*3.1. Determination of Vulnerability*

As mentioned, an environmental vulnerability evaluation along the BR 050 highway was accomplished by the authors of this study in a previous publication [15]. To complete the task, a Multi Criteria Analysis (MCA) embedded in a geographic information system was applied to the areas under direct influence (within a 210 m buffer from each margin of the highway) and indirect influence (within the limits of the micro basins along the 97 km segment) of spills of dangerous products. The MCA approach is a four-step process, which involves (1) selection of factors to describe vulnerability, with subsequent normalization of factor scales into a common dimensionless range, and elaboration of raster maps that describe the spatial distribution of normalized factors; (2) the allocation of a weight to each factor; (3) the weighted combination of factors to compose a final vulnerability map; and (4) a sensitivity analysis of vulnerability results based on scenarios [36]. The four steps are briefly outlined in the next paragraphs:

*Step 1*: In the study of [15] the vulnerability maps were based on the following factors: (1) drainage density, (2) distance from water courses, (3) ground slope, (4) soil type, (5) land use/occupation, and (6) geology (Figure 3; Table 1, part (a)). These factors were selected because they are comparable to key variables of drainage models and flow routing algorithms that describe the detachment and transportation of pollutants in catchments [5]. The normalization of factor classes was based on a byte-level interval (0 to 255), which linked a 0 level to the least important class and 255 level to the most important class (Table 1, part (b)). The association of factor classes to levels of importance (ratings) was based on the authors' personal experiences about vulnerability assessments.

**Figure 3.** Spatial distribution of vulnerability-relevant factors included in the Multi Criteria Analysis. Adapted from [15].

**Table 1.** (**a**) Factors used by [15] in the Multi Criteria Analysis of environmental vulnerability related to road accidents along the studied segmen<sup>t</sup> of BR 050 highway involving the transport of hazardous substances; (**b**) Normalization of factors within the Multi Criteria Analysis—MCA (step 2) designed to evaluate soil and water vulnerability along roads. The higher the value of a normalized factor the greater its importance for vulnerability. The MCA model was applied to a segmen<sup>t</sup> of BR 050 highway where transport of hazardous substances is intense and spills of those products following a road accident can cause severe damage to the surrounding environment. Adapted from [15].


*Step 2*: The allocation of weights was based on the Analytical Hierarchy Process (AHP; [37]) whereby the user (or a group of experts) assign a relative importance to each factor based on pairwise comparisons with the other factors, and then this hierarchy is processed in the AHP algorithm to obtain a set of optimized levels of importance (weights). Because the attribution of weights can be subjective, in the study of [15] a sensitivity analysis was performed (step 4 below) where various factors were given the largest relative importance and hence maximum weights.

*Step 3*: The overall vulnerability was calculated by Equation (1), implemented in the GIS software using map algebra tools and factor maps in raster format (Figure 4):

$$S\_i = \sum\_{i=1}^{p} w\_j X\_{ij} \tag{1}$$

where *Sj* represents the vulnerability at pixel *i*, *wj* represents the weight of the factor *j*, and *Xij* represents the normalized value of factor *j* at pixel *i*. The *Si* values are reclassified into five classes using the same byte-level range: Invulnerable (0–50), Weakly vulnerable (50–100), Vulnerable (100–150), Strongly Vulnerable (150–200), and Extremely Vulnerable (200–250).

**Figure 4.** Vulnerability maps of the intercepted water course catchments, highlighting the vulnerability at the road accident sites (also termed hazard; labeled circles). The maps are outcomes of a Multi Criteria Analysis where vulnerability-relevant factors ground slope (map (**a**)), drainage density (**b**), geology (**c**) and soil type (**d**) were given the largest weight [15]. The concomitant effects on vulnerability are reflexes of factor heterogeneity across the studied area. The largest effect occurs when factors geology or soil type are maximized highlighting the importance of these factors.

*Step 4*: The sensitivity of *S* to changing factor weights was evaluated through generation of four scenarios where one of these factors has been given the largest relative importance maximizing its weight. The factors that have been given maximum weights were drainage density, ground slope, soil type and geology. The scenarios were created because the aforementioned factors are heterogeneous across the studied region and in that context associated with an ample range of scores. For these reasons there is no easy way to define a universal hierarchy to describe the importance of each factor. For example, the studied segmen<sup>t</sup> of BR 050 highway is contrasting as regards ground slope, because the north and south sectors are occupied by plains linked to low vulnerability while the central part is mountainous and linked to high vulnerability. In a scenario that maximizes the importance of ground slope, these contrasting topographic features will be highlighted in the final vulnerability map, while being smoothed otherwise. The same rationale holds for the other factors as well. Vulnerability in the four scenarios is illustrated in Figure 4a–d.

As expected, when factor ground slope is maximized (Figure 4a) the vulnerability map shows a central area with high vulnerability bordered to the north and south by areas with low vulnerability. However, it is evident from analysis of Figure 4c,d that rising the role of geology or soils in the vulnerability assessment results in larger overall vulnerability. It is also worth to note that the BR 050 highway in the studied segmen<sup>t</sup> was built nearly along a water divide in the northern and central sectors but away from it in the southern part (Figure 1). For that reason, spills of dangerous substances in the northern-central parts of the segmen<sup>t</sup> will potentially affect the water courses in both sides of the highway while in the southern part spill drainage will primarily affect the western channels.

### *3.2. Occurrence Data Involving Hazardous and Potentially Harmful Products to the Environment*

The data on accidents involving hazardous and potentially harmful products in the studied segmen<sup>t</sup> was obtained from the Concessionaire who manages approximately 700 km of BR 050 highway. The collected data is summarized in Table 2. It is important to note that some products represented in the table are not classified as dangerous by the United Nations (http://www.unece.org/trans/danger). These products are identified as "not applicable" under the heading "UN Code" (Column 3). However, because the road spill of these products can contribute significantly to soil and water contamination in accident scenarios, they were used in the present study of risk management. The accident data was compiled from operational resources such as Operational Control Center (OCC), Traffic Inspection and Mechanical Rescue Vehicles, Emergency Medical Service Vehicles (rescue and salvage), Closed Circuit TV and Radio Communication System. Through the OCC, all the occurrence data are recorded using the software Kria Operational Control for Highways. Among other issues, the recording involves the generation of a GIS database through the conversion of site details (i.e., the exact kilometer of the occurrence) into geographical coordinates (last columns of Table 2). Besides generation of data the software releases managemen<sup>t</sup> reports according to the periodicity and type of occurrence desired, allowing analysis, treatment and decision making. The data for the present study spans the period from July 2014 to December 2017, which represents a 2 years and 6 months interval. Figure 1 shows the distribution of occurrences (red circles) involving hazardous products in the studied segment, obtained through the Concessionaire. We recognize that the data record is not long to provide a clear image of the situation, but are confident that enables a preliminary view.


**Table 2.** Occurrences involving dangerous products in BR 050 during the monitored period. Symbols: UN—United Nations; UTM—Universal Transverse Mercator (coordinate system); X, Y—planimetric coordinates of the accident.

### *3.3. Environmental Vulnerability at Occurrence Sites (Risk)*

The vulnerability around the road accident sites listed in Table 2 was assessed by the IDRISI Selva software [35], taking into account the four predefined scenarios (Figure 4). The IDRISI Selva software embeds a set of tools capable to determine the environmental vulnerability of an area according to the necessary steps. The vulnerability profile of each site was defined through the following steps: (1) a 200 m buffer was drawn around the site. This area covers the environmental resources immediately affected after the occurrence of spills; (2) The buffers were plotted over the four vulnerability maps (Figure 4a–d); (3) For each vulnerability scenario, the area related to a vulnerability level (e.g., strongly vulnerable) was determined using raster map operations.

Steps 1–3 were repeated for all the vulnerability levels and all sites, and aggregated per vulnerability level in each scenario. To distinguish the vulnerability evaluated within the studied segmen<sup>t</sup> (catchment scale) from the vulnerability evaluated around the road accident sites (buffer scale) the latter was termed hazard. Having determined the hazard area within the 14 buffers, the risk of soil and water contamination is estimated for every vulnerability level using the formula:

$$R\_j = \frac{H\_j}{V\_j} = \frac{Ab\_j}{A\_j} \tag{2}$$

where *Rj* is the risk for level *j*, *Hj* is the hazard for level *j* evaluated within the 14 buffers and represented by the corresponding area (*Ab*j, in percentage of total buffer area), and *Vj* is the vulnerability for level *j* evaluated within the studied segmen<sup>t</sup> and represented by its area (*Aj*, in percentage of segmen<sup>t</sup> area). If the *Rj* value is >1 then road accident sites are considered risky at that level. If sites are risky for the preoccupying levels (e.g., "strongly vulnerable" or "extremely vulnerable") then the implementation of prevention and alert systems should be mandatory. These systems should also be considered for the "vulnerable level". The analysis of risk can be refined, which means executed site by site. In this case, the *Abj* represents the area of hazard level *j* within the specific site, in percentage of buffer area. It is worth mentioning that, besides hazard incidence and medium vulnerability the risk of soil and water contamination by dangerous substance, including public health issues, also depends on the extension of contaminant propagation, the amount and chemical properties (toxicity) of the spilled product, and the proximity of human presence [38]. Toxicity and proximity to urban centers will not be addressed in this study, because the vulnerability assessment on which the risk analysis is standing has been focused on the protection of environmental resources, soils and water.
