Ara · Territorial's weather engine

Weather forecasts,
calibrated for Brazil.

Ara takes the best global forecast there is, checks it every day against what Brazil actually measured, and corrects it with what it learns. Maps, places and river basins, through one API.

0.1°

one grid over South America, about 11 km

4×

global model runs a day, single and ensemble

5 min

between radar frames

1 h

between analyses of what the stations measure

2024

every forecast kept and verified since February

Why calibrate

A global model sees Brazil in 9 km boxes.

ECMWF runs the best global model in the world, and its forecasts are open to everyone. Where a global model struggles is the tropics. Ara learns those local errors from years of Brazilian measurements and corrects them, place by place, and every correction has to show up in the scores.

01

Storms that build in one afternoon

Tropical rain is local and fast. Radar and hourly stations show where it really fell, so the forecast learns where it tends to miss.

02

Rain on one side of the hill

Terrain, land cover and distance to the coast change the weather within a few kilometres. Ara feeds them into the correction.

03

Frost the grid cannot see

A cold valley is smaller than a model box. Corrections per cell, per season and per lead day bring the minimum down to where it belongs.

How it works

From raw model to calibrated forecast, every run.

  1. 01 · Ingest

    Models and measurements

    ECMWF's single and ensemble forecasts, four runs a day. Rain and temperature grids from CPTEC/INPE, INMET's automatic stations, REDEMET's radar.

  2. 02 · Harmonize

    One grid, one language

    Everything on the same 0.1° grid, with the same variables and local days. Humidity, wind, evapotranspiration, water balance and degree days derived.

  3. 03 · Verify

    Checked against reality

    Every forecast paired with what was observed, by lead day and by region. The scores are kept, not sampled.

  4. 04 · Calibrate

    Corrected where it errs

    Corrections learned per cell, lead day and season. One ships only when it beats the raw model on a year it never saw.

  5. 05 · Serve

    Delivered at the edge

    An API with keys and scopes; maps and city forecasts as static files at the edge. The raw model always beside the corrected one.

Who it's for

For the people whose decisions depend on the weather.

01

Agribusiness

Farmers, cooperatives and banks: series for a point or a whole field, chance of rain, evapotranspiration, windows for spraying.

02

Energy and hydrology

Rain over river basins, forecast and observed, for reservoirs and inflow, on the national basin map (ANA's Ottobacias).

03

Civil defense

Heavy rain, radar every five minutes and an hourly picture of what the stations measure.

04

Apps and media

Animated forecast maps and a forecast for every city in Brazil, ready to build a product or a page on.

05

Data teams

Ara's grids and series themselves, raw and calibrated, for your own models and reports.

Built on Ara

TempoFirme

Our own forecast app for Brazil runs entirely on Ara: moving maps, every city's forecast, live stations and a page that puts what was forecast next to what happened.

tempofirme.com.br

Delivery

One hosted API, served from the edge.

  • Points, fields and basins: a time series for any place or area you send.
  • Maps as data, not pictures: values a browser can animate and recolour.
  • Keys with product scopes and a usage record per key.
  • Observations and verification on the same API as the forecast.
ara · /v1
GET /v1/points/forecast # a place's forecast series
GET /v1/points/observed # what was measured there
POST /v1/polygons/aggregate # a field or basin, your GeoJSON
GET /v1/stations/{id}/latest # a station's last hour
GET /v1/verification # forecast against observed
GET /v1/maps/{product} # the map, as data

Ara builds on ECMWF open data (CC BY 4.0), MERGE and SAMeT from CPTEC/INPE, INMET's automatic stations and REDEMET's radar, credited on every product.

Want Ara's forecast in your product?

Tell us the place, the variable and the decision it feeds. We answer within one business day.