How to Read Weather Model Run-to-Run Changes

A model run is a new forecast generated from updated atmospheric observations and model calculations.

Forecast initialization times are commonly shown in UTC using labels such as:

• 00Z
• 06Z
• 12Z
• 18Z

Each run produces a new set of temperature forecasts.

After those temperatures are processed, they produce updated HDD, CDD and TDD values for the forecast period.

Different models update on their own schedules, so comparisons should always use clearly identified run times.

A run-to-run change compares the latest forecast with an earlier run of the same model.

For example:

The latest GEFS run has added 8 HDD across the selected forecast period.

For HDD:

• A positive change represents a colder forecast
• A negative change represents a warmer forecast

For CDD:

• A positive change represents a hotter forecast
• A negative change represents a cooler forecast

TDD changes combine the heating and cooling adjustments.

A run-to-run comparison is meaningful only when the underlying scope remains consistent.

Compare:

• The same model
• The same forecast period
• The same HDD, CDD or TDD metric
• The same geographic weighting
• The same daily or cumulative view

Comparing different forecast windows can create a misleading result because one period may contain additional days.

A cumulative change shows the total HDD or CDD adjustment across the selected period.

A daily view shows where that change occurred.

For example, a model may add 10 HDD in two different ways:

The cumulative change is identical, but the practical interpretation is different.

Near-term changes are generally supported by more observational information. Changes at the distant end of the forecast are normally more uncertain.

The daily distribution also matters because several consecutive cold days can produce a different demand profile from one isolated extreme day.

Persistence means that a forecast signal remains visible across multiple runs.

For example:

• GEFS adds 5 HDD
• The next GEFS run holds the colder forecast
• ECMWF EPS also moves colder
• Ensemble spread begins to narrow

This sequence provides more evidence than a single isolated +5 HDD change.

Persistence does not guarantee that the forecast is correct, but it indicates that the signal is becoming more stable.

A change appearing in both GEFS and ECMWF EPS has broader support than a change appearing in only one forecast system.

The models do not need to show identical values.

What matters is whether they support the same broad direction, timing and demand implication.

For example:

The exact totals differ, but both models are confirming a colder demand outlook.

If one model adds substantial HDD while the other removes HDD, forecast confidence remains lower.

A run-to-run change and an anomaly versus normal answer different questions.

Run-to-run change asks:

Forecast versus normal asks:

A model can move colder while remaining below normal.

For example:

The forecast became colder, but it still represents milder-than-normal conditions.

Both comparisons are necessary for proper interpretation.

A mean forecast can change because the entire ensemble shifted, or because a smaller group of extreme members pulled the average.

The accompanying P10-P90 range helps distinguish between these situations.

A colder mean combined with a narrowing range suggests that more members are converging on the colder solution.

A colder mean combined with a widening range indicates that uncertainty is increasing even though the central forecast moved colder.

The run change and the spread should be evaluated together.

A colder winter forecast generally adds HDD, but the timing of the cold can materially affect its importance.

Questions to consider include:

• Does the change occur in the next few days or later in the outlook?
• Is it concentrated on weekdays or weekends?
• Does it affect one day or a sustained period?
• Is the forecast adding demand before or after an important storage reporting period?
• Is the adjustment occurring in a major demand region?

A small but well-timed change can be more relevant than a larger adjustment spread across less important dates.

Weather changes must be interpreted within the broader market context.

A colder forecast may have limited price impact if:

• The market already expected the change
• Production is increasing
• Storage inventories are comfortable
• LNG demand is weak
• The adjustment occurs far into the forecast
• The change is not supported by other models

Likewise, a warmer forecast is not automatically bearish if the market had already priced in even milder conditions.

Run-to-run weather changes are an important input, not a complete trading signal.