Skip to content

About

No description, website, or topics provided.

Resources

Stars

3 stars

Watchers

2 watching

Forks

Repository files navigation

DOI ProjStatus Lifecycle License

Part of maniva

Global multi-crop agricultural trial data supported by citizen science

The triadic comparison of technology options (tricot) is a citizen science approach for testing technology options in their target environments, which has been applied to on-farm testing of crop varieties. ‘Triadic’ refers to the sets of three technology options that are compared by each participant. In the approach, participants are invited to test an anonymous set of three technologies (out of a larger number, generally between 5 to 20) randomly assigned. Between 2011 and 2025 the tricot approach was applied in more than 25 countries across Africa, Asia, Europe and Latin America with more than 30 crops.


Crops and sites currently covered by the dataset

The map below shows the trial sites that are currently available for public use, coloured by crop group. Sites of the same trial within 10 km of each other are drawn as one point. The map uses the Equal Earth projection.

Trial sites currently in the release, by crop group

Find a dataset

datasets.csv is the finding aid. It lists every trial, one row each, with trial id, crop, taxon, country, data producer, start and end dates, type, management, objective, participants, sites, genotypes, collection moments, ranked traits, the path to fetch, file size and license. Start there, filter, then open only the files you need through their path. Never compose a file name yourself.

The current release holds 431 trials and 70,650 sites across 38 crops in 14 countries, from 16 data producers. Of these, 397 are farmer-managed tricot trials and 34 are researcher-managed mother trials of participatory varietal selection in India, linked to the tricot trials on the same genotypes. The release covers the 64 trials of the first release and adds trials from India, Ethiopia, Central America, Ghana, Nigeria and Laos, all in the harmonised format described below.

Data structure

Each trial is one JSON file at data/<crop>/<crop>-<iso3>-<trial_id>.json, for example data/potato/potato-NGA-88f82a3f073e.json. The trial id is the only identity of a trial. Trials run on ClimMob keep their ClimMob project id, and older trials from partner files have an id that ends in v. Crop and country in the name make a folder easy to scan, and pvs in the name marks researcher-managed companion data. Every file follows release schema v3.0 and has three parts.

  • metadata describes the trial, its producer, contributors, crop, participants, genotypes, funders and every variable in the file.
  • site_data has one row per site, the farm or household where a package of three genotypes was grown, with coordinates rounded to two decimals, administrative units as defined for the country (adm1 to adm5, then location for the village) where recorded, planting dates and participant-level answers. Categorical answers have one spelling across trials, in English.
  • plot_data is the long table of observations, one row per site, plot, collection moment and trait. Values are stored as text and read according to value_type (rank, numeric, date, categorical, multiple, or vs_local for the comparison with the farmer's own variety), with the unit where one applies. Measured values also carry value_standard and unit_standard, the same value in one unit per trait across trials (kg, kg/ha, cm). moment_days is the planned day of the collection moment and moment_date is the date it took place at the site, where recorded.

The files hold observations only. Model results such as Plackett-Luce estimates are not part of the release; fit them from the rankings in plot_data (see SKILL.md).

Coordinates. Site coordinates are rounded to two decimals, which places the published point within about 800 m of the recorded one. metadata.location_privacy states the method and the radius. This protects participants and keeps the points fit for climate and soil data.

Companion data. A tricot trial lists in metadata.companions the researcher-managed files with data on its genotypes, and those files point back through metadata.companion_of. The file in a link is relative to data/. Five groundnut trials from Tanzania link to on-station files that are not part of this release yet.

Site and plot rows link through site_id. metadata.variables lists every column of site_data and plot_data with its description, harmonised question id, Crop Ontology term, value type and unit. Rankings are the position of each genotype within its site, where 1 is best. Traits, site variables and collection moments have one standard name per concept across trials, after the review by Robert J. Hijmans. trait_original keeps the question code of the trial, metadata.variables[].original_name the name of a site variable before the standard name, and standard_moment in metadata.study.collection_moments gives the stage of each moment.

Small trials. 16 farmer-managed trials have fewer than 15 sites, 136 sites in all. A trial of that size cannot rank its varieties on its own. Treat it as a record of which varieties were tested where, or pool it with other trials of the same crop and programme that share varieties, as the ISSD trials of Ethiopia do. Seven of these trials, from the GEF project in India, share no variety with any other trial and are descriptive only. They stay in the release because they hold three crops (black gram, rice bean and small millets) and about 140 varieties that no other trial has. To leave small trials out, filter datasets.csv on n_sites. The 22 mother trials of participatory varietal selection with fewer than 15 sites are small by design, since researchers run them on a few plots.

Yield. yield_plot is the weight harvested from the plot and yield the yield per area in kg/ha. A yield with the flag derived was computed from the plot weight and the plot area. Where the trial reported no plot area, the area comes from the tricot protocol for the crop, and the yield also carries plot_area_assumed. Filter on that flag to keep only yields from reported or published plot areas. site_data.plot_area_source says where each plot area comes from.

├── metadata
│   ├── schema_version                      "3.0"
│   ├── changelog (version, date, notes, software)
│   ├── identifier                          concept DOI, https://doi.org/10.5281/zenodo.17112491
│   ├── license                             CC0-1.0
│   ├── study
│   │   ├── id                              trial id, the identity of the file
│   │   ├── code, title, description        ClimMob project code and the trial as described by the producer
│   │   ├── type, objective, experimental_site, unit_of_analysis
│   │   ├── country_iso2, country_iso3, country_name
│   │   ├── source_platform                 ClimMob version the trial came from
│   │   ├── collection_moments (moment, moment_days, standard_moment, visit, assessor)   moment_days is the planned day, counted from the start of the trial
│   │   ├── management, season, design      companion data only (researcher-managed)
│   ├── date (start, end as YYYY-MM-DD)
│   ├── geo_location
│   │   ├── boundingBox (xmin, xmax, ymin, ymax)
│   ├── location_privacy (method, radius_m, admin_names)   coordinates rounded to two decimals, radius 800 m
│   ├── admin_levels (adm1 ... adm5)         what each administrative level is in the country, for example region, zone, woreda
│   ├── data_producer (name, identifier ROR, principal_investigator, email, program, partner)
│   ├── contributors (name, orcid, affiliation, contributorType)
│   ├── crop (name, taxon)
│   ├── participants (total, women, men, not_stated)
│   ├── genotypes (genotype_name, role, release_year, crossing_year, origin, target_trait, remarks)
│   ├── funder (funderName, funderIdentifier ROR, awardTitle, awardNumber)
│   ├── companions                          tricot trials only, the companion files with data on its genotypes
│   ├── companion_of                        companion data only, the tricot trials the file relates to
│   ├── variables
│   │   ├── level                           site or plot
│   │   ├── variable_name, original_name, description
│   │   ├── harmonised_id, ontology_id      harmonised question id and Crop Ontology term
│   │   ├── value_type, unit, controlled_vocabulary
├── site_data                               one row per site
│   ├── site_id                             trial id plus package number
│   ├── trial_id, country_iso3, country, crop
│   ├── longitude, latitude                 WGS84, rounded to two decimals
│   ├── geo_source, geo_uncertainty (metres), geo_flag
│   ├── adm1 ... adm5, location             administrative units of the country (metadata.admin_levels), then the village
│   ├── planting_date, plot_area, plot_area_source, ...    and the survey variables listed in metadata.variables
├── plot_data                               one row per site, plot, moment and trait
│   ├── site_id, plot                       plot is the package letter (A, B, C)
│   ├── genotype_name                       null where the variety of the plot is not in the data
│   ├── collection_moment, moment_days      moment and its planned day, the same for every site
│   ├── moment_date                         date the moment took place at the site, null when not recorded
│   ├── trait, trait_id, trait_original
│   ├── value, value_type, unit, flag
│   ├── value_standard, unit_standard       measured values only, the value in the standard unit of the trait

Every field is defined in metadata/data-dictionary.csv, and metadata/tricot-trial.schema.json is the JSON Schema of a trial file, release schema v3.0.

Repository structure

├── data/           Trial files, one folder per crop, one JSON per trial
├── docs/           Data paper for Scientific Data and the README maps
├── metadata/       Schema, data dictionary, project and per-partner metadata
├── datasets.csv    Index of all trials, with the path to each file
├── SKILL.md        Instructions for AI assistants using the data
├── AGENTS.md       Short instructions for AI agents using the data
├── MAINTAINERS.md  Guide for the tricot data managers
├── NEWS.md         What each release added
├── script/         build-parquet.py, which builds the parquet tables attached to each release

Use with AI assistants

This repository is the data layer of the maniva, the AI-ready documentation for tricot. Share that link with an AI assistant to reach the data together with the published evidence, protocols and analytical approaches, or use SKILL.md on its own.

Using the ranking data

To analyse rankings, read the file path from datasets.csv, keep the plot_data rows where value_type is rank, convert them to rankings with gosset::rank_tricot2(), and fit Plackett-Luce models.

library("jsonlite")
library("gosset")
library("PlackettLuce")

base = "https://raw.githubusercontent.com/AgrDataSci/tricot-data/main/"

# Find the trial in the index
idx = read.csv(paste0(base, "datasets.csv"))
path = idx$path[idx$trial_id == "88f82a3f073e"]

# Load the trial
dat = fromJSON(paste0(base, path))

# Study metadata, and what adm1 to adm5 are in the country of the trial
dat$metadata$study
dat$metadata$admin_levels

# Sites, with the standard column names of schema 3.0
site = dat$site_data
head(site[, c("site_id", "country", "crop", "adm1", "adm2", "location",
              "longitude", "latitude", "geo_source", "farmer_gender", "planting_date")])

# Rankings by collection moment and trait.
# trait is the standard name, the same across trials. trait_original is the
# question code of the trial, and keeps apart two questions on the same trait.
R = dat$plot_data
R = R[R$value_type == "rank", ]
R$traitmoment = paste(R$collection_moment, R$trait, R$trait_original, sep = " - ")

R = split(R, R$traitmoment)
R = lapply(R, function(x) {
  rank_tricot2(
    data  = x,
    items = "genotype_name",
    input = "value",
    id    = "site_id"
  )
})

mod = lapply(R, PlackettLuce)

# Measured values. value_standard is the value in the standard unit of the
# trait (unit_standard), the same unit in every trial.
Y = dat$plot_data
Y = Y[Y$trait == "yield" & Y$value_type == "numeric", ]

# A yield flagged "derived" was computed from the plot weight and the plot area.
# "plot_area_assumed" marks a protocol plot size, not a reported one. Drop these
# to keep only yields from reported or published plot areas.
Y = Y[!grepl("plot_area_assumed", Y$flag), ]
Y$yield_kg_ha = as.numeric(Y$value_standard)

aggregate(yield_kg_ha ~ genotype_name, data = Y, FUN = median)

Moving code from schema 2.1 to 3.0

Code written for the earlier files needs these changes.

In schema 2.1 In schema 3.0
site_data$country_name, site_data$crop_name site_data$country, site_data$crop
coordinate_source, coordinate_uncertainty_m, coordinate_flag geo_source, geo_uncertainty, geo_flag
adm1, adm2, adm3 as region, district, village adm1 to adm5 as defined for the country in metadata$admin_levels, and location for the village
gender, plot_area_m2 and other site columns standard names such as farmer_gender and plot_area. The earlier name of each column is in metadata$variables$original_name
trait names such as overallchar or yieldperse standard names such as overall_performance, yield_plot (plot weight) and yield (kg/ha). The question code of the trial stays in trait_original
value and unit for measurements unchanged, with value_standard and unit_standard added. Use these to combine trials
one flag in flag one or more flags separated by ;, so test with grepl() and not with ==
collection moment labels unchanged, with standard_moment added in metadata$study$collection_moments

The table and column structure is the same. plot_data is still one row per site, plot, collection moment and trait, and links to site_data through site_id. For more information, check the gosset package documentation.

Analysis tables (parquet)

Each release also carries every trial in three parquet tables, built from the JSON by script/build-parquet.py:

Table One row per Source in the JSON
plots.parquet plot, collection moment and trait plot_data
sites.parquet site site_data
variables.parquet trial and variable metadata.variables

Columns are those of the JSON plus trial_id, so rank_tricot2() reads the plots table as it reads plot_data. Use them to pool trials or to filter on site variables. The JSON stays the full record, with the metadata the tables leave out.

Report an error

The data are maintained by the tricot data managers. If something looks wrong, open an issue naming the file, the field and the problem. Corrections are made in the data managers' processing repository and published in the next release.

Citation

The dataset is archived on Zenodo. This DOI always resolves to the latest version.

The dataset is dedicated to the public domain under CC0 1.0. You can use it without restriction. Please cite it, and the publications named in each trial's metadata, so that the farmers, data producers and funders behind the trials get credit.

de Sousa, K., Laporte, M.-A., Abate, L., Abdulmalik, R. O., Abidin, E., Abolore, B., Achigan-Dako, E. G., Adonia, J., Agawini, J., Aglinglo, L. A., Aguilar, A., Ahmed, H. O., Andrade Lopez, R. S., Angudubo, S., Arnaud, E., Asirifi, O., Assefa, T. M., Atieno, E. O., Atieno, M., Attamah, P., … van Etten, J. (2025). Global multi-crop agricultural trial data supported by citizen science [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.17112491

BibTeX (dataset)

@dataset{desousa_tricot_2025,
  author       = {de Sousa, Kauê and Laporte, Marie-Angélique and Abate, Legesse and Abdulmalik, Rekiya O. and Abidin, Erna and Abolore, Bello and Achigan-Dako, Enoch G. and Adonia, Joseph and Agawini, Jonathan and Aglinglo, Lys Amavi and Aguilar, Amilcar and Ahmed, Hauwa Oziohu and Andrade Lopez, Robert Santiago and Angudubo, Stephen and Arnaud, Elizabeth and Asirifi, Ophelia and Assefa, Teshale M. and Atieno, Elly Ouma and Atieno, Mary and Attamah, Patrick and Ayenan, Mathieu and … and van Etten, Jacob},
  title        = {Global multi-crop agricultural trial data supported by citizen science},
  year         = {2025},
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.17112491},
  url          = {https://doi.org/10.5281/zenodo.17112491},
  note         = {Dataset}
}

About

No description, website, or topics provided.

Resources

Stars

3 stars

Watchers

2 watching

Forks

Releases

Packages

Contributors

Languages