How warm could it get?Temperature forecasting
Study the remaining temperature path and the distribution of a day’s maximum. Update the forecast as new observations and weather guidance become available.
Weather · Evidence · Uncertainty
Making sense of
an uncertain sky.
Independent research into temperature forecasts and the decisions they inform. We build software to understand how the day might unfold.
Explore our researchProbabilistic forecasting & weather-market researchIndependent · Self-funded
Our focus
The forecast is a
range of possibilities.
A temperature reading tells us what has happened. A forecast asks what could happen next.
Heliomark Research develops probabilistic forecasts of intraday temperature and daily maximum temperature. We combine weather observations, forecast inputs and machine learning to study outcomes, uncertainty and their implications for weather-market decisions.
Research & developmentResearch directions
Three connected questions.
From the physical course of a day to a forecast that can be tested against the world.
What did we know, and when?Weather data & evidence
Preserve source identity, observation times and release times. Build model inputs that reflect the information available when a forecast would have been made.
What does that imply?Weather-market research
Study probabilities for weather outcomes and the rules used to resolve them. Develop internal tools for evaluating forecasts and trading decisions in our own business operation.
Research discipline
Uncertainty belongs
in the forecast.
Our aim is to estimate plausible outcomes, then test whether the probabilities hold up on later, unseen days.
- 01Define the exact temperature target.
- 02Use information available at forecast time.
- 03Evaluate against observed outcomes.
- 04Keep failed experiments and replay evidence.
One observation. Several possible paths.Conceptual
Clear targets. Honest tests. Reproducible evidence.Forecast skill remains under evaluation.
The questions that guide us
A changing day.
A changing probability.
Temperature is shaped by an evolving atmosphere. Our research asks how the evidence arriving through the day should change what we expect.
- 01How much warming remains after the latest observation?
- 02When should new weather guidance change the forecast?
- 03Do the estimated probabilities work on days the model has never seen?
Work in progress
From research inputs
to usable tools.
Our first version is being developed for internal use in our own weather-market business operation. The work is still in research and development.
Public development profileWeather inputs that retain their context
Prepare accepted observations and forecast data with their source identities and timing intact.
Model integration and evaluation
Connect prepared inputs to fitting and scoring workflows. Check predictions against observed outcomes and preserve experiment evidence.
Compute and storage on AWS
Use Amazon EC2 for data preparation and model experiments, and Amazon S3 for accepted research inputs and experiment artifacts.
View the development activity record
3,140
Merged pull requests
12,043
Default-branch commits
Repository totals checked on 8 October 2026 and reported by Heliomark. They include automated and assisted development and do not measure hours worked, forecast accuracy or commercial results. Implementation code, model artifacts and research data are maintained privately. GitHub’s public private-contribution counts span account activity and use a different counting method.
About Heliomark Research
An independent
research venture.
Heliomark is a self-funded venture developing forecasting software and tools for its own weather-market business operation. Heliomark Research is the public identity for that work.
Our current focus is probabilistic temperature forecasting, careful weather-data preparation and the evaluation of internal research tools. We are building toward an operational first version; we have no external customer deployments to report.
This website introduces the research. Forecast accuracy and commercial performance have not been established.