.png)
☰☰.png)
OUR TECHNOLOGY
From observation to decision
Seven connected stages turn raw atmospheric data into clear, actionable temperature intelligence at the level of cities, neighborhoods and street blocks.
Every output we deliver follows the same path, from raw observation through to product. You always know where the number came from.
Raw Data
Large Temperature Models
Temperature Catalogs
Temperature Analytics
Temperature Insights
Temperature Recommendations
Products
Raw Data
Large Temperature Models
Temperature Catalogs
Temperature Analytics
Temperature Insights
Temperature Recommendations
Products
What each stage does, and how it builds on the last.
STAGE
01
Three input streams, one fused record.
Three input streams are fused at this stage: proprietary collected data, satellite imagery, and simulated atmospheric reanalysis data. The output is one time-aligned record of atmospheric and surface conditions, ingested continuously and made available to every stage downstream.
Three streams cover what each one misses — proprietary collected data supply ground-truth at points, satellite imagery extends spatial coverage, and assimilated reanalysis provides physics-consistent fill where direct measurement is sparse.
Temperature travels alongside the variables it is physically coupled to, so downstream models resolve it within its full physical context.
Ingestion runs continuously at the refresh rate of each source; older records are retained, so historical and forecast workflows draw from the same feed.
STAGE
02
Deep learning, built for temperature.
A family of deep learning models trained specifically for temperature prediction. All models in the family share a transformer-based architecture; what differs between them is the temporal scale each is built for, from short-range nowcast to multi-year climate projection.
The family spans three model classes: downscaling models that resolve temperature down to 20-metre spatial detail, weather-forecasting models that predict up to 12 hours ahead, and climate models that project long-range trends up to 10 years out.
Time horizons span the next hour to the next decade, so nowcast, forecast, seasonal and multi-year climate projection are all served within one continuous model family.
Outputs are probabilistic — each value is paired with its uncertainty, which downstream stages carry through analytics and insights rather than drop at the model boundary.
STAGE
03
Unified datasets for all downstream stages.
Outputs from the model family are persisted into a versioned, queryable catalog. Every downstream stage in the pipeline reads from this catalog, so all stages share the same underlying numbers.
The catalog is stored in the cloud-native Zarr format, which supports partial reads — a query for a single AOI, time window or variable returns only the requested slice rather than the full dataset.
Analytics, insights, recommendations and the delivery products all draw from the same catalog, so values stay consistent across the pipeline and across queries — the catalog is the single source of truth.
Provenance is retained per entry — model version, input observations and ingest timestamp — so any value can be traced back to the data and model that produced it.
STAGE
04
From data to signal.
Analytics fall into two categories, distinguished by whether a criterion is part of the analytic itself. Observational analytics describe patterns in the temperature data without comparing against a threshold or baseline. Diagnostic analytics evaluate the data against an explicit criterion — a threshold, a reference baseline or a published index — so the output measures behaviour against that criterion.
Observational analytics cover a wide range of descriptive views — from instantaneous single-hour fields through to long-window aggregations and full distributional summaries — so the everyday behaviour of the temperature field is characterised across space and time directly from the data.
Diagnostic analytics evaluate the field against an explicit reference — from user-configurable thresholds through to published scientific standards and reference baselines — so any extreme or anomaly is reported in terms of how it compares to an operational or standards-anchored benchmark.
Beyond raw values, the analytic layer surfaces the geophysical structure underneath the field — from coastline edges and elevation transitions through to land-use and climate-regime boundaries — so the physical drivers behind the temperature signal become directly readable.
STAGE
05
From signal to interpretation.
Insights translate analytic signal into exposure context, expressing what the measured temperature implies for the populations, assets and operations located inside the AOI.
Insight outputs are aligned to published standards — WMO, NWS, ETCCDI and ASHRAE — so they reference the same definitions used by the agencies and stakeholders that consume them.
Spatial pattern detection covers a wide range of capabilities — from localised hotspot identification through to large-scale regime and boundary characterisation — so spatial structure is reported explicitly rather than left implicit in the grid.
Each cell's temperature is paired with the populations and assets co-located inside it, so exposure-and-risk readings report impact in terms of who and what is affected rather than as a raw thermal value alone.
STAGE
06
From interpretation to intervention.
Recommendations convert analytic and insight outputs into physical interventions that act directly on the temperature field at the locations identified by the prior stages. Interventions are organised by cooling mechanism, ordered by severity, with each action traceable back to the underlying signal.
Recommendations span four intervention categories — water-based interventions covering irrigation and evaporative cooling, green-based interventions covering vegetation and canopy cover, shade-based interventions ranging from built shade structures through to urban geometry, and material-based interventions covering high-albedo and reflective surfaces — so each recommendation targets a specific physical pathway from intervention to thermal effect, matched to the conditions that triggered it.
Within each category, recommendations are ranked by severity rather than presented as a flat list, so high-impact actions surface first under operational time pressure.
Every action carries provenance back to the analytic and insight that produced it, so the trigger condition for any recommendation is auditable.
Pick the surface that fits your team. Or use all three.
A map-first browser UI built for situational awareness. Explore the pipeline on the geography that matters to your team.
Programmatic access to every layer of the pipeline. Built to plug straight into the GIS, asset and operational systems your team already runs on.
Heat Intelligence Advisory that translates heatmaps into procurement-ready scopes, ROI logic, and measurement plans.
.png)
.png)
.png)
.png)
.png)