The Variable Years: Why Output Swings, and What P90 Really Means

One number, and no year that delivers it

Every solar proposal rests on an annual yield figure, and it is almost always quoted as a single number: so many kilowatt-hours per kilowatt, per year. The atmosphere does not work that way. The same site, the same panels, the same tilt will produce several percent more in one year and several percent less in the next, and the difference is not noise: it is the ocean and atmosphere cycling through modes that redistribute cloud and rainfall across entire hemispheres. The three that matter most for solar are the El Nino-Southern Oscillation (ENSO), the Indian Ocean Dipole (IOD) and the Southern Annular Mode (SAM). This report is about the swing they produce, and about the one number on a proposal that is designed to price it: P90.

The engine: El Nino and La Nina

ENSO is the largest year-to-year climate signal on Earth. Measured by the Oceanic Nino Index, it swings between warm El Nino and cool La Nina phases every few years, and the twenty-one-year record shows just how uneven those swings are:

Selected years from the 21-year climate-signals record (2005-2025); full annual table in the public dataset. ENSO peak is the strongest three-month ONI of the season; IOD peak is the September-November DMI; SAM is the annual mean. Across the record, ENSO peak correlates with global temperature (r = 0.42) and with the global precipitation total (r = 0.49): warm phase runs warmer and, on the global mean, wetter.

The pattern in the driver record is coherent. The strong El Nino years cluster at the warm, globally wet end (2015-16, 2023-24); the sustained La Nina of 2020-22 sits at the dry end, with the global precipitation total falling to 1007 mm in 2021 from 1059 mm in 2024, its wettest year. But the global mean hides the real story, because these modes do not add rain everywhere. They move it. An El Nino that floods the coast of Peru is the same El Nino that dries out eastern Australia and Indonesia, and a positive Indian Ocean Dipole, at its 2019 record, drives drought and the catastrophic fire season that followed into south-east Australia while soaking east Africa.

Why the global average looks calm and your site does not

Averaged over the whole planet, the annual precipitation total varies by only about 1.3% year to year, which makes the global climate look almost steady. Any individual site is far less steady, because the oscillations redistribute cloud rather than remove it: where one hemisphere clears, another clouds over. For solar, what matters is the variability of annual irradiance at your address, and the published solar-resource literature measures it as a coefficient of variation that depends strongly on climate type:

Representative interannual variability of annual GHI from the published solar-resource literature (SolarGIS, Vignola, Meteonorm), not computed from our per-site baseline, which is a multi-year climatology rather than a per-site time series. P90 as a share of P50 uses the normal-distribution approximation P90 = P50 x (1 - 1.28 x CoV).

What P90 is, and what it protects

A proper resource assessment does not quote one number; it quotes a distribution. P50 is the median expectation: half of years come in above it, half below. P90 is the level that nine years in ten will exceed, and it is the number a lender or a careful buyer plans against, because it answers the question a single figure cannot: how bad can a normal bad year be? The arithmetic follows directly from the variability above. In hyper-arid Atacama, where annual irradiance varies by about 2.5%, P90 sits near 97% of P50 and a bad year is barely felt. In monsoon Jodhpur, where it varies by about 8%, P90 falls to roughly 90% of P50, so a one-in-ten year delivers a tenth less energy than the headline figure, every projection built on the median silently overstates the hard years, and a battery or a loan sized to P50 is undersized for the year it matters.

The buyer actions are concrete. Ask whether the yield figure on your proposal is P50 or P90; if the salesperson does not know, it is P50, and in a variable-climate market that is optimistic by several percent in the years you can least afford it. Match the margin to the climate, not to a rule of thumb: a desert site can be planned close to its median, while a monsoon or maritime site needs a P90 buffer several times larger. Read a weak year against its driver before blaming the equipment, because a La Nina or a positive-dipole year can take a real bite out of a perfectly healthy system, and the fault log is the climate record, not the inverter. And treat these as swings around a stable mean, not a trend: unlike the brightening and dimming of part 02, ENSO and its siblings oscillate, so a run of strong years is not a new normal and a run of weak ones is not a failing array.

Method and limits

The driver record is twenty-one years (2005-2025) of published indices: NOAA's Oceanic Nino Index for ENSO, the Bureau of Meteorology's Dipole Mode Index for the IOD and its Southern Annular Mode series, joined to NASA GISTEMP global temperature and GPCP global precipitation by year. Three limits are worth stating plainly. The correlations here are global-mean relationships (ENSO to temperature and to total precipitation); the solar-relevant effect is regional and teleconnected, and a market-by-market ENSO-sensitivity map is a larger piece of work than this pillar contains. The P90 coefficients of variation are representative literature bands by climate type, not figures computed from our own per-site series, because the yield baseline is a climatology; they are the right order of magnitude for planning, and a site-specific assessment should use a site-specific number. And twenty-one years spans only a handful of full ENSO cycles, enough to characterise the oscillation but not to pin down its rarest extremes. Every index is public and named to its source.