Soil Comparison Casework and Statistical Inference

Forensic comparison is not simply a measurement. It is a structured experiment designed to answer a specific question about the provenance of a questioned sample. The experimental design determines the scope and the limits of the conclusion, and poor design produces results that cannot support a reliable opinion regardless of how technically precise the measurements are.

Three sample classes. Every forensic soil comparison involves three logically distinct sample classes. The questioned sample (Q) is the soil of unknown origin collected from the evidence item, whether a shoe, a tyre, a tool, a vehicle undercarriage, or a victim's clothing. The known sample (K) is a reference sample collected from the proposed source location, the scene of the alleged crime, the grave, the vehicle exclusion zone, or wherever the prosecution's theory places the suspect. The control sample (C) is a reference sample from a location that is not the proposed source, collected in the same way and from the same geographic region, to characterise the local background variability against which the comparison must be made.

The control sample is the most frequently omitted component in inadequate forensic soil comparisons. Without a control, the analyst cannot say whether the similarity between Q and K is distinctive or whether most soils in the region share the same properties. A comparison that reports "Q matches K in colour, texture, and mineralogy" without establishing that local background soils are different from K is not informative: it says the questioned soil could have come from the scene, but not whether the same result would have been obtained from any soil within a wide radius. The OSAC Trace Evidence Subcommittee draft standard (2023) and the ENFSI ENG-FG1 guideline both require documentation of at least three background control samples from the scene perimeter.

Scene sampling design. The known sample collection at the scene is not trivial. Soils can vary substantially over short distances, particularly in disturbed landscapes (building sites, gardens, allotments), along watercourses (where sediment transport creates lateral variability), and at terrain boundaries (where two geological units meet). The FBI Soil Examination Unit recommends collecting a minimum of three to five known samples from the scene at distances of 1 m, 3 m, and 10 m from the specific deposition location, to characterise the natural within-scene variability. This spatial variability data becomes part of the statistical comparison: the analyst asks whether Q falls within the range of within-scene variability defined by the K samples, or outside it.

Transfer and persistence. The comparison design must also account for the forensic context of transfer. Soil transferred from a scene to a shoe during a single contact will be dominated by A horizon surface material. Soil from a grave excavation will contain mixed horizon material. Soil from a vehicle tyre may have been mixed from multiple traversals of different locations. The analyst should document the expected transfer mechanism and its likely effect on sample composition when presenting the comparison opinion. Transfer physics are discussed in the ASTM E2937 standard for trace evidence, and the specific soil-transfer literature includes published studies on shoe-sole retention by Croft and Pye (2004) in Forensic Science International, which estimated that soil persists on footwear at forensically useful concentrations for 20-200 m of walking depending on substrate and footwear type.

Raymond Murray's involvement in forensic geology began with a 1969 case in Michigan that would become the foundational casework anchor for the field. Murray, then a geology professor at the University of Montana, was asked to compare soil samples in the investigation of John Norman Collins, convicted of the murder of seven young women near Ann Arbor and Ypsilanti. Soil from the basement of Collins's uncle's house, mixed with human hair and what appeared to be blood, was compared against soil from a known scene location. Murray's analysis showed that the basement soil was consistent with the known location soil in colour, texture, and mineral assemblage, while being distinguishable from background soils in the area. Collins was convicted. The case established that systematic, tiered soil comparison by a qualified geologist could produce court-admissible evidence.

The 1992 framework. Murray and Tedrow's 1992 book codified the comparison methodology that had evolved from that case and from subsequent work. The framework is organised into four principles that remain the basis for US forensic soil examination:

First, a soil comparison is always a probability statement, not a certainty. Two samples can be "consistent with a common origin" but can never be proven to have come from the same location by physical and chemical analysis alone, because the same combination of properties might theoretically occur at more than one location. The analyst's job is to quantify how unusual the match is.

Second, the probability statement requires a reference distribution. Without knowing how often the observed combination of properties occurs in soils across the relevant geographic area, the analyst cannot say whether the match is common or rare. The FBI Soil Examination Unit maintains a reference database of soil profiles from across the United States, cross-indexed to USDA Soil Taxonomy classifications and to published NCSS county soil surveys, to support this reference distribution.

Third, the comparison must be sequential and tiered. Cheap, fast techniques that can eliminate a match should be applied first; expensive, time-consuming techniques that can confirm or quantify a match are applied only when elimination fails. The tiered workflow prevents wasting resources on techniques that add little to an already-clear outcome.

Fourth, the opinion must be stated in terms that a fact-finder can understand and evaluate. The forensic geologist is not the trier of fact. Their role is to provide the probability that the observed data would occur if the questioned and known samples came from the same source compared to the probability that the observed data would occur if they came from different sources, the likelihood ratio that anchors the Bayesian inference framework now used in most forensic science reporting.

FBI Soil Examination Unit protocols. The FBI Soil Examination Unit, housed at the FBI Laboratory in Quantico, Virginia, has published its soil examination procedures in the SWGMAT guidelines (Scientific Working Group for Materials Analysis, now superseded by OSAC) and through individual expert-witness testimony in federal cases. The unit uses all four tiers of the Murray-Tedrow framework: colour and gross morphology, particle size and density gradient, PLM and XRD mineralogy, and palynology when indicated. Their database of reference soils from US federal lands and from internationally sourced reference collections supports the comparative frequency assessments required for probability statements.

The European Network of Forensic Science Institutes (ENFSI) published the ENG-FG1 European guidelines for forensic soil examination in 2018 through its Expert Working Group on Forensic Geology. The guideline reflects European forensic-science practice, which is more explicitly Bayesian in its reporting culture than the US tradition, partly because of the influence of the Forensic Science International paper by Curran, Hicks, and Buckleton (2000) on the likelihood ratio framework and of the European Court of Human Rights judgments requiring explicit uncertainty quantification in forensic opinions.

Core requirements. The ENFSI ENG-FG1 guideline requires: (1) a defined question framework in terms of competing propositions (the prosecution proposition: the questioned soil came from scene X; the defence proposition: the questioned soil came from some other location); (2) documentation of the specific comparison techniques applied and their sequence; (3) a qualitative or quantitative expression of the strength of evidence as a likelihood ratio; and (4) a statement of the limitations of the evidence and the assumptions on which the opinion depends.

WRB terminology requirement. All soil descriptions in ENFSI-compliant reports must use World Reference Base (WRB) nomenclature, which is the shared European vocabulary. Reports that use only national classification terms (UK Soil Survey classification, German AG Boden system, French AFNOR system) without WRB equivalents are non-compliant. This requirement reflects the multi-national character of European forensic cooperation, where a report produced in a German laboratory may be read by a French judge and assessed by a Swedish expert witness.

ISO 17025 accreditation. The ENFSI guideline, like all ENFSI guidelines, requires that the comparison laboratory operate under an ISO 17025 quality-management system with a validated soil examination scope. This means that the specific methods used (particular XRD instruments, specific hydrometer procedures, defined pollen-counting protocols) must be documented in method validation records that demonstrate their measurement uncertainty, their reproducibility within the laboratory, and their reproducibility across laboratories in proficiency testing exercises. The ENFSI Forensic Geology EWG organises biennial proficiency tests for European forensic geology laboratories.

UK application. In the UK, forensic soil evidence is admitted in Crown Court proceedings under the expert-evidence rules of the Criminal Procedure Rules Part 19. Since the closure of the Forensic Science Service in 2012, soil forensic analysis in England and Wales is provided primarily by Eurofins Forensics (formerly LGC Forensics), Cellmark Forensic Services, and by independent consultants, many of whom are academic geologists at universities. The UK Forensic Science Regulator's Code of Practice (2021) applies to all of these providers and requires ISO 17025 compliance consistent with ENFSI guideline requirements. In Scotland, the Crown Office and Procurator Fiscal Service commissions soil analysis through the University of Glasgow's School of Geographical and Earth Sciences when specialist casework arises.

The quantitative comparison of two soil samples using multiple physical and chemical parameters is inherently a multivariate problem. When the questioned and known samples are each characterised by, for example, ten measurements (three colour parameters, three particle-size percentiles, one organic-matter content, one density-gradient peak count, two XRD mineral-ratio indices), each sample is a point in a ten-dimensional data space. The question of whether the two points are the same, and whether their proximity is unusual, requires multivariate statistical analysis.

Principal Component Analysis (PCA). PCA is a dimensionality-reduction technique that transforms a set of correlated measurements into a smaller number of uncorrelated components (principal components) that capture the dominant axes of variation in a reference dataset. When applied to a reference database of known soils, PCA produces a map of the chemical-physical space occupied by those soils, with clusters of similar soils appearing as dense regions in the two- or three-component plot. A questioned sample can then be projected onto the same map to visualise where it falls relative to the reference population and relative to the known sample. If Q and K project to closely adjacent positions in a region that few reference samples occupy, the case for a common origin is strengthened. If Q projects far from K, the evidence supports different origins.

PCA for forensic soil comparison was pioneered by Pye and Blott (2004) at University College London's UCL Department of Geography, whose work on multivariate soil characterisation is the most cited academic reference in UK forensic geology. Their software toolset, described in "SoilFit" publications, applies PCA and discriminant analysis to XRD, laser-diffraction, and colour data simultaneously. The ENFSI ENG-FG1 guideline recommends PCA as the standard statistical approach when multiple quantitative measurements are available.

Mahalanobis distance. The Mahalanobis distance is a multivariate generalisation of the standard deviation that accounts for the correlation structure of the data. It measures the distance between a data point and the centre of a reference distribution in units of the distribution's natural variability, correcting for the fact that measurements in different dimensions may be correlated. A Mahalanobis distance of zero means the questioned point is at the mean of the reference distribution; a distance of two means it is two "standard deviations" away in multivariate space, with each standard deviation scaled to the joint covariance of all variables. The Mahalanobis distance is used in forensic soil comparison to quantify how unusual a particular Q-K proximity is: if Q and K are close to each other but both far from the reference population mean, the match is more significant than if Q and K are close but surrounded by many similar reference samples.

The FBI Soil Examination Unit has used Mahalanobis distance calculations in high-profile cases where the quantitative data (XRD mineral ratios, laser-diffraction size parameters, LOI) permitted their application. The ENFSI ENG-FG1 guideline recommends documenting the multivariate distance metric used and the reference database on which it was calculated.

Soil mineral ratio fingerprinting. A practical variant of multivariate comparison that does not require full PCA is mineral-ratio fingerprinting. The analyst calculates the ratio of two or more minerals in the XRD pattern (e.g., quartz/feldspar ratio, quartz/(quartz+calcite) ratio, clay mineral diversity index) for both the questioned and known samples and for the background control samples. These ratios are plotted as simple scatter plots or bar charts. If Q and K plot together in a region that is separated from the background controls, the evidence for a common origin is visually apparent and can be quantified by computing a one-dimensional Mahalanobis distance (equivalent to a z-score) for each ratio.

The likelihood ratio statement. The formal output of a probabilistic forensic soil comparison is the likelihood ratio (LR), defined as the probability of the observed data given the prosecution's proposition (Q came from scene K) divided by the probability of the observed data given the defence's proposition (Q came from some other location). An LR of 100 means the observed data are 100 times more probable if Q came from the scene than if it came from a random alternative location. An LR of 0.01 means the data are 100 times more probable if Q came from somewhere else. The verbal scale adopted by the ENFSI EWG on Forensic Geology converts LR values to statements: LR 1-10 "limited support," LR 10-100 "moderate support," LR 100-1000 "strong support," LR above 1000 "very strong support." This scale parallels the verbal probability scales used by ENFSI across other evidence types (DNA, glass, fibres).

Quantifying the LR numerically for soil evidence is difficult in practice because the reference database must be large and representative for the frequency estimates to be reliable. The FBI reference database covers US soils adequately; the ENFSI EWG maintains a European reference collection. For casework outside these covered regions, the analyst must acknowledge that the LR is a qualitative estimate rather than a calculated value, and must state the assumptions explicitly.

Forensic soil comparison has contributed to investigations across every continent and across a range of case types: homicide, rape, burglary, hit-and-run, agricultural fraud, environmental crime, and international crimes against humanity. Three anchor cases illustrate different aspects of the comparison methodology.

John Norman Collins (Michigan, 1969). The Collins case, already introduced in Section 2, is the foundational US forensic geology casework reference. Collins was convicted of the murder of Karen Sue Beineman in 1969. Soil and hair fragments found in the basement of his uncle's house were compared by Murray against soil from the burial scene and against soil from other areas. Murray's testimony that the basement soil was consistent with the scene soil and inconsistent with the background soil established the evidential framework that the FBI Soil Examination Unit subsequently formalised. Collins was convicted and received a life sentence. The case is cited in every major US forensic geology text, including Murray and Tedrow (1992, 2017), as establishing the admissibility of systematic soil comparison testimony under the pre-Daubert Frye standard.

The Lindbergh kidnapping (New Jersey, 1932). The 1932 kidnapping and murder of Charles Lindbergh's infant son is the earliest well-documented case in which physical evidence, including soil on a ransom ladder, was systematically examined for geographic provenance. Arthur Koehler's wood examination is the more famous forensic evidence from this case, but soil analysis also contributed: soil from the estate driveway was compared against soil on the defendant Bruno Hauptmann's vehicle and shoes. The methodology was unsystematic by modern standards, but the case established that soil transfer from a crime scene to a suspect's vehicle or person was a forensically recoverable evidential link, decades before the Murray-Tedrow framework formalised the comparison protocol. The Lindbergh case is cited in historical reviews of forensic geology as the template from which subsequent practice developed.

Ujjain casework (India, 2013). The Ujjain Bharat Bandh-related casework of 2013 involved arson and violence in which multiple crime scenes were spread across a semi-urban landscape with varied soil types reflecting the underlying Deccan Trap basalt (Vertisol black cotton soil) and alluvial deposits (Inceptisol). Soil comparison evidence was used by the FSL Madhya Pradesh to associate a suspect vehicle with a specific scene, using Munsell colour and texture analysis. The case is reported in the peer-reviewed literature by Sharma and Rishi (2014) in the Indian Journal of Forensic Sciences, and it represents one of the few published Indian casework reports documenting a fully tiered soil comparison methodology. The evidence was admitted in the sessions court without methodological challenge, a pattern typical of Indian criminal proceedings at that time, though the report itself met the standard documentation requirements of the DFSS SOP.

Srebrenica and ICTY proceedings. The most internationally significant forensic geology casework of the late twentieth century involved the linkage of secondary mass graves to primary graves in the investigation of the 1995 Srebrenica massacre. Geologists and soil scientists working for the ICTY and the ICMP used soil comparison, pollen analysis, diatom analysis, and material transfer evidence to demonstrate that bodies had been moved from primary graves to secondary graves in an attempt to conceal the scale of the killings. The soil evidence contributed to the conviction of Ratko Mladic and other Srebrenica perpetrators by the ICTY. This body of casework, documented in academic publications by Keil and colleagues (2004, 2009) in the Journal of Forensic Sciences and by Williams and colleagues, established the international standard for forensic geology in mass-atrocity investigations and was referenced in the ENFSI ENG-FG1 guideline as a precedent for the level of methodological rigour required in high-stakes casework.

Forensic soil comparison faces specific cross-examination challenges in courts across jurisdictions, and the forensic scientist who presents this evidence must be prepared to address them honestly.

The frequency of matching soils. The most fundamental challenge is: how often does the observed combination of properties occur in soils that are not the scene soil? Without a large, well-characterised reference database covering the relevant geographic area, the analyst cannot answer this question quantitatively. US federal courts have, in several post-Daubert cases, required the soil examiner to state the known error rate of the comparison method or to acknowledge that it is not formally quantified. The FBI's OSAC draft standard (2023) addresses this by requiring documentation of the reference database used and the basis for any frequency estimate.

Within-scene variability. If the soil at the scene shows significant compositional variability over small distances (as it does in disturbed landscapes, near drainage features, or at geological contacts), a match between Q and K may be unremarkable because most soils near the scene would also match K. The defence expert can demonstrate this by obtaining additional samples from the scene perimeter and showing that many of them also match the questioned sample. The prosecution's analyst must have collected background control samples to counter this argument; if they did not, the comparison is methodologically incomplete.

Transfer and secondary transfer. A suspect may argue that the soil on their footwear came not from the alleged scene but from a secondary transfer via a common indoor location (a mudroom, a vehicle interior, shared flooring). The forensic analyst must address whether the questioned sample composition is distinctive enough to exclude secondary transfer origins, which requires knowing the composition of candidate secondary-transfer soils. In practice, this is rarely possible to exclude definitively, and the analyst must acknowledge the uncertainty.

The Daubert challenge to palynology. In the United States, forensic palynology has faced Daubert challenges in several jurisdictions where opposing counsel argues that the technique lacks a formally validated error rate and that the reference pollen library is not sufficiently comprehensive. The Texas Court of Criminal Appeals discussed the scientific basis of palynology in Campbell v. State (2012) without reaching a definitive ruling on admissibility. The UK, New Zealand, and Australia have generally admitted palynology under their respective evidence acts without the same methodological scrutiny applied under Daubert. The OSAC Trace Evidence Subcommittee is developing a standard for forensic palynology that, when approved, will provide the Daubert-compliant validation basis currently missing.

In India, the admissibility framework under BSA 2023 Section 39 (expert opinion) is relatively tolerant of new forensic methods because Indian courts have not yet adopted a Daubert-equivalent analytical framework. Expert testimony on soil comparison, including palynology, is generally admitted if the witness has suitable qualifications and the opinion is internally consistent. The Supreme Court of India has, in judgments including Ram Chandra Singh v. Savitri Devi (2003) and more recent decisions on DNA evidence, emphasised that the credibility and methodology of expert witnesses should be scrutinised more rigorously, a trend that will eventually reach forensic geology as the profession matures.

Forensic soil comparison reports across jurisdictions share a common structural requirement: the report must be reproducible. Any qualified forensic geologist reading the report should be able to identify the methods used, understand the basis for the comparison criteria, and assess whether the opinion follows from the data. This reproducibility requirement is the operational definition of scientific validity in the forensic context.

Required elements under multiple standards. The OSAC Trace Evidence Subcommittee draft standard (2023) for forensic soil examination requires: sample description (provenance, packaging, condition on receipt), preparation procedures (drying, disaggregation, size fractionation), analysis procedures (each technique applied with reference to the method standard used), comparison criteria (what threshold was applied to determine consistency or inconsistency), comparison results (the specific values obtained for each technique), and the opinion statement (consistent with / inconsistent with a common source, with a verbal likelihood statement or quantitative LR where supported).

The UK Forensic Science Regulator's Code of Practice (2021) requires additionally that the report identify any limitations that would affect the interpretation, including chain-of-custody gaps, degraded sample quality, or limitations in the reference database. The Criminal Procedure Rules Part 19.4 requires expert witnesses in UK criminal proceedings to state whether their opinion is supported by the data, to identify the range of opinion in the field, and to state anything that might adversely affect the reliability of their opinion.

The ENFSI ENG-FG1 guideline requires a likelihood ratio statement, even if only qualitative, rather than a simple "match" or "no match" conclusion. The verbal scale (limited / moderate / strong / very strong support for the prosecution proposition) is now standard in European forensic geology reports.

In Australia, the Evidence Act 1995 and the Australian Law Reform Commission (ALRC) "Evidence Law" report on expert witnesses have driven a similar shift toward the likelihood ratio framework. The ANZFSS Standards and Guidelines for Forensic Geological Analysis (2016) recommend the ENFSI ENG-FG1 verbal scale as the reporting standard for Australian practitioners.

The Indian forensic community is in transition. The DFSS has issued internal SOPs for soil examination that require methodology documentation comparable to the OSAC standard, but these SOPs are not publicly available and are not yet consistently applied across all FSL and CFSL facilities. The BSA 2023 expert-opinion framework does not require a likelihood ratio statement, but individual court judgments increasingly expect expert witnesses to state the basis for their opinion with greater precision than was required under the IEA 1872 regime.