Versión en español info@attraktorcode.com

High-precision weather forecasting platform

Before the weather decides.

Terrain-adapted high-resolution models, autonomous sensor networks and artificial intelligence agents to anticipate wildfires, agricultural risks and renewable output in a detail conventional models can't reach.

Scroll
111mModel resolution
×22More detail than 2.5 km high-resolution operational models
≈100%Territorial coverage with NB-IoT and LoRa, even without a mobile network
24/7AI agents correcting and monitoring

Where other models see a uniform area, we resolve the hyperlocal scale.

  1. 2,500 m cell

    Conventional operational model. Terrain is averaged over cells of 2.5 km or more. Valleys, slopes and small hollows all but disappear, and with them much of the weather driven by the terrain. In complex terrain, temperature and wind are smoothed out too much and no longer reflect the real variability of the land.

  2. 1,000 m cell

    The terrain starts to take shape, but its features are still partly smoothed out. Local differences in temperature and wind can barely be resolved yet.

  3. 333 m cell

    The terrain clearly shapes the simulation. Cold-air pools, temperature differences between elevations and preferred wind corridors emerge, revealing patterns that stay hidden at lower resolution.

  4. 111 m cell

    Attraktor Labs. The simulation enters the micrometeorological scale. Cold air pools in hollows, wind is channelled through passes and gullies, and strong contrasts appear between places just a few hundred metres apart.

Surface temperature and wind streamlines at each model's resolution · 20 × 20 km domain · scroll to increase resolution

Four layers, one forecast.

Each layer feeds the next. The result isn't a weather map: it's decisions.

Sources
  • Global models
  • Remote sensing and satellite
  • Weather stations
  • Our own sensors
  1. 01

    Our own high-resolution model

    Physics adapted to the terrain and continuous calibration with local observations. Down to 111 m resolution, with ensemble forecasting to quantify uncertainty too.

    Core
  2. 02

    Autonomous sensor network

    Low-power NB-IoT and LoRa nodes, modular and configurable, ready to measure multiple variables, with coverage where mobile networks don't reach.

    Observation
  3. 03

    AI agents and machine learning

    They fuse model, sensors and satellite; they detect anomalies and correct biases with every real observation.

    Intelligence
  4. 04

    Decisions

    Alerts, indicators and recommendations by sector, in a web dashboard or built into your own systems.

    Output

For those who depend on the weather.

Three sectors, one platform. Scroll through them or jump straight to yours.

A

Wildfires

Simulate how a wildfire will spread before it does.

From early warning to decision: sensors that detect smoke within minutes, simulations that show where the fire will go and how likely it is, and alerts for those who need to act.

Calibrated with real wildfires

Wildfires reconstructed from official data and satellite observation to tune the model to real fire behaviour.

Ensemble fire-spread forecasting

From an ignition point or perimeter, the platform simulates how the fire evolves using high-resolution weather, terrain and fuel models. It doesn't run one simulation but dozens, with small variations in wind, humidity and fuel: the result isn't a single line, but the range of possible outcomes and their probability.

  • Hourly spread isochrones, with their spread
  • Probability of the fire reaching each area
  • Most likely scenario and extreme scenarios
  • Burnt area, rate of spread and fire-front intensity

Early detection

Sensors in the forest that detect the fire in its smouldering phase, before flames are visible, and raise the alert within minutes.

  • Combustion gases (CO, volatile organic compounds)
  • Particulate matter (PM2.5)
  • Air temperature and humidity
  • Solar-powered, maintenance-free

Fire–atmosphere coupling

The fire changes its own wind and the simulation accounts for it: the front responds to the atmosphere and the atmosphere to the front.

Daily risk by area

Danger indices using the weather on each slope and the microclimate measured by our own sensors.

  • Fuel moisture
  • Channelled wind
  • Danger indices by sector

Alerts and coordination

The information reaches those who need to act, when it matters.

  • In-app, email and push notifications
  • Map of sensors, alerts and simulations
  • Integration with emergency coordination centres

Network planning

We plan where to install each sensor and gateway based on terrain, vegetation and communications coverage.

  • LoRa and NB-IoT coverage simulation
  • 3D deployment map
  • Health monitoring for every node

Prevention and scenarios

Simulate possible ignitions to plan prevention, firebreaks and the deployment of resources.

  • Public forests and natural areas
  • Power lines, railways and infrastructure
  • Wildland–urban interface
Ensemble forecast: each hour, the spread of possible perimeters from 10 scenarios · click the map to start a fire
B

Agriculture

Hyperlocal forecasts for each field, not a regional estimate.

Hyperlocal forecasts for each field, combined with the sensors installed on the farm. On your phone and computer, with alerts and recommendations ready to decide what to do today and what can wait until tomorrow.

Field-by-field forecasts

Hour by hour for the coming days, at 111 m resolution and corrected with observations from your own sensors.

  • Temperature and dew point
  • Rain and probability
  • Humidity, wind and gusts
  • Solar radiation

Frost alerts

Advance warning of start time, duration and minimum temperature in each field, so you can switch on frost irrigation, wind machines or heaters in time.

  • Wet-bulb temperature
  • Alerts on your phone
  • Event history

Spraying guide

An hour-by-hour traffic light to spray effectively and safely.

  • Wind and gusts: drift risk
  • Forecast rain: wash-off risk
  • Temperature and humidity: evaporation
  • Night-time temperature inversion

Precision irrigation

Water balance for each field to decide when and how much to irrigate.

  • Reference (ET0) and crop evapotranspiration
  • Soil moisture and tension probes at several depths
  • Dose recommendation

Pests and diseases

Risk models driven by the weather in your field, so you only treat when needed.

  • Leaf-wetness hours
  • Downy mildew, powdery mildew and other models
  • Growing degree days

Sowing, fertilising and harvest

Work windows based on the real state of soil and crop.

  • Soil temperature for sowing
  • Best time to fertilise
  • Grain moisture for harvesting

Your data, always available

Your own sensor network in your fields, and a virtual station anywhere in the territory.

  • Rain gauge, anemometer, thermo-hygrometer
  • Leaf wetness and soil probes
  • History, reports and API
72-hour forecast for one field: frost, wind, rain, irrigation and spraying windows
C

Renewable energy

Know how much you'll produce — and when it might fail.

Wind and solar resources change at local scale with terrain, exposure and atmospheric conditions. We model them at that level of detail for operators, portfolio managers and traders.

Asset-level production forecasts

Wind and solar for each plant, calibrated with its production and availability data to minimise imbalances.

  • The plant's actual power curve
  • SCADA integration
  • Portfolio aggregation

Every horizon

From the next few hours up to two weeks ahead, updated several times a day.

  • Nowcasting with real-time data
  • Intraday
  • Day-ahead for the auction
  • Medium range up to 14 days

Probabilistic forecasting

Not just a number: the range of possible output, so you can manage risk.

  • P10–P90 percentiles
  • Ensemble scenarios
  • Quantified imbalance risk

Ramps and extreme events

Anticipate what makes output drop or surge.

  • Up- and down-ramps
  • High-wind cut-outs
  • Blade icing
  • Storms, snow and dust on panels

Operations and maintenance

Plan work on site safely.

  • Weather windows for work at height and cranes
  • Lightning and gust alerts
  • Site access in snow or ice

Resource assessment and new projects

Complex terrain is where conventional models fail most — and where ours adds the most.

  • Wind and solar resource assessment
  • High-resolution historical series
  • Micrositing support

Delivery and integration

Forecasts where you need them, 24 hours a day, 365 days a year.

  • API, FTP and web dashboard
  • Accuracy reports (MAE, RMSE)
  • Format tailored to your system
48-hour forecast: wind with uncertainty band, wind power output and solar radiation

We measure where coverage doesn't reach.

Low-power autonomous nodes, powered by solar panel and battery, with NB-IoT and LoRa connectivity. Their modular design takes weather, environmental and soil sensors. With NB-IoT coverage they transmit directly, even where conventional mobile coverage doesn't reach; beyond it, LoRa sends the data to the nearest connected node, which acts as a gateway.

NB-IoT nodes with their own antenna, and LoRa nodes relaying through the nearest NB-IoT node

One node, many variables.

From a full weather station to nodes specialised in soil, radiation, air quality, snow or process-specific variables. A modular design adapts each deployment to the real measurement needs. Some examples:

  • Temperature and humidity
  • Wind
  • Precipitation
  • Solar radiation
  • Pressure
  • Soil moisture
  • Soil temperature
  • Leaf wetness
  • Fuel moisture
  • Water level
  • Flow rate
  • Air quality
  • Smoke and gases (CO, VOC)
  • Vibration
  • Energy consumption
  • Cameras
  • …and whatever you need

They don't check the sky once a day. They watch it nonstop.

  • They analyse large volumes of data from models, sensors and remote sensing.
  • They detect anomalies and relevant patterns before they become a problem.
  • They correct biases by comparing every forecast with the real observation.
  • They turn data into action: alerts, indicators and recommendations for each sector.
Agents · sample activity

    Calibrated in the Serranía de Cuenca. Ready for your territory.

    One of the most demanding terrains in the Iberian Peninsula: sharp contrasts, deep valleys and microclimates a few metres apart. The model has been calibrated and validated there with observations from meteovegadelcodorno.es and AEMET (Spain's national weather service).

    For each new territory we adapt the model configuration and deploy the observation network it needs.

    Attraktor Labs is Attraktor Code's weather intelligence platform.

    Attraktor Code designs systems that connect software, artificial intelligence, cloud, data, IoT and scientific modelling. Attraktor Labs is where that engineering becomes a ready-to-use product: operational weather forecasting for the sectors that need it most.

    Discover Attraktor Code

    Request a demo with your territory.

    Tell us where you work and what you need to anticipate. We'll show you the platform with data from your area.

    info@attraktorcode.com
    What do you need?

    Concepts

    What is a model grid cell?

    The grid

    A weather model doesn't compute the weather at every point of the land: it divides it into a regular grid of squares, the cells, and solves the equations of the atmosphere in each one. The side of the cell is the model's horizontal resolution .

    One value per cell

    For each cell the model gets a single value of each variable — temperature, wind, humidity, rain — representing the average over its whole area. What happens inside the cell is invisible: it's averaged out.

    The terrain is averaged too

    Terrain is represented by one elevation per cell. With 2.5 km cells, a 1 km-wide valley fits inside a single cell: the valley floor, slopes and peaks are blended into one average elevation, and the model can't represent the cold air pooling in the valley or the wind channelled along it.

    What changes at 111 m

    One 2.5 km cell is equivalent to about 506 cells of 111 m (22.5 × 22.5). At that resolution the valley spans dozens of cells, the terrain looks like the real thing and local phenomena appear: temperature inversions, frost in the hollows, channelled winds and speed-ups over the ridges.

    More resolution takes far more computing: dividing the cell side by 22 multiplies the number of cells by about 500 and also forces shorter time steps. That's why general operational models work at the kilometre scale — and why our model is configured for each territory.