Integrated Resource Planning (IRP) has always been an exercise in managing uncertainty. Load growth projections carry multi-year error bands. Fuel price forecasts are notoriously unreliable past five years. Retirement timing for aging thermal assets is driven by regulatory actions that no model can fully anticipate. Planners have developed frameworks — scenario analysis, probabilistic load forecasting, reserve margin standards tied to LOLE/LOLP targets — to manage that uncertainty systematically.
But there is a structural gap in most IRP processes when it comes to variable renewable energy (VRE) capacity contribution. The solar or wind capacity factor assumptions embedded in production cost models are typically derived from historical TMY (Typical Meteorological Year) data or from a few years of site actuals. That single-value capacity factor carries a precision that the underlying data does not support, and it embeds a silent assumption about inter-annual climate variability that can cause a plan to fail its adequacy target in stress years even when it passes in the median.
What a capacity factor assumption actually implies
When an IRP production cost model assigns a 26% annual capacity factor to a proposed 200 MW solar installation in the Southern Rocky Mountain region, it is implicitly assuming that the 52 MW of average annual output from that asset is the right number to use for multi-year adequacy planning. But annual capacity factors for utility-scale solar are not fixed physical constants — they are outcomes of the atmospheric forcing in a given year, which varies with large-scale climate patterns including ENSO phase, Pacific Decadal Oscillation state, and persistent ridging or troughing patterns over the western US.
Published analysis of NSRDB-derived irradiance at representative Mountain West sites shows inter-annual variability in GHI (global horizontal irradiance) of approximately 4–7% around the long-term mean, expressed as a coefficient of variation. For a site with a long-term mean capacity factor of 26%, a 5% GHI inter-annual CV implies a realistic annual capacity factor range of approximately 23–29% across a 20-year planning period. The distribution is not symmetric — low-irradiance years tend to be driven by persistent cloud patterns associated with La Niña conditions, and those patterns affect multiple sites in the same geographic region simultaneously.
An IRP that uses 26% as a deterministic input is not wrong in the median — it is wrong in its representation of risk. The plan will look adequate against its LOLE target in typical years and will have surprise shortfalls in the correlated low-irradiance years that stress the system exactly when load is also elevated or when other resources are unavailable.
How probabilistic solar forecasts enter the IRP workflow
The connection between operational probabilistic forecasting (P10/P50/P90 at 15-minute resolution) and long-range IRP planning is not immediately obvious, because the tools and timescales are so different. Day-ahead probabilistic forecasts are used for reserve commitment. IRP modeling runs multi-decade capacity expansion scenarios. These processes seem disconnected.
The bridge is in how the capacity contribution of solar assets is characterized in the IRP model. In PROMOD, PLEXOS, or Aurora (the dominant production cost modeling platforms), solar capacity is typically represented as a fleet of hourly dispatch profiles that determine the effective load carrying capability (ELCC) of the resource. The ELCC is the contribution that solar makes toward meeting peak load in reliability-constrained periods.
Probabilistic forecast output, particularly the P10 production distribution, is directly relevant to ELCC estimation: a solar resource's ELCC is determined by its contribution in the hours when the system is most stressed, which are high-load low-generation tail events. Using a median-year TMY profile to represent solar in ELCC calculations understates the risk in low-irradiance years and overstates the system's ability to meet its adequacy standard without additional dispatchable capacity.
Stress-testing the plan: a practical approach
A resource adequacy plan that passes its LOLE target under median-year solar assumptions should be tested against an alternative scenario representing a P90 (low-production) solar year. The P90 annual energy scenario can be constructed from the historical irradiance distribution at each proposed solar site — it represents a year where annual GHI falls at approximately the 10th percentile of the multi-decade historical record. For Mountain West sites, this might correspond to a La Niña year with persistent winter cloud cover and reduced spring insolation.
Consider a regional utility in the Colorado/Wyoming footprint planning to add 800 MW of solar capacity as part of a coal retirement replacement program. Under median-year assumptions, the new solar contributes approximately 208 MW of ELCC (26% capacity factor) during the summer peak hour. Under P90 assumptions, the same 800 MW nameplate delivers approximately 172 MW of ELCC (21.5% capacity factor) — a 36 MW difference that may or may not push the system below its reserve margin threshold depending on the rest of the portfolio.
If the reserve margin is 15% and the system is 1,800 MW of peak load with 2,070 MW of available capacity under median-year assumptions, a 36 MW ELCC reduction from solar stress takes available capacity to 2,034 MW — a 13% reserve margin, below the 15% planning threshold. The plan passes in the median and fails in the P90 solar year. That is a real adequacy risk that a median-only IRP missed entirely.
We're not saying every IRP needs to be planned against P90 solar conditions — that would result in systematic over-procurement of dispatchable capacity. We're saying that the IRP should explicitly quantify the LOLE impact of a low-solar year, make that risk visible, and decide whether additional adequacy hedges (battery storage, demand response contracts, extended gas peaker availability) are warranted based on that quantified risk rather than treating median-year assumptions as conservative planning.
Correlated weather risk across the portfolio
A single-site capacity factor stress test understates the risk in a regional portfolio, because low-irradiance years are driven by regional-scale climate patterns that affect all sites in a geographic footprint simultaneously. A utility with 800 MW of solar spread across ten sites in Colorado and New Mexico is not holding ten statistically independent capacity factor risks — the sites are highly correlated in their annual energy output because they are all under the influence of the same synoptic-scale forcing.
The correlation structure between sites in a regional portfolio should be explicitly characterized in the IRP stress test. Two sites 40 km apart in similar terrain may have a 0.85–0.90 correlation in annual GHI. Two sites 400 km apart across a mountain range may have a 0.60–0.70 correlation — lower, but still substantially correlated relative to what a diversification-assumes-independence model would imply. Portfolio-level ELCC in a stressed year should be computed using correlated production profiles, not as a simple sum of uncorrelated individual ELCCs.
What this means for capacity procurement decisions
The practical implication for IRP process is that solar capacity needs to be represented in the model with a distribution over annual performance outcomes, not a single central estimate. That distribution should be derived from site-specific irradiance history, ideally 20+ years, supplemented by climatological analysis of the relationship between large-scale climate indices and regional irradiance anomalies.
Battery storage paired with solar changes the analysis further: the battery's contribution to ELCC depends partly on how often it can be fully charged during peak irradiance hours, which is itself a function of the solar capacity factor distribution. A storage-plus-solar system's combined ELCC in a P90 solar year is more complex to characterize than either asset separately, and it requires production profiles for both assets under correlated atmospheric forcing.
The quality of the long-range capacity planning output is bounded by the quality of the renewable performance characterization going into the model. Getting that characterization right — using probabilistic distributions rather than single-point capacity factor assumptions, characterizing regional correlation structure, and stress-testing against tail climate scenarios — is where the investment in better data and better methodology has its highest return in IRP work.