The forensic pedigree of soil evidence is older than most people realize. In 1904, the German forensic pioneer Georg Popp used soil recovered from the pant cuffs of a suspect in the Eva Disch murder case, comparing it to soil from the bean field where the victim was found and to soil along a pathway leading from the suspect’s home to the crime scene. Confronted with the soil analysis and other evidence, the suspect Karl Laubach confessed. What has changed since Popp’s era is not the principle but the technology: where early investigators relied on visual comparison of color and texture, modern analysts can interrogate the molecular bonds within soil particles and let algorithms do the discriminating.

FTIR spectroscopy works by shining infrared light across a sample and measuring which wavelengths are absorbed. Every chemical bond vibrates at characteristic frequencies in the micron range, and when those vibrations match the incoming infrared radiation, the energy is absorbed and recorded as a dip in the spectrum. Soil is a complex mixture of organic matter, sand, clay, minerals, and salts, and it may also contain heavy metals introduced by mining, tanning, and other industrial activities depending on the pH and land use. Compounds such as humic and fulvic acids in the natural organic matter, along with clays that carry permanent dipoles, all interact with the infrared beam. The result is a spectrum packed with absorptions at specific wavenumber values that can be assigned to functional groups using correlation charts, from the hydroxyl stretches of clay minerals near 3,700 wavenumbers to the silicate stretches below 1,000.

The technique’s appeal for forensic work is practical as much as scientific. FTIR is non-destructive, rapid, and requires little to no sample preparation, making it ideal for analyzing the trace quantities of soil that typically survive on footwear, tools, or vehicle mats. Eleven of the fourteen studies reviewed used attenuated total reflectance FTIR, in which the sample is pressed against a crystal and the evanescent wave penetrates only a few microns, while three older studies prepared samples as potassium bromide pellets. One study compared ATR with diffuse reflectance infrared Fourier transform spectroscopy and found the DRIFTS approach yielded notably lower classification error, 27 percent versus 57 percent with ATR data under linear discriminant analysis, suggesting that how the spectrum is collected matters as much as what is collected.

The geographic scope of the reviewed research is genuinely global. Studies analyzed soils from Australia, India, the United States, the United Kingdom, China, Turkey, Brazil, and North Macedonia, with sample sizes ranging from five to several hundred. The sampling protocols varied enormously. Chauhan and colleagues collected ten samples from within a 10 by 10 meter area in eighteen cities across five Indian states, while de Caritat and colleagues analyzed 324 samples from 268 locations across North Canberra, collecting five samples from each 1 by 1 meter plot and combining them. Sangwan’s team gathered 232 samples across 29 locations in Haryana, India, and e Silva’s group collected 284 samples across the Curitiba Metropolitan Region of Brazil. Sampling depths ranged from the top half inch to a full meter, and some studies deliberately collected both surface and sub-surface samples, reasoning that surface soil is what embeds in shoe treads while deeper soil might indicate digging.

Sample preparation is where the field shows its cracks. Drying temperatures ranged from gentle air drying over several days to ovens set at 110 degrees Celsius, and one study heated fine fractions at 650 degrees Celsius in a muffle furnace to strip away organic material entirely. Sieving mesh sizes ranged from 38 micrometers to 2.8 millimeters, and some groups ground and homogenized samples while others did not. Newland’s team in Western Australia went further, hand-picking quartz grains from sieved samples under a microscope and subjecting them to sonication, centrifugation, and resuspension before drying them into thin films. These choices are not trivial: high temperatures and long drying times can alter the constituency of the organic portion of soil, and the review warns that such alterations could compromise the very spectral features that make each soil unique.

Once spectra are collected, typically across the 4,000 to 400 wavenumber range at 4 wavenumber resolution with 64 scans, the real analytical power emerges from chemometrics. Twelve of the fifteen papers applied statistical methods rather than relying on visual comparison alone. Principal component analysis was the most common tool, used in eleven studies, often combined with linear discriminant analysis or partial least squares discriminant analysis. Preprocessing steps such as normalization, baseline correction, and smoothing were applied before modeling, and several groups removed spectral regions associated with ATR crystal interference or deemed non-relevant. The review notes that statistical analysis removes interpreter variation and leads to higher discriminating power, whereas visual inspection, though sometimes successful, depends on the skill and training of the examiner and can be unreliable.

The reported results are remarkable in places. Woods and colleagues achieved 99.7 percent discrimination across 29 Australian samples using PCA on FTIR spectra and their first derivatives. Idrizi’s team in North Macedonia reported 96 to 100 percent classification accuracy with supervised self-organizing maps, and, notably, twelve samples collected three years after the initial sampling were correctly grouped with their corresponding urban parks, suggesting the spectral fingerprints are stable over time. Newland’s mock casework study in Perth achieved 100 percent classification accuracy when matching a simulated suspect-recovered soil sample to the crime scene using PCA-LDA. Chauhan’s team in India reported 99.35 percent discriminating power for surface samples through visual analysis and 100 percent accuracy for sub-surface samples with LDA models. Sangwan’s group in Haryana reported 100 percent accuracy for surface samples and 98.85 percent for sub-surface samples, with blind samples classified correctly in most cases.

But the picture is not uniformly rosy. De Caritat’s provenancing study across North Canberra found that FTIR alone achieved only 26.6 percent average accuracy in predicting the origins of blind samples under a simultaneous multivariate approach, though combining FTIR with magnetic susceptibility raised accuracy to 61.3 percent and precision to 80.7 percent under a sequential approach. Two studies, one in Turkey near the Ömerli Dam and one in Brazil’s Curitiba region, concluded that FTIR did not provide sufficient information for discrimination on its own. Several groups recommended pairing FTIR with complementary techniques such as X-ray diffraction, X-ray fluorescence, or laser-induced breakdown spectroscopy. The review observes that discrimination accuracy improved when the locations in question were more varied, spanning sand, clay, and loam across short and long geographic distances, and that well-developed databases are expected to improve prediction accuracy further.

The review’s authors close with a call that reads like a roadmap for the field’s next decade: a standardized and consensus approach to sampling, preprocessing, and data collection would accelerate comparisons across laboratories worldwide. They envision an interlaboratory study in which all participants follow a specified procedure and compile results into a searchable, GPS-mapped database, allowing broad conclusions about any sample set and clarifying whether surface and sub-surface soils can be differentiated in a given region. The standard FTIR recording range was most frequently 4,000 to 400 wavenumbers at 4 wavenumber resolution with 64 scans run in triplicate, and the authors suggest these settings as a baseline. If that consensus emerges, the humble dirt on a suspect’s shoes could become one of forensic science’s most powerful and objective witnesses, capable of placing a person at a specific patch of earth with a confidence that no visual inspection could ever match.

Subject of Research: FTIR spectroscopic analysis of soil in forensic science

Article Title: FTIR spectroscopic analysis of soil in forensic science

Article References: Force, J., & Elkins, K. M. (2025). FTIR spectroscopic analysis of soil in forensic science. Frontiers in Analytical Science, 5, Article 1716867. https://doi.org/10.3389/frans.2025.1716867

Image Credits: AI Generated

DOI: 10.3389/frans.2025.1716867

Keywords: FTIR, spectroscopic, analysis, soil, forensic, science, scientific research