Plan the next five years of a field with two decades of Earth observation.

Mark a field, set what matters most and compare crop rotations scored against the climate, soil and vegetation record of that exact location.

Open the demo field Mark your own field

Travertine supports farmer judgment and local expertise; it does not replace them. Check its suggestions with people who know the field.

Try it: move a priority

The engine's rotations for the Denizli demo field, re-ranked as you change how much saving water matters. Other priorities stay at 50 %.

Save water

Less irrigation need and water deficit

Top rotation now

Loading data

Open all rotations and priorities

Checked against the satellite record

Loading data

See the full back-test, method and caveats

How it works

1
Mark the field
Draw it on the map or drop a point. The location goes into the link, so the view can be shared.
2
Read its record
Daily NASA POWER weather since 2001, FAO-56 reference evapotranspiration, SoilGrids soil and MODIS vegetation for that location.
3
Compare rotations
Every rotation is replayed through a daily FAO-56 soil-water balance over past seasons and ranked by what matters to you.

The problem

Farms in dry regions such as the Büyük Menderes basin face hotter and more variable growing seasons. Which crops to grow, and in what order, decides how much water a field needs and how often it runs short. Long climate and satellite records exist for every field on Earth, but they are rarely turned into an answer for one specific field.

What do you grow now?

Tell Travertine the rotation the field grows now. It is scored with the same model and every alternative shows its difference from it, along with any rotation rule your current sequence breaks.

Compare your current rotation

Data sources

NASA POWER
Daily weather and solar radiation since 2001 (NASA Langley Research Center), 0.5° × 0.625° grid.
MODIS MOD13Q1
Growing-season vegetation greenness (NDVI) at 250 m (NASA LP DAAC).
HLS v2.0 (Landsat + Sentinel-2)
Field-scale (30 m) vegetation greenness since 2020 from Landsat 8/9 and Sentinel-2 (NASA LP DAAC).
ISRIC SoilGrids 2.0
Soil texture, organic carbon, pH and available water at 250 m (ISRIC).
FAO-56
Reference evapotranspiration and crop coefficients (FAO Irrigation and Drainage Paper 56).

How Travertine answers the challenge

Use NASA and other Earth observation data: Daily NASA POWER weather since 2001, MODIS NDVI at 250 m and HLS at 30 m (when the engine serves it), SoilGrids soil, with provenance on every number.
Local soil information and crop characteristics: SoilGrids texture, organic carbon, pH and available water for the field; crop coefficients cited from FAO-56 for every crop.
Farmer priorities: Four priority sliders (water, income, soil health, risk) re-rank the strategies; the current rotation can be entered as a baseline.
Adaptive crop-rotation strategies: Every valid 3- to 5-year rotation is replayed through a daily FAO-56 water balance over past seasons and ranked.
Explain the reasoning: Each strategy carries the engine's own explanations and, when the engine provides them, five parts: why suggested, evidence, benefits, information gaps and risks for local review.
Show uncertainty and limits: A pre-registered back-test against satellite vegetation with confidence intervals, a simple baseline to beat, and caveats shown as returned.

Meet Cosmo, the guide

Cosmo walks you through the screens and explains the terms Travertine uses. It answers only from fixed, reviewed text: no language model, no generated advice. Open it from the button at the bottom right.

Evapotranspiration

Evapotranspiration is the water that leaves a field as vapour: evaporation from the soil plus transpiration through crop leaves. Reference evapotranspiration (ET₀) is the rate for a standard, well-watered grass, computed here with the FAO-56 Penman-Monteith equation from NASA POWER weather. A crop's need is ET₀ multiplied by its crop coefficient (Kc).

Soil-moisture anomaly

A soil-moisture anomaly is how much wetter or drier the soil is than usual for that place and time of year, measured against the long-term average. Travertine does not compute soil-moisture anomalies yet. It shows NASA POWER root-zone wetness (0 dry, 1 saturated), a modelled value for a grid cell of about 55 × 55 km, not a probe reading from your field.

NDVI

NDVI (Normalised Difference Vegetation Index) compares the near-infrared and red light a surface reflects: (NIR − Red) / (NIR + Red). Green, dense vegetation gives high values; bare soil gives low values. Travertine uses MODIS NDVI at 250 m and, where the engine provides it, 30 m HLS NDVI.

Frost risk

Frost risk is the chance that the temperature falls to 0 °C or below while a crop is sensitive, for example at flowering. It is judged from daily minimum temperatures on record. On the field page Travertine shows, from NASA POWER, the median last spring and first autumn frost, the frost-free season and the number of frost days; the plan page shows the share of frost days in each month a crop is in the field. These are past frequencies for a grid cell about 50 km across, not a forecast for your field.

Water balance

A water balance tracks the water in the root zone day by day: rain and irrigation add water, crop evapotranspiration removes it, and water beyond what the soil can hold drains away. When stored water runs low the crop is stressed. Travertine's rotation scores come from this FAO-56 daily balance replayed over past seasons.

Satellite versus ground measurements

Satellites and models give averages over an area, not a point: a 250 m pixel for MODIS NDVI and a grid cell of about 55 × 55 km for NASA POWER weather. A ground station or soil probe measures one spot precisely. Satellite and model data cover every field and many years; ground measurements are more exact where they are taken. Your own field records are better where you have them.

What it cannot tell you

Weather is an average over a grid cell of about 55 × 55 km, not a measurement in your field.
Soil values are model predictions for a 250 m pixel with wide uncertainty; your own soil test is better.
The vegetation check uses one 250 m pixel whose land use is unknown, and so far one site.
Every number is replayed from past weather. Travertine makes no forecast of future climate.
Scores compare rotations with each other; they are not yield or price forecasts.

Travertine supports farmer judgment and local expertise; it does not replace them. Check its suggestions with people who know the field.

Roadmap

PlannedClimate trend tests (Mann-Kendall, Sen's slope) and a season-so-far view.
PlannedOutcome ranges (P10, P50, P90) per rotation and the Pareto set: rotations that no other rotation outscores on every objective.
PlannedYield response with published Ky factors (FAO-33, FAO-66).
PlannedField-scale vegetation back to 2013 and IMERG rainfall, which need a NASA Earthdata login.
PlannedA load test and one command that regenerates every result and figure in the docs.

Team

Mustafa Babayev
Product and web lead
Kemal Gencel
Science engine lead
Azra Özkan
Research support

Built with AI coding tools under human review: Claude Code (Anthropic) agents for the engine, contract and web app, and OpenAI Codex, which a teammate used to build the current interface and the Cosmo guide. Humans review every change before it is merged. Cosmo itself uses no language model: every answer is a fixed, reviewed text. The full statement is in docs/AI_USE.md.

Source code on GitHub