Multispectral reflectance dataset of medicinal Cannabis sativa cultivars for varietal classification
Universidad Católica de Manizales, Faculty of Engineering and Architecture, Cra 23 No 60-63, Manizales, 170001, Colombia
Universidad Autónoma de Manizales, Faculty of Engineering, Antigua Estación del Ferrocarril, Manizales, 170001, Colombia
CUBIKAN GROUP, Research Division, Cll 69B No 27-37, Manizales, 170001, Colombia
Abstract
The dataset comprises 800 unique spectral signatures collected from four distinct cultivars: Embera CBD, Yocoto CBD, Limón CBD, and Valle de los Umbras. Data acquisition was performed using an Ocean Optics RED TIDE USB650 spectroradiometer with an optical fiber probe, covering a wavelength range from 410 nm to 890 nm. Measurements were taken under controlled ambient light conditions within a greenhouse, with cultivation parameters standardized. Spectral signatures were obtained from 30 plants of Embera CBD and Valle de los Umbras, and 20 plants of Limón CBD and Yocoto CBD varieties. Data collection spanned three distinct phenological stages (early vegetative, early flowering, and mid-flowering) and included measurements from both upper and lower leaves. Each signature represents an average of three measurements per leaf sample, calibrated with a white reflectance reference panel. Raw data were processed and normalized using Spectroscopy OceanView software, R, and JASP. The dataset is structured with columns for date (phenological stage), variety, plant ID, leaf location, and 2442 reflectance values corresponding to specific wavelengths. This publicly available dataset serves as a valuable resource for developing and training machine learning models for non-destructive and cost-effective classification of Cannabis varieties based on spectral characteristics, and for exploring correlations between spectral signatures and plant properties like cannabinoid content.
Untitled section
Keywords: Cannabis sativa, Spectroscopy, Machine learning, Varietal classification, Reflectance, Plant phenology
Article notes
Untitled section
Received 2025 Aug 4; Revised 2025 Sep 2; Accepted 2025 Oct 8; Collection date 2025 Dec.
Specifications Table
| Subject | Computer Sciences |
| Specific subject area | Development of machine learning models for the non-destructive classification of plant varieties using multispectral data. |
| Type of data | Processed dataset |
| Data collection | Spectral signatures were acquired using an Ocean Optics RED TIDE USB650 spectroradiometer with an optical fiber probe. Individuals of each variety were randomly marked in the crop to ensure data collection at all three vegetative stages. A min-max normalization process was applied to each spectral signature. |
| Data source location | Universidad Católica de Manizales, Manizales, Caldas, Colombia and Cubikan Group, Manizales, Caldas, Colombia |
| Data accessibility | Repository name: Mendeley Data Data identification number: 10.17632/ctbys33fzh.1 Direct URL to data: https://data.mendeley.com/datasets/ctbys33fzh/1 |
| Related research article | Vladimir Henao-Céspedes, Oscar Cardona-Morales, Jolián Andrés Vargas-Alzate, Julián Ricardo León-Zuleta, Eddy Mackniven Guzman-Buendia, and Yeison Alberto Garcés-Gómez “Hyperspectral reflectance for classification of medicinal Cannabis varieties using machine learning algorithms,” Optical Engineering 63(6), 064,103 (25 June 2024). https://doi.org/10.1117/1.OE.63.6.064103 |
1.Value of the Data
- •This dataset offers a unique and detailed collection of multispectral reflectance signatures from four medicinal Cannabis sativa varieties, providing a foundational resource for the development of non-destructive analytical techniques in the Cannabis industry. It is particularly valuable as it allows for the study of spectral characteristics without requiring plant material extraction or laboratory testing, which can significantly reduce production costs. Researchers can utilize this data to develop more efficient and cost-effective methods for quality control and varietal identification.
- •The data facilitates the training and validation of machine learning algorithms for the classification of Cannabis sativa varieties based on spectral signatures, as demonstrated in a related research article. This enables other researchers to replicate existing studies, enhance their analyses, and explore alternative methodological approaches to improve classification metrics and reduce false positive or negative classifications.
- •The inclusion of data collected across three distinct phenological stages (early vegetative, early flowering, and mid-flowering) and from different leaf locations (upper and lower leaves) provides a comprehensive view of spectral variations over a plant's lifecycle and within its structure. This allows researchers to investigate how spectral signatures change with plant development and environmental conditions, potentially leading to new insights into plant health, growth, and stress detection.
- •The standardized data acquisition methods under controlled greenhouse conditions ensure high data quality and consistency, making the dataset reliable for comparative analyses and robust model development. This also provides a baseline for comparing data collected under different cultivation environments or with other types of sensing technologies.
- •Given the increasing prominence of Cannabis cultivation for medicinal purposes, this dataset supports advancements in crop technification and precision agriculture management, allowing producers to standardize quality processes. The approach can be applied to other plant species or crops after their spectral characterization, broadening its impact across agricultural research.
2.Background
The compilation of this dataset was motivated by the growing global interest in medicinal Cannabis cultivation [1,2] and the need for non-destructive, cost-effective methods for crop classification and quality assessment [3]. Traditional methods often involve destructive sampling and laboratory analysis, which can be time-consuming and expensive. Hyperspectral reflectance, as a remote sensing technique, offers a promising alternative by providing detailed spectral signatures that can be correlated with various plant properties. This work builds upon the theoretical understanding that distinct varieties of the same plant species may exhibit unique spectral characteristics due to variations in their physiological and chemical compositions [4]. The data collection aimed to create a robust foundation for developing machine learning models capable of identifying Cannabis sativa varieties solely based on their spectral fingerprints.
This data article supports the related research article “Hyperspectral reflectance for classification of medicinal Cannabis varieties using machine learning algorithms" [3]. While the research article presented the application of machine learning algorithms for varietal classification using these spectral signatures, this data article provides a detailed description of the dataset itself [5], including the comprehensive data collection methodology, structure, and associated metadata. It enhances the reproducibility and transparency of the original research by making the raw and processed data publicly available, allowing other researchers to validate findings, explore new analytical avenues, or apply the dataset to different research questions related to Cannabis or other plant species.
3.Data Description
The dataset is organized into a single compressed file within the Mendeley Data repository, titled “Multispectral Reflectance Database of Medicinal Cannabis sativa” [5]. This file contains the primary data in .csv format, accompanied by a readme.txt file that provides a comprehensive overview of the database.
The core of the dataset is contained in the .csv files, where each row represents a unique spectral signature. The columns are structured to provide essential metadata alongside the reflectance values, enabling a clear understanding of each measurement.
Table 1 summarizes the distribution of collected spectral signatures across the four Cannabis sativa varieties. It provides the number of sample units (individual plants) for each variety, and the total number of spectral signatures derived after averaging multiple measurements per leaf and replicate. This table helps the reader understand the overall balance and size of the dataset for each cultivar.
| Cannabis variety | Number of sample units | Total signatures per variety |
|---|---|---|
| Embera (EMB) | 30 | 720 |
| Limón (LIM) | 20 | 480 |
| Valle de los Umbras (VALL) | 30 | 720 |
| Yocoto (YO) | 20 | 480 |
Fig. 1 displays the average spectral reflectance curve for each of the four Cannabis sativa varieties (Embera, Limón, Valle de los Umbras, and Yocoto) across the primary measured wavelength range (410 to 890 nm). The shaded areas around each average curve represent the standard deviation, illustrating the variability within each variety's spectral signatures. This visual representation allows readers to observe general trends and potential distinctions in reflectance patterns among the different cultivars.
Dataset File Structure: The dataset file (in .csv format) are structured as follows:
- •date: Indicates the phenological stage at which the sample was taken (Early Vegetative, Early Flowering, Mid-Flowering).
- •variety: Identifies the Cannabis sativa variety using abbreviations (EMB, LIM, VALL, YO).
- •plant_ID: A unique code for each specific plant, allowing tracking across stages and leaf locations.
- •leaf_location: Indicates the leaf's position on the plant, with “0” for lower leaf and “1” for upper leaf.
- •Remaining Columns (410 nm - 890 nm): These columns contain the 2442 reflectance values, with each header being the wavelength in nanometers (e.g., “410 nm", "410.2 nm", …, "890 nm”).
4.Experimental Design, Materials and Methods
4.1.Materials
Four varieties of Cannabis sativa were analyzed: Embera (EMB), Limón (LIM), Valle de los Umbras (VALL), and Yocoto (YO). These biological materials are of high scientific interest in Colombia, as several ongoing investigations have attributed therapeutic and functional cosmetic potential to the extracts derived from these varieties [6,7]. Furthermore, these varieties represent a seed source exclusively owned by Cubikan Group and have been granted a National Cultivar Registry, approved by the Colombian Agricultural Institute (ICA). These plants were cultivated in a controlled greenhouse environment in Colombia to maintain standardized cultivation parameters including temperature, humidity, lighting, and nutrients. Certified seeds were used to ensure varietal authenticity and compliance with quality control requirements for plant material export. Samples were randomly selected from each variety, with 30 individuals chosen for Embera (EMB) and Valle de los Umbras (VALL), and 20 for Limón (LIM) and Yocoto (YO). Selected plants were marked at the beginning of the study to maintain measurement standards.
4.2.Data collection
Spectral signatures were acquired using an Ocean Optics RED TIDE USB650 spectroradiometer with an optical fiber probe. This instrument covers a wavelength range from 350 to 1000 nm with an optical resolution of 1.5 nm. The spectroradiometer was connected to a computer via USB for data storage. The edges of the signatures must be trimmed (410–890) because of noise is generated at these wavelengths due to environmental conditions, even though in a controlled greenhouse environment. Therefore, these edges have no value for the purpose of classification algorithms
Before each measurement session, the spectrometer was calibrated using a standard reflectance calibration target with 99 % reflectance and a dark object with minimal reflectance to reduce noise. Data collection was performed on sunny days to ensure direct sunlight illuminated the leaf surfaces, given the greenhouse cultivation.
For each selected plant, four replicates were collected, corresponding to three vegetative stages sampled on separate dates (early vegetative – two dates, early flowering, and mid-flowering). In each replicate, six spectral signatures were recorded—three from upper leaves and three from lower leaves. Measurements were taken approximately 1 cm from the leaf surface at a 45° angle to minimize shadows. The three upper-leaf signatures were averaged to yield a single upper-leaf signature, and the three lower-leaf signatures were averaged likewise. This averaging is part of the instrument-setup procedure to reduce variability from uncontrollable environmental factors and to ensure each replicate represents an independent sample. This procedure was repeated for all three replicates.
4.3.Data processing and normalization
Initial preprocessing steps were carried out using Microsoft Excel, Spectroscopy OceanView software, R, and JASP. The noisy extremes of the spectral signatures were removed, resulting in a consistent wavelength range from 410 to 890 nm for all signatures. Subsequently, the signatures underwent linear scaling normalization to standardize the reflectance values and enhance the performance of subsequent analyses. Finally, the multiple signatures collected per plant (averaged from upper and lower leaves across replicates) were further consolidated to obtain the final set of 800 unique multispectral reflectance signatures, with each representing a specific plant at a given phenological stage and leaf location.
The .csv data file include metadata for each signature, specifically: date (phenological stage), variety (cultivar), plant_ID (unique plant identifier), and leaf_location (upper or lower leaf). The remaining columns correspond to the 2442 reflectance values for each measured wavelength from 410 nm to 890 nm.
Limitations
While the data collection was conducted under controlled greenhouse conditions, subtle variations in ambient light or other environmental factors not explicitly controlled may have introduced minor noise into the spectral signatures. The dataset is based on four specific Cannabis sativa varieties, which should be considered a limitation when generalizing the results, as spectral responses could differ in untested varieties. Although the number of samples per variety is statistically sufficient for classification purposes, expanding both the sample size and varietal diversity in future studies would allow for a more comprehensive representation of intraspecific variability. Additionally, since the study was conducted in a single geographical location (Colombia) under specific cultivation practices, spectral characteristics might vary under different climatic conditions or cultivation methods.
Ethics Statement
The authors have read and follow the ethical requirements for publication in Data in Brief and confirm that the current work does not involve human subjects, animal experiments, or any data collected from social media platforms. All plant material used in this study was sourced ethically and legally, in compliance with all relevant national and international regulations regarding Cannabis sativa cultivation for research purposes.
Vladimir Henao, Oscar Cardona, Jolián Vargas, Julián León and Yeison Garcés
Acknowledgments
Acknowledgements
The authors would like to acknowledge the technical team at Cubikan Group for their invaluable support during the data collection process. This work was supported by the Universidad Católica de Manizales and the Cubikan Group.
Declaration of Generative AI and AI-assisted Technologies in the Writing Process
During the preparation of this work the authors used “DeepL Write®” in order to improve the quality of written language. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Data Availability
References
Untitled section
References
- 1.M.M. Islam, Z. Rengel, P. Storer, K.H.M. Siddique, Z.M. Solaiman, Industrial hemp (Cannabis Sativa L.) varieties and seed pre-treatments affect seed germination and early growth of seedlings. Agronomy 2021, 12, 6, doi:10.3390/agronomy12010006.
- 2.Torkamaneh D., Jones A.M.P. Cannabis, the multibillion dollar plant that No genebank wanted. Genome. 2022;65:1–5. doi: 10.1139/gen-2021-0016.
- 3.Henao-Céspedes V., Cardona-Morales O., Vargas-Alzate J.A., León-Zuleta J.R., Guzman-Buendia E.M., Garcés-Gómez Y.A. Hyperspectral reflectance for classification of medicinal cannabis varieties using machine learning algorithms. Opt. Eng. 2024;63 doi: 10.1117/1.OE.63.6.064103.
- 4.Wan L., Li H., Li C., Wang A., Yang Y., Wang P. Hyperspectral sensing of plant diseases: principle and methods. Agronomy. 2022;12:1451. doi: 10.3390/agronomy12061451.
- 5.Garcés-Gómez Y.A., Henao-Céspedes V., Cardona-Morales O., Vargas-Alzate J.A., León-Zuleta J.R. Multispectral reflectance database of cannabis sativa. Mendeley Data. 2025 doi: 10.17632/ctbys33fzh.1.
- 6.Aguirre-Ortega E., Henao-López M., Vargas-Alzate J.A., Castro Rios K., Montoya-Estrada C.N. Identification of Aspergillus Westerdijkiae and its potential risk of ochratoxin A synthesis in cannabis inflorescences. Sci. Agropecu. 2024;15:45–54. doi: 10.17268/sci.agropecu.2024.004.
- 7.Henry Osorio J., Bolaños N. Management of chronic pain with cannabis extract in a group of patients in the City of Manizales. Biosalud. 2019;18:22–25. doi: 10.17151/biosa.2019.18.1.2.