Planting IntelligenceAbout the data
Back to cities

About the data

What the scores mean, and what they do not.

What a score means

Each crop is compared with the crops people have recorded growing near the city. A crop that looks like those has a higher rank. It is a guide to what is likely to suit a place, not a promise of yield.

Where the information comes from

  • Crop records: GBIF, within 100 km of each city.
  • Soil: the WoSIS soil database. Where a city has no measurement, we show it as not measured instead of guessing.
  • Climate: ten-year averages from NASA.
  • Companion crops and crop needs: a hand-written list. Every entry shows its source, including when it has not yet been checked against a paper.

How well it works

The model beats the best simple baseline by 0.085 AUC (95% range 0.034 to 0.136), and the gain holds when cities are resampled.

1886 crop and city pairs across 46 cities were tested; 17% of them were recorded nearby. Each city was scored by a model that had never seen it.

Keep in mind

  • A score says how much a crop looks like the crops people recorded near this city. It is not a yield forecast.
  • Recorded means someone logged the crop on GBIF within 100 km. A crop nobody logged can still grow well.
  • The reasons describe how the model scored the crop, not proof of what makes it grow.
  • Some crop requirements come from a general agronomic reference that has not been checked against a paper yet; each crop shows its source.