Sectors Architecture Why it works Team Careers Contact Request a demo

Every minute. Every metre.

We forecast the temperature. You raise the line rating.

Point-resolved forecasting, and the hazard models each sector acts on.

LAT/LON
50.0755° N  14.4378° E
T
−2.4 °C
FF
11 m/s
RR
0.8 mm/h
N
6/8

Sectors

Seven sectors where weather sets the risk.

Pick a sector to see what comes out of the top of the stack: a hazard layer over one landscape, at one hour, with a threshold attached and a decision already written against it.

Sector 01 · Energy Intelligence

Icing & dynamic line rating · 400 kV network

  • Conductor temperature
  • Dynamic line rating
  • Icing risk
  • Storm risk
  • Asset health
  • Maintenance planning
Shaded relief map of the Czech Republic showing the 400 kV transmission network with icing hazard states at each substation.

Hradec

V441 · Hradec — Výškov

LAT/LON
50.3630° N  13.2810° E
Alt
322 m
T conductor
−1.8 °C
Icing rate
1.2 mm/h
Line rating
−2 % headroom

Alert

De-rate V441 to 2 060 A and hold until the icing rate falls below 0.2 mm/h.

Select a site

  • Normal
  • Watch
  • Alert

Simulated data · real terrain, real 400 kV sites

Sector 02 · Wildfire Intelligence

Spread forecast · 32-member ensemble

  • Fire danger index
  • Ignition probability
  • Spread modelling
  • Corridor risk
Shaded relief of the Labe canyon near Hřensko, with a simulated wildfire spread forecast and wind field.

T+12 h

Mezní Louka · 50.8790° N 14.3230° E

Expected
0.47 km²
90 % core
0.17 km²
10 % envelope
1.49 km²
Head ROS
168 m/h
Wind
7.5 m/s from 235°

Spreading

Hold the line on the eastern ridge. The 10 % envelope is three times the expected perimeter — commit resources to the head, not the flanks.

Lead time

  • 90 % core
  • Expected
  • 10 % envelope
  • Wind

Simulated data · real terrain, modelled fire

Sector 03 · Agriculture Intelligence

Radiative frost · apple, full blossom

  • Irrigation planning
  • Frost risk
  • Disease pressure
  • Growing degree days
  • Yield forecasts
Shaded relief of the Litoměřice orchard basin with a simulated overnight minimum temperature field showing cold air pooled on the valley floor.

Sink

50.5440° N  14.0508° E

Min temp
−4.8 °C
At
04:12
Below −2 °C
4.7 h
Elevation
147 m
Aspect
N · 14.5 %

Severe

Cold air ponds here all night. Protection must already be running before the pool forms — once it has, wind machines have no inversion left to mix down.

Block

    Simulated data · real terrain, modelled frost

    Sector 04 · Transport Intelligence

    Black ice · road surface temperature

    • Road icing prediction
    • Crosswind warnings
    • Fog forecasting
    • Aviation support
    Shaded relief of the D1 over Vysočina with the road network drawn on it and simulated black ice highlighted on the bridge decks and valley crossings.

    most Vysočina

    D1 · bridge deck

    Surface
    −2.4 °C
    Approach
    +0.5 °C
    Dew point
    +1.8 °C
    Air
    +2.4 °C
    Deck length
    430 m

    Black ice

    The deck is below freezing while the approach either side is not, and the dew point is above both — moisture is depositing onto the span and freezing clear. Treat the structure ahead of the route.

    Site

    • Black ice
    • Marginal
    • Above freezing

    Simulated data · real road geometry, modelled surface

    Sector 05 · Water Intelligence

    Vltava flood extent · Prague

    • Flash flood potential
    • Drought monitoring
    • Reservoir inflow
    • Catchment forecasts
    Bare-earth relief of central Prague with simulated Vltava flood extent covering Karlín, Libeň, Holešovice and Kampa.

    Karlín

    Aug 2002 · 5 160 m³/s

    Water depth
    3.7 m
    District under
    100 %
    Water level
    190.2 m
    Ground
    186.3 m
    Extent
    6.64 km²

    Inundated

    Water stands 3.7 m over Rohanské nábřeží and covers the whole district. The last two metres arrive between Q100 and peak, so the evacuation decision has to be taken at the Q100 threshold, not at the peak forecast.

    Scenario

    • Flood extent
    • Lower scenarios
    • Vltava

    Simulated data · real lidar terrain, modelled flood

    Sector 06 · Cities & Climate

    Urban heat island · 03:00 air temperature

    • Urban heat island
    • Heat-health warnings
    • Energy demand
    • Air quality
    Central Prague with a simulated heatwave night temperature field: the dense blocks several degrees above the rural reference, the parks and river corridor cooler.

    Old Town

    Medieval core

    At 03:00
    22.5 °C
    Above rural
    +5.1 °C
    Sky view
    0.22
    Built
    63 %
    Mean height
    18 m

    Tropical night

    The canyon sees a fifth of the sky, so the heat it stored today has nowhere to radiate. It never drops below 20 °C — residents get no overnight recovery, which is the exposure that fills hospital beds, not the afternoon peak.

    District

      Simulated data · real lidar city fabric, modelled night

      Sector 07 · Defence & Government

      Ground trafficability · route selection

      • Route & mission planning
      • Ground trafficability
      • Sub-cloud terrain monitoring
      • Personnel exposure limits
      Lidar relief of the Libavá training area with simulated no-go ground and a routed line between an origin and an objective.

      +60 mm

      VVP Libavá · wheeled, 30 % limit

      Route
      12.9 km
      Moving time
      40 min
      Detour
      +68 %
      Max gradient
      29 %
      No-go ground
      17 %

      Route degraded

      The valley floors have lost bearing capacity and the direct line is 15 % no-go. The routed line adds 3.6 km and 28 minutes but stays on ground that will take a wheel — the terrain did not change, the soil did.

      Rainfall

      • Routed line
      • Direct line
      • Slow going
      • No-go

      Simulated data · real lidar terrain, modelled ground state

      Architecture

      One engine. One platform. One intelligence per sector.

      Everything we ship sits on a single vertical stack. Atmospheric physics at the bottom, an operational decision at the top — and every layer above the engine inherits its resolution.

      LAYER 01

      Hyperlocal Forecast Engine

      Point-resolved, any latitude, longitude and altitude

      Seven variables resolved to a coordinate and an altitude rather than a grid cell — including at sites tens of kilometres from the nearest station.

      • Temperature
      • Wind
      • Humidity
      • Solar Radiation
      • Cloud Cover
      • Precipitation
      • Snow / Ice
      LAYER 02

      Environmental Intelligence Platform

      Variables in, consequences out

      Five capabilities stand between a raw variable and something a duty officer can act on: fusion, models that have learned the physics, hazard models, validation, delivery.

      • Data Fusion
      • AI / ML
      • Hazard & Risk Models
      • Validation & Calibration
      • Decision Support
      LAYER 03

      Sector Intelligence

      Hazard layers, not raw variables

      A consequence layer per sector. Fire danger, not humidity. Line rating headroom, not temperature.

      Energy
      • Conductor temperature
      • Dynamic line rating
      • Icing & storm risk
      Wildfire
      • Fire danger index
      • Ignition probability
      • Spread modelling
      Agriculture
      • Irrigation planning
      • Frost & disease risk
      • Yield forecasts
      Transport
      • Road icing
      • Crosswind warnings
      • Fog forecasting
      Water
      • Flash flood potential
      • Drought monitoring
      • Reservoir inflow
      Cities & Climate
      • Urban heat island
      • Heat-health warnings
      • Energy demand
      Defence
      • Route & mission planning
      • Ground trafficability
      • Personnel exposure
      LAYER 04

      Decision makers

      Where the forecast becomes an action

      A line de-rated, a crew pre-positioned, a road salted, a reservoir drawn down before the rain arrives.

      • TSOs & DSOs
      • Fire & Rescue Services
      • Civil Protection
      • Forest Agencies
      • Farmers & Agribusiness
      • Road Authorities
      • Railway Operators
      • Airports
      • Water Authorities
      • Municipalities
      • Insurance
      • Industry

      Why it works

      Interdisciplinary by construction, not by accident.

      Meteorology and machine learning in the same room. That is why we use AI to learn the physics of the atmosphere first, and carry what it learns into every model we ship — and why the outputs are scored against observations rather than asserted.

       
      Classic NWP forecast
      Tempest hyperlocal
      Resolution
      10 × 10 km to 50 × 50 km grid cell
      Resolved to a coordinate and altitude
      Update cycle
      Hourly at best
      Continuous, driven by live point measurements
      Coverage gaps
      Blind where radar and stations are absent
      Satellite-derived nowcasting fills radar-scarce regions
      Output
      Standard meteorological variables
      Sector hazard indices and decision thresholds
      Terrain
      Smoothed to the model orography
      Real elevation, slope, aspect and land cover
      Physics

      Learned first, then applied

      We use AI to learn how the atmosphere actually behaves — and then apply what it learns inside the forecast, so outputs stay physically plausible in regimes the training record never covered.

      Choma, Šimánek & Bartel — Improving deep learning precipitation nowcasting by using prior knowledge, 2023 ↗

      Global

      Including where radar ends

      Geostationary imagery plus AI that has learned the physics extends precipitation nowcasting across Africa, South America, Asia and open sea.

      Pihrt, Raevskiy, Šimánek & Choma — WeatherFusionNet. 1st place, NeurIPS 2022 Weather4Cast ↗

      Scored

      Verified, not asserted

      Models are checked against withheld observations and reference forecasts, with skill tracked per variable, per lead time and per region — and recalibrated when it drifts.

      Methodology and per-variable skill scores available on request, and under NDA for pilot evaluations.

      Who we are

      Core Team

      AI, data science, physics and engineering — assembled out of the Czech Technical University in Prague, where most of this research started.

      Alexander Kovalenko
      Alexander Kovalenko
      CEO, Co-Founder
      Google Scholar

      PhD in theoretical physics. AI research scientist at CTU.

      Petr Šimánek
      Petr Šimánek
      CSO, Co-Founder
      Google Scholar

      AI researcher in meteorology, CTU.

      Vojtěch Rybář
      Vojtěch Rybář
      COO, Co-Founder
      Google Scholar

      AI researcher at CTU. Consultant in energy trading.

      Miroslav Čepek
      Miroslav Čepek
      CTO, Co-Founder
      Google Scholar

      PhD in informatics. AI research scientist at CTU.

      Jiří Pihrt
      Jiří Pihrt
      ML Development Lead, Co-Founder
      Google Scholar

      PhD student at CTU.

      Filip Miškařík
      Filip Miškařík
      Lead Engineer

      PhD student at CTU.

      Careers

      Come and work on the hard part.

      AI that learns the physics, satellite nowcasting and hazard modelling for operators who act on the answer. Prague-based, with the research group next door.

      Talk to us

      Tell us the decision weather is putting at risk. We’ll show you the layer that answers it.

      Or email info@tempestlabs.ai
      Prague, Czech Republic