Forecasting with Astronomical Calculations

What This Step Produces

By now you have three charts. Temperature, built from solar cardinal signs adn outer planet positions. Air movement, from Mercury's zodiacal sign. Moisture, from lunar phases and their historical correlations. Each chart tells part of the story. This step is where you combine them into a single forecast statement for the target month.

A forecast statement covers four things:

• Temperature tendency: above, below, or near the long run monthly average
• Moisture tendency: wetter, drier, or near average
• Dominant air mass type: the most likely wind regime for the period
• Confidence level: how much the analogue set agrees with itself

That's it. Four things. Not a daily breakdown, not an hourly temperature curve. A monthly tendency statement with an honest confidence rating attached.

Calculating Planetary Positions

You need the ecliptic longitude for each outer planet at the start of the target month. NASA's JPL Horizons system gives you this for free. Jean Meeus's Astronomical Algorithms is the reference if you want to compute it yourself. Either way, record longitude to the nearest degree. The pattern matching tolerance in this method does not warrant anything finer than that.

For Mercury: find the zodiacal sign it occupies for the majority of the target month. Mercury moves fast. If it changes signs partway through the month, note both and weight the one with the longer occupancy.

For the Sun: find which cardinal period the target month falls into.

• Aries: March equinox through June solstice
• Cancer: June solstice through September equinox
• Libra: September equinox through December solstice
• Capricorn: December solstice through March equinox

Most months sit cleanly inside one period. If the month straddles a transition, note it. It's one of the genuinely uncertain configurations, adn the forecast should reflect that.

Worked example: October 2024

The table below shows what the position recording looks like for a real month. October 2024 fell in the Libra quarter. Jupiter was actually in Gemini at 21°, Saturn in Pisces at 14°, Uranus in Taurus at 26°. Mercury occupied Scorpio for the majority of the month after entering on 3 October.

From this configuration you go to your historical library and search for years where the outer planets were in similar positions. Similar means within roughly 30° for Jupiter and Saturn. The analogue search is covered in Step 6. What we're doing in this step is translating those analogues into forecast values.

Translating the Pattern Match into Forecast Values

From Step 6 you have a set of analogue years. Each one has an outcome on record. Now you average those outcomes.

Take the temperature anomalies from your analogues. If four of your five analogue Octobers ran above average for MaxTemp, with anomalies of +0.8, +1.4, +0.6, +1.1, and one ran at +0.2, your average anomaly is about +0.8°C. That becomes your temperature forecast: MaxTemp likely above the long run October mean by around 0.8°C.

Do the same for MinTemp and for rainfall. Express rainfall as a percentage departure from the BoM climate normal for that month at the Dayboro station. If your analogues show rainfall at 95%, 130%, 115%, 88%, and 120% of the long run October mean, your forecast range is roughly 90%–130% of mean. The central tendency is slightly above average.

These departures get anchored to BoM's published climate normals for Dayboro. You're not making up a number. You're saying "given this configuration, the historical analogues ran X% above or below the 30-year average." That's a testable claim.

Expressing Uncertainty Honestly

Confidence in the forecast depends on three things: how many good analogues you found, how consistent they were, and how unusual the current configuration is. An unusual configuration with sparse analogues is a genuinely low confidence situation. Say so.

I use three levels:

• High: 5 or more close analogues, low variance in the outcomes, no unusual aspects in the configuration
• Moderate: 3 or 4 analogues, some variance, or one unusual feature in the current configuration
• Low: fewer than 3 analogues, high variance across outcomes, or a configuration that's genuinely rare in the record

In the forecast text I state it plainly. "Based on 4 analogue years, this month tends toward above average rainfall. Confidence is moderate. The analogues don't all agree on timing." That's more useful to a reader than false precision.

We do not give numeric confidence percentages. The data does not support that level of precision. Expressing it as 67% confident would imply a statistical rigour we haven't earned.

What the Forecast Document Looks Like

The actual output is a short document. Two to three paragraphs for the target month. No more than that. If the forecast needs five paragraphs to explain itself, something has gone wrong in the underlying process.

It covers:

• Temperature tendency: warmer, cooler, or near average compared to the historical monthly mean, with the anomaly estimate in degrees
• Moisture tendency: wetter, drier, or near average, with an indicative range in millimetres (for example, 80–140mm against a long run October mean of 110mm at Dayboro)
• Dominant air mass type: the most likely wind regime, adn which parts of the month are more likely to see the dominant pattern break down
• Key dates: any dates where a planetary aspect and a lunar phase coincide, where the historical record shows a higher frequency of significant weather events
• Confidence statement: one sentence, honest about the analogue count and variance

What the Process Looks Like Month to Month

Every month, about 4 weeks before the target period, I run through the calculation steps. Pull the planetary positions. Search the analogue library. Average the outcomes. Write the forecast. It takes 3 to 4 hours for a month I've done before. Unusual configurations take longer because I'm reading further back in the historical record and verifying the analogue set more carefully.

The forecast gets written, distributed to members, and locked. No revisions once it's out. That's the deal. If I could keep revising it, I could eventually revise it into whatever happened, and the whole verification exercise becomes worthless.

After the month ends, the forecast is compared against the actual data from the Lyndhurst Hill station. That comparison goes on the accuracy page and stays there. Good months and bad months both.

Simple process. Transparent output. Either the forecast was right or it wasn't.