A year of data: what the seasons taught me

The first year you watch your solar production data closely, every day is interesting because every day is new. The second year, you start noticing that what you thought was unusual is actually a pattern. By the third or fourth year, you have absorbed the rhythm of the seasons in a way that no textbook on solar irradiance could have taught you, and you have probably also figured out which patterns are real and which were just one weird year.

I have been logging continuously since the system went live, which means several full annual cycles of data, hour by hour, with consumption and weather alongside. Most of it is exactly what you would expect from the physics of latitude 51°N. Some of it surprised me. The interesting bits are the surprises.

This article is a tour through the year as the data actually shows it, with the small observations that only become visible after you have watched the same months repeat themselves a few times. It is less technical than the articles that came before it. The point is the rhythm, not the diagnostic.

The dark months

November to February in Belgium is when solar owners learn humility. Daily production in late December can drop below 1 kWh per kWp of installed capacity, which on a typical 6 kWp residential system means an entire day's production comes to maybe 3 to 4 kWh, less than what a single warm shower costs to heat. The widget on the lock screen shows numbers that look like rounding errors. The cumulative annual chart, which has been climbing energetically all summer, flatlines for weeks.

The first winter, this felt wrong. By the third winter, it felt obvious. Belgium is at the same latitude as Calgary, Newfoundland, and the southern tip of Kamchatka, and we get the daylight that goes with that latitude. The winter solstice gives us roughly eight hours between sunrise and sunset, and a meaningful fraction of those hours are overcast, low-angle, or both. There is simply not much sun to harvest. The system is not broken in December. The sun is broken in December, if you live up here.

What you can do with the dark months, if you are inclined, is use them to think about consumption instead of production. December is when your heating bill arrives, and the relationship between what your panels produce and what your household consumes is at its most lopsided. If you have a heat pump or an EV, December is also when you discover that the panels are paying for a slice of those costs even when their daily output looks invisible. The cumulative kilowatt-hours since installation still tick upward; they just do so very, very slowly for about ten weeks.

The shortest day passes around 21 December, and within a week or two you can already see daily totals starting to inch upward. The change is invisible day-to-day but unmistakable over a fortnight. By late January, on a clear cold day, the panels can have a good morning. The system is waking up earlier each week, by minutes at first, then by tens of minutes.

The March awakening

If you have to pick one moment in the year when solar production goes from "barely worth checking" to "actually exciting", it is somewhere in the first half of March. The exact week varies, some years the shift happens in late February, some years not until the third week of March, but it always happens, and once it does, the dashboard starts feeling alive again.

The mechanism is straightforward: the days are getting noticeably longer, the sun is climbing higher in the sky, and the cells are still cool. A clear March day in Belgium can produce 75 to 85% of what the same panels would produce in midsummer, on the same kind of clear weather, simply because cool cells produce more efficiently than hot ones. If the air is around 8 to 12°C and the sun is out, the daily peak power can come within striking distance of the annual maximum.

The cumulative chart does what mathematicians call a derivative-discontinuity here. The flat winter line takes a noticeable bend upward, and from then until October it climbs steeply. The first time you saw this on your own data, if you are like me, you spent a moment just looking at it.

This is also the moment in the year when the widget on the lock screen reappears as a thing worth glancing at. The shape of the daily curve grows from a shallow ripple to a proper bell, and you start being able to predict, around noon, roughly what the day's total is going to be by mid-afternoon. Solar production becomes a presence in the day again rather than a background trickle.

The April-May surprise

Here is something I genuinely did not expect when I first installed solar panels: the highest annual production day, on most Belgian roofs, falls in late April or May rather than in June or July. This goes against everything intuition tells you about midsummer being the peak season for everything sunny.

The reason is the temperature coefficient of crystalline silicon, which we covered in Estimates versus reality. Panels are most efficient at 25°C cell temperature, and they lose about 0.4% of their output for every degree above that. On a clear May day in Belgium, the air is often around 18°C, the cells run at maybe 35 to 40°C, and the efficiency penalty is small. On a clear July day at 32°C ambient, the cells can hit 60 to 65°C, and the efficiency penalty is twice as large. The May day produces more, despite the sun being a few degrees lower in the sky and the day being slightly shorter.

I have a folder somewhere with the dates of my own best production day each year. They cluster heavily in April and May. The single best day across my multi-year history is from late April, on a year when a high-pressure system parked itself over Belgium for an entire week of cloudless, cool spring weather. Production hit roughly 110% of the installer's predicted "ideal day" output for two days running. Both of those days, the cells were running cool enough that the inverter clipping that usually rounds the top of summer curves did not happen.

This window also tends to coincide with the cherry blossoms in Brussels parks and the lambs in the meadows, which is one of the smaller pleasures of solar monitoring: the data and the world outside agree, for a few weeks, that this is the best time of year.

The trade-off, of course, is that this window also overlaps with the heaviest pollen season in Belgium, particularly birch in April and grasses in May. On heavy pollen weeks, you can see a thin film build up on the panels that the next significant rain mostly washes off. The dip in production from pollen alone is small, usually 1 to 2% across the worst weeks, but it is visible if you compare two otherwise similar clear days a fortnight apart.

The summer plateau

June, July, and August deliver the volume rather than the peak efficiency. The days are long, the sun is high, the totals add up rapidly. But the daily peak power is often a little lower than spring, and the daily curve is wider and flatter. The total for any given clear summer day is high because the curve is long; the height of the curve is what April was offering.

What I find interesting in summer is the variability. Belgian summer weather is dominated by westerlies, which means cloud cover can come and go on timescales of hours. A typical July week alternates between brilliant clear mornings, hazy afternoons, and the occasional thunderstorm that drops production to 10% of expected for an afternoon. The cumulative chart climbs but in a jagged way, not the smooth steep line that April and May provide.

The other summer phenomenon worth knowing about is heatwaves. Belgium has had several since 2018 that broke local temperature records, and the production data on these days is counter-intuitive. The day you remember as "the hottest day of the year, brilliantly sunny" usually produced less than a cooler clear day from a month earlier. The thermal coefficient grinds the cells down, and on the worst days you can lose 15 to 18% of theoretical peak just to temperature. The widget shows numbers that feel underwhelming for the brightness outside, and friends ask why your system is not making more on such a perfect day. The answer is that solar cells, like most of us, prefer slightly cool weather to genuine heat.

The post-thunderstorm production bump is also a summer specific phenomenon. After a hot dry fortnight, the next significant rain washes pollen, dust, and atmospheric haze off the panels, and the following clear day can produce noticeably more than the previous one would have. Sometimes you can see a 3 to 4% jump from rain alone. I have ten or twelve notes in my data flagging these moments over the years, and they almost always cluster in late June and August.

September's quiet excellence

September is the unsung month of Belgian solar. The sun is still relatively high, the days are still relatively long, the cells are running cool again, and the weather is often dominated by stable high-pressure systems that give long stretches of clear cool days. A good September can produce within 15% of what a good June produces, despite the day being two hours shorter.

This is also the month where, if you have a tilt and orientation that handles low sun angles well, you can see some surprisingly strong daily peaks. The angle of incidence on a 35-degree tilt is actually closer to optimal in September than it is in midsummer, which compensates somewhat for the shorter day. I have notes on September days in my data where the daily peak power matched or beat days from August.

The Belgian "indian summer", those stretches of cool clear weather in the first half of September, has become my favourite solar season for this reason. The numbers come in strong, the weather is comfortable, and the panels are doing exactly what they should be doing without the heat penalty of midsummer.

October's cliff

If March is the awakening, October is the falling off. Daily totals that were comfortably above 20 kWh in late September can be down to 8 or 10 kWh by mid-October, and you can see the change happening week by week on the cumulative chart, which starts to flatten visibly. The transition is steep enough that the first October after installation feels jarring; by the third October you are expecting it and watching for the specific day where the curve really bends.

What is happening is several factors lining up at once. The day length is dropping by about three minutes per day, fast enough to add up to almost half an hour per week. The sun angle is dropping noticeably, putting more atmosphere in the path of the light. The cloud cover increases as the autumn weather settles in. And the temperature is dropping into the range where the air starts producing the kind of low overcast that defines a Belgian November.

By the end of October, the daily peak power has dropped from summer's 4-plus kW to barely 2 kW on a clear day. The system is still working perfectly. The data is showing exactly what it should. But the curve is now narrow, low, and short, and you are heading back into the dark months.

What several years of data actually teach you

The first surprise is how stable the annual total is despite the variability of any given month. On a system in Belgium, the year-on-year variance is typically 8 to 12%. Two years can have completely different individual months, a cloudy April here, a brilliant May there, and still come in within a few hundred kilowatt-hours of each other. The weather averages itself out over a year, even when it does not feel like it from day to day.

The second surprise is how non-random the variability is. Belgian weather does not produce a uniform distribution of clear and cloudy days. It produces clusters: stretches of bright weather followed by stretches of overcast weather, both lasting days or weeks. The cumulative chart over a single year is bumpy in ways that have meaning, those flat stretches were a cloudy week, this steep climb was the spring high-pressure stretch, and looking back at multi-year history, you can see the same pattern of clustering repeating, just with the clusters in different places each year.

The third surprise is how much you start to enjoy the dark months once you have been through a few. The first winter, dropping below 1 kWh per kWp per day felt like the system had broken. By the fourth winter, it felt like a known character trait of the system, and the steep climb in March became something to anticipate. The annual rhythm becomes its own pleasure, not in spite of the dark months but partly because of them. The contrast is what makes the summer feel like a gift rather than just a baseline.

The fourth, smaller surprise is the conversations. After several years, you become the person at the dinner table who knows that December produces 3% of the year, that the best production day is usually in late April, that thunderstorms can clean panels and produce visible jumps. These are not useful facts, individually. Collectively, they become a small grammar of the world that you would not have without the data, and that other people sometimes find quietly interesting.

The annual report habit

Once you have a few years of data, the most satisfying thing to do with it is to look at one year as a single document. The annual report PDF we mentioned in Reading your solar charts: what the numbers actually mean does this automatically on the system's anniversary, but you can do it manually too: take the calendar heatmap for the year, the year-on-year comparison, the specific yield, the peak power, the consumption overlap if you have it, and look at them together for a few minutes.

The result is not a report card for the system, which is what most owners initially treat it as. It is a record of a year. The cloudy fortnight in May that you had forgotten. The heatwave week when the peak dropped despite the brightness. The September stretch when production stayed high longer than you expected. The first frost morning where the cells were so cold the daily peak power tied the annual maximum.

After several years, those reports start being interesting to compare. The summers are roughly equivalent. The winters too. The shoulder seasons, the months where the weather can go either way, are where the year-to-year differences live, and they are also where the most interesting individual days hide.

This is what monitoring is for, after years rather than months. Not the diagnostic. Not the optimisation. Just the record. A small annual book of how the sun behaved over your specific roof for one cycle, kept against the time when you want to compare it to another year and notice that the spring was different this time, or that the November fog lifted earlier, or that the heatwave came in July rather than August.

A closing thought

The system on the roof is fundamentally a converter of one form of energy into another. Sunlight goes in, electricity comes out, the meter ticks. That is the basic value proposition, and it is genuinely most of what matters financially.

But there is a secondary pleasure that develops slowly, and only with multi-year data, which is the sense of being plugged into the weather and the seasons in a way that you were not before. The forecast becomes more interesting. The cherry blossoms become an indicator of a specific kind of solar week. The first frost morning has a meaning beyond "wear warmer clothes". The autumn cliff is not just darkening days but a measurable bend in a chart you have learned to read.

That extra layer is what the data adds, eventually, if you stay with it. Not on day one, or day thirty. Somewhere around year three. The panels were producing kilowatt-hours all along; what changes is your relationship to the cycle they are responding to.

This is, in the end, why I built HelioPeak the way I did. The numbers matter, but the numbers in their context matter more, and the context only emerges when the data has been around long enough to teach you what the seasons actually look like through your particular roof.