Distinguishing Hemp from Marijuana by Mid-Infrared Spectroscopy

Published on: 
Cannabis Science and Technology, July/August 2020 , Volume 3, Issue 6
Pages: 24-38

A novel mid-infrared spectrometer that measures total THC in dried, ground cannabis plant material in 2 min is discussed as well as the applications of this analyzer for law enforcement, forensic laboratories, and regulatory agencies.

Federal law defines legal hemp as cannabis plant material that contains not more than 0.3% dry weight total tetrahydrocannabinol (THC). Because of confusion over the law, a number of false arrests and seizures of legal hemp have occurred. Additionally, the United States Department of Agriculture recommends about one sample per acre of hemp should be tested. This means regulatory agencies may be swamped with samples come harvest season and might have difficulty keeping up with the demand. There is a need then for law enforcement and regulatory agencies to measure total THC in cannabis plant material to discriminate between legal hemp and illegal marijuana accurately, quickly, and easily to alleviate these problems. We have developed a novel mid-infrared spectrometer that measures total THC in dried, ground cannabis plant material in 2 min. By applying a unique total THC hierarchical classification algorithm to the data, high total THC cannabis is sorted from low THC cannabis with a success rate of 99.4% based on a challenge set of 491 samples. The algorithm distinguishes low total THC marijuana from hemp with a success rate of 95.1% based on a challenge set of 284 samples. The applications of this analyzer for law enforcement, forensic laboratories, and regulatory agencies are discussed.

For the purposes of this paper these terms will have the following meanings.

Cannabis: Plants of genus cannabis including the species Cannabis Sativa, Cannabis Indica, and Cannabis Ruderalis.

Cannabis plant material: The dried, ground buds, flowers, stems, stalks, leaves, and seeds of genus Cannabis plants.

THCA: Tetrahydrocannabinolic acid

THC: Δ9-tetrahydrocannabinol

Total THC: 0.877*(THCA) + THC

High THC cannabis: Cannabis plant material containing equal to or more than 5% dry weight percent total THC.

Low THC cannabis: Cannabis plant material containing less than 5% dry weight percent total THC.

Marijuana: Cannabis plant material containing more than 0.3% dry weight percent total THC as defined by Federal law (1,2).

Hemp: Cannabis plant material containing 0.3% or less dry weight percent total THC as defined by Federal law (1,2).

Weight %, Wt. %, and wt.%: The weight percent of an analyte in a sample.

Statement of the Problem

The 2018 United States Farm Bill as passed by the United States Congress (1) and a subsequent Interim Final Rule as promulgated by the United States Department of Agriculture (USDA) on October 29, 2019 (2) makes it legal to grow and possess hemp in the United States so long as the total THC content by dry weight does not exceed 0.3%. Undried or “wet” hemp is not analyzed because the law clearly states that dry hemp must be analyzed (1,2). Additionally, drying hemp changes cannabinoid concentrations, and we are not aware of any peer-reviewed literature that correlates the composition of wet hemp with dried hemp. Until this happens, the industry should continue to test dry hemp only.

Since passage of the Farm Bill, several legal hemp shipments have been seized by law enforcement around the country under the mistaken assumption that the material was illegal marijuana (3–7). Places where this has happened include Idaho (3), Oklahoma (4), New York City (5), Texas (6), and Colorado (7). The latter is surprising since medicinal and recreational marijuana are legal in Colorado. However, one must still possess the proper paperwork and licenses to transport hemp and marijuana in that state (7). As a result of these seizures, innocent people have spent time in jail, thousands of dollars of valuable hemp have been seized, bad press for law enforcement has ensued, and lawsuits have been filed (3). Another outcome of the change in the law is that forensic laboratories are now being swamped with cannabis plant material samples, many of which are legal hemp, tying up resources and spending money that is better spent elsewhere.

The law and interim final rule task state departments of agriculture with developing programs for hemp testing (1,2). The interim final rule contains sample calculations that recommend that about one cutting per acre of hemp be tested (2). This means come harvest season each state must test thousands of samples during a short harvest season. We fear that many states may not have the testing capacity to perform the Federally recommended amount of testing.

There is a need then for law enforcement to analyze seized cannabis plant material in the field to prevent false arrests, for forensic laboratories to weed out nonprosecutable cases to help streamline their operations, and for state departments of agriculture to increase their cannabis testing capacity so hemp farmers receive results in a timely fashion. Some current testing technologies are nonquantitative and hence not congruent with the law since they do not produce a total THC value. Other technologies produce a total THC value, but require expensive consumables, significant sample preparation, skilled operators, and are slow with a high cost per analysis. Ideally, then there would exist a method for distinguishing hemp from marijuana that is fast, easy-to-use, portable, inexpensive, and accurate. Our progress in using mid-infrared (mid-IR) spectroscopy and a unique hierarchical total THC classification algorithm towards this goal is discussed.

Criteria for Evaluating Chemical Analysis Methods

Figure 1 (click to enlarge) shows the “Golden Rectangle of Chemical Analysis,” four figures of merit that should be used when evaluating any analytical chemical method.

The criteria listed are accuracy, speed, cost, and the ability to perform representative sampling. Accuracy is a measure of how far away an analysis is from its true value (8). Speed includes the time to prepare a sample and how long it takes for an instrument to analyze it. Cost includes the initial cost of the instrument, the cost to analyze each sample, and the cost of maintenance and upkeep of the system. The representative sampling criterion comes into play when analyzing inhomogeneous substances such as cannabis plant material. The proper way to analyze these materials is to collect many random samples, analyze them, and report the average result with an appropriate margin of error (9,10). This criterion dovetails off the others since for representative sampling a technique must be fast enough and inexpensive enough to allow the requisite number of samples to be analyzed at a realistic cost in a reasonable amount of time.

In a perfect world, an analytical method would be accurate, fast, low cost, and allow for representative sampling. The reality is that there are trade-offs between these criteria. For example, the most accurate methods, such as typical laboratory analyses, can be slow and expensive. Field analyzers, while perhaps not as accurate, may be faster, cheaper, and more readily allow representative sampling than laboratory methods. Keeping these criteria in mind, we will review the technologies currently available for distinguishing hemp from marijuana.

Technologies for Distinguishing Hemp and Marijuana

Nonquantitative Technologies

For cannabis plant material the presence of THC is legal so long as it contains less than 0.3% total THC (1,2). This means that for a chemical analysis technique to be used by regulatory and law enforcement officials to distinguish between hemp and marijuana it must report a weight percent total THC value.

We draw the analogy to drunk driving laws. These laws typically state a blood alcohol value below which it is legal to drive and above which it is illegal to drive. An analyzer that only tests for the presence of alcohol in one’s body without generating a blood alcohol level would be pointless because it cannot distinguish between legal and illegal drivers. Similarly, a cannabis plant material total THC test that only tests for the presence of THC and does not generate a total THC value is pointless because it cannot distinguish between legal hemp and illegal marijuana as defined by Federal law. There are several methods in place that claim to be able to distinguish between hemp and marijuana that do not generate a total THC value. These include organoleptic analysis, color analysis, genetic analysis, and Raman spectroscopy.

Organoleptic Analysis

Organoleptic analysis is the use of the human senses to analyze a material (11). For cannabis there are people, particularly botanists, who are trained to distinguish cannabis from other plants by look and smell. Before the legalization of marijuana anywhere in the United States, this would have been probable cause for arrest. However, now that cannabis plant material with upwards of 0.3% total THC is legal this is no longer sufficient because legal cannabis plant material may contain some THC. We believe some of the false arrests documented above were caused by failure of law enforcement to understand this problem. We have seen no peer-reviewed data showing a human can organoleptically determine total THC levels. Hence this method should be avoided.

Color Tests

There exist color tests that disclose the presence of THC in a sample (12). These tests do not produce a weight percent total THC result so are not compliant with what the law requires. Also, practically every cannabis sample contains THC, so this technology would not be useful in distinguishing different types of cannabis plant materials from each other because of the large number of positives from legal hemp samples.

Genetic Testing

A recent paper in this journal disclosed the use of genetic testing to distinguish hemp from marijuana (13). Marijuana cultivars are bred for high THC values, whereas hemp cultivars are bred for low THC values. It makes sense then that these strains would have different genotypes. However, DNA testing is not compliant with the law since it does not measure a total THC value. Additionally, a cannabis plant’s genotype is not the only predictor of its total THC concentration; growing conditions play a role as well. It is the combination of genotype and growing conditions that determines the cannabinoid profile of a cannabis grow. Thus, genetic testing may be useful to classify cannabis plants into strains, but it does not give us the total THC measurement needed to determine legality.

Raman Spectroscopy

In a recent paper, claims were made that a Raman spectrometer successfully distinguished between hemp and marijuana 100% of the time (14,15). However, their reports contain no total THC measurement data on the samples analyzed, and no science supporting the ability of their spectrometer to quantitate total THC in cannabis plant material. Thus, we must conclude like the other techniques listed here, that Raman spectroscopy is not congruent with the law and should not be used to distinguish hemp from marijuana.

In summary then nonquantitative methods may be fast, inexpensive, and allow a high sample throughput enabling representative sampling, but are inaccurate since they do not provide a total THC value.

Quantitative Methods

For the quantitative determination of total THC in cannabis plant material chromatography and spectroscopy have been used (16–19).


In chromatography, cannabis plant material is prepared by weighing, grinding, extracting, filtering, and diluting to obtain a sample of purified cannabinoids dissolved in solvent (16). In high performance liquid chromatography (HPLC) (17,18), the sample is injected into a flowing stream of liquid that passes through a column containing a bed of separating medium (16). Different molecules interact with the medium with different strengths and leave the column at different times (17,18). These purified materials are then quantitated. For the determination of cannabinoids by HPLC, a common detection scheme is diode array ultraviolet-visible spectroscopy (17,18). In gas chromatography (GC), the sample is vaporized in a heated injection port, swept into a bed of separating medium by a flow of gas, this bed is heated, cannabinoids vaporize based in part on their boiling points, and are swept towards a detector for quantitation. Flame ionization is one type of GC detection scheme that has been used to analyze cannabis (19).

Chromatography is certainly accurate, but it is slow (16-19). Between the sample preparation steps mentioned above and running the instrument it can take 20 min or more to analyze a single sample (17–19). Chromatography is also expensive. Each analysis requires expensive consumables such as solvents, vials, filters, and syringes. The required chemical operations and the use of a chromatograph is difficult enough that state regulations require cannabis laboratory technicians using these devices be degreed and highly trained (20). These people do not work cheap, driving up the cost per analysis. There is also the cost of legally disposing of the solvents used in HPLC. In total, between the expense of the instrument, consumables, time, labor, and waste disposal it can cost $30 or more per sample to analyze cannabis samples (16). The slowness and expense of using chromatographs prevents the analysis of the large number of samples needed for appropriate representative sampling (9,10). Thus, when considering the Golden Rectangle of Chemical Analysis chromatography is certainly accurate but suffers when considering cost, speed, and throughput.

Infrared Spectroscopy

Infrared spectroscopy is the study of the interaction in infrared light, otherwise known as heat, with matter (21,22). Two different wavelength regions, the near infrared (NIR) and the mid-infrared (mid-IR) are commonly used to analyze samples. NIR wavelengths are shorter than mid-IR wavelengths (21). NIR is used to determine analytes such as fat, protein, and moisture in agricultural crops including hay, forage, and grain (27,28). Given the history of NIR spectroscopy’s use on agricultural crops, its extension to cannabis analysis seems logical. NIR has been used to determine cannabinoid content in hemp (24).

However, NIR suffers from some problems. Firstly, NIR is not selective, it can typically only see chemical functional groups that contain O-H, N-H, and C-H bonds (27,28). This is fine for gross measurements, such as total protein, but NIR struggles to distinguish cannabinoids of similar structure such as THCA and THC. Another problem NIR suffers from is lack of sensitivity. Because of the physics of the situation, NIR absorbances are 100 times weaker than mid-IR absorbances (23,28). This means NIR can struggle with accurately determining total THC at low levels in cannabis plant material.

Mid-IR is used throughout the food, forensic, pharmaceutical, and many other industries (21). Mid-IR has been used to determine cannabinoids in marijuana (25), hemp (26), cannabis distillates (29), extracts (30), and to study the rate and mechanism of potency degradation in cannabis oils (31). Mid-IR has excellent selectivity since it can see hundreds of different kinds of chemical bonds (22). Therefore, it can determine total THC and it can quantitate both THC and THCA in hemp and marijuana (25,26). Since mid-IR absorbances are 100 times more intense than NIR absorbances (22,27,28), this allows mid-IR to accurately determine low levels of total THC in hemp (26). Additionally, mid-IR requires no complex sample preparation or consumables so the cost per analysis is $0, a single cannabis plant material sample can be analyzed in 2 min (26), and both of these advantages lend themselves well for handling the large number of samples needed for true representative sampling. For these reasons, we chose to study the utility of mid-IR spectroscopy for distinguishing hemp from marijuana. We will show that mid-IR is sufficiently accurate for reliably distinguishing hemp from marijuana while maintaining the advantages of speed and low cost needed for representative Sampling.


Dried cannabis plant material samples (<10% moisture) were prepared for analysis by taking ~5 g aliquots and grinding them for 1 min in a standard coffee grinder. Samples were scanned using a BSS 3000 Cannabis Analyzer from Big Sur Scientific. It is a general purpose quantitative mid-IR spectrometer and can be used in other applications besides cannabis (32).


The analyzer is 4-in. x 6-in. x 8-in. and weighs 2 lbs so it is compact, lightweight, and portable. Using its carrying case the unit can be taken into the field and used almost anywhere including laboratories, offices, greenhouses, barns, trade shows, and the bed of pickup trucks. It can be powered by a battery giving portable power. The optics are pinned in place at the factory yielding a rugged design that is insensitive to vibrations. The analyzer is precalibrated and validated and works out of the box with a few minutes of setup and testing. So long as the device is kept dry, has a power source, and a horizontal surface to sit upon it will work well.

The spectrometer design has been disclosed in detail before (32); a brief description will be provided here. Figure 2 (click to enlarge) shows a schematic layout of the spectrometer.

In the figure, the number 1 denotes a solid state mid-IR source that is pulsed electronically to discriminate between IR light coming from the source and ambient heat. Number 2 is a transfer optic that directs the infrared beam to number 3, a zinc selenide (ZnSe) triple bounce attenuated total reflectance (ATR) crystal (21). The top surface of the ATR crystal is where the sample, number 5, is placed, and hence is called the sampling surface. The infrared beam, which is internal to the ZnSe crystal, will internally reflect from a spot on the sampling surface (21,33,34). At this spot, some infrared amplitude will appear above the crystal surface forming an “evanescent wave” (21,33,34). Samples placed on the sampling surface can absorb the evanescent wave, and the infrared beam is then analyzed to determine the sample’s spectrum (21).

To analyze liquids by ATR, they are spread on the sampling surface and their spectra are scanned (21). To ensure reproducible interaction between solid samples and the evanescent wave, a clamp, 4 in Figure 2, is used to secure these samples to the sampling surface. Figure 3 (click to enlarge) shows a sample of ground cannabis plant material being applied to the sampling surface of the analyzer prior to clamping.

The clamp features a slip-clutch mechanism that will slip when a specific pressure is reached, and the user is notified of this by an audible click. This ensures all solid samples have the same amount of pressure applied to them, which is important for reproducible sample-window contact and hence quantitative analysis (21,23). The ATR method was chosen because it requires little sample preparation, is fast, easy-to-use, and has been shown to be quantitative when applied to cannabis plant material (25,26). Additionally, the technique is nondestructive, which is good if the sample is particularly valuable, or needs to be preserved as evidence.

After the infrared beam interacts with the sample it leaves the ATR crystal and impinges upon transfer optic (number 6),which directs the beam to the spectrometer-on-a-chip (number 7). The spectrometer is a Fabry-Perot interferometer (FPI), which consists of an optical cavity with two reflective surfaces (mirrors) (32). A scan is performed by moving one of the mirrors away from the other. At any given mirror position only a fixed bandwidth of wavelengths will pass through the spectrometer and onto the detector, number 8, and changing the mirror position changes the wavelengths that impinge on the detector. Scanning a spectrum is a matter of moving the mirror through a given set of positions and measuring the amount of light hitting the detector at each position. The detector material used was lithium tantalate. The spectrometer and detector comprise a monolithic unit about the diameter of a dime. We call the spectrometer a Fabry-Perot Interferometer-Attenuated Total Reflectance spectrometer or FPI-ATR for short. It has been the tradition in spectroscopy to name spectrometer designs after their inventors. For example, the well-known Czerny-Turner spectrometer is named after its inventors (35). Hence, we will use the terms “FPI-ATR” spectrometer and “Smith spectrometer” interchangeably for this device.

A Microsoft Surface Pro computer (number 9) running the Windows 10 Pro operating system was used. Big Sur Scientific’s Cannabis Analyzer software controlled and scanned the spectrometer, measured spectra, applied calibration models to the spectra to predict concentrations, and displayed results. Communications between the analyzer and computer are via a USB cable.

Mid-IR spectra were measured from 1250 to 960 cm-1 at 12 cm-1 instrumental resolution. This spectral region was chosen because THC and other cannabinoids absorb strongly in this wavenumber range, enhancing sensitivity. A background spectrum with no sample on the sampling window is measured prior to the analysis of each sample and takes 1 min. After cannabis plant material is clamped to the sampling window, the sample scan takes an additional 1 min. It thus takes 2 min to analyze a single sample. The sampling window is cleaned before and after each analysis with a paper towel wetted with a small amount of alcohol.

After each analysis the results are presented to the user on the computer’s screen. The user has the option to print the results, copy and paste the results into a different application, or generate a certificate of analysis in Adobe Acrobat PDF format. Each sample spectrum measured is automatically saved in an industry standard spectroscopy file format (Thermo GRAMS *.SPC format), and each set of results is automatically saved as a text file. The unit comes with a system check sample that is run daily to ensure the analyzer is functioning properly.

HPLC reference data from a state licensed, ISO registered laboratory (36) were used to calibrate the Smith spectrometer. The laboratory did not correct the measured weight percent cannabinoid values for variations in moisture content of the cannabis plant material. Pure cannabinoid standards were not used for calibrating the mid-IR analyzer because this presents a calibration applicability problem (23). For chromatographic cannabis potency analyses the final sample is a solution of cannabinoids (17–19), and hence it is appropriate to use solutions of cannabinoids as standards. However, in practice the mid-IR spectrometer analyzes dried, ground cannabis plant material not solutions of cannabinoids. For the spectroscopic models to be applicable to cannabis plant samples, the spectra and weight percent cannabinoids as determined by HPLC of standard cannabis plant samples were used to build calibrations.

The partial least squares algorithm (PLS1) was used to build the calibration models used (23). An advantage of this algorithm over traditional single peak Beer’s Law analyses is that it works well even if spectral peaks contain overlaps from multiple analytes (23). Thus, there is no need for each analyte to have a spectrally resolved peak for quantitation to be achieved (23).

The Hierarchical Classification algorithm employed involves using spectroscopic calibrations to measure the total THC in a cannabis sample, and then sorting the sample into an appropriate category based on this result. Here is how it works. In an initial step, a broad concentration range calibration is used to determine the total THC in cannabis plant material, and the samples are sorted into high and low total THC cannabis categories. In a second step, a narrower range calibration is applied to the low total THC cannabis plant material and this number is used to sort samples into legal hemp or illegal marijuana.

The quality of the total THC spectroscopic calibrations was measured by analyzing a validation set by mid-IR and applying a calibration that did not include spectra of the validation samples. The validation samples were analyzed in triplicate by mid-IR; the predicted values reported here are the average of these values. The predicted total THC values for the validation set were compared to the known values for these samples as measured by HPLC (36). The accuracies of the spectroscopic calibrations were determined as a standard error of prediction (SEP) (23). The SEP is the best measure of calibration accuracy since it determines how well a calibration performs on samples it has not seen before, which is how real-world calibrations work (23). The ability of the Hierarchical Algorithm to distinguish between hemp and marijuana correctly was measured with a set of challenge samples that were known to be either hemp or marijuana. The percentage of samples correctly sorted is reported as the success rate.

Results and Discussion

Spectral Analysis

Previous work (17,18) has shown that for cannabis plant material much of the contribution to the total THC value is from THCA rather than THC. The mid-IR spectrum of pure THCA is seen in Figure 4 (click to enlarge). The portion of the spectrum from 1250 to 960 cm-1 is shown, the region scanned with Smith spectrometer used in this study.

The peak labeled A falls at 1250 cm-1 here because it is the shoulder of a larger peak that tops out at 1256 cm-1, beyond the range scanned. Peak B is at 1190 cm-1 and peak C is at 1120 cm-1. In going from hemp with low total THC values to recreational marijuana with high THC values for quantitation to be possible there should be mid-IR spectral features whose size changes with total THC content.

The same three mid-IR features present in the spectrum of pure THCA as seen in Figure 4 are clearly present in the spectra of cannabis plant material seen in Figure 5 (click to enlarge) and are labeled the same in both figures. As the total THC content goes from 0.1%, to 14%, to 30% in Figure 5, the mid-IR features of THCA clearly get larger. This correlation between peak size and total THC value means Beer’s Law (23) can be used as the basis for using mid-IR to quantitate total THC in cannabis plant material.

The Hunt for Quality Reference Data

Obtaining accurate HPLC reference data for calibrating the spectrometer proved difficult. The problem of inter-laboratory error, different cannabis laboratories obtaining varying results on the same samples, has been researched and is well documented (29,37–42). Causes of this problem include lack of standard reference materials, laboratories using different sample extraction methods, and lack of adequate training (29,41–43).

At first, we attempted to use reference data from multiple laboratories in our spectroscopic calibrations. However, the inter-laboratory error problem caused the resultant calibrations to be of poor quality. When we built calibrations using data from a single laboratory, calibration quality improved markedly. This meant we had to choose a laboratory whose reference data we would use and trust. We literally roamed the country visiting many cannabis analysis laboratories, and vetted them by interviewing their laboratory directors, reviewing their standard operating procedures, and watching them perform analyses. Additionally, at some laboratories we ran the same cannabis plant samples by mid-IR spectroscopy and HPLC to provide an orthogonal test of the quality of the chromatographic results. We took the correlation between mid-IR and HPLC data as a measure of chromatographic quality, assuming that mid-IR is an unbiased and independent method. Based on our work we chose a state licensed, ISO certified laboratory (36), and used their weight percent cannabinoid values to calibrate the mid-IR analyzer.

The Search for Cannabis Standard Reference Materials

In the United States, the National Institutes of Standards and Technology (NIST) is the Federal agency tasked with generating standard reference materials (SRMs) for industry (43). SRMs are materials of known and agreed upon composition that can be used in round robin studies to evaluate the accuracy of different laboratories, the quality of different methods, and to evaluate instrumentation. For example, NIST supports the green tea industry with a standard reference material (44). This means that NIST knows how to create SRMs for agricultural crops. It would make sense then that NIST should create SRMs for the cannabis industry.

However, recall that in the United States at the Federal level all cannabis plant material with a total THC value of greater than 0.3% is considered marijuana and is illegal (1,2). Thus, it would be against Federal law for NIST to develop SRMs for marijuana analysis. But there is nothing preventing NIST from developing SRMs for legal hemp. NIST is in fact developing such materials (45). However, it may be some time until these SRMs are ready.

Fortunately, the University of Kentucky (UKY) has developed their own set of hemp standard reference materials as part of their Hemp Proficiency Program (HPP) (46). This sample set consists of four legal hemp samples, each of which has been analyzed by more than 50 cannabis analysis laboratories. The certificates of analysis for these samples contain the average total THC, total CBD, and other cannabinoid weight percents as determined by these laboratories. Barring the introduction of any NIST cannabis SRMs, we believe the UKY samples are the best validation set now available in the cannabis industry, and we encourage all cannabis analysis laboratories to use these samples to vet their methods and to participate in the HPP.

Independent Validation Results

Taking our own advice, we used the University of Kentucky sample set in a validation study of how well mid-IR can quantitate total THC in dried, ground hemp. The results are seen in Table I (click to enlarge).

Note that the standard error of prediction is ±0.05 wt. % total THC, consistent with the accuracy for total THC determination of ±0.04 wt. % in dried, ground hemp previously published for this analyzer (26). These results mean that mid-IR spectroscopy is accurate enough to determine if hemp is above or below the 0.3% total THC legal limit.

The Hierarchical Total THC Classification Algorithm

The challenge in analyzing cannabis plant material for total THC is we encountered samples that ranged from 0.1% to 33% total THC. In a perfect world a single, global spectroscopic calibration would accurately measure total THC values across this entire concentration range. However, the accuracy of spectroscopic calibrations is a function of the number of data points used in the calibration, the range of concentrations of the standard samples, and the distribution of the calibration points throughout the concentration range (23). Therefore, it was not possible to construct a single spectroscopic calibration that was sufficiently accurate throughout the entire cannabis plant material total THC concentration range.

Therefore, a hierarchical total THC classification was implemented. The flow chart for this algorithm is seen in Figure 6 (click to enlarge).

After the mid-IR spectrum of a cannabis plant material sample is measured, a global total THC calibration is applied to the spectrum to determine if the sample contains greater than or equal to 5% total THC or less. If the reading is 5% or above the material is sorted as high total THC marijuana and the analysis is stopped; this sample is marijuana and is clearly illegal by Federal law.

If the material contains less than 5% total THC it may be legal hemp. In this case a special high accuracy total THC calibration is applied to the spectrum. If the reading comes back at 0.34% total THC or less (the acceptable hemp THC level for this analyzer [2]) the material is classified as hemp. If the total THC reading falls above 0.34% the material is classified as marijuana.

The Global Total THC Calibration and Challenge Sample Set Results

The purpose of the global total THC calibration is to sort cannabis plant material into high and low total THC classes. Figure 7 (click to enlarge) shows a correlation chart for total THC values for this calibration. Total THC values as measured by HPLC are on the x-axis, total THC values for the same samples as measured by mid-IR are on the y-axis. Approximately 129 data points were used in creating this calibration.

The correlation coefficient (R2) is a measure of how well the readings on two instruments correlate for the same sample set, where 1 is perfect and 0 is the worst possible correlation (23). Note from Figure 7 that the R2 value between total THC measurements determined by HPLC and mid-IR on the same samples is 0.94, which we consider excellent for a spectroscopic calibration of a highly inhomogeneous material over such a large concentration range. Note that the samples self-segregate nicely, with the high total THC marijuana clustering above 5% total THC inside the green oval, whereas low total THC cannabis clustering around 1% total THC or less inside the red oval. This indicates marijuana breeders have done a good job of maximizing the total THC in their plants, and hemp breeders have done a good job of minimizing the total THC in their plants. Note the significant gap between the green and red ovals. Because of this gap we chose to define high total THC cannabis as containing more than 5% total THC, and anything below this was defined as low total THC cannabis.

The global total THC calibration was applied to a challenge set of 491 samples that were known to be either high total THC cannabis or low total THC cannabis. The algorithm sorted the samples into their correct classes 488 out of 491 times for a success rate of 99.4%. We consider this very impressive given the speed, ease of use, and the $0 cost per analysis of the mid-IR method.

Low Total THC Calibration and Challenge Sample Set Results

According to the flow chart in Figure 6, if upon application of the global total THC calibration a sample tests less than 5% total THC, a special low total THC calibration is applied to the sample’s spectrum and its total THC value is predicted. The correlation chart for low total THC cannabis is seen in Figure 8 (click to enlarge).

This calibration contains 63 data points over a total THC range from 0.07% to 0.7%. It is this calibration whose SEP for total THC was ±0.05% for the UKY validation samples shown in Table I. The excellent accuracy of this calibration is a result of the large number of data points over a narrow concentration range (23).

The purpose of this calibration is to sort cannabis plant material with less than 5% total THC into two categories, hemp with a total THC level of 0.34% or less, or marijuana with a total THC value above 0.34%. The 0.34% total THC value was used because the acceptable hemp THC level as calculated for this analyzer is 0.34% (2). The low total THC calibration was challenged with a set of 284 samples that were known to be either hemp or low total THC marijuana. The calibration correctly sorted the hemp from the low total THC marijuana samples 270 out of 284 times, for a success rate of 95.1%.

A summary of the results of the hierarchical total THC classification method are seen in Table II (click to enlarge).

The results in Table II show that the measurement of cannabis plant samples by mid-IR spectroscopy when combined with our total THC hierarchical classification algorithm accurately distinguishes between hemp and marijuana.

Implications for Law Enforcement, Forensic Laboratories, and Departments of Agriculture

In addition to classifying cannabis plant material accurately, the analyzer system we developed is small, lightweight, portable, features little sample preparation, and comes precalibrated so it works out of the box. It analyzes cannabis plant material samples in 2 min. This means the system is well suited for use by police officers roadside for when there is suspicion someone is transporting cannabis plant material illegally, hopefully preventing the false arrests and seizures documented above (3–7). Police officers receive training in how to organoleptically identify cannabis plant material. Once this identification is made, mid-IR could be used to determine the legality of the cannabis plant material quickly and accurately. Any material thought to be illegal should be confirmed by chromatography.

Because of the lack of a roadside test, samples of cannabis plant material are being sent to forensic laboratories to determine whether they are hemp or marijuana. Analysis of these samples is typically done by chromatography, which is slow and expensive (16–19). If forensic laboratories screened these samples using mid-IR, they could weed out the legal samples, and only test those samples by chromatography that are suspected of being illegal. This will save time, streamline operations, and save taxpayer money.

The Federal government has tasked state departments of agriculture with establishing hemp testing programs (1,2). The USDA also recommends that about one cutting per acre of hemp planted be analyzed (2). Applying this idea to the state of Kentucky hemp crop for 2019, where licenses were issued for the planting of 42,000 acres of hemp (47), tens of thousands of samples may be generated. The challenge is that all these samples must be tested during a brief harvest season of about 60 days.

Our observation is that in the hands of a skilled technician, and with automated sample injection, an individual HPLC can process about 50 samples per day. At 42,000 tests and 60 days, this means 700 tests/day must be performed during harvest season. This would require 14 HPLC systems and trained technicians at the ready come hemp harvest season to analyze the number of samples generated. That is well over a $1 million investment in equipment and personnel. Other states may not have this many acres planted but will face a similar testing challenge.

To avoid this problem, a mid-IR system could be used in the field by state department of agriculture personnel to screen cannabis plant material samples. After a day’s worth of cuttings is collected, these could be dried overnight, and the samples analyzed the next day. Material that is determined legal would require no further analysis, and the hemp farmer would be cleared to sell their crop. Samples that tested illegal should be confirmed by chromatography. This approach will give hemp farmers the accurate and timely results they need and reduce the sample load at state laboratories saving taxpayer dollars.

Ultimately, the decision to prosecute a case is up to district attorneys and other law enforcement personnel in a given jurisdiction. It is the job of scientists to provide these officials with the best data we can. Our goal has been to develop a system that easily and accurately distinguishes between marijuana and hemp so law enforcement, forensic laboratories, and state departments of agriculture have the data they need to make their important decisions. We believe our results show mid-IR spectroscopy is up to the challenge.


There is a need amongst law enforcement, forensic laboratories, and state departments of agriculture to distinguish hemp from marijuana in an accurate, fast, easy, and inexpensive manner. We have invented a novel general purpose quantitative mid-IR spectrometer, the FPI-ATR or Smith spectrometer, and applied it to the problem. With use of a hierarchical total THC classification method, high total THC cannabis can be sorted from low total THC cannabis successfully more than 99% of the time. Low total THC marijuana can be properly sorted from legal hemp more than 95% of the time. This means mid-IR can be used by law enforcement, forensic laboratories, and state departments of agriculture to reduce false arrests, streamline laboratory operations, and more easily and cheaply perform representative sampling.


  1. 115th United States Congress, Senate Bill S.2667,” Hemp Farming Act of 2018.
  8. B.C. Smith, Cannabis Science and Technology 1(4), 12–16 (2018).
  9. B.C. Smith, Cannabis Science and Technology 2(1) 14 (2019).
  10. P. Atkins, Cannabis Science and Technology 2(2) 26 (2019).
  13. A. Hilyard, S. Lewin, S. Johnson, P. Henry, and C. Orser, Cannabis Science and Technology 2(6) 42–46 (2019).
  14. L. Sanchez, C. Filter, D. Baltensperger, and D. Kurouski, RSC Advances 10, 3212 (2020).
  15. D. Kurouski, Cannabis Science and Technology 3(3), 53–54 (2020).
  16. B.C. Smith, Cannabis Science and Technology 2(6), 10–14 (2019).
  17. M.W. Giese, M.A. Lewis, L. Giese, and K.M. Smith, Journal of AOAC International 98(6), 1503 (2015).
  18. C. Giroud, CHIMIA Intl. Journal of Chemistry 56, 80 (2002).
  19. T. Ruppel and M. Kuffel, "Cannabis Analysis: Potency Testing Identification and Quantification of THC and CBD by GC/FID and GC/MS," PerkinElmer Application Note (2013).
  20. California Bureau of Cannabis Control Regulations, Section 5719.
  21. B. C. Smith, Fundamentals of Fourier Transform Infrared Spectroscopy, 2nd Ed., (CRC Press, Boca Raton, Florida, 2011).
  22. B. C. Smith, Infrared Spectral Interpretation, A Systematic Approach (CRC Press, Boca Raton, Florida, 1999).
  23. B.C. Smith, Quantitative Spectroscopy: Theory and Practice (Elsevier, Boston, Massachusetts, 2002).
  24. C. Sánchez-Carnerero Callado, N. Núñez-Sánchez, S. Casano, and C. Ferreiro-Veraa, Talanta 190, 147–157 (2018).
  25. B.C. Smith, M. Lewis, and J. Mendez, “Optimization of Cannabis Grows Using Fourier Transform Mid-Infrared Spectroscopy,” PerkinElmer Application Note (2016).
  26. B.C. Smith, Cannabis Science and Technology, 2(6), 28–33 (2019).
  27. Handbook of Near Infrared Analysis, D. Burns and E. Ciurczak, Eds., (Marcel Dekker, New York, New York, 1992).
  28. J. Workman and L. Weyer, Practical Guide to Interpretative Near-Infrared Spectroscopy (CRC Press, Boca Raton, Florida, 2008).
  29. B.C. Smith, P. Lessard, and R. Pearson, Cannabis Science and Technology 2(1), 48 (2019).
  30. B.C. Smith, Terpenes and Testing, Jan.-Feb., 32 (2018).
  31. B.C. Smith, Terpenes and Testing, Nov.-Dec., 48 (2017).
  32. B.C. Smith, United States Patent 10,451,480.
  33. N. Harrick, Internal Reflection Spectroscopy (Wiley, New York, New York, 1967).
  34. F. Mirabella and N. Harrick, Infrared Reflection Spectroscopy: Review and Supplement (Marcel Dekker, New York, New York, 1985).
  35. M. Czerny and A. Turner, Zeitschrift fur Physik 61, 792 (1930).
  37. M.O. Bonn-Miller, M.J.E. Loflin, B.F. Thomas, J.P. Marcu, T. Hyke, and R. Vandrey, Journal of the American Medical Association 318, 1708 (2017).
  38. B. Young, The Seattle Times, January 5, 2016.
  40. B.C. Smith, Cannabis Science and Technology 2(2), 12-17 (2019).
  41. B.C. Smith, Cannabis Science and Technology 2(3), 10–14 (2019).
  42. B.C. Smith, Cannabis Science and Technology 3(2), 10–15 (2020).

About the Authors 

Brian C. Smith and Christopher A. Fucetola are with Big Sur Scientific in Aptos, California. Direct correspondence to: