Journal of Forensic Science and Medicine

: 2015  |  Volume : 1  |  Issue : 2  |  Page : 99--108

Chemical Analysis of Tire Traces in Traffic Accidents Investigation

Line Gueissaz, Genevieve Massonnet 
 Faculty of Law, Criminal Justice and Public Administration, School of Criminal Justice, University of Lausanne, Batochime, CH-1015 Lausanne-Dorigny, Switzerland

Correspondence Address:
Line Gueissaz
School of Criminal Justice, University of Lausanne, UNIL–Batochime, CH-1015 Lausanne-Dorigny


The aim of the forensic investigation of traffic accidents is to help establish the nature and/or the circumstances of the event. This might be done with the purpose of determining the legal responsibilities of each person involved or to provide families, with a reconstruction of the events, to help understand why their relatives were injured or killed. A methodology for the comparison of chemical profiles of tire traces and tire tread samples obtained by pyrolysis-gas chromatography/mass spectrometry has been developed. Chemical profiles are represented by relative abundances of 86 compounds. The variability of the tread within and between 12 tires was assessed. Considering the level of the source as "brand and model" the intra-variability was found to be smaller than the inter-variability, leading to the complete discrimination of the 12 tires of the sample set. Braking tests were carried out on a racetrack in order to produce tire traces which origin was known. The results obtained with a supervised classification method showed that more than 94% of the replicates of the traces were correctly assigned to the class membership (i.e., brand and model) of the tire at their origin. These results support that the chemical profile of one trace does not differ from the chemical profile of the tire at its origin but differs generally from the other chemical profiles of the sample set.

How to cite this article:
Gueissaz L, Massonnet G. Chemical Analysis of Tire Traces in Traffic Accidents Investigation.J Forensic Sci Med 2015;1:99-108

How to cite this URL:
Gueissaz L, Massonnet G. Chemical Analysis of Tire Traces in Traffic Accidents Investigation. J Forensic Sci Med [serial online] 2015 [cited 2022 Oct 2 ];1:99-108
Available from:

Full Text


Traffic accidents occur every day worldwide. These can be more or less severe in terms of injured persons and/or costs.[1] According to the World Health Organization, traumatisms due to traffic accidents represent the 8th cause of death in the world and the leading cause of death for young people aged from 15 to 29 years old.[2] The number of serious traffic accidents is very high, and a certain percentage of these cases can lead to a forensic investigation in order to help determine what happened.

The aim of the forensic investigation of traffic accidents is to help establish the nature and/or circumstances of the event. This might be done with the purpose of determining the legal responsibilities of each person involved or to provide families, with a reconstruction of the events, to help understand why their relatives were injured or killed.[3] Whatever the reasons for the investigation, the result to be achieved is the same: "Understand what happened."

To reach this goal, a reconstruction of the accident is primordial. In this process, the scene investigation is a crucial step. The examination and comparison of the different traces collected (e.g., paints, polymers, skid marks, etc.) with reference materials of each element involved (e.g., vehicle, traffic sign) is necessary for the reconstruction of the trajectories and the events of the accident like the points of impact between vehicles. The search of all traces on the scene and/or the vehicles for comparison purposes is an essential part in determining relevance and position. When traces are further analyzed chemically, their exploitation helps reconstruct the accident, given the nature, relative position, and source. The analysis of these traces may also show whether or not a suspected vehicle is involved in the accident.

Among the different traces (paints, glass, lamps, oil, brake fluid, etc.) encountered in traffic accidents, tire traces1have a strong potential given that several information can be extracted to help reconstruct the sequence of events. The type, pattern, length, and location of the tire traces on the accident scene can in particular be used to reconstruct the collision points, the minimum speeds before the collision and the trajectories after the collision.[1],[4] However, a primary question concerns the determination of which vehicle is the source of a particular tire trace. Measurements between two tire traces (left/right or front/rear) can be conducted for comparison purposes, respectively with the track or the wheelbase of the suspected vehicles to help answer this question.[5],[6],[7] Nevertheless, this may not be obvious given the quality of traces and lead to errors. Moreover, in some cases only one tire trace can be found. The morphological features of the tire trace can be compared to the ones of the tire treads of the suspected vehicles.[5],[6] However, tire traces left by the abrasion of the tread can be of poor quality in terms of physical features, leading to serious difficulties for a comparison process based on such characteristics. Indeed, tire traces resulting from braking or skidding are more often in the form of an agglomeration of small rubber abrasions.[8] The tread pattern is thus reduced or may even be absent. Chemical analysis of the tire trace and the tread of the suspected tires offers a potential solution. The chemical profiles obtained can be compared to help determine if a suspected tire is the source of a particular trace. The exploitation of the chemical profile of tire traces obtained by pyrolysis coupled to a gas chromatograph with a mass spectrometer as detector (Py-GC/MS) have shown initial promises.[8] This article provides further information and description of tire traces from their general aspect to the morphological features of the rubber particles deposited on the pavement. Additional statistical treatments on the data of the chemical profiles enhance the quality of the information provided. In particular, linear discriminant analysis (LDA) helps in predicting class membership (brand and model) of the tire traces. Comparison methodology and results are thoroughly discussed and provide relevant details to enable an extended comprehension of the utility of the chemical profile of a tire trace.

 Materials and Methods

Study of tread variability


Twelve summer tires of the same size (205/55 R16) were used for this research [Table 1]. Nine brands are represented. There were two tires from Goodyear but of different models (tires 1 and 3) and three tires from Continental with two from the same model (tires 11 and 12). The other tires came from distinct brands (and were thus of distinct models). The 12 tires come from European countries. The numbers in the last column [Table 1] are linked to the inscription of the Department of Transportation, which is a hallmark of USA standard. This marking is a combination of numbers and letters and indicates that the tire meets the applicable safety standards. Since 2000, the last four digits are related to the date of manufacture of the tire. The first two digits indicate the week of manufacture and the next two relate to the year. The tires of the sample set were thus fabricated between the 5th week of 2008 and the 8th week of 2009.{Table 1}

Tires 1 to 10 were provided by the Touring Club Suisse2 (TCS), a Swiss leader in the field of assistance and road rescue. Each year, the TCS completes performance tests on new tires of different brands and models. Tires 1 to 10 were used by the TCS in 2009. Hence, these tires were not new for the present research but still in good condition for driving. Tires 11 and 12 were also still in use by their owner when samples were collected from their tread for this study.

Sample preparation and analysis

To assess the variability within the treads (i.e. intra-variability), several specimens were cut (size ≥ 200 mm 3) from the blocks of the treads. The variability was evaluated according to the width and the circumference of the tread. Tires 1 to 10 were used for the evaluation regarding the width and the specimens were collected in order to cover this width (from one side to the other of the tread). Tires 11 and 12 were used for the circumference and specimens were collected at the center of the tread at each quarter turn.

Small pieces of approximately the same size (<1 mm 3–

representing around 30–35 µg) were then cut from each tire specimen without controlling precisely their shape and thickness despite the fact that the sample size and thickness may have an influence on the repeatability of the pyrograms.[9],[10],[11] These parameters cannot be controlled for tire traces and, rather than controlling these parameters for the tire samples, it was chosen to take their influence into account. For each analytical run, one of the small pieces was introduced into a quartz tube, which was previously half filled with quartz wool. Each tread specimen was analyzed at least in duplicates (i.e. two analytical runs). The tire treads were therefore represented by several replicates.

Instrumentation and method

Pyrolysis was conducted with a resistively heated filament (platinum coil), Pyroprobe Series 5150 using the software version 3.21 from CDS Analytical, Inc., Oxford, PA, USA. The separation and the detection of the pyrolysis products were carried out using a GC-6890N coupled with an MS-5973 network using the software G1701DA MSD ChemStation version D.00.01.27 from Agilent Technologies. The capillary column used was an Agilent HP5-MS (length 30 m, internal diameter 0.32 mm, film 1.0 µm). The optimization of the method can be found in [8], and the details of the method used are given in [Table 2].{Table 2}

Data acquisition and statistical treatment

Eighty-six compounds were selected to compare the analyses [Table 3]. These ones covered the pyrograms from 1.47 to 25.25 min. An attempt of presumed identification of these 86 compounds was performed with the help of NIST05 mass spectra library and by comparison with the literature.[10],[12] A presumed identification is reported in [Table 3] for compounds presenting a match quality equal or superior to 90 with the mass spectra library. Unidentified compounds were named: Unidentified + their retention time.{Table 3}

These compounds were integrated using the software enhanced data analysis MSD ChemStation version D.02.00.275 from Agilent Technologies. The area of the selected target ions was integrated, normalized on the total sum of areas, and pretreated by a fourth root. The chemical profile of each analysis is thus represented by the relative abundances of the 86 selected compounds.

Descriptive statistics

For each tire tread, the relative standard deviations (RSD) were calculated for each compound using all its replicates (RSDintra-variability). These RSD were compared to a threshold of repeatability fixed at 7%. This threshold was determined by previous analyses of reference materials conducted over several weeks with the same Py-GC/MS method as for tires. Compounds showing RSD value below 7% were judged repeatable and then selected for the comparison step. Compounds having RSD values higher than the threshold were considered unrepeatable. However, before rejecting them for further comparisons, their RSD between the 12 tires were computed (RSDinter-variability). These RSD were calculated taking into account all analyses (i.e., 131 analyses). If the RSDinter-variability were higher than the RSDintra-variability, the corresponding compounds were retained for comparison purposes.

Principal component analysis

The preprocessed data of the 12 tires (131 analyses of 86 variables) were simultaneously treated by principal component analysis (PCA) using the software Unscrambler ® X version 10.1 from CAMO software AS (Oslo, Norway). This statistical method simplifies the data by reducing their dimension, which leads to the possibility of studying their structure with the help of graphical representations.[13] The initial variables are transformed into a new set of uncorrelated variables, called latent variables or principal components.[14] The latter are linear combinations of the initial variables and are calculated in such a way that the first components explain most of the variation of the original variables.[15] The scores of the principal components can be plotted by pairs or by three allowing the visualization of the multivariate data. The study of the coefficients of the linear combinations (i.e., the loadings) conducts to highlight the most influential variables.

Study of the tire traces

Production of tire traces: Braking tests and sampling

Braking tests were performed in collaboration with the TCS in order to obtain tire traces produced by the abrasion of the tread. The racetrack in asphalt from the technical center of TCS in Emmen (Switzerland) was used to perform the emergency braking. This was a loop circuit measuring approximately 400 m and the surfacing was dry during the tests. The tests were carried out with an Opel Signum 2.2 DTI, equipped with an anti-lock braking system. As these braking tests were carried out by the TCS, only tires provided by this association were used for the production of tire traces. Thereby tires 1 to 10 [Table 1] were used for the braking tests. One after the other, each of these tires was mounted on the left front wheel of the vehicle to perform the braking tests. The other three wheels were fitted with tires Goodyear EfficientGrip with the same characteristics that tire 1 (i.e., size, country, and date of fabrication).

For each test, an area of the racetrack was selected and delimited with cones. The areas were different for each test to minimize potential background noise. In order to minimize contaminations due to prior brakings to this study, each area was thoroughly brushed by two people prior to the tests.

The vehicle completed three laps of the track at a speed of about 80 km/h to heat up the tires. After this distance, the vehicle did an emergency braking from 80 to 0 km/h with the anti-lock braking system operating. The vehicle was equipped with a system to perform the braking automatically to ensure the same braking conditions for each test. Once stopped, the vehicle stayed in place. A blank sheet of gelatine was then applied on the asphalt behind the left front wheel (position of the tested tire). The gelatine sheet was applied perpendicularly to the direction of the trace to ensure that the width of the trace was completely covered. Finally, gelatine sheets were scanned with a resolution of 600/600 dpi to record the tire traces collected.

Morphological characteristics

The gelatines were observed with a stereomicroscope (Leica M205C, from 32 to 650 times magnification) to characterize the physical aspect of the tire traces and measure the particles making up these traces.

Sample preparation, analysis, and data acquisition

Each tire trace was analyzed 3 times. For each analytical run, one particle was taken from the gelatine with tweezers and then placed into a quartz tube as for the tires. The sampling of the particles was undertaken under a stereomicroscope (Leica M205C) in order to visually check that no gelatine was present on the particle. The same Py-GC/MS method than for the tire samples was used for the traces. Finally, the 86 compounds shown in [Table 3] were integrated for each replicate of the traces, normalized, and pretreated according to the same procedure as for tires.{Table 3}

Comparison with the tire collection

Visual comparison

As a first step, the pyrograms of the traces were visually compared with those of the tire at their origin. A superimposition was performed, and the comparison was based on the overall profile of the pyrograms, the number of detected peaks and their retention times. This qualitative comparison was performed in order to highlight the potential differences between the traces and the tires at their origin.

Linear discriminant analysis

The LDA was performed with the software Unscrambler ® X version 10.1 from CAMO software AS. This statistical method was used to predict the class membership of the traces.

LDA belongs to the supervised classification methods. The aim of these methods is to provide a model which gives the optimal discrimination between several classes (or categories) in terms of predictive performance.[14]

Supervised classification methods establish rules, also called classifiers, by adequately summarizing the multivariate structure of the data, in order to correctly assigning new observations (or objects) to a specific class for which the affiliation to one category is unknown.[16] The aim of the LDA is to find one or several linear functions of the predictive variables. These linear functions are calculated in such a way that the variance is maximized between classes and minimized within each category.[17] The model which gives the smallest number of misclassified objects is searched. The establishment of the linear functions and the evaluation of the model require objects whose class membership is known.

If the number of objects per category is sufficient, it is advised to randomly split the data into two groups: One group for the construction of the model (training data set) and one group for the validation (test data set). The obtained classification model can thus be tested with data that were not used for the construction of the model. This procedure can be conducted several times with different segmentations of the objects to get a mean rate of correct classifications. This process avoids getting too optimistic rates of correct classification.[16]

In this study, the relative abundances (Normalized and pretreated) of the 86 selected compounds were the predictive variables. It was however not possible to directly use those for the calculation of the linear functions because of their large number. Indeed, for LDA computation, the number of predictive variables must be lower than the number of objects per category. A solution consists in proceeding to a prior reduction of the predictive variables, and a PCA is advised.[16] LDA is then applied to the scores of a number of principal components.

 Results and Discussion

Study of the tread variability

Descriptive statistics

[Figure 1] presents the RSD of each compound for the 12 tires of the samples set. Only four compounds, on the 86 considered, have an RSD superior to the repeatability threshold of 7%, and this only for some tires. Thereby, the large majority of the selected compounds are repeatable and present a weak intra-variability. If the four compounds are excluded, the tire tread (circumference and width) can be judged homogenous with the size of the specimens analyzed for the tires of the sample set.{Figure 1}

The four compounds which present a RSD superior to 7% are: n°44 (aniline), n°49 (DL-limonene), n°85 (styrene dimer) and n°86 (unidentified 25.25). The compound n°44 has an RSD superior to the threshold for the tires n°2 and 4. This compound is therefore considered as nonrepeatable and points out that there is an inhomogeneity for this compound given the size of the specimens analyzed for these two tires. All tires present a relatively high RSD for this compound, the interval going from 3.2% (tire 11) to 11.9% (tire 4). These values support that this compound does not seem to be a good candidate for comparison purposes because its variability within a same tire tread is high and as well as for several tires. The aniline was used as an accelerator for the vulcanization process as early as 1906. However, because of its toxicity, the aniline had to be combined for example with carbon disulfide to form the thiocarbanilide which was used as an accelerator in 1907.[18] Other important developments followed, and the 2-mercaptobenzothiazole and the 2-benzothiazole di-sulfide were the first delayed-action accelerators introduced in 1925.[18] Other accelerators have been developed since but, for their majority, their chemical formula contains a nitrogen atom bonded to a benzene ring. Aniline detected in all the tires of the sample set could be explained as a product of pyrolysis of the vulcanization accelerators. Kaminsky and Mennerich [19] have also suggested that the aniline and other aromatic amines probably come from the cure system and other additives.

The RSD for compound n°49 (DL-limonene) is of 8.3% and 16.9% for, respectively tire 7 and tire 8. For this compound, the other tires have values of RSD ranging from 1.6% to 4.2%. This compound is thus not repeatable only for tires 7 and 8. An observation of the data of tires 7 and 8 showed that the relative abundance of this compound varies from one replicate to another and that the high value observed for the RSD is not due to a single analysis (i.e. outlier). These observations support that the tread of the tires 7 and 8 are not homogenous for this compound considering the size of the specimens analyzed.

Tire 3 presents an RSD slightly above 7% for compound n°85 while the RSD of the other tires are below the threshold for this compound. This dimer of styrene is thus nonrepeatable only for tire 3.

Compound n°86 has an RSD above the threshold for tires n°9, 10, 7, 8, 6, and 3 (in descending order of RSD value). An observation of the data showed that the relative abundance of this compound varies widely from one replicate to another for these tires. This compound is thus considered nonrepeatable for these six tires (considering the size of the specimens analyzed) and is not a good candidate for future comparisons.

The RSDs between the 12 tires were computed (RSDinter-variability) for these four compounds. The obtained values (RSDinter-variability for compounds: n°44 [aniline] =10.3%; n°49 [DL-limonene] =41.6%; n°85 [styrene dimer] =8.8%, and n°86 [unidentified 25.25] =11.3%) showed that the variability between the treads for these four compounds were close or largely superior to the RSDintra-variability of the treads which presented a high value for these compounds. It was decided that, even if these four compounds were not repeatable within some treads, they should be kept for further statistical treatments because their variability between the treads of the sample set was higher or equivalent.

Principal component analysis

The first three principal components of the preprocessed data explain more than 80% of the variance, and this result is considered to be good. [Figure 2] presents the scores of the first two principal components plotted against each other for all data of the 12 tires (i.e., 131 analyses). On this figure, the replicates of a same tire tread are of the same color (and symbol) and the number above each replicate refers to the number of the tire [Table 1].{Figure 2}

Each tire has its replicates plotted close to each other and those form groups with a relatively low dispersion, except tire 8 whose replicates are more spread out. These results support that the tire treads investigated can be considered homogeneous even if some variability between the replicates of a same tire (i.e., intra-variability) is observed. The groups of tires are well-separated from each other, excluding two groups. The first is composed by tires 2 and 8, which show an incomplete overlap of their replicates. Nevertheless, these two tires from different brands are completely separated by the third principal component. The second group is formed by tires 11 and 12. Their replicates are completely overlapped, and the use of supplementary principal components failed to separate them. The methodology used could not differentiate these two tires. It can be concluded that these two tires present similar chemical profiles. This result is very interesting given that tires 11 and 12 are from the same brand and model, produced in the same country and were built up within the same week. In contrast, tire 10, which is from the same brand than tires 11 and 12 but of a different model, is well-separated from these two tires. This leads to suppose that the tread composition can vary between models of the same brand, resulting in different chemical profiles. This observation is supported by the complete separation of tires 1 and 3 from the brand Good year but of different models.

Considering the level of the source by the "brand and model," a complete discrimination is obtained using the first three principal components. This result argues that the intra-variability is lower than the inter-variability for the treads of the sample set given this level of the source. It can be inferred that the tires of different brands and/or models of the sample set present different chemical profiles.

The study of the loadings allowed identifying the compounds which have an important influence for the linear combinations of the first two principal components. 10 from 86 compounds used present a great influence: Compounds n°3, 33, 36, 41, 45, 46, 49, 50, 52, and 79. For the third component, eight more compounds have a great influence: Compounds n°44, 71, 72, 76, 80, 81, 83, and 86.

For the first principal component, compounds n°3 (isoprene) and n°49 (DL-limonene) are the most important. These two pyrolysis product come mainly from the polyisoprene (i.e., natural rubber or synthetic polyisoprene). Natural rubber is employed for the tread formulation generally blended with polybutadiene and styrene-butadiene rubber.[20] Tires which are well-separated by the first principal component should thus have different relative abundances of these two compounds (isoprene and DL-limonene) as illustrated in [Figure 3] for tires 1 and 6, which are the most separated by the first principal component.{Figure 3}

For the second principal component, compounds n°52 (indene or benzene, 1-propynyl), n°41 (unidentified 11.67) and n°45 (alpha-methylstyrene) are the most important (in decreasing order of influence). Tires which are well-separated by the second principal component should thus have different relative abundances of these three compounds as illustrated on [Figure 4] for tires 2 and 4, which are well-separated by this principal component.{Figure 4}

Study of the tire traces

Braking tests and sampling

For each of the 10 braking tests with the anti-lock braking system operating, continuous tire traces were visible on the road. This result differs from the literature [21] which reports that vehicles with a system of anti-lock braking deposit interrupted traces. The tire traces were not well-marked but for some of them a general pattern was distinguishable by the naked eye.

For each of the 10 tire traces, the sampling has been successfully carried out using gelatine. For six traces, the general pattern was still visible on the gelatine as shown in [Figure 5] for the traces deposited by tires 3 and 8. For all traces presenting a general pattern, the number of dark lines of each trace corresponded to the number of lines of the blocks from the tread at the source of the trace. This method of sampling was thus suitable to preserve the physical characteristics of the traces.{Figure 5}

Morphological characteristics

Macroscopic observations showed that all tire traces are mainly formed by numerous dark, elongated and rough particles [Figure 6] located at the surface of the gelatine. These particles look like small rubber rolls whose length is <1 mm. The biggest particles presented a size approximately of 0.004 mm 3 and weighted around 15–18 µg. The quantity of particles depends on the trace, but each is composed of at least several hundreds of particles. The number of particles is thus large enough for chemical analyses. The particles were not randomly arranged on the gelatin but showed a high density in some areas. For some traces, these agglomerations of particles form the distinctive parallel lines separated from each other.{Figure 6}

Several other dark particles were also observed on the gelatines, but these were more roughly of cubic or round shapes. These particles were less numerous than the particles of interest and randomly arranged on the entire surface of the gelatines. These particles were likely background from the asphalt itself but are easily distinguished from those of the tire traces under a stereomicroscope.

Comparison with the tires

Visual comparison

The visual comparison between the pyrograms of the traces and the tires at their origin showed that they do not differentiate on the basis of their general profile and the number of detected peaks between 1 and 28 min. However, the traces showed a shift of the retention time (+0.05 to about + 0.10 min) for all the compounds compared to their tire. This shift can be explained by the delay as well as the change in the column since traces were analyzed 8 months after the tires on a new chromatographic column (same make and model). However, the shift is weak and only visible when the pyrograms are superimposed. [Figure 7] shows the pyrograms of tire 8 and its trace in juxtaposition. For a better visualization, only one pyrogram for the tire and for its trace are illustrated on this figure.{Figure 7}

Linear discriminant analysis

Given that tires 1 to 10 were used for the production of the tire traces, a PCA was performed with only the data of these 10 tires (i.e., 115 analyses). The principal components of this PCA were then used to calculate the scores of the traces (i.e., projected scores). The scores of the first three principal components were selected to proceed to the LDA. It was chosen to do the classification at the level of source "brand and model." Indeed, the PCA results clearly showed that the replicates are grouped together in function of the brand and model. The choice to consider the class membership as "brand and model" led to classify the data into 10 categories.

Around two-thirds of the tire data, randomly selected, were used to create the model (i.e., 75 analyses = training data set) and the last third to test it (i.e., 40 analyses = test data set). This procedure was conducted 10 times with different randomized segmentations of the tire data, and the mean rate of correct classification was computed.

Finally, each of the 10 LDA-models was used to predict the class membership of the tire traces (i.e., 30 analyses), which were not used for the construction of the model.

Each training data set led to 100% of correct classifications for the calibration models. A rate of 100% of correct classifications was also obtained for the 10 test data sets (i.e., validation). The created models are thus faultless since no error was encountered during the calibration step and during the validation.

For the traces, an average rate of correct classifications superior to 94% was obtained. This result is considered as good. It means that more than 94% of the replicates of the traces could be correctly assigned to the class membership (i.e., brand and model) of the tire at their origin. All the 10 LDA-models created led to mistakes: At least one analysis and maximum three analyses were misclassified. The error rate thus varies from 3.3% to 10%. By investigating the results, it could be noted that it was always the same analyses of the traces which were misclassified: One analysis of the trace deposited by tire 8 classified in the category of tire 7 and two analyses of traces deposited by tire 6 classified in the category of the tire 9. Thereby, it is not a misclassification of all replicates of the same trace but only a part of the replicates of two traces. This result is interesting because it illustrates the importance of having performed several replicates per trace. For example, for the trace deposited by tire 8, the fact that two of the three analyses performed are classified in a class and the last one in another pushes the experimenter to reconsider the obtained results for this trace and possibly perform additional analyses. The misclassified traces were assigned to tires that are close to their tire source according to the scores of the three first principal components. Thus, the chemical profiles of these tires, although different, are close, and this leads to errors of classification between them.

Finally, it can be concluded that the chemical profile of one trace does not differ from the chemical profile of the tire at its origin but generally differs from the other chemical profiles of the sample set. The LDA can thus be used to help to determine if the chemical profile of a questioned trace is differentiable or not from the chemical profile of a tire suspected to be at the origin of this trace. To apply this technique for real caseworks, LDA-models should be created (and validated) with tires data also including the data of the suspected tire(s) of the case in question. The models could then be used to predict the class membership of the trace(s) collected on the scene. The major drawback of the LDA is that a class membership is always attributed to an unknown object. In other words, a class membership will always be provided to a questioned trace by the model even if the real class membership of this trace is not contained in the model. This is problematic given that the classes cannot be exhaustive. New models of tire come out frequently on the market and a continuous update of the classification models is, therefore, necessary.

Despite this limitation, the results of the LDA-models supported that it was possible with a high degree of confidence to determine the class membership "brand and model" of a questioned trace. This statistical tool can be used to compare chemical profiles but also as an investigative technique to help the investigators which brand and model of the tire should be searched for example in a hit and run cases.


This research studied a number of tires of different brands and models and their traces in order to evaluate their chemical profiles. The aim was to determine whether the chemical profile of a trace differ or not from the chemical profile of its tire and from the chemical profiles of tires of different brands or models. First, the variability of the tires was assessed according to the width and the circumference of the tread with the help of several replicates per tread. The results indicate that some variation, concerning the relative abundance of the 86 selected compounds, exists within each of the treads of the sample set. This variation depends on the compound and the tire considered. However, this variation was rather weak for the 12 tires tested compared to the variation between the tires, and their tread could thus be judged homogenous (in width and in circumference). This is supported by the PCA results which clearly showed that the variation within a tread (i.e. intra-variability) was lower than the variation between treads from tires of different brands and models (i.e., inter-variability). Given that the sample set was small, it is not possible to draw general conclusions (for all tires on the market). Nevertheless, the obtained results allow a conclusion that the key is to take into account this variation for future comparisons with the help of several replicates per analyzed tread.

The tires of the sample set could be fully discriminated regarding the level of source "brand and model" by the PCA. This level of the source was thus considered as class membership for the LDA computations. The results obtained using this statistical technique were faultless for the step of calibration, as well as for the validation for the 10 computations, giving confidence in the results. The constructed models are thus valid for prediction. The obtained results also support that the tires could be completely discriminated. For the traces, the LDA-models provided a mean rate of correct classification superior to 94%. This result is good and highlights the possibility to correctly classify a tire trace given the category "brand and model."

From the presented results, the chemical profile of a tire (and a trace) seems to be dependent on the brand and model but further research on a larger sample set is needed to support this hypothesis.

The findings support that the chemical profiles of the traces are not different from the chemical profiles of the tires at their origin but generally differ from the other tires of the sample set. The results encourage the development of a model for comparing a questioned trace and a suspected tire to determine if their respective chemical profiles are different or not. This model should be directly applicable to the chemical profiles of a case and should not be dependent on other data.


This study would not have been possible without the technical support of the TCS, and the authors are very grateful to this organization for its collaboration and interest in this research. The authors would like to especially thank Anton Keller, Marcel Bachmann, Robert Emmenegger and Stefan Lehmann from the TCS team in Emmen (Switzerland) for their contribution to this project.

Financial support and sponsorship


Conflicts of interest

There are no conflicts of interest.


1Van Kirk DJ. Vehicular Accident Investigation and Reconstruction. Boca Raton, FL: CRC Press, LLC; 2000.
2World Health Organization. Global Status Report on Road Safety 2013: Supporting A Decade of Action. Geneva: World Health Organization; 2013.
3Rudram D. Accident investigation determination of cause: Overview. In: Siegel JA, editor. Encyclopedia of Forensic Sciences. 1st ed. Oxford: Elsevier; 2000. p. 9-16.
4Wang YW, Lin CN. A line-based skid mark segmentation system using image-processing methods. Transp Res Part C Emerg Technol 2008;16:390-409.
5Bodziak WJ. Pattern evidence vehicle tire marks and tire track measurement. In: Siegel JA, editor. Encyclopedia of Forensic Sciences. 1st ed. Oxford: Elsevier; 2000. p. 1228-35.
6Bodziak WJ. Tire Tread and Tire Track Evidence: Recovery and Forensic Examination. Boca Raton, FL: CRC Press; 2008.
7Martin JC, Delémont O, Esseiva P, Jacquat A. Investigation de Scène de Crime: Fixation de L'état des Lieux et Traitement des Traces D'objets.3r d ed. Revue et Augmentée ed. Lausanne: Presses Polytechniques et Universitaires Romandes; 2010.
8Gueissaz L, Massonnet G. Tire traces – Discrimination and classification of pyrolysis-GC/MS profiles. Forensic Sci Int 2013;230:46-57.
9Hancox RN, Lamb GD, Lehrle RS. Sample size dependence in pyrolysis: An embarrassment, or a utility? J Anal Appl Pyrolysis 1991;19:333-47.
10Moldoveanu SC. Analytical Pyrolysis of Synthetic Organic Polymers. Amsterdam: Elsevier; 2005.
11Wampler TP. Applied Pyrolysis Handbook. 2nd ed. Boca Raton, FL: CRC Press, Taylor and Francis; 2007.
12Tsuge S, Ohtani H, Watanabe C. Pyrolysis-GC/MS Data Book of Synthetic Polymers: Pyrograms, Thermograms and MS of Pyrolyzates. Amsterdam: Elsevier; 2011.
13Brereton RG. Applied Chemometrics for Scientists. Chichester, England: John Wiley and Sons; 2007.
14Wehrens R. Chemometrics with R: Multivariate Data Analysis in the Natural Sciences and Life Sciences. Berlin, Heidelberg: Springer; 2011.
15Everitt B, Hothorn T. An Introduction to Applied Multivariate Analysis with R. New York: Springer Science+Business Media, LCC; 2011.
16Varmuza K, Filzmoser P. Introduction to Multivariate Statistical Analysis in Chemometrics. Boca Raton, FL: CRC Press; 2009.
17Otto M. Chemometrics: Statistics and Computer Application in Analytical Chemistry. 2nd ed. Weinheim: Wiley-VCH; 2007.
18Coran AY. Vulcanization. In: Mark JE, editor. The Science and Technology of Rubber. 3rd ed. Amsterdam: Elsevier, Academic Press; 2005. p. 321-66.
19Kaminsky W, Mennerich C. Pyrolysis of synthetic tire rubber in a fluidised-bed reactor to yield 1, 3-butadiene, styrene and carbon black. J Anal Appl Pyrolysis 2001;58-59:803-11.
20Rodgers B, Waddell W. Tire engineering. In: Mark JE, editor. The Science and Technology of Rubber. 3rd ed. Amsterdam: Elsevier, Academic Press; 2005a. p. 619-61.
21Noon RK. Forensic Engineering Investigation. Boca Raton, FL: CRC Press; 2001.