. This formula is substantiated and enhanced in the manual for policy makers and planners of the United Nations Agency regarding disaster mitigation, which in addition to hazard and vulnerability inserts element at risk (UNDRO 1991, Diley et al. 2005, BMRC 2006):

Methodologies applied in risk determination include the stochastic and the systematic approach (Ferrier and Haque 2003, UNDP 2004, Dilley et al. 2005). The stochastic (or quantitative) method involves estimating the probability of occurrence and intensity of a hazard, based on historical data. One of the major weaknesses of this approach is the insufficient length of historical information, and even its non-existence in some areas (Huppert and Sparks 2006). In addition, for non-frequent events, some physical processes such as deforestation, urban sprawl, extension of impervious areas, and construction in high slopes may affect vulnerability of places to a certain hazard. Thus, projection of zones susceptible to disasters based solely on past occurrences may be misleading. The systematic or deterministic method depends on prior knowledge about the physical conditions and processes that control the chance of the occurrence of a hazard (El Morjani 2007). This type of information may be more accessible. An integrated approach was used by Diley et al. (2005) in a study sponsored by the World Bank. The application of either or both methodologies relies heavily on the information situation at hand. Given the strict limitation of historical and detailed spatial data regarding environmental health hazards we are dealing with for the study area, this paper relies mainly on the deterministic approach, which offers the advantage of being integrative, and does not necessitate factual and historical information.

2.5 Hazard identification or delineation

Hazards can be characterized by their event frequency and associated characteristics, their probability of exceeding a certain threshold, and their probability of occurrence based on a range of physical factors (UNDP 1994, Dilley et al. 2005). These concepts were previously generalized in Hansen (1984) and Hansen and Frank (1991) which classified hazards determination into two approaches: indirect/causal and direct/occurrence. The earlier is based on a priori knowledge of the underlining factors of hazards in the area under study and involves two sub-approaches: heuristic and statistical. In the heuristic approach factors are ranked and weighted based on their assumed importance in causing the hazards; in the statistical approach the role of each factor is determined in comparison to the observed relations with past/present distribution of the hazard. The fundamental principle of the direct/occurrence approach relies on the observed distribution of the hazards over time.

Whereas the advantages of these techniques are unquestionable, some drawbacks are data availability particularly for small areas, validation of the information, time consumption for data collection, and precision. Error in mapping can influence the predictive ability of the model that may not be possible to be extrapolated to other areas. But the most important limitations are spatial scale and availability of reliable historic data (Dilley et al. 2005). The intricate context of data collection at spatial and temporal scale in health hazards may make it even more inappropriate.

2.6 Environmental health factors

According to Corvalan et al. (1999), two types of environmental threats exist: traditional hazards, which are linked to lack of development, and modern hazards, related to unsustainable development. While the former is mainly related to the household's immediate surrounding standard quality and are quickly expressed as disease, the later is associated to outdoor conditions lacking health and environment safeguards, and their health effects that take a long period before they show. Ehrenberg and Ault (2005), classifying the determinants of health into intrinsic and extrinsic, retained poverty, vector ecology, and human activities and the environment as part of the extrinsic factors that affect health. But this definition doesn't outline the differences between hazards and vulnerability.

More specifically, factors suggested as environmental health hazards comprise indoor air pollution from solid-fuel use, pollution from traffic, power line, wastes, and slum-like settlements (Harper et al. 2003, Maheswaran and Elliott 2003, Nsiah-Gyaabah et al 2004, Greene and Pick 2006, Campbell and Campbell 2007, Campbell-Lendrum and Corvalan 2007). It should be noted that unclean settlements may be considered simultaneously as a hazard and a physical vulnerability. The proximity of slums to each other along with waste concentration within the interior of or very close to the houses, or the proximity of human excreta to residences represents a hazardous condition. On the other hand, high density housing occurring in an environment where air circulation is poor and compact increases the vulnerability of inhabitants of these neighborhoods. As underlined by Nsiah-Gyaabah et al. (2004), an unhealthy environment and overcrowded housing in the slums expose the urban poor to high rates of infectious diseases such as pneumonia, tuberculosis and diarrhea. Other environmental risk factors for elevated blood levels in human body are polluted soil and dust in urban surroundings, and the huge number of automobiles consuming leaded gasoline (Harper et al. 2003). Consequences identified associated with air pollution also include higher disease rates, death, reduced lung function, and neurobehavioral issues (Greene and Pick 2006). Unplanned growth and rapid urbanization are responsible for the degradation of the environment, the destruction of watersheds and wetlands, traffic congestion, contamination of water, and increasing population demands for service which exceed the supply capacity. These conditions place human health at risk, both physically and mentally (WHO 2001; Moore et al. 2003; Latkin and Curry 2003; Fernandez et al. 2003; Nsiah-Gyaabah et al. 2004). Many of these conditions also represent vectors for communicable and non-communicable diseases and are underpinning of human health deterioration (WHO 2001). The health effects of these hazards are countless and it is almost impossible to provide an exhaustive list.

As already stated, the probability aspect in hazard determination is not easily assessed for many reasons. First, even in developed countries, it is recognized that most cities do not regularly produce indicators of health conditions at the neighborhood level, and where they exist, detailed information is very limited (Pettit et al. 2003). Second, deaths or disease occurrence are not usually or systematically attributed to a particular health hazard, the hazard-disaster transmission process is not always materially perceptible. Another characteristic of health hazards is that they are man-made, continuous phenomena with no time boundary until their triggers disappear. Consequently they are not perceived as salient hazards with immediate life-threatening properties. The deterministic approach is well-suited to delineate areas where these hazards are likely to occur.

2.7 Approaches to Vulnerability Assessment

The assessment of the state of vulnerability of an area to natural disasters has traditionally paid attention to the intensity or the scale of a natural event (Hamza and Zetter, 1998). Obviously, as a possible source of danger, the more powerful a hazard is the greater the likelihood that a catastrophe will result. However, the living conditions that frame a population's neighborhood before a disaster occurs is a key factor of the vulnerability of this population to the event (idem). Blaikie et al. (1994) relate vulnerability to the capacity of a person or a group to foresee, to deal, to resist, and recuperate from the impact of a catastrophe. As this approach suggests, people's social and economic characteristics are the center of disaster assessment. According to Bakrim (2001) two extents of vulnerability can be measured: collective and individual. The collective vulnerability result from the conditions prevailing in the economy as a whole, which determines the Gross Domestic Product (GDP), the institutional framework, the financial resources available, and the infrastructures (Adger, 1999). At the individual level the vulnerability is measured by the access of a person or groups to the resources (Bakrim 2001). In this context, at either level, poverty is one of the determinants of vulnerability.

Based on these aforementioned characteristics of vulnerability, several models have been proposed for its assessment. Smit and Pilifosova (2003) represent vulnerability as a function of the exposure (E) and the adaptive capacity (AC) of a given community, in a given location, for a given climatic stimuli, and at a given period of time:

Though the specific mathematical form of this relationship is not stated, the direction of variation is known. E is positively correlated to V while AC has a negative correlation with V. That is, the vulnerability increases with the exposure level, and the greater is the potential to cope with the hazard, the lesser the vulnerability (Mcleman and Smit 2006, Ferrier and Haque 2003).

Furthermore, the components of vulnerability have been identified as access to various forms of capital, financial, physical, social, and human (Sorensen et al. 2006), which some studies crystallize in the GDP per capita or the population density (e.g. Schmidt-Thomé 2006, UNDP 2004). As a result, poor people face greater exposure to hazards because of lower housing standards, location, and lack of access to capital and information (Sorensen et al. 2006, Goodyear 2000).

The Disaster Risk Indexing (DRI) program of the UNDP in partnership with UNEP-GRID (UNDP 2004) uses two measures of human vulnerability, which are 1) the relative vulnerability calculated as the ratio between mortality and population exposed to a hazard; 2) a step-wise multiple regression of disaster mortality as the independent variable and a set of socio-economic dependent variables including economic status, economic activities, environmental quality, demography, health and sanitation, education and human development. On the other hand the Hotspots indexing project implemented by Columbia University and the World Bank under the umbrella of the ProVention Consortium, represented vulnerability by the historical disaster mortality and economic losses resulting from each hazard type (Dilley et al. 2005). Finally the Americas Programme of IDEA in partnership with the InterAmerican Development Bank (IDB) generated four indicators among which, the Disaster Deficit Index (DDI) vulnerability, function of a country's financial exposure to disaster loss and resiliency; and the LDI vulnerability, which is the proneness of a country to significant disasters and their cumulative effects.

Some issues that are worth consideration in these proposed models are, first of all, the access to reliable and consistent information about past events and their correlated casualties. We reiterate that this information is not always collected and in our case simply does not exist. Therefore the probability that the current area of study will be harmed can not be assessed. The second issue is one of scale. Some of the models can be applicable at national or regional scale but not at the small communities scale for which macro-economic or social indicators are not available. Furthermore, including an important number of independent variables in multi-regression analysis results in increasing the level of complexity of these models (Schmidt-Thomé 2006), yet these variables may be inter-correlated - an issue that is likely to overemphasize the model.

Several studies of vulnerability assessment were realized in Haiti, with the purpose to evaluate human and structural vulnerability to natural and human-induced disasters, one at national level, one at departmental level, another one at communal level, and the last one in selected sites (CDERA 2003). However, none has paid attention to environmental health issues, and none was focused on this specific area with very high population density.

The current study evaluates vulnerability at fine geographical scale and integrates physical exposure such as proximity of dense populations to potential sources of hazards. It is recognized that poor communities are more likely to occupy hazardous locations and are forced to use inadequate materials to build their houses, which adds up to their vulnerability (UNDP 2005, Mathee 2002, WHO 2001).

2.8 Multi-Criteria Evaluation (MCE) and Weighted Linear Combination (WLC)

Multi-Criteria Evaluation consists of a set of procedures whose purpose is to facilitate decision making by investigating a number of alternatives in light of multiple conditions and conflicting objectives (Voogd 1983). It has been used in multiple fields, such as land suitability evaluation (Janssen and Rietveld 1990; Pereira and Duckstein 1993), urban planning (Voogd, 1983), and residential quality assessment (Can 1993). In MCE multiple layers can be scaled, weighted and summed into one stratum representing levels of suitability for an investigated issue (Eastman et al. 1995, Jankowski 1995). This process can be brought out by experts, interest groups and/or stakeholders founding their evaluation on the degree of suitability or the importance of the within variables of the issue considered (Dodgson et al. 2000, Malczewski 2004).

According to Eastman (2001), three techniques are generally used to implement MCE. The first option is Boolean Overlay in which the criteria are reduced to two logical suitability statements. This technique is deemed lacking of flexibility regarding the number of choices permitted (Mahini and Gholamalifard 2006). The second procedure, the Weighted Linear Combination (WLC), offers more flexibility than the Boolean approach. Hopkins (1977) portrayed WLC as the most common techniques to integrate multi-criteria evaluation in GIS for land suitability. The last technique is known as Ordered Weighted Average, which is a stronger extension of the two previous techniques, and which addresses uncertainty in modeling interaction between various criteria (Bell et al. 2007).

Another typology of MCE classifies it into two approaches: concordance-discordance analysis and WLC (Voogd 1983, Carver 1991, and Eastman et al. 1995). In the concordance-discordance approach each pair of cell is compared on specific criteria in order to determine which cell outweighs the other, while WLC is based on multiplying a designated weight to the multiple factors that are subsequently summed and ranked (Aly et al. 2005). Whereas the earlier is computational impractical for a large raster dataset, the later is suitable for solving problems which involves multiple factors with a raster geodatabase (Aly et al. 2005). The implementation of WLC is made simple within GIS, using map algebra operations and cartographic modeling (Tomlin 1990). In addition, it is easy to understand and appealing to decision makers (Massam 1988).

Weights assigned in a WLC process may derive from different techniques including Analytical Hierarchy Process (AHP) introduced by Saaty (1977) and based on pair-wise comparison of factors or alternatives; experts judgment or opinion used in many studies (e.g. Beaton 1986, Dakin and Armstrong 1989, Steptoe and Wardle 1994, Clevenger et al. 2002, Tobias 2004), and public opinion. AHP presents the inconvenient of being «too data-hungry», not intuitive, and inconsistency-prone (Bailey and Grossardt 2006). Wherever the pair wise comparison is to be achieved on an important quantity of variables, the process becomes cumbersome and can easily result in inconsistency related to the scores assigned. Surveying the public to collect its view about a matter to which it is unfamiliar would result in collecting erratic and meaningless data. The Expert Opinion Weighting (EOW) is considered more suited to this study for its flexibility and communication accessibility. Yet the designation of experts may be cautiously used since no similar study experience could be accounted to the respondents to the survey, though they all have working experience in health and environment.

2.9 Classification methods

Classification procedures are utilized in various map production software to facilitate user interpretation (Longley et al. 2005). However, the statistical algorithm used to classify a range of continuous values can strongly influence the visual impression (Evans 1977), the analysis (Smith et al. 2007) and consequently the conclusions of a study. Based on the way a thematic map is created, the characteristics of the original data might be overlooked, or there might be a risk of misjudgment about the characteristic of the original data (Osaragi 2002). Natural breaks (Jenks), Quantile, Equal Interval, and Standard Deviation classifications are among the most popular used in GIS software (Osaragi 2002, Longley 2005). ESRI (1996) provides a conceptual framework of the different classification techniques along with some of their advantages and drawbacks.

In the quantile technique, an equal number of features is allocated to each class. While this arrangement is suitable for linearly distributed data it can be misleading since comparable values can be grouped in adjacent classes or diverging values can be put in the same category. On the other hand, the natural breaks method, by looking at big jumps between values overcome this weakness and ensures that similar values are placed in the same class. The equal interval scheme divides the range of values into equal-length sub-ranges and helps determining the number of intervals into which the values are distributed. The algorithm used in the geometric interval insures that there is a good distribution of values in term of quantity between classes. Likewise, this technique makes reliable the change between intervals. This approach is deemed convenient to accommodate continuous data and can generate cartographically comprehensive results. Finally, the Standard Deviation method shows the extent to which an attribute's values depart from the mean of all the values.

A study conducted by Brewer and Pickle (2002) in which they asked the respondents to evaluate seven classification methods recognized the quantile technique as the best for conveying patterns of mapped rates. To investigate the characteristics of different classification algorithms, Osaragi (2002) applied them to seven different datasets. The results suggest that the Natural Break method can be applied to different types of data for its relatively lower loss of information compared to the other, but it is not suitable for data with unclear division. Osaragi recommends examining the distribution of data before choosing a particular method. Alternatively some cartographers suggest to generate several maps for one dataset to allow the reader to compare them (Dramowicz and Dramowicz 2004). The present study compares the classification performed by four of these techniques on the spatial dataset used.

3. METHODS

3.1 Study Area

Port-au-Prince (Figure 1) is the administrative, commercial, and political capital of Haiti, but regarding the size it is the second smallest commune of the country. It measures 36 km²(IHSI 2003). The study area, which is the populated areas within Port-au-Prince, is about 28 km². Elevation in the study area ranges between the sea level and 600 meters. The last census realized in 2003 indicated that more than 730,000 inhabitants (9% of the country's population) populated this place, which represents about 20,500 people per km². This pressure of dense population on this narrow strip of land is not without negative impacts on the environment. The Atlantic Ocean forms the northwesternmost boundary of the study area, which in turn verges on several slums. The study area was obtained by cutting off the south section of the base map approximately at latitude UTM 2049000 meters as indicated on Figure 1(left). This cut was done for several reasons. First, the topographic map sheet used for digitization missed a portion of the south section of port-au-Prince. Efforts to find the missing part at the same scale (1:12,500) were fruitless. The largest scale found, 1:50,000 would difficultly allow digitizing the contours 10 meters apart. Another fundamental reason was the fact that the missing area was poorly inhabited with the density of housing close to zero. Since we wanted to assess health risks in populated places, we felt that the exclusion of this area in the study would not substantially affect the study. The last reason concerned time-efficiency. The elevation at these excluded areas was the highest (about 600 meters). Consequently, a lot of contours needed to be digitized, which would add to the burden of digitizing tasks without contributing to the improvement of the study. Therefore, the most convenient choice was to take this section off of the study area.

Figure 1: left: Base map of Port-au-Prince and the study area; right: Port-au-Prince's view from the southeast hills.

3.2 Data collection

The features included in the dataset derive from two main sources: a) data readily available from the Remote Sensing and GIS Unit of Haiti's Planning Ministry (formerly UTSIG, currently CNIGS) and IHSI; b) digitization of multiple layers from topographic map, scale 1:12,500 prepared in 1994 by the Defense Mapping Agency, Hydrographic/Topographic Center, Bethesda, MD. The first source category includes the administrative boundary of Port-au-Prince, the habitat density, and the land use. The IHSI's SDE delimitation contributed to the reattribution of the habitat density layer. Features digitized within an ArcMap interface included: contours, rivulets and other waterways, high voltage power lines relay-centers and power energy centrals, the main roads and other high-traffic-density streets, the waste collection network, formal and informal marketplaces, the main hospitals, the cemetery, the seashore, and the very important points (VIPs), which are landmarks found on the topographic map. All the layers were standardized to UTM projection, NAD83, Zone 18N, unit in meters. As can be seen most of the features were obtained by the laborious digitization process.

After digitization the features where edited in accordance to pre-established set of topologic rules in order to ensure the integrity of the database. Overlaps, dangles, unwanted intersections, wrong attributions, and any other topologic errors revealed by the topology validation tool were corrected with the editing tools provided in ArcMap until all the errors were adjusted.

3.3 Data limitation

During the digitization, attributes were partially collected and put into the associated tables. In spite of our knowledge of the area, which we used to fill the gap of missing information on the topographic map, this task could not be completed. A field data collection would be necessary to correct this deficit. However due to time and resource limits, this was beyond the scope of this study. In addition to this, an independent data set would be required to validate the digitized features. In fact many errors of digitization inherent to human might have escaped the topologic validation but without compromising the data integrity. As noticed by Murphy (2005), digitizing contours... is a tedious and mistake ridden process. Nonetheless we don't feel that this significantly affected the results of this study.

Essentially, the biggest concern was the lack of data that restricted the insertion of some important factors in the model. The last population census was built upon the SDE unit and contains data about the number of people and other demographic characteristics. Nonetheless, the format of this data has not been made available to the public. We estimate that it is an important step toward comprehending the reality at micro-spatial scale and we strongly encourage researchers to adopt the SDE in future assessments. Finally, during the data collection process it was difficult, if not impossible, to discover any national government entity's website providing access even for purchasing spatial data. The consequence was a loss of much time and energy that could be allocated elsewhere.

3.4 Construction of the model

3.4.1 Process Overview

The choice of variables affecting environmental health hazards and vulnerability arises from the literature review, data available for the study area, and ground-specific reality that might not be in line with any known theory. Generally, tools such as buffer and Euclidean distance were applied to measure people's exposure level to the hazards considered. A raster structure was utilized to facilitate the integration of the multi variables through Boolean operations and overlay combination. Nine factors were included in the model and each was assigned a weight between 0 and 1, based on its relative importance in affecting health. Since we could not access any specific study providing weights for the study area different weighting approaches were brought out. The sub-variables contributing to the making of one factor, such as in the case of traffic pollution, waste pollution, and pollution from rivulets, were weighted on the basis of our own perception of their respective importance. Again this approach was used because of lack of support from the literature.

The entire modeling process was compiled, validated and run within the ArcMap Model Builder through multiple iterations. The model's outline and the script of its execution are provided in Appendix C and Appendix D.

1) The first stage consisted of transforming the basic parameters into factors either by aggregating sub-variables or calculating distance where applied. The general form of this process is as follow:

with F_i: Factor i; V_j: sub-variable j; w_j: weight of sub-variable j, and w₁+w₂+...+w_j = 1

2) Subsequently, grids with continuous values were standardized into discrete values from 1 to 4 using the geometric classification technique.

3) In the final stage the weighted sum of factors was reclassified into discrete values representing the four risk levels, using four different reclassification techniques.

3.4.2 Environmental Health Factors

3.4.2.1 Air pollution from traffic

Vehicular traffic is recognized to be one of the main sources of air pollution in urban cities. Carbon monoxide, hydrocarbons, nitrogen oxides are the major air pollutants generated by motor vehicles, and are the underline causes of lung malfunction, lung cancer, cardiovascular diseases, respiratory symptoms, stroke, neurobehavioral problems, premature mortality, and possibly exacerbation of asthma, which ultimately results in deaths (UNEP 1994a, Watkiss et al. 2000, Maheswaran and Elliott 2003, Nafstad et al. 2003, Greene and Pick 2006). The main roads of Port-au-Prince are associated with high traffic concentration particularly at peak hours. Wargny (2004) describes the traffic situation in the downtown area in these terms: «The traffic jam is quasi permanent. The third-hand vans that provide public transportation spit black smoke, and the carbon monoxide aggregates to the fecal dust...» Refining measures of exposure to air pollution takes into account proximity to the source of pollution (WHO 2005). Other factors that exacerbate pollution concentration in Port-au-Prince are the occasional maintenance of the vehicles, the lack or absence of motor vehicle emission control, leaded gas consumption, and in general the deficient control of cars by the authorities. In addition to housing close to heavy traffic, which also creates indoor pollution, commercial activities take place for long hours in the streets. Therefore, vehicle traffic impact both outdoor and indoor pollution.

Among various approaches, the qualitative method is recommended to assess the exposure to air pollution from traffic (WHO 2005). Different studies used a distance analysis approach to model air pollution from traffic (Elliott et al. 2001, Hoek et al. 2001, 2002, Wilhelm and Ritz 2003, Ferguson et al. 2004, Schikowski et al. 2005). The final determination of traffic pollution was built upon the linear combination of four variables: land use, elevation, traffic density, and distance to roads with high traffic volume. This approach assumes that there is no spatial variability excluding elevation within the area and doesn't take into account different coping strategies and capacity for the households exposed. Likewise the mean level of concentration and the mean amount of time exposed were used in this model.

Though buffers applied differ from one study to another, depending on the situation at hand, a Euclidean distance grid was generated to the main roads and important traffic streets, and distances comprised between 0 and 300 meters were considered in the delimitation of four vulnerability levels ranging from low to very high. Afterward three raster surfaces were constructed for traffic density, elevation, and land use. The elevation raster was derived from the DEM itself created from the contours, the rivulets, and the VIPs. One thousand points were randomly generated for the study area and input as location data to be sampled for DEM values. The elevation points obtained were subsequently converted to raster using the Inverse Distance Weighting (IDW) within the Spatial Analyst tool. IDW was chosen for its simplicity of use and the large amount of points that were available for its creation (1000). Though IDW has the effect of flattering peaks and valleys, the utilization of the VIPs (representing sample of high and low elevation in the area) in the DEM surface creation, this disadvantage was minimized.

High risks were associated with distance less than 50 meters, high traffic density areas, high residential areas and low elevation as shown on Table 13 in Appendix A. These sub-variables were aggregated using weights in relation to their assumed implication in creating air pollution. No data, specific studies or expert opinions were available to support this assumption. We had to make a personal decision about these weights.

Greater weights (0.4 and 0.3) were assigned to the exposure factors (distance to road and traffic density) and a smaller weight (0.15) was applied to the two other factors as seen below:

Traffic Pollution Risk = 0.40*(Distance to Roads) + 0.30* (Traffic Density) + 0.15*(Land Use) + 0.15*(elevation)

The final step involved reclassifying the resulting grid into different thresholds using the geometric interval reclassification algorithm^2(*). This scheme was used for its ability to deal both with the number of values in each class range and to establish consistent change between intervals (ESRI 1996). More details are provided in Appendix in Table 14.

3.4.2.2 Waste Pollution

The United Nations Conference on Environment and Development (1992) pointed out that rapid urbanization and demographic concentration have shocking implications for shelter and sanitation, and especially the disposal of wastes. Waste is responsible for the transmission of agents of infectious disease from human and animal excreta, the breeding of disease vectors, and exposure to toxic chemicals in human and animal excreta (WHO 2006). Port-au-Prince is the perfect example of this statement. Waste production and disposal outweigh the institutional, structural, and managerial capacity of institutions in charge. Two problems arise from waste collection in Port-au-Prince. First, the collection of waste is absent in slums due to inaccessibility to the narrow streets and alleys (World Bank 2005). Second, where it occurs, waste collection is not reliable. For instance, a lack of fuel or money to buy it, broken or lack of vehicles, and employees' strikes may cause waste to store up for days, even weeks, spreading all over the streets, blocking vehicle transport, and giving off offensive smells. In fact the garbage is fully exposed to the air and the wind, along with mosquitoes, cockroaches, and rodents facilitating the spatial propagation of the pollutants. No less shocking is the presence of hogs in some neighborhoods streets where waste is dumped.

Though these factors could not be accounted in the determination of waste incidence, we assumed that the spatial extent of this hazard is inversely proportional to the proximity to the sources. The waste collection network was digitized based on information gathered from SMCRS (Service Metropolitain de Collecte des Résidus Solides). The geometry of the network being linear, we had to consider whether the hazard manifests across the network or in dumping at specific points. However, no such information was available: the collection locations may change at any time and, in spite of the presence of some established posts along the network, there is no way to control the emergence of new unplanned ones. After calculating the Euclidean distance of the waste network feature, the resulting raster was divided into 4 classes from 0 to 400 meters with increment of 100 between each threshold.

To take into consideration the inaccessibility of some areas to waste collection, neighborhoods located at least 400 meters from the waste collection network were also integrated into the model. In low-to-medium residential areas where mostly people of middle economic class or above reside, some private services ensure the pick up of garbage. But people living in high-to-extremely high density housing neighborhoods may be struggling to get rid of waste, may have used non-hygienic or non-conventional ways to eliminate their garbage (for instance burning, or depositing it in the streets when it rains, or dumping it into the channels) , and consequently create a hazard for human's health. Those areas away from the network were given values in reference to the housing density. Then, both distances less than 400 meters and greater 400 meters were summed.

Regarding the sanitation aspect of the neighborhoods, it would not be reasonable to assume that the waste network conditions are uniform everywhere. For instance, in areas involved with intensive commercial activities like informal markets places, waste production is far much greater than in areas with little activity. The waste network at La Saline does not weigh against Lalue's circuit. This consists of a different aspect influencing waste accumulation and conditions. Based on this evidence a raster that stands for waste conditions was created and was integrated into the waste factor determination.

Furthermore, elevation was believed to play a role in neighborhoods' exposure to waste effects. On one hand, waste from upslope is carried down by wastewater from the canals and by runoff; on the other hand, in the absence of an efficient collection system, waste accumulates in lower elevation and is mixed with water from artificial ponds and obstructed canals. The same rationale illustrated for the pollution from traffic vehicle may be also true for waste: better air circulation in higher elevation acts as barrier to attenuate the pollution's impact of the garbage.

The different factors were aggregated using weights of 0.5, 0.3 and 0.2 for distance from waste collection network, conditions, and elevation respectively. Once again those weights were personally chosen since I could not access any example in the literature review for the study area. The result was standardized to values comprised from 1 to 4 as displayed in Table 2.

3.4.2.3 Public formal and informal market places

Public formal and informal markets are another potential source of waste and neighborhood pollution. Recent evidence has suggested a close relationship between the activities of the informal sector and the degradation observed in the urban environment of Port-au-Prince (Howard 1998). The so-called public markets contain unspeakable deleterious hygienic conditions harmful either for the attendants or people living in their proximity. The informal market places pack the downtown area of Port-au-Prince in a chaotic manner and at a point that is impossible to delineate its extent. According to Wargny (2004), everybody attempts to sell something. In the past, Port-au-Prince had some market places. Nowadays, port-au-prince is a market place, continuous, insisting, unstoppable, and obsessing.

Market features were digitized both as polygons and points. Yet the list provided by the communal administration's office was not exhaustive. Many spontaneous and dispersed markets are established along the streets and generate tons of waste for which no recovery plans exist. Some of them are the extended version of a traditional market whose capacity has exploded with the astounding population increase. The nature of these commercial exchange sites make it difficult if not impossible to delineate their physical extent and thus to integrate them in a model.

After calculating the Euclidean distance from markets provided in the SMRCS' list, point and polygon features were combined using the maximum function within the single output map algebra tool, and were reclassified as shown in Table 3 below:

3.4.2.4 Hospitals and the main cemetery

Though there is not much theoretical support and previous studies available for validation, another factor affecting neighborhood pollution in Port-au-Prince is the major hospitals and the central cemetery. Not integrating them into the model would result in ignoring an important part of the specific environmental context in the study area. The conditions of sanitation within and around the hospitals, and more particularly the Sanitarium and the State University Hospital, make them hazardous and unsafe places to be exposed for long time. This situation is aggravated first, by the adjacent location of the mortuary to the general hospital often affected by the common lack of power to adequately maintain the equipment; second, the general belief and its concomitant negative impact that, since the type of service is public, its management will be inefficient. Consequently the level of care allocated is marked with negligence and falls far below the normal level it would be in a private structure.

EPA (2005) provides a list of air pollutants that may derive from hospitals even in normal operating situations. This list comprises mercury, found for instance in thermometers. Mercury emits toxic vapors that go to the lungs, and impact the kidneys, liver, respiratory system, and central nervous system. It can escape into the outdoor through any openings, and even its incineration doesn't prevent it from reaching the outdoor air. Polyvinyl chloride, used in plastic products, is another source of air pollution even when it is incinerated. It emits dioxin, which is a strong carcinogen that hampers normal reproduction and development even in small amounts.

However, the environment in question in this study area is still more complex than the normal conditions described by EPA, though quantitative information is lacking. The sanitation conditions in the main cemetery are precarious. Thus far the cemetery suffers from overpopulation. Often dead bodies are exposed to the air for hours to allow new corpses to be buried in the same grave or in the same hole. Moreover, cases of vandalism and stealing of fresh coffins are reported, which also expose corpses to the air and contributes to pollution.

For the cemetery the same distance approach was used to generate thresholds separated by 100 meters from 0 to 400 meters. Regarding the hospitals, first buffers from 10 to 100 meters were built in relation to the comparative sanitation conditions at these facilities. Then the result of the Euclidean distance on these buffered areas was reclassified into very high risk (4), high risk (3), moderate risk (2), and low risk (1) using corresponding distance thresholds of 50, 100, 200 and 300 meters.

3.4.2.5 Housing density

High density housing poses serious threats to human health by forming sinks for outdoor pollutants, which is aggravated by the fact that the houses are not well-ventilated (Ezatti et al. 2005, YTV 2007). Housing density also makes up the primary source of indoor pollution due to unsafe human excreta disposal, and household fuel consumption (Ezatti et al. 2005). Consequently these unsafe settlements with poor environmental quality expose their residents to social instability and communicable and non-communicable diseases, such as skin and eyes infections, diarrhea, respiratory infections and environmental hazards (WHO 2001a). Slums populate different locations of Port-au-Prince's sea coast, built on and in unstable, stinky and unhealthy waste, in the dust or in the mud. They are called «Brooklyn» or «Manhattan», using a supreme sense of the sarcasm (Wargny, 2004).

We acknowledge that housing density embodies both aspects of hazards and vulnerability. Though present as a vulnerability factor in almost all the variables considered, and acting as the main trigger of the different risks considered, it was suitable to reduce the redundancy that would result by adding it each time. We use the term risk for the reason stated above that housing density represents a factor of air pollution hazard and at the same time a vulnerability feature. That is, even if there is no other risk factor present in a location, high and very high housing density are deemed sufficient conditions for the existence of high risk in these places.

Housing density was originally based on a grid 0.5x0.5 kilometers in size, which classified density in 6 categories from null to extremely high (Table 4). Though it was unclear what theoretical background this categorization was based upon, it was found inappropriate to use it in this study. A first example of this inconvenience provided in Figure 2 indicates that extremely high density represents more than 72% of the area, including slums, commercial and residential neighborhoods. Another illustration is that the difference between average and very high density is only 6 houses per km². Hence it was too general and would not allow us to discriminate neighborhoods with environmental patterns significant for health assessment. Then a grid of higher resolution (0.3x0.3 km, Figure 3) was generated. This was manually reclassified based on SDE count of housing published by IHSI and with the assistance of a live image from Google Earth, which clearly displayed the density difference. Table 5 provides a summary of risk levels associated with different thresholds of housing density.

3.4.2.6 Pollution from water bodies

Urban and peri-urban population health is equally affected by wastewater from urban drains and municipal dumping of waste, especially human excreta (Nsiah-Gyaabah et al 2004). River pollution is particularly found to be worse where rivers pass through cities and the most widespread is the contamination from human excreta, sewage and oxygen loss (UNEP 1986). Canals and rivers are also used as an outlet for trash, particularly at locations where waste collection is lacking. Stagnating water transmits bilharziose or triggers malaria. Additionally, where potable water is rare, poor people often use the polluted rivers to wash their clothes, contributing to the pollution, and exposing themselves to multiple pollution hazards. Lastly, polluted water bodies represent a source for the breeding of disease vectors (WHO 2001).

In addition to distance to rivulets, which was classified into four levels from 0 to 400 meters, slope was integrated as a vulnerability factor. Low slopes were deemed more favorable to water pooling, mosquitoes breeding, and a higher concentration of pollutants, therefore correspond to a higher risk. That is, for the same threshold distance to the channel (e.g. 100 meters) people living in areas located in lower slope were more at risk than those living on steeper slopes. These two factors were summed with a greater weight assigned to distance to channels. The result was reclassified with the geometric interval technique as it appears on Table 6.

Distance(0.7) Slope (0.3)		>200	101-200		51-100		0-50
		1	2		3		4
> 25%	1	1.0	1.7		2.4		3.1
15 - 25%	2	1.3	2.0		2.7		3.4
5 - 15%	3	1.6	2.3		3.0		3.7
0 - 5%	4	1.9	2.6		3.3		4.0
Risk Levels
1 - Low		2 - Moderate		3 - High		4 - Very High
1 - 2.4		2.4 - 3.05		3.05 - 3.35		3.35 - 4.0

3.4.2.7 Proximity to the sea

The area immediately contiguous to the seashore is the last repository for all type of waste, mud, household and human refuse, and industrial waste. In developing countries, people in poor households living in proximity to the sea and who do not have a toiletor latrine don't have any other alternative than to unload their defecation eitherin the sea shore orvacant plots and open drains (Olanrewaju 1990). In addition, discarded cans, bottles, waste incoming from all activities are sinks and places for mosquitoes breeding, which spread diseases such as malaria and dengue (SKAT 2002). In Haiti, as reported by UNEP (1996), several shantytowns have been built in proximity to the seashore in recent years. Some of them are erected on dumping sites which often block water flowing to the sea, and generating ponds. The wind blowing from the sea may dissipate the effects of pollutants or it may spread them across a larger area. These locations have a high dose of pollutants and expose their residents to very harmful effects. A 2001 Panamerican Health Organization (PAHO 2001)) report on health in Haiti stated that the transmission of malaria occurs mainly in coastal areas coincident to altitude lower than 300 meters.

A Euclidean distance surface was calculated with a threshold of 1000 meters from the sea divided into four vulnerability levels as shown in Table 7. The impact may not be linear and may vary in function of some factors such as the disposition of the slums, local variation in elevation, canalization or conditions of the canalization (which is a temporal variable), amount and types of debris (vehicle carcasses for instance may block the movement of waste towards the sea), etc. For simplicity, we assumed that the phenomenon is linear in its materialization.

3.4.2.8 Proximity to high voltage power line

Power lines are considered a noticeable source of electromagnetic pollution to which extended exposure to some specific frequencies may result in cancer, birth defects, decreased immunity to disease, even new sicknesses (Michrowski 1991). A study conducted by Fews et al. (1999) observed an increased exposure to pollutant aerosols under high voltage power line. But another study conducted by Draper et al. (2005) and leading to the conclusion of association between the proximity of birth residence to high voltage power lines and children with leukemia (e.g.) has been the occasion of much debate (Day et al. 2005).

The delineation of this hazard was built on a uniform buffer of 500 meters from the power cables (line geometry) and 1000 meters around the distribution centers (point geometry). The distance ranges considered are shown in the Table 8.

Both distances were combined using the max function within the single output Map Algebra tool. In areas where the two grids intersect, the max function assigns the greatest value to the output grid. The use of this tool was necessary to put in evidence extreme cases of risk.

3.4.3 Linear Combination of the Variables

The Expert Opinion survey was conducted with the participation of ten local professionals having a certain familiarity with health and environmental issues in the study area. The respondents comprised 5 professionals in public health, 2 in environment, 2 economists and one agronomist. They were requested to assign a score from 1 to 9 (with 1 being very low and 9 very high) in order to rank the relative importance of the environmental variables of the model (see results in Table 15 of Appendix A). The ten scores were summed for each factor and divided by the total scores for all the factors. This provided a normalized weight comprised between 0 and 1. Hereafter this approach will be referred to Expert Opinion Weighting (EOW).

The second weighting approach assumed equal influence of the factors on health risks and assigned the same weight of 1/9 to each. The last weighting scheme was based on our own perception of the spatial extent and intensity of the different hazards of the model. Henceforth, either of these designations will be used for this approach: Own weighting or personalized weighting. Finally, to put in evidence those areas exposed uniquely to high risks for any factor, the maximum combination was applied. The max operator is a local function within the Spatial Analyst tools, which uses several input rasters to calculate the highest value on a cell-by-cell basis within the Analysis window.

Further the linear regression technique was used to validate the rationale under the Expert Opinion weighting and the personalized (own) weighting scheme. The intent was to discover the degree of correlation existing between the assigned coefficient in these two schemes and the factors in terms of area covered and the proportion of high and very high risks.

3.4.4 Classification Schemes

The final classification was implemented with four classification techniques namely, Quantile, Natural Breaks (or Jenks), Equal Interval, and Geometric interval. To compare the linear combination methods, the results provided by these classification algorithms were averaged and compared to another manual classification (we call it standardized classification). This process was utilized to isolate the effect of a given classification scheme. The manual classification regrouped the features into four discrete classes 0 -1, 1 - 2, 2 - 3, and 3 - 4 representing the four risk levels.

3.5 Model's Summary

Here are the four combination schemes employed to aggregate the environmental health risks factors considered in the model.

EHR = 0.13(Traffic Pollution) + 0.14Waste + 0.08(cemetery + hospitals + High Voltage Power) + 0.13Markets + 0.13(Housing Density) + 0.12(Polluted water) + 0.11(Distance to the sea)

EHR = 0.14(Traffic Pollution) + 0.16Waste + 0.06(Distance to Cemetery + Distance to the Hospitals) + 0.10*(Distance to market places) + 0.14(Housing Density) + 0.10(Pollution from Streams) + 0.16(Distance to the sea) + 0.08(High Voltage Power)

EHR = 0.1111*(Traffic Pollution + Waste + Distance to Cemetery + Distance to the Hospitals + Distance to markets places + Housing Density +.Pollution from Streams + Distance to the sea + High Voltage Power)

EHR = Max(Traffic Pollution, Waste, Distance to Cemetery, Distance to the Hospitals, Distance to markets places, Housing Density, Pollution from Streams, Distance to the sea, High Voltage Power).

The spatial model presented in this study is defined by a series of input factors, parameters, variables, weights, and combination techniques subject to many sources of uncertainty. These sources comprise lack of data, insufficient support of literature review for the study area, and error of measurement, and weight attribution, which impose limit on our confidence in the output model. The range of alternatives available in implementing the overlay provides the flexibility to come to any desired conclusion (O'Sullivan and Unwin 2003). Therefore, it is important to evaluate our confidence in the model and assess the uncertainties associated with the modeling process and with the output. A sensitivity analysis evaluates the effects of modifying one or several input parameters on the model output (Akçakaya 1996, Franklin et al. 2002) and examines the overall variability in the possible output (O'Sullivan and Unwin 2003). The weight of factors believed to be of greater magnitude in the model such as traffic and waste pollution was modified while adjusting the coefficient of the other variables to keep the total sum of the weights to 1. Three different scenarios were assessed. First, using the weights obtained from the expert opinion survey as basis, the coefficient of traffic pollution was increased from 0.13 to 0.3. Second, the weight of waste pollution was increased from 0.14 to 0.3. The last approach was to allocate weights in proportion to the area extent of each factor. We observed the change in proportion occurred between the four risk levels and compared the arithmetic sum of changes in the three cases to determine the sensitivity of the model.

4 RESULTS AND DISCUSSION

The results of this modeling exercise are divided into three sections: the first section presents the summary results of the model as a whole along with the different aggregation and classification schemes; then statistics are shown for the multiple factors that make up the model; finally, the effect of changing one assumption about the model is analyzed.

4.1 Results by Linear Combination Schemes

Participants in the Expert Opinion survey ranked waste as the major source of pollution (0.14) above traffic, markets and housing density, which equally came second with a weight of 13%. A lesser weight was allocated to cemetery, hospitals, and high voltage, which accounted for 8%. Indeed, the largest gap between the factors' weight was at most 6%.

4.1.1 Expert Opinion Survey, Equal Influence (Equal Weight), and Personalized Weightings

Irrespective of the combination or the classification technique used, about 41% of the study area was found exposed to high and very high environmental health risks, all factors considered. This result was brought out by simultaneously averaging the Expert Opinion survey, the equal influence, and the personalized combination schemes on one hand, the quintile, the natural breaks, the equal interval, and the geometric interval techniques on the other hand, so that this number is freed of the influence of any single combination method or classification technique. The averaged results are displayed in the table below. A more complete set of results is provided in Table 16 in Appendix A.

Table 9: Percent of areas per vulnerability level - Average score for the four classification techniques

Combination scheme	Very High	High	Moderate	Low	Total
Expert Opinion	15.6	25.4	35.9	23.1	100.0
Equal Weight	14.4	21.6	38.3	25.7	100.0
Personalized assignment	14.9	25.9	36.5	22.7	100.0
Average	14.9	25.9	36.5	22.7	100.0

The standardized classification scheme, which terms as a control (Table 10), led to a far different result with no areas exposed at very high risk and at most 15.9% of areas with high risks. In both cases the largest percentage was found in the category moderate risks, which amounts up to 59.3% in the standardized classification versus 36.5% in the first case. This perceptible discrepancy brings forward how the analysis built on one or another technique may lead to erroneous conclusions.

Table 10: Percent of areas by risk level and aggregation scheme using a standardized classification

Combination scheme	Very High (3-4)	High (2-3)	Moderate (1-2)	Low (0-1)	Total
Expert Opinion	0.0	16.6	60.2	23.2	100.0
Equal Weight	0.0	11.7	59.7	28.6	100.0
Personalized assignment	0.0	19.5	58.0	22.5	100.0
Average	0.0	15.9	59.3	24.8	100.0

Comparing the three methods used to combine the multiple factors, EOW and the personalized approach provided very similar results displaying a maximum difference of 0.7% for areas at very high risks (Figure 4). Whereas the gap between the EOW and the personalized weighting was minimal, a greater percentage (maximum 4.3%), discriminated them from the equal influence technique. The equal weighting technique was prone to assign more cells in low and moderate levels and consequently promoted the influence of low values over high values than did the two other techniques. Despite this difference the general trend showed a larger percentage of areas at moderate and low risk irrespective of the classification technique utilized, as it appears on the figure below. What may explain this?

Most of the parameters did not have a spatial domain covering the entire area. The average area extent for all the parameters was 52%, with only housing density having 100% assignment. For a given factor, areas with no risk participating in the combination reflect on the result. Large scores (values greater than 3) can be obtained only if the largest possible number of cells participating in the combination contributes with a maximum score. In other words, the various factors considered are dispersed over the area of study with few cases of high value overlap. Since the equal weighting method does not discriminate the variables, the final grid calculation was heavily affected by these zeros. This fact is evident in the standardized classification technique which established the classes based on discrete values at fixed interval. The category 4 (values ranging from 3 to 4) was virtually absent in all the combination schemes.

The map displayed on Figure 5 results from the EOW combination using the geometric interval reclassification technique. It indicates a concentration of high risk areas in the downtown area and heading toward the coast. Conversely, areas at moderate and low risks are located away from the downtown toward the periphery mostly made of high slopes. This observation confirms the assertion that high population density associated to untidy urbanization in a context of weak institutional framework is mainly responsible for the daily occurrence of life-threatening hazards.

Figure 5: Environmental Health Risks in Port-au-Prince - EOW classified with the geometric interval technique

4.1.2 The Maximum Weighting Scheme

The maximum weighting scheme so far was not taken into consideration in the previous discussions since it has a different connotation and could not bear a reasonable comparison with the other approaches. This function tends to maximize risk. Because of this a high percentage, 92.5%, of the area was found exposed at high and very high risks for all the hazards combined, which is a huge contrast to the results above. In this approach, since no weight was allocated, the cell with the biggest value rules over the others. Simply put, the result signifies that for this specific cell, there is at least one health hazard whose risk level corresponds to this maximum value. This method was chosen intentionally so as to bring forth the diversity of high risks prevailing in almost the entire area (Figure 6). This situation also reveals the lessening of individual factors' significance in aid of the collective importance of the model. Policies interested rather in solving environmental problems that arise from a specific field may find this approach more attractive and more practical.

Figure 6: Environmental Health Risks in Port-au-Prince - Maximum combination technique using the Geometric Interval classification method

This approach also outlines the specific location and extent of each hazard considered. For instance, someone accustomed to the area can clearly notice the path of high voltage power lines and the prism of risks in its neighborhood. Likewise, pollution associated to the proximity of the sea and of vehicle traffic is obvious.

The opposite procedure consisted of determining whether some areas were exposed simultaneously at high risk for all the hazards overlaid. A conditional statement was performed to superimpose all the areas where the conditions «high risk and very high risk» were met. This situation was not present in any area. This might be because of the spatial dispersion of the factors over the entire area.

4.2 Comparison of the Classification Techniques

This section presents a comparison of the performance of the classification methods, namely quantile, natural breaks (Jenks), equal interval, and geometric interval, in distributing values among risk levels. Once again to attenuate the effect of the combination schemes, the scores of each classification were averaged for the 3 approaches. A summary of the results is displayed in Appendix A, Table 17, while Figure 7 facilitates the comparison of the distribution of the area among the four risk stages.

Figure 7: Environmental Health Risks in Port-au-Prince - Percent of area at-risk by classification technique

Two different patterns are exhibited regarding the distribution of risk levels. First, whereas the maximum gap among risk levels is very important for Natural Breaks and Equal Interval (21 and 41%), the Quantile and the Geometric Interval techniques show a maximum difference of 16%. The second pattern relates to the cluster of values in opposite categories (low and high). In this regard, though all the classification techniques have a higher proportion of values classified as low and moderate risk, equal interval and geometric interval place a smaller proportion of areas in high and very high risk (36%) than do Quantile and Natural Breaks (43% and 41%). The former techniques appear to be pessimistic while the latter is rather optimistic. The intent of this observation is not to generalize the comparison. This result might depend on the specific distribution of values and the way these values were classified beforehand. A visual comparison is offered in Figure 8 and Figure 9.

For this specific study, if the goal was to highlight the severity of environmental health risks, the Quantile or the Natural breaks methods would be appropriate. Conversely, Equal interval or Geometric interval would be suitable to provide a certain sense of environmental stability and improvement.

These discussions and the analysis of the combination schemes made above illustrate the influence the choice of a specific model or technique can have on the final results. These results can be manipulated in a certain way to respond to decision makers' willingness to impress either about assumed progress accomplished by an implemented project or to attract funding for a prospective project by drawing a somber cliché of the state of the environment. As suggested by Osaragi (2002), the selection of a classification method depends on the nature of data and what we want to communicate about data.

4.3 Neighborhoods Exposed at High Risks

Another important step was to focus on areas identified as highly prone to environmental health hazards and assess their most salient characteristics. While we can't display here all the combinations for the different schemes, only the results obtained with the personalized weighting scheme coupled with the natural break (chosen randomly) are displayed in Figure 10.

Figure 10: Environmental Health Risks in Port-au-Prince: Percent of areas at-risk using the Own weighting scheme and the natural breaks classification method

Twenty two percent (22%) of areas was found overlapped for high and very high risk from one aggregation method to the other. That is, these areas always presented high risks irrespective of the approach used to combine the nine factors. This forms a space of certitude about the results. Most of these areas are located in the downtown neighborhood and include all the shanty towns. In particular these areas are well known for high traffic volume, high population density (so housing density), are located in neighborhoods of public facilities such as markets, hospitals, cemetery, in low elevation and near water bodies and the sea coast. These neighborhoods deserve the greatest attention from the government and/or the non-governmental agencies for interventions aimed at preventing or lessening impacts on health resulting from these identified hazards. The complete list of figures is provided in Appendix B, Figures 21-37. Through these maps the same geographic pattern can be seen for the different techniques considered. The differentiation, mainly quantitative, is not substantial.

4.4 Environmental Health Hazards

Table 18 in Appendix A and Figure 11 present a synopsis of the results by health hazard included in the model. As can be seen, waste pollution portrays the factor whose severity reaches the greatest physical extent, 53.4% of high and very high risks. Following this were pollution from streams (49.1%), high density housing (45.7%), and traffic (42.1%) (Table 18, Appendix A). The ranking of six factors coincides both for the percent of area covered and the proportion of area at high and very high risks. This would indicate a strong association between the physical extent of the factors and their perceived strength. Indeed the correlation coefficient between these two outputs exceeds 0.98. Correlation between EOW and Own weights over these two results was also tested. This yielded to a coefficient of regression (R²) of 41% (Table 20, Appendix A) and 33% (Table 21, Appendix A) when running successively single regressions of the EOW (column 2, Table 19) and the own weighting (column 3, Table 19) over the proportion of high/very high risk areas (column 6, Table 18). For both regressions the t-statistic was insignificant at 95% confidence limits and the p values significantly greater than zero. This result would suggest that the weight assigned by the experts had little to do with the perceived strength of these hazards. Similarly, when we ran a regression with the percent of study area covered by each factor (column 4, Table 19) as independent variable, correlations of 46% and 35% were found respectively for EOW and Own weighting with the t-statistic significant and p values acceptably low (close to zero) (see regression results in Tables 22 & 23, Appendix A). This would signify that the experts' judgment and our perception were more correlated to the physical extent of health hazards, which can be more easily quantified, than the perceived influence of those hazards. Although the participants in the expert opinion survey were not cautioned to provide their rankings based on the physical occurrence of these hazards, this result implies that this characteristic had to a certain extent influenced the weight the respondents attributed. Consequently, the experts did a better job than us in ranking the factors.

If we were to average the results obtained by individual factor, 52% of the study area would be at risk at all levels, of which 47% at high and very high risk. However, what these percentages did not unveil is the population exposed, though housing density is an unambiguous surrogate of population. Applying the 47% rate on the total population would be misleading since the areas at high risk are also the most populated. Not only are locations with high and very high housing density inherently exposed to high risks, the superimposition of low and moderate housing density areas to any other hazards where the vulnerability is high induces a high risk for the populations living in these areas. This was one of the issues the maximum approach attempted to address. Endless combinations could be done to put in evidence the severity of health risks in the study area. But for now, let's look at individual hazards and their associated risk levels.

4.4.1 Traffic Pollution

Vehicle traffic forms a major source of indoor and outdoor air pollution in the study area. It is due to high population density that exposes both, to an increasing number of public and private vehicles using road surfaces not designed for this volume of traffic, and to the disorganized development of informal commercial activities which take over the streets. In addition to these environmental and physical settings, conditions of maintenance, regulation, and controls of the vehicle fleet exacerbates pollution. Nevertheless there is no obvious sign of concern either from the exposed population or from the entities in charge of the environment and traffic regulation and management, or again public and private organizations promoting health.

As displayed on Table 18 in Appendix A, about 76% percent of the area was found exposed to low to very high pollution risks. Neighborhoods with high and very high risks accounts for 42% of the study area while low and moderate risk amount to 13% and 21% respectively. The north-northeast section of the map displayed in Figure 12 makes abstraction of high traffic activities in some sections of Delmas Road adjacent to the border of Port-au-Prince. Only when Delmas Road intersects or crosses the border of the study area risks originated from this road appeared in the analysis. Nevertheless, most of this boundary section was found prone to high risk from traffic pollution.

Waste Pollution

We must concede that the modeling of zones susceptible to waste pollution hazard was arduous. Though the final approach adopted was not completely satisfactory and necessitates scrutiny, the assumptions made and the mechanisms developed were likely to provide results that are coincident to the reality of waste, which is wide spread over the study area. This factor covers more than 95% of the area and received the greatest weight in the EOW survey. About half of the study area (53%) was found exposed to at least high risks, which pinpoints to the severity of this issue and the challenge it poses to decision makers particularly in a context of very dynamic demographic variables, and in a setting of weak institutional capacity of management.

Figure 13 shows that neighborhoods away from the collection network but having high density housing, and those very close to the network but again with high density housing and located in lowlands were found to exhibit the highest risk to waste pollution. Conversely people living in high elevation and where the density of houses is relatively low enjoy a better environment.

4.4.3 Housing Density

Housing density was deemed one of the most important factors, which not only embodies a source of pollution and serves as sink for other sources, but also underlines vulnerability to all the hazards considered. Though concessions were made to classify some areas at low density where they would be considered medium-to-high density by international standard classification, about 46% of the areas were estimated to be high-to-extreme density (Figure 14). However these densities are unique and incomparable to those found in developed countries' urban neighborhoods. To illustrate this sharp contrast, an average of 10,583 homes by square kilometer was counted in Port-au-Prince by IHSI in the last national census (IHSI 2003). Even more striking was the maximum density of more than 164,500 dwellings found in one district. As a result the ventilation is very poor inside and outside these structures; the distance between the toilets and the living spaces is negligible, a situation that constantly exposes people to nearly physical contact with feces and its manifestation; cooking with charcoal is done inside or adjacent to the house; the effortless transmission of bacteria through insects, rodents, and cockroaches and many other impacts for the physical and the mental health.

4.4.4 Pollution from Market Places

A common phenomenon that escapes authorities' control in the study area is the extension of public markets beyond their original assigned location. This expansion in Port-au-Prince resembles water run-off quickly turning to overflows due to excess above the retention capacity. It has become impossible to delimit the physical extents of the market places. In addition to the official markets, there are the informal markets that have been accepted as a fact. This study covered only those recognized by the city's municipality. From low to very high, the hazards they represented formed more than 22% of the area, of which about 36% (8% in total) where at high to very high risk (Figure 15). The buffer built around the markets assumes that the polluted air shed spreads homogeneously in all direction without consideration of wind direction, transmission vectors traveling patterns, and obstacles in the vicinity.

4.4.5 Pollution water bodies

Though the study area is not crossed by any major stream, the topographic map exhibits multiple rivulets including some with uncertain physical destination. They might have been relayed by mechanical pathways built purposely in an urban drainage plan, or obstructed by unexpected obstacles that make it difficult to determine their conduit. Regardless of the situation at hand, it represents a serious source of pollution representing the receptacle of all kinds of municipal waste, and as such poses serious concerns for health.

As reflected in the model, the total extent of pollution from water was about 88% of which 56% was at high and very high risk, 18% moderate, and 26% at low risk (Figure 16). The modeling of these areas integrated slope to account for the accumulation factor across the path down to the sea and also the likelihood of water pooling at low slopes atop of high slope. As can be seen on the map less areas in high elevation were exposed to high risk. However, neighborhoods in low elevations and even not near channels experience high risks.

A large part of the downtown area was not part of this assessment as shown on the light blue section in the northwest. This was because ponds and water pooling in canals could not be precisely digitized. However this area is known for poor drainage and channels obstruction by waste, which allow very unclean water to flow on the streets.

Once again, rivulets that flow near the border of the study area but that did not cross or intersected the area, were not included in the determination of this hazard, though they might have had a partial impact.

4.4.6 Pollution from the coast

Pollution incoming from the sea encompasses about 23% of the study area, among which more than half is at high and very high risk (Figure 17). This area is the densest in term of habitat and traffic, and hosts most of the slums, commercial activities, some formal and informal market places. In addition, it comprises the path of one branch of high voltage power line as well as a power distribution center, and the cemetery and some of the hospitals are also located close to this area. Consequently it is deemed as one of the most unsafe neighborhood to live. Nevertheless, as noticed above, this is the area with the densest population. Any sanitation, urban conversion program, dislocation and relocation program, health intervention, should initiate in this area located in the storm's eye in term of environmental health risks.

4.4.7 Pollution from high voltage electric power

This hazard was mainly identified along transects in the south and the northwest section and extended over more than half of the study area. Neighborhoods under high and very high risks represented over 30% of the total area. However, though the high and very high risk of the south section of this hazard is very obvious on Figure 18, it did not appear on the result of EOW combination. The EOW combination weakens its impact by the fact that part of this area is isolated in high elevation and low housing density and poorly influenced the combination. This underlines the importance of approaching the analysis of the factors both globally and individually.

4.4.8 Pollution from the hospitals

Buffers of different sizes were applied in relation to the size and the level of care used at the hospitals. This information was gathered from health professionals accustomed to these facilities. The hazards they represent account for more than 7% of the study area, of which only 30% was considered at high to very high risk (Figure 19). This low percentage was expected because of the small buffer applied around three out of 5 hospitals. Nonetheless the high rate of low and moderate risk levels does not mean absence of risk, yet a direct measurement of pollution concentration within this delineation is required.

4.4.9 Pollution from the cemetery

This hazard represents the smallest in terms of area extent: 4% of the entire area. High and very high risks were identified in less than 2% of the study area (Figure 20). However, as mentioned earlier, it is believed that a reasonable extension of the buffer westward would result in a greater physical incidence. We assumed that this part was partially covered in the pollution from water bodies, traffic and waste pollution. An inspection of these maps at the west side of the cemetery confirms this hypothesis.

4.5 Sensitivity analysis

Though the comparison of the different aggregation schemes constitutes a sort of analysis of sensitivity, this section introduces a more in-depth consideration of this issue. To keep the comparison of the results as simple as possible, reference was made only to the Expert Opinion weighting using the geometric interval classification. The maps are displayed in Appendix B, Figures 38 to 40.

Traffic Pollution Influence

The assumption of an increase of traffic pollution's weight to 30% results in decreasing the coefficient of all the other parameters of the model while keeping the total weight to 1. Table 11 shows a substantial alteration of the percent area in each risk level. The most significant change was observed in the lower risk level which declined by 67%. At the opposite, high and very high risk levels increased by 50% and 77% respectively. The overall change is almost 60%. This reveals the strong influence traffic pollution has on health risk in the study area. To this regard, traffic pollution represents one of the most urgent issues that policies should address. Let's recall that traffic pollution occurs in about 76% of the study area.

	EOW	Increase in Traffic weight	Difference (%)	Absolute Difference
Low	36.0	15.8	-56.1	-20.2
Moderate	27.1	24.0	-11.5	-3.1
High	19.3	29.0	50.2	9.7
Very High	17.7	31.3	76.8	13.6
Sum of changes			59.4

Waste Pollution Influence

The geographical incidence of waste represents 95% of the area. With the alteration of waste coefficient, the model records a decline in low classes of about 54% while high and very high risk levels indicate a positive joint change of 100%. The aggregate change is 45%, 15 points smaller than changes induced by traffic pollution, though the physical occurrence of waste is greater than traffic pollution (Table 12). Nevertheless, this influence on the model is still substantial.

	EOW	Increase in waste weight	Difference (%)	Absolute Difference
Low	36.0	19.1	-46.9	-16.9
Moderate	27.1	25.2	-7.0	-1.9
High	19.3	32.9	70.4	13.6
Very High	17.7	22.8	28.8	5.1
Sum of changes			45.3

4.5.3 Proportional Spatial Incidence of the factors

This weighting, brought out based on the relative physical extent of each model's parameters, assigned a greater coefficient to housing density (0.21), followed by waste (0.20), rivulets (0.19), and traffic pollution (0.16). Unlike the previous two scenarios, the area incidence of high and very high risk levels increased only by 60% and the aggregate change is less than 30% (Table 13). This weighting shrank the gap between risk levels and enables a better balanced distribution than did EOW.

	EOW	Proportional Weighting	Difference (%)	Absolute Difference
Low	36.0	23.8	-33.9	-12.2
Moderate	27.1	27.0	-0.3	-0.1
High	19.3	31.3	62.2	12
Very High	17.7	17.9	1.1	0.2
Sum of changes			29.1

To summarize, the alteration of some of the coefficients of the model's parameters influenced the results in different ways. The increase of the traffic pollution weighting had the greatest impact on the model, altering drastically the categories' rank from top to bottom. This exercise definitely illustrates the responsiveness of the model to changes of its parameters.

5 CONCLUSIONS

This study took place in a context of unavailability and/or inaccessibility of data. Even the basic spatial information could not be obtained from official sources on the Internet. The web pages of the institutions in charge of the production and distribution of spatial data, or information in general were projected to be constructed soon or were under perpetual construction. It took a great amount of effort to generate most of the features through a long and arduous process of digitization of most of the data. Often assumptions had to be made to avoid having to collect more data. The difficulties encountered reveal the requirements for a greater flexibility of data exchange conditions between institutions, and between institution and the public, as well as the needs for a better integration of spatial data available. Furthermore, it is a pressing demand to produce geographic information at higher resolution in a way to enable researchers to address issues at the level of communities. Very general spatial information can not lead to anything but this congenital practice of confounding semantics and pragmatism. Information, particularly in such disinherited and poor environment as Haiti, may not be used to confer power to individuals or institutions. Rather it may constitute a power to promote changes starting at individual and community levels. We advocate cost efficient strategies to produce spatial data and make them available to anyone anywhere through the Internet, even for purchasing.

This study, a first of its kind for Port-au-Prince, attempted to delineate at-risk neighborhoods for nine environmental health hazards. Though traditionally the focus has been oriented toward natural hazards, which in nature are more discreet and whose manifestation leads to direct observation and quantification, this study reveals that health hazards are not a mental invention. These hazards are continuous, constant, perilous, and daily affect a large part of the population. The lack of information and the complexity posed by quantitative assessment must not lead to the minimization or the abstraction of such a precarious and deadly phenomenon. This work may represent a first geographic inventory of environmental health-related hazards in Port-au-Prince. By relating its occurrence to the underlying causes, it offers the ability to intervene in specific areas and targeting a specific factor. To this respect, we strongly encourage the use of the SDE as a suitable spatial study unit for the collection of data pertaining to a large array of fields. This approach can strongly facilitate the insight of phenomena at small community level and promote participation.

High housing density, traffic, waste, streams, and the sea coast represent the main sources or pollution in Port-au-Prince. However the impact of the other sources must not be overlooked. They are very serious in the specific environment and institutional setting of Port-au-Prince.

While the geographic delineation of the hazards may not be exact in its spatial shape and extent, one confidence remains: the issues addressed exist and exist in the neighborhoods indicated. Without any field data the categorization brought out by the ordinal approach remains subjective and depends on the researcher's perception of the reality under study. Nevertheless it has the merits of pinpointing areas where environmental health risks might reasonably be incident and offers the ability to make different assumptions and adjusting the results on the fly. Various combination and reclassification techniques likely to influence the output were assessed and demonstrated the high sensitivity of the model to change in its parameters.

As a specific outcome of this study, the prime interest of any program aiming to reduce population's vulnerability to health hazards should focus on these areas identified as at very high risk. They are located particularly in neighborhoods with very high housing density, which also host an array of other environmental hazards such as air pollution from traffic, waste, water bodies and the coastline. By relating these risked areas specific hazards that threaten them the strategy to adopt becomes less ambiguous.

This study is not complete without a ground truth survey consisting of collecting information and taking specific measurements on various pollution factors included in the model. Very limited in time, budget and other resources, the scope of this study would not allow such investigation. Information gathered on the field not only would allow validating the results but equally would enable establishing strong rationale for hazards delineation and vulnerability scaling. At least the outcome is deemed a first but important step by defining specific area where distinct interventions can be done.

APPENDIX A - TABLES

Roads		Land Use		Traffic Density		Elevation (meters)
Distance	Value	Types	Value	Threshold	Value	Threshold	Value
0 - 50 meters	4	Dense Urban	4	Extreme	4	0-100	4
51 - 100 meters	3	Low Residential and Business areas	3	High	3	101-200	3
101 - 200	2	Rocks & Bare soil	2	Moderate	2	201-400	2
> 200 meters	1	Vegetation and other uses	1	Low	1	>400	1

Factors	Rank 1-9 (1=Low; 9=High)										Total	Average (row)	Average (Total)
Grader	1	2	3	4	5	6	7	8	9	10	Total	Average (row)	Average (Total)
Air pollution from traffic	4	9	7	8	7	5	4	7	8	8	67	0.67	0.13
Garbage dumped in the streets	6		8	9	8	8	8	9	9	8	73	0.81	0.14
Cemetery (Main cemetery)	4	8	3	6	4	5	3		5	7	45	0.50	0.08
Hospitals (particularly the main hospitals)		9	3	4	4		6	3	5	7	41	0.51	0.08
Public market places (formal and informal)	3	9	8	8	5	8	8	8	7	8	72	0.72	0.13
Housing density	5	9	8	9	7	7	8	9	3	6	71	0.71	0.13
Polluted Water from rivers (not for drinking	4	7	8	8	5	8	8	7	2	8	65	0.65	0.12
Proximity to the sea (think about slums)	3	5	4	8	8	6	6	5	8	7	60	0.60	0.11
High voltage power lines	6	3	2	7	3	3	3	4	4	7	42	0.42	0.08

Classification Combination	% of areas per risk levels and classification scheme
	Quintile				Natural Breaks (Jenks)				Equal Interval				Geometric Interval
	4¹	3	2	1	4	3	2	1	4	3	2	1	4	3	2	1
Expert Opinion	18.3	18.6	27.8	35.3	14.4	23.7	32.9	29	9	28	44.5	18.5	17.7	19.3	27	36
Equal Influence	18.8	22.7	28.2	30.3	15.4	17.5	36.7	30.3	5.9	21	50.8	22.3	15.4	11.5	26.1	46.9
Maximum assignment	0	85.4	10.6	4.0	0	85.4	10.6	4.0	60.5	24.9	10.6	4.0	60.5	24.9	10.6	4.0
Personalized Assignment	18.7	20.1	28.9	32.3	13.8	25.9	36.1	24.2	9.0	28.2	43.8	19.0	16	21.2	27.4	35.4
Average (1, 2, 4)	18.6	20.5	28.3	32.6	14.5	22.4	35.2	27.8	8.0	25.7	46.4	19.9	16.4	17.3	26.8	39.4

Quantile

Natural Breaks

Equal Interval

Geometric

Interval

4¹

Average (%)

19.4

23.8

30.8

26.1

15.2

25.8

37.1

21.9

8.3

28.2

49.2

14.4

17.0

19.4

30.6

33.1³

Factors	Very High	High	Moderate	Low	Area Coverage (%)	High and Very high risks combined (% of total area)
Cemetery	22.6	19.8	25.8	31.7	4.0	1.7
Hospitals	18.0	12.0	31.0	39.0	7.3	2.2
Sea Pollution	33.1	19.2	18.9	28.8	22.7	11.9
Traffic Pollution	33.6	22.1	27.6	16.7	75.7	42.1
Streams Pollution	28.9	27.2	17.8	26.1	87.6	49.1
Markets Pollution	14.0	21.6	29.2	35.2	22.4	8.0
Housing Density	12.1	33.6	24.6	29.7	100	45.7
Electric Power	30.7	25.5	19.0	24.8	54.4	30.6
Waste Pollution	27.8	25.1	24.8	22.3	95.1	53.4
Average	24.5	22.9	25.6	30.1	52.13	27.2

Factors (1)	EOW (2)	Own Weight (3)	(High & Very High Risks) % of study area (4)	Ranking of Area coverage (5)	High and Very high risks (% of total area) (6)	Ranking of High/Very High risks ranking (7)
Traffic Pollution	0.13	0.14	42.1	4	42.1	4
Waste Pollution	0.14	0.16	53.4	2	53.4	1
Cemetery	0.08	0.06	1.7	9	1.7	9
Hospitals	0.08	0.06	2.2	8	2.2	8
Markets Pollution	0.13	0.10	8.0	7	8.0	7
Housing Density	0.13	0.14	45.7	1	45.7	3
Streams Pollution	0.12	0.10	49.1	3	49.1	2
Sea Pollution	0.11	0.16	11.9	6	11.9	6
Electric Power	0.08	0.08	27.2	5	30.6	5

Regression Statistics
Multiple R	0.64165
R Square	0.411715
Adjusted R Square	0.327674
Standard Error	19.44877
Observations	9
ANOVA
	df	SS	MS	F	Significance F
Regression	1	1853.065	1853.065	4.898988	0.062486
Residual	7	2647.783	378.2548
Total	8	4500.849

	Coefficients	Standard Error	t Stat	P-value	Lower 95%
Intercept	-38.9955	31.57875	-1.23486	0.25672	-113.667
% High/Very High Risks	615.6591	278.1552	2.213366	0.062486	-42.0734

Regression Statistics
Multiple R	0.57574
R Square	0.331477
Adjusted R Square	0.235974
Standard Error	0.035085
Observations	9

ANOVA
	df	SS	MS	F
Regression	1	0.004272	0.004272	3.470846
Residual	7	0.008617	0.001231
Total	8	0.012889

	Coefficients	Standard Error	t Stat	P-value
Intercept	0.081591	0.019694	4.142963	0.004333
Results	0.001084	0.000582	1.863021	0.104746

Regression Statistics
Multiple R	0.677797
R Square	0.459409
Adjusted R Square	0.382182
Standard Error	30.48402
Observations	9

ANOVA
	df	SS	MS	F
Regression	1	5528.073	5528.073	5.9488
Residual	7	6504.927	929.2753
Total	8	12033

	Coefficients	Standard Error	t Stat	P-value
Intercept	-66.0182	49.49654	-1.33379	0.224032
% Area covered	1063.364	435.9805	2.439016	0.044824

Regression Statistics
Multiple R	0.589602
R Square	0.347631
Adjusted R Square	0.254435
Standard Error	33.48765
Observations	9

ANOVA
	df	SS	MS	F
Regression	1	4183.041	4183.041	3.73012
Residual	7	7849.959	1121.423
Total	8	12033

	Coefficients	Standard Error	t Stat	P-value
Intercept	-11.1655	34.62315	-0.32249	0.756502
% Area covered	569.6897	294.9694	1.931352	0.094734

APPENDIX B - FIGURES

APPENDIX C - MODEL'S OUTLINE

APPENDIX D - MODEL'S EXECUTION SCRIPT

Executing (Topo to Raster): TopoToRaster "'M:\RISK Source Data\Vector\Study_area\Rivulets.shp' # Stream;'M:\RISK Source Data\Vector\Study_area\peaks.shp' Altitude PointElevation;'M:\RISK Source Data\Vector\Study_area\SA_contours.shp' Elevation Contour;'M:\RISK Source Data\Vector\Study_area\Sea_Coast_.shp' Elevation Contour" "M:\RISK Source Data\Process\dem2" 10 "777513.9969 2048923.9338 786313.9969 2054993.9338" 20 # # ENFORCE CONTOUR 40 # 1 0 2.5 100 "M:\RISK Source Data\Process\synth_streams.shp" "M:\RISK Source Data\Process\sinks.shp" # #

Executing (Buffer (5)): Buffer "M:\RISK Source Data\Vector\Study_area\Streets_roads.shp" "M:\RISK Source Data\Process\Streets_roads_Buf.shp" "10 Meters" FULL ROUND NONE #

Executing (Euclidean Distance): EucDistance "M:\RISK Source Data\Process\Streets_roads_Buf.shp" "M:\RISK Source Data\Process\Roads_eucdist" # 10 #

Executing (Euclidean Distance (2)): EucDistance "M:\RISK Source Data\Vector\Study_area\Waste_coll.shp" "M:\RISK Source Data\Process\EucDist_Waste" # 10 #

Executing (Euclidean Distance (4)): EucDistance "M:\RISK Source Data\Vector\Study_area\Sea_Coast_.shp" "M:\RISK Source Data\Process\EucDist_Sea_" # 10 #

Executing (Buffer (3)): Buffer "M:\RISK Source Data\Vector\Study_area\Hospitals.shp" "M:\RISK Source Data\Process\Hosp_Buf.shp" Buffer FULL ROUND NONE #

Executing (Euclidean Distance (5)): EucDistance "M:\RISK Source Data\Process\Hosp_Buf.shp" "M:\RISK Source Data\Process\EucDist_Hosp" # 10 #

Executing (Buffer (4)): Buffer "M:\RISK Source Data\Vector\Study_area\Cemetery.shp" "M:\RISK Source Data\Process\Cemetery_Buf.shp" "50 Meters" LEFT ROUND NONE #

Executing (Euclidean Distance (6)): EucDistance "M:\RISK Source Data\Process\Cemetery_Buf.shp" "M:\RISK Source Data\Process\EucDist_Cem" # 10 #

Executing (Merge): Merge 'M:\RISK Source Data\Vector\Study_area\Rivulets.shp';'M:\RISK Source Data\Vector\Study_area\SA_UTSIG_rivlines.shp' "M:\RISK Source Data\Process\streams.shp" "SHAPE_Leng SHAPE_Leng true false false 19 Double 0 0 ,First,#,M:\RISK Source Data\Vector\Study_area\Rivulets.shp,SHAPE_Leng,-1,-1,M:\RISK Source Data\Vector\Study_area\SA_UTSIG_rivlines.shp,Shape_Leng,-1,-1;Name Name true false false 50 Text 0 0 ,First,#,M:\RISK Source Data\Vector\Study_area\Rivulets.shp,Name,-1,-1,M:\RISK Source Data\Vector\Study_area\SA_UTSIG_rivlines.shp,Name,-1,-1"

Executing (Euclidean Distance (7)): EucDistance "M:\RISK Source Data\Process\streams.shp" "M:\RISK Source Data\Process\EucD_streams" # 10 #

Executing (Euclidean Distance (9)): EucDistance "M:\RISK Source Data\Vector\Study_area\SA_high voltage.shp" "M:\RISK Source Data\Process\Pow_Line_dist" # 10 #

Executing (Euclidean Distance (10)): EucDistance "M:\RISK Source Data\Vector\Study_area\High_volt_pts.shp" "M:\RISK Source Data\Process\Pow_cent_dist" # 10 #

Executing (Euclidean Distance (8)): EucDistance "M:\RISK Source Data\Vector\Study_area\Markets_poly.shp" "M:\RISK Source Data\Process\EDist_Markpol" # 10 #

Executing (Euclidean Distance (11)): EucDistance "M:\RISK Source Data\Vector\Study_area\Markets_pts.shp" "M:\RISK Source Data\Process\EDist_Markpts" # 10 #

Executing (Polygon to Raster): PolygonToRaster "M:\RISK Source Data\Vector\Study_area\pap_hab_dens_cod.shp" Dens_code "M:\RISK Source Data\Process\pap_hab_dens.img" CELL_CENTER NONE 10

Executing (Single Output Map Algebra (8)): SingleOutputMapAlgebra con(pap_hab_dens.img >= 5, 4, pap_hab_dens.img == 4, 3, pap_hab_dens.img == 3, 2, pap_hab_dens.img <= 2, 1, 0) "M:\RISK Source Data\Process\hab_dens_rec" 'M:\RISK Source Data\Process\pap_hab_dens.img'

Executing (Single Output Map Algebra (7)): SingleOutputMapAlgebra con(EucDist_Cem <= 50, 4, EucDist_Cem <= 150, 3, EucDist_Cem <= 250, 2, EucDist_Cem <= 350, 1, 0) "M:\RISK Source Data\Process\Cemetdist_rec" 'M:\RISK Source Data\Process\EucDist_Cem'

Executing (Single Output Map Algebra (6)): SingleOutputMapAlgebra "con(EucDist_Hosp <= 50, 4, EucDist_Hosp <= 100, 3, EucDist_Hosp <= 200, 2, EucDist_Hosp <= 300, 1, 0) " "M:\RISK Source Data\Process\HospDist_rec" 'M:\RISK Source Data\Process\EucDist_Hosp'

Executing (Single Output Map Algebra (24)): SingleOutputMapAlgebra con(EDist_Markpol <= 100, 4, EDist_Markpol <= 200, 3, EDist_Markpol <= 300, 2, EDist_Markpol <= 400, 1, 0) "M:\RISK Source Data\Process\mkpoldist_rec" 'M:\RISK Source Data\Process\EDist_Markpol'

Executing (Single Output Map Algebra (25)): SingleOutputMapAlgebra "con(EDist_Markpts <= 100, 4, EDist_Markpts <= 200, 3, EDist_Markpts <= 300, 2, EDist_Markpts <= 400, 1, 0) " "M:\RISK Source Data\Process\mkptsdist_rec" 'M:\RISK Source Data\Process\EDist_Markpts'

Executing (Single Output Map Algebra (5)): SingleOutputMapAlgebra max(mkpoldist_rec, mkptsdist_rec) "M:\RISK Source Data\Process\markets_pol" 'M:\RISK Source Data\Process\mkpoldist_rec';'M:\RISK Source Data\Process\mkptsdist_rec'

Executing (Single Output Map Algebra (12)): SingleOutputMapAlgebra con(Pow_cent_dist <= 300, 4, Pow_cent_dist <= 500, 3, Pow_cent_dist <= 700, 2, Pow_cent_dist <= 1000, 1, 0) "M:\RISK Source Data\Process\Pow_cent_rec" 'M:\RISK Source Data\Process\Pow_cent_dist'

Executing (Single Output Map Algebra (19)): SingleOutputMapAlgebra con(Pow_Line_dist <= 150, 4, Pow_Line_dist <= 300, 3, Pow_Line_dist <= 400, 2, Pow_Line_dist <= 500, 1, 0)

"M:\RISK Source Data\Process\Pow_Line_rec" 'M:\RISK Source Data\Process\Pow_Line_dist'

Executing (Single Output Map Algebra (20)): SingleOutputMapAlgebra max(Pow_cent_rec,Pow_Line_rec) "M:\RISK Source Data\Process\Power_factor" 'M:\RISK Source Data\Process\Pow_cent_rec';'M:\RISK Source Data\Process\Pow_Line_rec'

Executing (Single Output Map Algebra (21)): SingleOutputMapAlgebra con(EucDist_Sea_ <= 300, 4, EucDist_Sea_ <= 500, 3, EucDist_Sea_ <= 700, 2, EucDist_Sea_ <= 1000, 1, 0) "M:\RISK Source Data\Process\sea_dist_rec_" 'M:\RISK Source Data\Process\EucDist_Sea_'

Executing (Single Output Map Algebra (9)): SingleOutputMapAlgebra con(EucD_streams <= 100, 4, EucD_streams <= 200, 3, EucD_streams <= 300, 2, EucD_streams <= 400, 1) "M:\RISK Source Data\Process\Str_dist_rec" 'M:\RISK Source Data\Process\EucD_streams'

Executing (Fill): Fill "M:\RISK Source Data\Process\dem2" "M:\RISK Source Data\Process\dem_fill" #

Executing (Extract by Mask): ExtractByMask "M:\RISK Source Data\Process\dem_fill" "M:\RISK Source Data\Vector\Study_area\sa_mask" "M:\RISK Source Data\Process\pap_dem"

Executing (Slope): Slope "M:\RISK Source Data\Process\pap_dem" "M:\RISK Source Data\Process\PAP_Slope" PERCENT_RISE 1

Executing (Single Output Map Algebra (23)): SingleOutputMapAlgebra con(PAP_Slope <= 5, 4, PAP_Slope <= 15, 3, PAP_Slope <= 25, 2, PAP_Slope > 25, 1, 0) "M:\RISK Source Data\Process\pap_slop_rec" 'M:\RISK Source Data\Process\PAP_Slope'

Executing (Single Output Map Algebra (10)): SingleOutputMapAlgebra "0.7*(Streams distance reclassified) + 0.3*pap_slop_rec" "M:\RISK Source Data\Process\Streams_pol" 'M:\RISK Source Data\Process\Str_dist_rec';'M:\RISK Source Data\Process\pap_slop_rec'

Executing (Reclassify): Reclassify "M:\RISK Source Data\Process\Streams_pol" VALUE "1 2.4100469267072113 1;2.4100469267072113 3.0567306046542977 2;3.0567306046542977 3.3533163220529132 3;3.3533163220529132 4 4;NODATA 0" "M:\RISK Source Data\Process\str_pol_rec1" DATA

Executing (Feature to Raster (2)): FeatureToRaster "M:\RISK Source Data\Vector\Study_area\Land_use.shp" L_U_Code "M:\RISK Source Data\Process\land_use" 10

Executing (Single Output Map Algebra (18)): SingleOutputMapAlgebra con(Roads_eucdist <= 50, 4, Roads_eucdist <= 100, 3, Roads_eucdist <= 200, 2, Roads_eucdist <= 300, 1) "M:\RISK Source Data\Process\roadist_rec" 'M:\RISK Source Data\Process\Roads_eucdist'

Executing (IDW): Idw "M:\RISK Source Data\Vector\Study_area\pap_elevs.shp" PAP_DEM "M:\RISK Source Data\Process\Elev_4_40" 10 4 "VARIABLE 12 50" #

Executing (Reclassify (7)): Reclassify "M:\RISK Source Data\Process\Elev_4_40" Value "-2 100 4;100 300 3;300 500 2;500 600 1;NODATA 0" "M:\RISK Source Data\Process\elev_reclass" DATA

Executing (Polygon to Raster (2)): PolygonToRaster "M:\RISK Source Data\Vector\Study_area\Streets_roads_Buffer.shp" Traf_dens "M:\RISK Source Data\Process\traf_buf_den" CELL_CENTER NONE 10

Executing (Single Output Map Algebra): SingleOutputMapAlgebra "0.35*(1.00*roadist_rec) + 0.15*(1.00*Land_Use) + 0.15*(1.00*elev_reclass) + 0.35*(1.00*traf_buf_den)" "M:\RISK Source Data\Process\traf_pol_risk" 'M:\RISK Source Data\Process\land_use';'M:\RISK Source Data\Process\roadist_rec';'M:\RISK Source Data\Process\elev_reclass';'M:\RISK Source Data\Process\traf_buf_den'

Executing (Reclassify (2)): Reclassify "M:\RISK Source Data\Process\traf_pol_risk" VALUE "0.94999998807907104 2.1672228247818373 1;2.1672228247818373 2.8767925732977973 2;2.8767925732977973 3.2904302514840391 3;3.2904302514840391 4 4;NODATA 0" "M:\RISK Source Data\Process\traf_pol_fin1" DATA

Executing (Feature to Raster (5)): FeatureToRaster "M:\RISK Source Data\Vector\Study_area\Waste_coll_Buf.shp" Net_cond "M:\RISK Source Data\Process\Wastecond_ras" 10

Executing (Reclassify (21)): Reclassify "M:\RISK Source Data\Process\Wastecond_ras" VALUE "1 1;1 2 2;2 3 3;3 4 4;NODATA 0" "M:\RISK Source Data\Process\waste_condrec" DATA

Executing (Euclidean Distance (3)): EucDistance "M:\RISK Source Data\Vector\Study_area\Waste_coll.shp" "M:\RISK Source Data\Process\Waste_Edist" # 10 #

Executing (Single Output Map Algebra (11)): SingleOutputMapAlgebra con(Waste_Edist > 400, 1, 0) "M:\RISK Source Data\Process\WasteDist_400" 'M:\RISK Source Data\Process\Waste_Edist'

Executing (Single Output Map Algebra (14)): SingleOutputMapAlgebra con(WasteDist_400 == 1 & pap_hab_dens.img == 1, 1, WasteDist_400 == 1 & pap_hab_dens.img == 2, 2, WasteDist_400 == 1 & pap_hab_dens.img == 3, 3, WasteDist_400 == 1 & pap_hab_dens.img == 4, 4, 0) "M:\RISK Source Data\Process\WastD400_hden" 'M:\RISK Source Data\Process\WasteDist_400';'M:\RISK Source Data\Process\pap_hab_dens.img'

Executing (Single Output Map Algebra (2)): SingleOutputMapAlgebra con(EucDist_Waste <= 100, 4, EucDist_Waste <= 200, 3, EucDist_Waste <= 300, 2, EucDist_Waste <= 400, 1, 0) "M:\RISK Source Data\Process\Waste_dist" 'M:\RISK Source Data\Process\EucDist_Waste'

Executing (Single Output Map Algebra (22)): SingleOutputMapAlgebra "WastD400_hden + Waste_dist" "M:\RISK Source Data\Process\Wastdist_sum" 'M:\RISK Source Data\Process\WastD400_hden';'M:\RISK Source Data\Process\Waste_dist'

Executing (Single Output Map Algebra (13)): SingleOutputMapAlgebra con(Wastdist_sum > 0, Wastdist_sum) "M:\RISK Source Data\Process\Waste_sum_rec" 'M:\RISK Source Data\Process\Wastdist_sum'

Executing (Single Output Map Algebra (3)): SingleOutputMapAlgebra "0.5*(1.0*Waste_sum_rec) + 0.3*(1.0*waste_condrec) + 0.2*(1.0*elev_reclass)" "M:\RISK Source Data\Process\Waste_factor" 'M:\RISK Source Data\Process\elev_reclass';'M:\RISK Source Data\Process\waste_condrec';'M:\RISK Source Data\Process\Waste_sum_rec'

Executing (Reclassify (3)): Reclassify "M:\RISK Source Data\Process\Waste_factor" VALUE "0.69999998807907104 1.9015878793008327 1;1.9015878793008327 2.6912910404759467 2;2.6912910404759467 3.210296838824886 3;3.210296838824886 4 4;NODATA 0" "M:\RISK Source Data\Process\waste_risks1" DATA

Executing (Single Output Map Algebra (15)): SingleOutputMapAlgebra "0.13 * (1.00*traf_pol_fin1) + 0.14 * (1.00*Waste_Risks1) + 0.08 * (1.00*Cemetdist_rec) + 0.08 * (1.00*HospDist_rec) + 0.13 * (1.00*markets_pol) + 0.13 * (1.00*hab_dens_rec) + 0.12 * (1.00*Str_pol_rec1) + 0.11 * (1.00*Sea_dist_rec_) + 0.08 * (1.00*power_factor)" "M:\RISK Source Data\Process\EOW_output" 'M:\RISK Source Data\Process\hab_dens_rec';'M:\RISK Source Data\Process\Cemetdist_rec';'M:\RISK Source Data\Process\HospDist_rec';'M:\RISK Source Data\Process\markets_pol';'M:\RISK Source Data\Process\Power_factor';'M:\RISK Source Data\Process\sea_dist_rec_';'M:\RISK Source Data\Process\str_pol_rec1';'M:\RISK Source Data\Process\traf_pol_fin1';'M:\RISK Source Data\Process\waste_risks1'

Executing (Reclassify (4)): Reclassify "M:\RISK Source Data\Process\EOW_output" VALUE "0.12999999523162842 1.0815625195391476 1;1.0815625195391476 1.5300000309944153 2;1.5300000309944153 1.9346875413320959 3;1.9346875413320959 2.9300000667572021 4" "M:\RISK Source Data\Process\EOW_Rec_Quant" DATA

Executing (Single Output Map Algebra (17)): SingleOutputMapAlgebra "0.14*(1.00*traf_pol_fin1) + 0.16*(1.00*Waste_Risks1) + 0.06*(1.00*Cemetdist_rec) + 0.06*(1.00*HospDist_rec) + 0.10*(1.00*markets_pol) + 0.14*(1.00*hab_dens_rec) + 0.10*(1.00*Str_pol_rec1) + 0.16*(1.00*Sea_dist_rec_) + 0.08*(1.00*power_factor)" "M:\RISK Source Data\Process\Own_rank_out" 'M:\RISK Source Data\Process\hab_dens_rec';'M:\RISK Source Data\Process\Cemetdist_rec';'M:\RISK Source Data\Process\HospDist_rec';'M:\RISK Source Data\Process\markets_pol';'M:\RISK Source Data\Process\Power_factor';'M:\RISK Source Data\Process\sea_dist_rec_';'M:\RISK Source Data\Process\str_pol_rec1';'M:\RISK Source Data\Process\traf_pol_fin1';'M:\RISK Source Data\Process\waste_risks1'

Executing (Reclassify (6)): Reclassify "M:\RISK Source Data\Process\Own_rank_out" VALUE "0.14000000059604645 1.0996874694828875 1;1.0996874694828875 1.5737499541137367 2;1.5737499541137367 2.013124939869158 3;2.013124939869158 3.0999999046325684 4" "M:\RISK Source Data\Process\OwnWght_quan" DATA

Executing (Reclassify (8)): Reclassify "M:\RISK Source Data\Process\EOW_output" VALUE "0.12999999523162842 0.83000001311302185 1;0.83000001311302185 1.5300000309944153 2;1.5300000309944153 2.2300000488758087 3;2.2300000488758087 2.9300000667572021 4" "M:\RISK Source Data\Process\EOWrec_EqInt" DATA

Executing (Reclassify (9)): Reclassify "M:\RISK Source Data\Process\EOW_output" VALUE "0.12999999523162842 0.99406251730397344 1;0.99406251730397344 1.497187530156225 2;1.497187530156225 2.0550000444054604 3;2.0550000444054604 2.9300000667572021 4" "M:\RISK Source Data\Process\EOWrec_Jenks" DATA

Executing (Reclassify (10)): Reclassify "M:\RISK Source Data\Process\EOW_output" VALUE "0.12999999523162842 1.107699205033057 1;1.107699205033057 1.5300000309944153 2;1.5300000309944153 1.9523008569557736 3;1.9523008569557736 2.9300000667572021 4" "M:\RISK Source Data\Process\EOWrec_GeoInt" DATA

Executing (Single Output Map Algebra (4)): SingleOutputMapAlgebra max(sea_dist_rec_,str_pol_rec1,waste_risks1,markets_pol,traf_pol_fin1,Power_factor,Cemetdist_rec,hab_dens_rec,HospDist_rec) "M:\RISK Source Data\Process\Max_risks_out" 'M:\RISK Source Data\Process\sea_dist_rec_';'M:\RISK Source Data\Process\str_pol_rec1';'M:\RISK Source Data\Process\waste_risks1';'M:\RISK Source Data\Process\markets_pol';'M:\RISK Source Data\Process\traf_pol_fin1';'M:\RISK Source Data\Process\Power_factor';'M:\RISK Source Data\Process\Cemetdist_rec';'M:\RISK Source Data\Process\hab_dens_rec';'M:\RISK Source Data\Process\HospDist_rec'

Executing (Reclassify (14)): Reclassify "M:\RISK Source Data\Process\Max_risks_out" VALUE "1 1.99609375 1;1.99609375 2.9921875 2;2.9921875 4 3" "M:\RISK Source Data\Process\MaxRisk_quant" DATA

Executing (Reclassify (15)): Reclassify "M:\RISK Source Data\Process\Own_rank_out" VALUE "0.14000000059604645 0.87999997660517693 1;0.87999997660517693 1.6199999526143074 2;1.6199999526143074 2.3599999286234379 3;2.3599999286234379 3.0999999046325684 4" "M:\RISK Source Data\Process\OwnWght_EqInt" DATA

Executing (Reclassify (16)): Reclassify "M:\RISK Source Data\Process\Own_rank_out" VALUE "0.14000000059604645 0.97249997360631824 1;0.97249997360631824 1.5506249548634514 2;1.5506249548634514 2.1749999346211553 3;2.1749999346211553 3.0999999046325684 4" "M:\RISK Source Data\Process\OwnWght_Jenks" DATA

Executing (Reclassify (17)): Reclassify "M:\RISK Source Data\Process\Own_rank_out" VALUE "0.14000000059604645 1.154653732663663 1;1.154653732663663 1.6199999526143072 2;1.6199999526143072 2.0853461725649516 3;2.0853461725649516 3.0999999046325684 4" "M:\RISK Source Data\Process\OwnWght_GeoIn" DATA

Executing (Reclassify (18)): Reclassify "M:\RISK Source Data\Process\Max_risks_out" VALUE "1 1.75 1;1.75 2.5 2;2.5 3.25 3;3.25 4 4" "M:\RISK Source Data\Process\MaxRisk_EqInt" DATA

Executing (Reclassify (19)): Reclassify "M:\RISK Source Data\Process\Max_risks_out" VALUE "1 1;1 1.99609375 2;1.99609375 2.9921875 3;2.9921875 4 4" "M:\RISK Source Data\Process\MaxRisk_Jenks" DATA

Executing (Reclassify (20)): Reclassify "M:\RISK Source Data\Process\Max_risks_out" VALUE "1 1.75 1;1.75 2.5 2;2.5 3.25 3;3.25 4 4" "M:\RISK Source Data\Process\MaxRisk_geoIn" DATA

Executing (Weighted Overlay): WeightedOverlay "('M:\RISK Source Data\Process\HospDist_rec' 11 'VALUE' (0 Restricted; 1 1; 2 2; 3 3; 4 4;NODATA NODATA); 'M:\RISK Source Data\Process\Cemetdist_rec' 11 'VALUE' (0 Restricted; 1 1; 2 2; 3 3; 4 4;NODATA NODATA); 'M:\RISK Source Data\Process\str_pol_rec1' 11 'VALUE' (0 Restricted; 1 1; 2 2; 3 3; 4 4;NODATA NODATA); 'M:\RISK Source Data\Process\Power_factor' 11 'VALUE' (0 Restricted; 1 1; 2 2; 3 3; 4 4;NODATA NODATA); 'M:\RISK Source Data\Process\sea_dist_rec_' 11 'VALUE' (0 Restricted; 1 1; 2 2; 3 3; 4 4;NODATA NODATA); 'M:\RISK Source Data\Process\markets_pol' 11 'VALUE' (0 Restricted; 1 1; 2 2; 3 3; 4 4;NODATA NODATA); 'M:\RISK Source Data\Process\traf_pol_fin1' 11 'VALUE' (1 1; 2 2; 3 3; 4 4;NODATA NODATA); 'M:\RISK Source Data\Process\waste_risks1' 11 'VALUE' (1 1; 2 2; 3 3; 4 4;NODATA NODATA); 'M:\RISK Source Data\Process\hab_dens_rec' 12 'VALUE' (1 1; 2 2; 3 3; 4 4;NODATA NODATA));1 4 1" "M:\RISK Source Data\Process\Equal_influ1"

Executing (Single Output Map Algebra (27)): SingleOutputMapAlgebra con(waste_risks1 >= 2 & hab_dens_rec >= 2 & traf_pol_fin1 >=2 & markets_pol >= 2 & sea_dist_rec_ >= 2 & Power_factor >= 2 & str_pol_rec1 >= 2 & Cemetdist_rec >= 2 & HospDist_rec >= 2, 1, 0) "M:\RISK Source Data\Process\Highrisk_Int" 'M:\RISK Source Data\Process\waste_risks1';'M:\RISK Source Data\Process\hab_dens_rec';'M:\RISK Source Data\Process\traf_pol_fin1';'M:\RISK Source Data\Process\markets_pol';'M:\RISK Source Data\Process\sea_dist_rec_';'M:\RISK Source Data\Process\Power_factor';'M:\RISK Source Data\Process\str_pol_rec1';'M:\RISK Source Data\Process\Cemetdist_rec';'M:\RISK Source Data\Process\HospDist_rec'

Executing (Single Output Map Algebra (16)): SingleOutputMapAlgebra "0.12*traf_pol_fin1 + 0.11*Waste_Risks1 + 0.11*Cemetdist_rec + 0.11*HospDist_rec + 0.11*markets_pol + 0.11*hab_dens_rec + 0.11*Str_pol_rec1 + 0.11*Sea_dist_rec_ + 0.11*power_factor" "M:\RISK Source Data\Process\Equalweight" 'M:\RISK Source Data\Process\hab_dens_rec';'M:\RISK Source Data\Process\Cemetdist_rec';'M:\RISK Source Data\Process\HospDist_rec';'M:\RISK Source Data\Process\markets_pol';'M:\RISK Source Data\Process\Power_factor';'M:\RISK Source Data\Process\sea_dist_rec_';'M:\RISK Source Data\Process\str_pol_rec1';'M:\RISK Source Data\Process\traf_pol_fin1';'M:\RISK Source Data\Process\waste_risks1'

Executing (Reclassify (5)): Reclassify "M:\RISK Source Data\Process\Equalweight" VALUE "0.10000000000000001 0.98991839663358405 1;0.98991839663358405 1.3262563003227115 2;1.3262563003227115 1.7710903019760735 3;1.7710903019760735 3 4" "M:\RISK Source Data\Process\Quan_Eqweight" DATA

Executing (Reclassify (12)): Reclassify "M:\RISK Source Data\Process\Equalweight" VALUE "0.10000000000000001 1.1316237692284663 1;1.1316237692284663 1.4998500570654871 2;1.4998500570654871 1.868076344902508 3;1.868076344902508 3 4" "M:\RISK Source Data\Process\GeoIn_eqWeigh" DATA

Executing (Reclassify (13)): Reclassify "M:\RISK Source Data\Process\Equalweight" VALUE "0.10000000000000001 0.98991839663358405 1;0.98991839663358405 1.4347523982869461 2;1.4347523982869461 1.8795863999403082 3;1.8795863999403082 3 4" "M:\RISK Source Data\Process\Jenks_EqWeigh" DATA

Executing (Reclassify (11)): Reclassify "M:\RISK Source Data\Process\Equalweight" VALUE "0.10000000000000001 0.80547503009438515 1;0.80547503009438515 1.4998500570654869 2;1.4998500570654869 2.1942250840365887 3;2.1942250840365887 3 4" "M:\RISK Source Data\Process\EqInt_Eqweigh" DATA

Executing (Reclassify (22)): Reclassify "M:\RISK Source Data\Process\EOW_output" VALUE "0 1 1;1 2 2;2 3 3;3 4 4" "M:\RISK Source Data\Process\EOW_defined" DATA

Executing (Reclassify (23)): Reclassify "M:\RISK Source Data\Process\Own_rank_out" VALUE "0 1 1;1 2 2;2 3 3;3 4 4" "M:\RISK Source Data\Process\Own_defined" DATA

Executing (Reclassify (25)): Reclassify "M:\RISK Source Data\Process\Max_risks_out" VALUE "0 1 1;1 2 2;2 3 3;3 4 4" "M:\RISK Source Data\Process\Max_defined" DATA

Executing (Reclassify (26)): Reclassify "M:\RISK Source Data\Process\Equalweight" VALUE "0 1 1;1 2 2;2 3 3;3 4 4" "M:\RISK Source Data\Process\EqWgh_defined" DATA

Executing (Single Output Map Algebra (30)): SingleOutputMapAlgebra "0.10*traf_pol_fin1 + 0.30*Waste_Risks1 + 0.05*Cemetdist_rec + 0.05*HospDist_rec + 0.10*markets_pol + 0.10*hab_dens_rec + 0.10*Str_pol_rec1 + 0.10*Sea_dist_rec_ + 0.10*power_factor" "M:\RISK Source Data\Process\Sens_An_Waste" 'M:\RISK Source Data\Process\hab_dens_rec';'M:\RISK Source Data\Process\waste_risks1';'M:\RISK Source Data\Process\traf_pol_fin1';'M:\RISK Source Data\Process\markets_pol';'M:\RISK Source Data\Process\Cemetdist_rec';'M:\RISK Source Data\Process\Power_factor';'M:\RISK Source Data\Process\str_pol_rec1';'M:\RISK Source Data\Process\HospDist_rec';'M:\RISK Source Data\Process\sea_dist_rec_'

Executing (Reclassify (24)): Reclassify "M:\RISK Source Data\Process\Sens_An_Waste" VALUE "0.40000000596046448 1.0536195691580446 1;1.0536195691580446 1.5097352912200621 2;1.5097352912200621 2.1633548544176424 3;2.1633548544176424 3.0999999046325684 4" "M:\RISK Source Data\Process\Waste_sensit" DATA

Executing (Single Output Map Algebra (29)): SingleOutputMapAlgebra "0.16*traf_pol_fin1 + 0.20*Waste_Risks1 + 0.01*Cemetdist_rec + 0.02*HospDist_rec + 0.05*markets_pol + 0.21*hab_dens_rec + 0.19*Str_pol_rec1 + 0.05*Sea_dist_rec_ + 0.12*power_factor" "M:\RISK Source Data\Process\Sens_An_pro" 'M:\RISK Source Data\Process\hab_dens_rec';'M:\RISK Source Data\Process\sea_dist_rec_';'M:\RISK Source Data\Process\markets_pol';'M:\RISK Source Data\Process\traf_pol_fin1';'M:\RISK Source Data\Process\waste_risks1';'M:\RISK Source Data\Process\Power_factor';'M:\RISK Source Data\Process\Cemetdist_rec';'M:\RISK Source Data\Process\str_pol_rec1';'M:\RISK Source Data\Process\HospDist_rec'

Executing (Reclassify (27)): Reclassify "M:\RISK Source Data\Process\Sens_An_pro" VALUE "0.40999999642372131 1.4893779366976032 1;1.4893779366976032 2.0149999409914017 2;2.0149999409914017 2.5406219452852001 3;2.5406219452852001 3.619999885559082 4" "M:\RISK Source Data\Process\Prorat_sensit" DATA

Executing (Single Output Map Algebra (28)): SingleOutputMapAlgebra "0.30*traf_pol_fin1 + 0.10*Waste_Risks1 + 0.05*Cemetdist_rec + 0.05*HospDist_rec + 0.10*markets_pol + 0.10*hab_dens_rec + 0.10*Str_pol_rec1 + 0.10*Sea_dist_rec_ + 0.10*power_factor" "M:\RISK Source Data\Process\Sens_An_traf" 'M:\RISK Source Data\Process\hab_dens_rec';'M:\RISK Source Data\Process\waste_risks1';'M:\RISK Source Data\Process\traf_pol_fin1';'M:\RISK Source Data\Process\markets_pol';'M:\RISK Source Data\Process\Cemetdist_rec';'M:\RISK Source Data\Process\Power_factor';'M:\RISK Source Data\Process\str_pol_rec1';'M:\RISK Source Data\Process\HospDist_rec';'M:\RISK Source Data\Process\sea_dist_rec_'

Executing (Reclassify (28)): Reclassify "M:\RISK Source Data\Process\Sens_An_traf" VALUE "0.20000000298023224 0.88484027420232358 1;0.88484027420232358 1.3087329934890177 2;1.3087329934890177 1.9935732647111089 3;1.9935732647111089 3.0999999046325684 4" "M:\RISK Source Data\Process\Traf_sensit" DATA

References

Adger, N. W., 1999. Social Vulnerability to Climate Change and Extremes in Coastal Vietnam. World Development, 1(2), 249-269.

Akçakaya, H. R., 1996. Linking GIS with Models. Paper presented at the Third International Conference on Integrating GIS and Environmental Modeling, Santa Fe, NM.

Aly, M. H., 1997. Integration of Remote Sensing and GIS for Geo-environmental Assessment and Monitoring of the New Minia City, Egypt: MS Thesis, Geology Department, Faculty of Science, Zagazig University, Zagazig, Egypt, 139 p.

Bailey, K. and Grossardt, T., 2006. Structured public involvement in context-sensitive noise wall design using casewise visual evaluation. Transportation Research Board, 1984, 112-120.

Bakrim, H., 2001. Reassessment of Natural Disasters: Living Conditions and Peoples' Vulnerability to Natural Disasters - Hurricane Mitch and Tegucigalpa, Honduras, working paper, posted at http://www.duyure.org/reassessment.pdf, accessed on 07/09/2007.

Beaton, P. 1986. On the use of expert opinion in housing quality analysis: an application of the «Five City Housing Survey». Journal of Planning Education and Research, 5(3), 178-190.

Bell, N., Schuurman, N., and Hayes, M. V., 2007. Using GIS-based methods of multicriteria analysis to construct socio-economic deprivation indices. International Journal of Health Geographics, 6(17).

Blaikie, P.; Cannon, T.; Davis, I.; Wisner, B. 1994. At Risk: Natural Hazards, People's Vulnerability, and Disasters. Routledge, London and New York.

Bureau of Meteorology Research Centre (BMRC), 2006. Global Guide to Tropical Cyclone Forecasting: Chap. 7: Warning Strategies. Accessed from http://www.bom.gov.au/bmrc/pubs/tcguide/ch7/ch7_5.htm on 07/13/2006.

Brewer, C.A., Pickle L., 2002. Evaluation of methods for classifying epidemiological data on chloropleth maps in series. Annals of the Association of American Geographers, 92, 662-681.

Campbell, T. and Campbell, A., 2007. Emerging Disease Burdens and the Poor in Cities of the Developind World. Journal of Urban Health: Bulletin of the New-York Academy of Medicine, 84(1), i54-i64.

Campbell-Lendrum, D., and Corvalan, C., 2007. Climate Change and Developing-Country Cities: Implications for Environmental Health and Equity. Journal of Urban Health: Bulletin of the New York Academy of Medicine, 84(1), i109-i117.

Can, A., 1993. Residential quality assessment. Alternative approaches using GIS in M.M. Fischer and P. Nijkamp (Eds.): Geographic Information Systems, Spatial Modeling and Policy Evaluation. Berlin, Springer-Verlag, 199-212.

Carver, S., 1991. Integrating multi-criteria evaluation with geographical information systems: International Journal Geographical Information Systems, 5, 321-339.

Clark, G. E., Moser, S. C., Ratick, S. J., et al. 1998. Assessing the vulnerability of coastal communities to extreme storms: the case of Revere, MA, USA. Mitigation and adaptation strategies for global change, 3, 59-82.

Corvalan, C. F., Kjellstrom, T., and Smith, K. R., 1999. Health, Environment and Sustainable Development: Identifying Links and Indicators to Promote Action. Epidemiology, 10(5), 656-660.

Cutter, S. L., 1996. Vulnerability to environmental hazards. Progress in Human Geography, 20, 529-539.

Cutter, Susan L., Boruff, B. J., and Shirley, W. L. 2003. Social Vulnerability to Environmental Hazards. Social Science Quaterly, 84(2), 242- 261.

Dakin, S. and Armstrong, J. S., 1989. Predicting job performance: a comparision of expert opinion and research findings. Marketing Papers, University of Pennsylvania. International Journal of Forecasting, 5(2), 187-194.

Day, N., Eden, T., Mckinney, P., Roman, E., and Simpson, J., 2005. Chilhood cancer and power lines: what do the data mean? BMJ, 2005, 331:634.

Department of Human Resources, city of Long Beach, California, 1998. Glossary of common risk management terms. Accessed from http://cms.longbeach.gov/hr/employees/riskmgmt/hr9b.htm on 7/31/06

Smith, M. J., Goodchild, M. F., and Longley, P. A., 2007. Geospatial Analysis - The comprehensive guide to principles, techniques and software tools. Webversion (Oct. 2007), accessed from http://www.spatialanalysisonline.com/ on 10/31/2007.

Diley, M., Chen, R. S.,Deichmann, U., Lernar-Lam, A. L., and Arnold, M., 2005. Natural Disasters hotspots: a global risk analysis, Washington, DC: The International Bank for Reconstruction and Development/The World Bank and Columbia University, 132 p.

Dodgson, J., Spackman, M., Pearman, A.D. and Phillips, L.D., 2000. Multi-Criteria Analysis: a Manual. Department of the Environment, Transport and the Regions, London, pp. 158.

Dramowicz, E. and Dramowicz, K. 2004. Chloropleth mapping with exploratory data analysis. Directions Magazine, Dec 29 issue. Accessed from http://www.directionsmag.com/article.php?article_id=718 on 10/31/07.

Draper, G., Vincent, T., Kroll, Mary E., and Swanson, J., 2005. Childhood cancer in relation to distance from high voltage power lines in England and Wales: a case-control study. BMJ, 330, 1290-

Eastman, J. R., Weigen, J., Kyem, P., and Toledano, J., 1995. Raster procedures for multi-criteria/multiple objective decisions. Photogrammetric Engineering and Remote Sensing, 61, 539-47.

Ehrenberg, J.P. and Ault, S.K., 2005. Neglected diseases of neglected populations: Thinking to reshape the determinants of health in Latin America and the Caribbean. BioMed Central Public Health, 5(119).

Elliott, P., Briggs, D.J., Morris, S., De Hoogh, D., Hurt, C., Jensen, T.K., et al., 2001. Risk of adverse birth outcomes in populations living near landfill sites. British Medical Journal, 323, 363-368.

Environmental Protection Agency (EPA), 2005. Reducing Air Pollution from: Hospitals - Community Information Sheet. Last updated 9/12/05/. Accessed from www.epa.gov./air/toxicair/community/guide/healthcare_comm_info.pdf on 07/23/07.

Environmental Systems Research Institute (ESRI), 1996. ArcView GIS - The Geographic Information System for Everyone. ESRI, USA.

Evans, I. S., 1977. The selection of class intervals. Transactions of the Institute of British Geographers, New Series, 2(1), 98-124.

Ezatti, M., Utzinger, J., Cairncross, S., Cohen, A. J., and Singer, B. H., 2003. Environmental risks in the developing world: exposure indicators for evaluating interventions, programmes and policies. Journal of Epidemiology and Community Health, 2005(59),15-22.

Ferguson, E. C., Maheswaran, R., and Daly, M. 2004. Road-traffic pollution and asthma - using modeled exposure assessment for routine public health surveillance. International Journal of Health Geographics 3(24).

Fernandez A, Mondkar J, Mathai S., 2003. Urban slum-specific issues in neonatal survival. Indian Pediatry, 1161-6.

Ferrier, N., Haque, C. E., 2003. Hazards Risk Assessment Methodology for Emergency managers: a Standardized Framework for Application. Natural hazards, 28, 271-290.

Fews, A.P., Henshaw, D.L., Keitch, P.A., Close, J.J., and Wilding, R.J., 1999. Increased exposure to pollutants aerosol under high voltage power lines. International Journal of Radiation Biology, 75(12), 1505-21.

Franklin, J.P., Waddell, P., and Britting, J., 2002. Sensitivity Analysis Approach for an Integrated Land Development and Travel Demand Modeling System. Paper for presentation at the ACSP 44^th Annual Conference, Baltimore, MD, Nov 21-24, 2002.

Goodyear, E. J., 2000. Disaster Mitigation: Challenges to raise the capacity of at-risk publications in coping with natural, social, and economic disasters, Austral. J. Emergency Management, 15(3), 25-30.

Greene, R.P. and Pick, J.B., 2006. Exploring the Urban Community - A GIS Approach. Keith C. Clarke Series editor, Prentice Hall series in GIS, NJ.

Hamza, M. and Zetter, R., 1998. Structural Adjustment, Urban Systems, and Disaster Vulnerability in Developing Countries. Cities, 15(4), 291-299.

Harper, C. C., Mathee, A., Von Schirnding Y., De Rosa, C.T., and Falk, H., 2003. The health impact of environmental pollutants: a special focus on lead exposure in South Africa. International Journal of Hygiene and Environmental Health, 206(4-5), 315-22.

Hoek, G., Brunekreef, B., Goldbohm, S., Fischer, P., Van Den Brandt, P., 2002. Associations between mortality and indicators of traffic-related air pollution in the Netherlands: a cohort study. The Lancet, 360, 1203-1209.

Hoek, G., Fischer, P., Van Den Brandt, P., Sandra Goldbohm, S., and Brunekreef, B., 2001. Estimation of long-term average exposure to outdoor air pollution for a cohort study on mortality. Journal of Exposure Analysis and Environmental Epidemiology, 11, 459-469.

Hopkins, L. D., 1977. Methods for generating land suitability maps: a comparative evaluation. Journal of the American Institute of Planners, 43 (4), 386-400.

Hossain, S. M. N. and Singh, A., 2002. Application of GIS for assessing human vulnerability to cyclone in India. ESRI Professional Papers. Accessed from gis.esri.com/library/userconf/proc02/abstracts/a0701.html on 06/25/07

Howard, P., 1998. Environmental Scarcities and Conflict in Haiti, in Republique d'Haiti et al., Cadre Physique de la Region Metropolitaine, Volume VI.

Hupper, H. E. and Sparks, R. S. J., 2006. Extreme natural hazards: population growth, globalization and environmental change. Philosophical Transactions of the Royal Society, 364, 1875-1888.

Huppert, H. and Sparks R. S., 2006. Extreme natural hazards: population growth, globalization and environmental change. Philosophical Transactions of Royal Society A, 364, 14 p.

Institut Haitien de Statistique et d'Informatique (IHSI), 2003. Fourth population census.

International Strategy for Disaster Reduction (ISDR), 2004. Basic terms of disaster risk reduction. Last update 03/31/04. Retrieved from http://www.unisdr.org/eng/library/lib-terminology-eng%20home.htm on 06/26/07.

Jankowski, P., 1995. Integrating geographical information systems and multiple criteria decision making methods. International Journal of Geographical Information Systems, 9, 251-73.

Janssen, R. and Rietveld, P., 1990. Multicriteria analysis and Geographical Information Systems: an application to agricultural land use in the Netherlands. Scholten H.J. and Stillwell, J.C.H., eds., Geographical Information Systems for Urban and Regional Planning, 129-139.

Latkin, C.A., Curry, A.D., 2003. Stressful neighborhoods and depression: a prospective study of the impact of neighborhood disorder. Social Behavior, 44(1) 33-44.

Liverman, D. 1990. Vulnerability to global environmental change, in Kasperson, R.E., Dow, K., Golding, D. and Kasperson, J.X. (Eds), Understanding Global Environmental Change: The Contributions of Risk Analysis and Management, Clark University, Worcester, 27-44.

Longley, P. A., Goodchild, M. F., Maguire, D. J. and Rhind, D. W. 2005. Geographic Information Systems and Science. Wiley, 2^nd Edition, 517 p.

Maheswaran, R., Elliott, P., 2003. Stroke mortality associated with living near main roads in England and Wales: a geographical study. Stroke, 34(12), 2776-80.

Malczewski, J. 2004. GIS-based land-use suitability analysis: a critical overview.

Mahini, A. S. and Gholamalifard, M., 2006. Sitting MSW landfills with a weighted linear combination methodology in a GIS environment. International Journal of Environmental Science and Technology, 3(4), 435-445.

Massam, B. H., 1988. Multi-criteria decision making (MCDM) techniques in planning. Progress in planning, 30, 1-84.

Mathee, A. 2002. Environment and Health Implications of Urbanization, with special reference to Johannesburg. WHO/SDE/HDE - Meeting of Senior Officials and Ministers of Health, Summary Report 11-12.

Mcleman, R. and Smit, B., 2006. Vulnerability to climate change hazards and risks: crop and flood insurance. The Canadian Geographer, 50(2), 217-226.

Moore, M., Gould, P., Keary, B. S., 2003. Global urbanization and impact on health. International Journal of Hygiene and Environmental Health, 206 (4-5), 269-78.

Myrphy, B., 2005. Modernizing Floodplain map production with GIS. 2005 ESRI User Conference Paper # 2502.

Nafstad, P., Haheim, L.L., Oftedal, B., Gram, F., Holme I., Hjermann, I., and Leren, P. 2003. Lung cancer and air pollution: A 27-year follow up of 16,209 Norwegian men. Thorax, 58, 1071-1076.

Nsiah-Gyabaah and al. 2004. Urbanization Processes - Environmental and health effects in Africa, Accessed from www.populationenvironmentresearch.org/papers/nsiah_gyaabah_contribution.pdf on 05/26/2006.

Olanrewaju, D., 1990. Soak-away Systems and Possible Groundwater Pollution Problems in Developing Countries. The Journal of the Royal Society for the Promotion of Health, 110(3), 108-112.

Osarangi, T., 2002. Classification methods for spatial data representation. Centre for Advanced Spatial Analysis (CASA), University College London. Working papers series paper 40.

O'Sullivan, D. and Unwin, D., 2003. Geographic Information Analysis. John Wiley & Sons, Hoboken, NJ.

OXFAM- GB, 2001. Etude des Risques, de la Vulnerabilité et des capacités locales de réponses. Port-au-Prince, Haiti.

PanAmerican Health Organization (PAHO), 2001. Country health profile, Haiti. Data updated for 2001. Accessed from http://www.paho.org/English/SHA/prflHAI.htm on 10/08/2007

Pereira, J.M.C. and Duckstein, L. 1993. A multiple criteria decision-making approach to GIS-based land suitability evaluation. International Journal of Geographical Information Systems, 7(5), 407-424.

Pettit, Kathryn L. S., Kingsley, G. T., and Coulton, Claudia J., 2003. Neighborhood and health: building evidence for local policy. The Urban Institute, Washington DC. Accessed from www.urban.org/url.cfm?ID=410819 on 07/05/2007.

Saaty, T. L., 1990. Decision Making for Leaders, Vol. II, EOW Series, RWS Publ., 315p.

Schikowski, T., Sugiri, D., Ranft, U., Gehring, U., Heinrich, J., Wichmann, H-E., and Krämer U., 2005. Long-term air pollution exposure and living close to busy roads are associated with COPD in women. Respiratory Research, 6(152).

Schmidth-Thomé, P. 2006a. Integration of Natural Hazards, Risk and Climate Change into spatial Planning Practices. Academic Dissertation.

Schmidt-Thomé, P., Klein, J., Aumo, R., & Hurstinen, J. 2006. Report: Technical Glossary of a Multi-Hazard Related Vulnerability and risk Assessment Language. Unpublished deliverable 4.2 of ARMONIA project, 6^th framework project funded by the European Community, Contract No. 511208. Espoo, 33 p.

Swiss Ressource Centre and Consultancies for Development, 2002. Introduction to solid waste management. Accessed from www.sanicon.net/titles/topicintro.php3?topicId=4 on 07/23/07.

Smit, B. and Pilifosova, O., 2003. From adaptation to adaptive capacity and vulnerability reduction. Climate Change, Adaptive Capacity and Development, ed J.B. Smith, R.J.T. Klein and S. Huq (London: Imperial College Press) 9-28 (in Mcleman and Smit 2006, Vulnerability to climate change hazards and risks).

Smith, K., 1999. Environmental Hazards: Assessing Risk and Reducing Disaster, 2^nd ed., Routledge, London.

Smucker, G. R., Bannister, M., D'Agnes H., Gossin, Y., Portnoff, M., Timyan, J., Tobias, S., and Toussaint, R., 2007. Environmental Vulnerability in Haiti: Findings and Recommendations. USAID/Haiti.

Sorensen, J., Vedeld, T., Haug, M., 2006. Natural Hazards and Disasters: Drawing on the international experiences from disasters reduction in developing countries. Norwegian Institute for Urban and Regional Research (NIBR).

Steptoe, A. and Wardle J., 1994. What the experts thinks: a European survey of expert opinion about the influence of lifestyle on health. European Journal of Epidemiology, 10(2) 195-203.

Stockholm Environment Institute (SEI), 2005. The SEI Poverty and Vulnerability Program. Last modified on 08/05/2005. Accessed from www.sei.se/risk/poverty.html on 06/26/2007.

Sui, D., 2004. Forum: Tobler's first law of geography: A big idea for a small world? Annals of the Association of American Geographers, 94, 269-77.

The Caribbean Disaster Emergency Response Agency (CDERA), 2003. Status of Hazard Maps/Vulnerability Assessments and Digital Maps, Haiti Country Report. Accessed from http://www.cdera.org/projects/cadm/docs/haiti_hmvadm.pdf on 03/16/2006.

Tobias, S., 2004, Quality in the performing arts: aggregating and rationalizing expert opinion. Journal of Cultural Economics, 28, 109-124.

Tomlin, C.D., 1990. Geographical Information Systems and Cartographic Modelling. Englewood Cliffs, NJ, Prentice-Hall.

UNDP, 2004. Reducing Disaster Risk: A Challenge for Development. A global report. Bureau for Crisis Prevention and Recovery. Accessed from www.undp.org/bcpr on 06/22/07.

UNDP/DHA, 1994. Vulnerability and Risk Assessment. Disaster Management Training Program. 2^nd edition, 1994.

UNDRO, 1979. Natural Disasters and Vulnerability Analysis. Report of Expert Group Meeting. 9-12 July 1979, Office of the United Nations Disaster Relief Coordinator, Geneva.

UNDRO, 1991. Mitigating Natural Disasters. Phenomena, Effects and Options. In Manual for Policy Makers and Planners, United Nations.

United Nations, 1992. Conference on Environment and Development (UNCED), Agenda 21, Rio de Janeiro.

United Nations Environment Programme (UNEP) 1994a. Air Pollution in the World's Mega-cities. Environment 36, 2:4.

UNEP, 1996. International Source Book on Environmentally Sound Technologies for Municipal solid Waste Management. UNEP Technical Publications 6 Nov. 1996.

Voogd, H., 1983. Multicriteria Evaluation for Urban and Regional Planning, Pion, Londres.

Wargny, C. 2004. Haiti n'existe pas. 1804-2004: deux cents ans de solitude. Editions Autrement, Paris, France, 191p.

Watkiss, P., Brand, C., Hurley, F., Pilkington, A., Mindell, J., Joffe, M., and Anderson, R., 2000. Informing transport health impact assessment in London. London's Health, accessed from http://www.phel.gov.uk/hiadocs/informing_transport_hia_in_london.pdf on 07/22/07.

Weichselgartner, J. 2001. Disaster mitigation: the concept of vulnerability revisited. Disaster Prevention and Management, 10, 85-94

World Health Organization (WHO), 2001. Health and Sustainable Development. Meeting of Senior officials and Ministers of Health, Oslo, Norway, 29 November - 1 December 2001.

WHO, 2001. Health in the context of sustainable development - Document for the WHO meeting. Oslo-Norway, WHO/HDE/HID/02.6

WHO, 2005. Air Quality Guidelines, Global Update 2005. WHO Regional Office for Europe.

Wilhelm M. and Ritz, B., 2003. Residential Proximity to Traffic and Adverse Birth Outcomes in Los Angeles County, California, 1994-1996. Environmental Health Perspectives, 111(2), 207-216.

World Bank, 2005. Brazil Funds Clean-up of Haiti. Davos, Switzerland. Accessed from http://web.worldbank.org, 07/21/2007.

YTV, Helsinki Metropolitan Area Council, 2007. Air quality in the Helsinki Metropolitan Area. Helsinki Metropolitan Area Council Publication series.