About the Data

How the GRAND database is built — unifying science from around the globe.

Where it comes from

The GRAND database brings together temporally-resolved N2O flux measurements and their covariates from several major public sources — GRACEnet, Ag Data Commons, the NANORP N2O Network, and the Global N2O Database — plus roughly 45 additional peer-reviewed publications. Every record carries its original source citation, delivered with each download.

Getting that data into one consistent, queryable form took several steps:

1

Extraction

Where data was published as tables, we ingest it directly. Where it lived only inside a figure, we digitize it point-by-point with WebPlotDigitizer, then add covariates and metadata by hand from the manuscript text.

Values read from figures carry the small reading uncertainty inherent to digitizing a plot.

2

Harmonization

Source datasets rarely agree on units, names, or structure. Python pipelines convert every measurement to common units and map inconsistent terminology onto a single vocabulary: crops as Scientific (Common) names, and so on.

3

Treatment averaging

Measurements from multiple replicate plots within an experimental treatment are averaged to a single value, so each record represents a treatment's mean flux (and soil moisture, etc) on a given date, rather than one individual chamber reading. We capture the variance of N2O flux measurements by computing the standard error, which we also provide when possible to compute.

4

Covariate enrichment

Soil texture, pH, bulk density, organic carbon, and site climate (MAP/MAT) are taken from the source study wherever reported. Where a study didn't report a value, we fall back to gridded global products — SoilGrids for soil, NASA POWER for climate — and flag every such value as modeled rather than measured, so its origin is always clear.

Good to know

A few characteristics of the data worth keeping in mind as you use it:

Before you model

  • Measurement gaps. Most data comes from static chambers, which typically sample every 1–2 weeks. GRAND fills the calendar so dates are continuous, but it never invents flux values. There is no gap-filling by default (though this feature could optionally be added if the community requests it).
  • Treatment-averaged. Records are treatment means, not raw replicate-level data.
  • Measured vs. modeled. Where a study didn't report soil pH, bulk density, organic carbon, or site climate (MAP/MAT), we fill the gap from gridded datasets (SoilGrids, NASA POWER). Every download includes MODELED_VALUES.md, listing which sites and treatments carry estimates so you can exclude them.
  • Figure-digitized values carry a small uncertainty from reading points off a plot.

Full column definitions and complete per-record citations are provided in the README bundled with every download. For the complete methodology, see Ackett et al. (2026).

Start exploring About the project