(Case study · 2026)

GeoPlant

A bilingual decision tool that tells a landowner which tree and shrub species suit a specific plot, using climate, elevation and soil data, and is explicit about what it does not know.

Role
Product, UX/UI and build (solo)
Platform
Web app, Persian and English
Scope
Pilot region: Mazandaran, Iran
Status
Beta 1.045

Open the GeoPlant beta

GeoPlant

In short

  • Problem: people planting on their own land choose species by habit or a nursery's advice. A wrong choice costs years, not days: the trees die and the money is gone.
  • What I did: defined a deliberately narrow first version, designed the flow and built a working beta that pulls live climate, terrain and soil data for any point.
  • Where it is: a public beta covering one validated pilot region, with the species catalogue under review.

The problem

The data needed to choose a tree well already exists: decades of temperature records, elevation models, soil maps and records of where species grow. It is just scattered across scientific services and written for specialists. A landowner who wants shade or fruit on 200 square meters cannot use any of it.

The risk is uneven, too. In northern Iran, conditions change quickly between the humid coastal lowlands and the Alborz slopes a few kilometers away. A species that thrives in one place can fail in the next.

The principle: being wrong is expensive

Most of the product decisions follow from one sentence: a wrong recommendation means dead trees and wasted money. That made caution a feature, not a disclaimer.

  • One pilot region, on purpose. Version one covers the Mazandaran lowlands and foothills, where the rules can be checked. The map still shows all of Iran so people can see where coverage ends; outside the pilot it shows climate only, with a clear notice.
  • "Don't know" is always an answer. Every question accepts it, and a "don't know" makes the result more cautious. The tool never fills a gap with a guess.
  • AI helps with the form, never with the verdict. You can describe your land in a sentence and AI fills in the questionnaire. It does not rank or recommend species; the rules do.
  • Protected land is excluded. Hyrcanian forests and protected areas are left out of planting advice entirely.
GeoPlant's explanation of why version one covers only Mazandaran

The flow

Four steps, shown as a progress bar: Location → Questionnaire → Results → Species file.

1. Location

Tap the map, type coordinates, paste a map link or pick a city. The tool confirms whether the point is inside the validated zone and previews elevation, rainfall and the coldest night in 25 years before you answer anything.

GeoPlant location step: map of Mazandaran with the pilot zone, and the selected point's elevation, rainfall and coldest night

2. Questionnaire

Eight short questions about the land, each with plain options: what you want from the trees, type of land, summer water, sunlight, drainage after rain, space per tree, nearby buildings or pipes, and how much care you can give. Answers that change the result are labelled as such.

GeoPlant questionnaire with an AI free-text fill and plain answer options, each with a Don't know choice

3. Results

The result leads with the answer, then shows its working. For the sample plot, a first screening of 1,203 tree and shrub species rules out 597 as unsuitable for the climate. Of the 46 species in the reviewed catalogue, 14 are suitable, 28 conditional, 3 to avoid and 1 lacks evidence, and the reasons are ranked (unknown drainage is the most common one).

Under the verdict, What this result is built on lists every answer, with the uncertain ones highlighted and editable in place, so the user can see exactly which answer to improve.

GeoPlant results: verdict, the share of suitable, conditional and rejected species, the reasons, and the inputs the result is built on

4. Species file

Each species card shows whether it is native, introduced or naturalized, its water need and mature height, and a short local note. Every environment figure on the page names its source and its certainty; elevation, for example, is cross-checked against a second terrain model.

GeoPlant environment cards with sources and certainty, and suitable species cards

Making the data visible

A tool that asks people to plant on its word has to show where its numbers come from. A Data sources page lists every source: 8 connected and live (including NASA POWER climate data, Copernicus elevation, ISRIC SoilGrids, GBIF plant records and Sentinel-2 imagery), 9 built-in and provisional, 10 planned and 25 for later. Results can be exported as JSON or CSV.

GeoPlant data sources page: live connections with what each is used for and when

What is next

  • Expert review of the species catalogue: 50 to 100 reviewed species are safer than thousands of unreviewed ones.
  • Soil tests as an input, since the soil model is an estimate and is not used by the rules yet.
  • New regions only after their zones and species are validated, not just by changing code.
  • What I will measure: how often each question gets a "don't know" (to find the questions people cannot answer), how many plots fall outside the pilot (to choose the next region), and how many results are exported or revisited.