At What Scale Does a Storm Start Making Sense?

A light switch is partly responsible for why I am now reading about meteorological satellites.

During my training in heavy-equipment mechanics, electricity and hydraulics changed what I saw in an ordinary action. Pressing a switch became an invitation to follow the circuit. Something happened somewhere else because a whole arrangement of relationships allowed it to happen.

I have been troublesome around simple explanations ever since.

That curiosity followed me into software, then mathematics, two books, educational tools, and a growing collection of questions about how humans and AI can investigate something together. Over the last few days, it has found a particularly unreasonable object to become interested in: the atmosphere.

Reasonable opening task: understand one observation. Long-term ambition: contribute useful mathematics to meteorological satellites. My brain apparently sees these as neighbouring items on a checklist.

I have just published Math Dev Journal 01: Reading the Living Atmosphere / Leer la atmósfera viva. The full English and Spanish editions are here:
DOI: 10.5281/zenodo.22848503.

There is also a short companion podcast.

This post opens the notebook behind it. Bring your curiosity. I have already supplied the excessive number of questions. The question I keep returning to is this:

at what scale does a storm start making sense?

I can look at a cyclone and recognize its organization. Then my attention starts moving inward. A spiral. A boundary. A smaller disturbance. Something developing inside something larger. I want to know which relationships survive as I change the scale, and which ones I lose by looking too closely.

My first instinct is to zoom in. The mathematics asks me to be more specific.

There is the scale of the picture on my screen. There is the spatial support of the data behind it. And there is the scale over which I calculate something. A beautifully detailed coastline can sit underneath a much coarser atmospheric field. The coastline belongs to the map; the weather information retains its own resolution. The mouse wheel has considerable authority over the picture and remarkably little authority over the atmosphere.

That distinction is already shaping the observatory I want to build. When I change the view, I want to see what changed in the calculation, what stayed fixed in the measurements, and what the display merely interpolated. It would make the interface an instrument for asking better questions.

Here is a small mathematical example from the paper that captures why I enjoy this work. Imagine two equal-area cells and two equally likely rainfall scenarios. These are invented numbers for an exact example:

Scenario	Cell A	Cell B	Regional average
1	0 mm	20 mm	10 mm
2	20 mm	0 mm	10 mm

For either cell, the 95th percentile of this two-outcome probability distribution is 20 mm. Average those two local percentiles and you get 20 mm. But the regional average is 10 mm in every scenario. Its 95th percentile is therefore 10 mm.

Same starting numbers. Different order of operations. A different answer to a different question.

The relationship between the cells carries information. If I summarize each location separately and then combine the summaries, I can lose that relationship. For a regional statistic, I need to calculate the regional value within each coherent scenario first, then summarize those results. This is the kind of mathematics that makes me sit up. Two rows are enough to expose a mistake that could hide comfortably inside a very impressive dashboard.

My books, Fractal NeutroGeometry and Mapping the Invisible Infinite, brought me toward questions about structure, indeterminacy, and representation. Now I want to give those questions something concrete to push against. If I propose a descriptor of a cloud boundary, I need to specify the boundary, the preprocessing, the range of scales, and what useful information the descriptor might add.

I also want to examine support, unresolved information, and contrary evidence separately. That is where neutrosophic representations interest me. Plithogenic methods raise further questions about relationships among attributes. Their practical value here will depend on comparisons with simpler ways of organizing the same evidence.

I can be attached to a question and still let an experiment change my mind about the method. After investing years in an idea, that takes a little practice. I am getting some.

My recent agent-development training has been unexpectedly useful in getting here. I spent time separating roles, tools, sessions, persistent memory, and the context an agent receives for one task. I kept applying the lessons to educational companions and to the practical problem of returning to work after a handoff.

I know that problem rather personally. Switching tools or models can leave me reconstructing decisions I already spent energy making. Somewhere in the conversation, an important qualification disappears. I remember that we resolved it. Finding exactly what we resolved becomes the next job. For an atmospheric investigation, that missing qualification could be the forecast cycle, a unit, the source snapshot, or which time was actually available when a prediction was made.

That gives my Context Continuity Protocol, CCP, a precise job to investigate: carry enough of the scientific task across a handoff that the next human or agent can continue faithfully. Keep the question, the relevant data references, the transformations already applied, the uncertainties, and the next action together.

In the proposed observatory, one agent might retrieve records, another invoke a numerical tool, and another review inconsistencies. Each would receive tools suited to that responsibility. Selected application actions could be exposed through WebMCP, alongside the human controls. The calculation should retain the same meaning whichever route invokes it.

I want that same continuity for a learner using an educational side panel. Coming back tomorrow should mean recovering the question and the understanding we were building. Remembering a transcript is only part of that experience.

The forecasting foundation I am studying is WeatherNext. My proposed contribution sits around existing forecasting work: organizing a particular investigation, checking transformations, testing a mathematical addition, and making the result understandable.

The next experiment I want to prepare is deliberately small: one historical cyclone, one defined target, an identified forecast and observation record, an unchanged reference, and one candidate addition. If the extra mathematics helps, I want to be able to explain where. If it fails, I want a result precise enough to teach me something.

That work still lies ahead. What I have completed is the first bilingual entry in this journal: nine chapters, eleven equations, and a registry of sixty-two external sources. It gives the investigation a starting point that another person can inspect and challenge.

And now the satellites are asking for homework.

For example, the GOES-R series' Advanced Baseline Imager observes in sixteen spectral bands spanning visible, near-infrared, and infrared wavelengths. Choosing a band already changes the question I can ask of the image. I want to understand what the instrument measures, how a useful product is derived from it, and how time, resolution, and uncertainty enter the interpretation.

My first exercise could be as simple as following one event through one documented channel, then explaining each step from observation to displayed result. I would like to earn the next layer of complexity by understanding the previous one.

Eventually, I want to explore whether a small, bounded analytical function could be useful aboard an instrument: a quality check, a feature-selection step, or a compact report prepared for transmission. That brings execution time, memory, energy, and recovery into the mathematics. Flight integration would require its own engineering and qualification with the relevant specialists.

I find that prospect exciting because it gives a mathematical idea a demanding destination. Somewhere between an equation and an instrument, every convenient assumption has to meet the world.

Héctor Fernando Aguilar also has room to change the direction. The planned second journal entry is for his perspective, in his own voice. The third will take the shared question toward implementation. These entries are part of the longer path toward Book III, with space for results that surprise us and revisions that improve the question.

The NASA Space Apps Challenge, scheduled for November 14–15, 2026, is the first public milestone I am aiming toward. Hackathons have a useful effect on me: they make an expanding idea meet another person's time and attention. Eventually, someone has to be able to open the thing and understand what it does.

For this direction, I would like that encounter to be simple. Choose a region, inspect the available evidence, follow one calculation, and see what the result supports. A person arriving fresh should be able to ask a difficult question and find the relevant source. That would give the mathematics, the interface, and the agent workflow a shared purpose.

For the entry itself, I will select an official challenge and follow its build window and reuse rules. This journal is the research preparation. The exact submission still has to earn its scope. What attracts me is the chance to meet people who know different parts of the problem and build something useful together. A good hackathon can turn a private obsession into a conversation with working parts.

That is what I want this journal to feel like: the pleasure of learning something difficult with enough care that somebody else can pick it up, question it, and take it further. I come to this as a builder who loves mathematics and still enjoys the moment when an ordinary thing opens into a whole system.

A switch did it once. A satellite image might do it again.

I would love to hear from people who have already made a few mistakes in this territory:

1. If you were teaching a developer to read satellite data, which instrument, channel, and historical event would you start with—and what misunderstanding would you tackle first?

2. Have you ever changed the scale or order of a calculation and watched a convincing result fall apart? I would enjoy hearing the actual example, especially the small one that taught you a large lesson.

3. When research passes between humans, tools, or AI agents, what is the one piece of context you insist must survive? I am particularly interested in something you learned to preserve after losing it once.