This week, Sunairio is officially launching our Trading Hub Price Forecast Ensemble directly in the Sunairio User Interface (first in ERCOT, other ISO regions to follow soon).

For a while now, customers have used our price forecasts via API to quantify market risk. By moving this capability into the UI, we’re making a truly probabilistic, fundamentals-driven view of ISO power prices significantly more accessible — no code required.

To mark the occasion, we want to lift the hood on our methodology and share how we cracked the code on modeling highly non-linear energy prices—even in the notoriously volatile gas basis markets of the Northeast US.

A Quick Refresher: The Grid Stress Index (GSI)

You can't accurately forecast power market prices without understanding the underlying physical reality of the grid. Traditional price models try to track this by building massive, unit-by-unit generation stacks. But in a modern grid dominated by variable renewables and unpredictable weather, that traditional approach breaks down.

Our solution is the Grid Stress Index (GSI)^TM. Instead of tracking thousands of individual power plants, we compress the entire grid’s supply-and-demand tension into a single, elegant metric in percentage space:

This is fundamentally a relationship between [the energy that needs to be supplied by dispatchable capacity] to [the amount of dispatchable capacity available to generate]. 

By mapping GSI to Delivered Spark Spreads (using a standard 6.5 HR to filter out commodity fuel noise), we uncover the true, empirical supply curve of a market, as seen in Figure 1.

Figure 1 plots GSI vs delivered North Hub delivered spark spreads in ERCOT for Dec 2025 - May 2026. What we observe is a learned market supply curve that reflects the enormously skewed behavior of most power markets: relatively controlled increases in price for much of the supply-demand regime, followed by extreme increase after an inflection point (about GSI = 75% in ERCOT).

Figure 1. The relationship between grid stress and power market prices

How the Sunairio Fundamental Price Model Works

Unlocking the relationship between grid stress and spark spreads was only step one.

Translating that relationship into highly accurate, hourly forward prices requires a clever mathematical approach. Here is how the Sunairio Fundamental Price Model operates under the hood.

1. Handling DA and RT Prices as a DA+RT Average

There are significant differences in the way that the DA market is solved compared to the RT market. For example, the RT market is fundamentally physical (physical supply must match physical demand), while the DA market is essentially a financial construct. Moreover, the DA energy market doesn’t even utilize a load forecast – it clears supply offers against demand bids. 

Power markets also utilize specific products known as virtuals (virtual supply and virtual demand) that work as a converging force to drive DA and RT prices together. Over the long term, the spread between DA and RT prices is exceptionally low in most power markets, though hourly and daily spreads can be quite wide.

Given the difficulty of fully reconstructing the DA market inputs (you would have to know demand bids and exact virtuals participation), and the existence of virtuals as a converging force, Sunairio doesn’t model the DA and RT markets separately. Instead we model an average of DA and RT prices, which represent the overall economic balance for that day. 

2. Temporal Bucketing

Power markets behave differently depending on the clock and the calendar. The grid dynamics of a Tuesday afternoon are vastly different from a Sunday morning – driven by both the hourly/daily cycle of demand and the physical constraints (like ramp rates and minimum run times) of many thermal generators.

For example, units with high startup costs or long minimum run times may self-schedule themselves overnight (essentially offering the plant as a market-price taker with $0 marginal cost rather than an economic resource at actual marginal costs) in order to more effectively compete for the prime daytime peak prices the next day. This means the overnight (7x8) stack is different from the weekday peak (5x16) stack.

To capture these operational nuances, we segment our historical grid data into distinct time buckets:

• 5x16 (Weekday peak)
• 2x16h (Weekend peak and holidays)
• 7x8 (Off-peak/night)

3. Outage Regime Bucketing

The Grid Stress Index is dependent on knowing the amount of dispatchable generation offline.  While we can never know with 100% certainty exactly what units will be unavailable in the future, Sunairio has built highly accurate outage forecasts that can predict the aggregate amount of MW offline.

The challenge is that not all generation outages are made the same. Removing 100 MW of baseload power from the generation stack could affect every hour’s price (by forcing the grid operator to dispatch more expensive units to meet baseload demand), while removing 100 MW of a peaker might not affect anything if demand doesn’t rise high enough.

We can clearly observe this dynamic in the plot below, which colors the GSI vs spark spread plot for ERCOT according to the percentage of dispatchable units offline. As Figure 2 shows, the inflection point in the empirical market supply curve is shifted left (occurs at lower levels of apparent GSI) when total nonrenewable outages are greater than 25% of installed nonrenewable capacity.

Figure 2. The effect of nonrenewable (thermal) generation outages on the relationship between GSI and delivered spark spreads

4. Emulating the Stack via Isotonic Regression

Once the data is bucketed, we fit the GSI vs. Spark Spread relationship using Isotonic Regression.

Unlike standard linear regressions that force a smooth line through data points, isotonic regression specifies only one constraint: the function must be non-decreasing. This mathematical choice yields an extraordinary side effect—it naturally produces step-wise models.

Figure 3. Using isotonic regression to construct an empirical power market supply curve

Because physical grid operators dispatch generators in blocks of capacity from lowest to highest marginal cost, the true supply curve is literally a staircase. Isotonic regression allows our model to learn the physical supply stack empirically from market behavior, without us having to model individual power plant heat rates or fuel contracts.

5. The PJM Paradox: Solving the "Messy Blob"

In a market like ERCOT, mapping GSI to spark spreads with time buckets generally gets the job done beautifully. But when we applied this exact framework to Northeast markets like PJM, we ran into a wall.

When you plot Western Hub LMPs against local northeast gas prices using a standard 6.5 heat rate, the data doesn't look like a nice hockey stick. It looks like a messy, unreadable blob. There appeared to be no relationship between grid stress and spark spreads.

The breakthrough came when we realized that in the Northeast, extreme fuel price variations don't just shift the cost of power—they fundamentally alter market bidding behavior. To fix this, we introduced Fuel Price Regimes.

We segmented the PJM data by delivered gas price bands. When you filter the data through these regimes, the magic happens: the blob vanishes, and distinct, beautifully defined curves emerge.

Fuel Price Regime	Observed Market Behavior	Impact on Curve
Low Gas	Abundant cheap fuel; flat bidding behavior.	Flatter, lower inflection point.
Moderate Gas	Standard economic dispatch behavior.	Moderate, predictable "hockey stick."
High Gas	High opportunity costs; aggressive scarcity bidding.	Steeper, highly aggressive price acceleration.

When gas prices are high, generator bidding behavior becomes significantly more aggressive for the exact same level of physical grid stress. By building separate isotonic regressions for both different time buckets, outage regimes, and fuel price regimes, our model successfully masters the complexity of PJM and the Northeast.

Figure 4. Modified empirical market supply curves according to local delivered natural gas prices

6. Running the 1,000-Path Hourly Ensemble

With our mathematical curves established, we feed the model our probabilistic inputs: Sunairio ONE’s 1,000-path hourly GSI ensemble.

Because our weather-driven ensemble correctly forecasts thousands of fully correlated scenarios—perfectly capturing the joint probabilities of a heatwave, low wind speeds, and forced thermal outages—passing those GSI paths through our price regressions yields a true, mathematically rigorous probability distribution of hourly energy prices.

Conclusions

1. Empirical Supply Stacks: By utilizing isotonic regression, Sunairio's price model naturally replicates the step-wise nature of a physical dispatch stack without the data burden of traditional unit modeling.
   
   
2. Fuel Price Conditioning: In volatile markets like PJM, looking at grid stress alone isn't enough. Sunairio isolates distinct market behaviors by segmenting forecasts into specific fuel price regimes.
   
   
3. True Probabilistic Pricing: Power price risk lives in the tails. By pairing our fundamental price model with our 1,000-path weather ensemble, we provide users with a clear view of the frequency, duration, and severity of potential price spikes.

Now in the UI: Beginning now, users no longer need code to access these insights. The full power of our trading hub price forecast ensemble can be completely visualized and interactive right inside the Sunairio platform.

By reconciling high-resolution weather ensembles with asset-specific Digital Twins, Sunairio provides the missing data needed to quantify curtailment, congestion, and basis risk across the grid, for thousands of utility-scale wind and solar farms throughout the US.

Baltimore, MD. — March 26, 2026 — Sunairio, pioneer of next-generation grid forecasting software, today announced the launch of its hourly Asset-Level Generation Potential Forecasts. This new capability provides hourly predictions for every utility-scale wind and solar farm across major power market regions, beginning with ERCOT.

Asset-level Generation Potential forecasts provide the high-fidelity visibility required to navigate the ever-evolving grid, helping to quantify the supply-side volatility that drives congestion, curtailment, and basis risk in modern power markets.

Sunairio generation potential forecasts differ from traditional regional wind and solar generation forecasts in two important ways:

1. They provide individual energy forecasts for every utility-scale project — not just regional aggregations — allowing users to anticipate local variability down to the nodal level.
2. They forecast potential generation compared to traditional renewable forecasts that are designed to forecast actual generation after curtailment.  

By forecasting weather-driven potential generation, rather than post-curtailment energy, Sunairio Asset-Level Generation Potential Forecasts can be used as high-quality inputs for power flow and trading models. Traditional as-generated renewables data, on the other hand, can’t be easily used to model the grid because they represent a curtailed solution of the grid operator’s dispatch model, not an input to it.  

"Trying to anticipate transmission congestion with traditional renewables forecasts is essentially guesswork to find the missing MWs," said Rob Cirincione, CEO, Sunairio. "Our Asset-Level Generation Potential Forecasts isolate the weather-driven energy potential from grid-enforced physical limits, allowing users to quantify the volume of energy at risk of curtailment before it happens."

The forecasts are powered by two proprietary Sunairio technologies:

• Sunairio Digital Twins: High-fidelity, physics-based models of over 1,400 wind farms and 7,500 solar farms.
• Sunairio ONE: A high-resolution weather forecast ensemble that provides the hyperlocal inputs necessary to drive the Digital Twins.

The launch of Asset-Level Generation Potential Forecasts provides critical, nodal-specific insights for a variety of stakeholders navigating today's increasingly complex power markets:

• For nodal traders: By aligning generation potential with historical nodal price data, traders can predict the likelihood of nodal congestion.
• For hub traders: Zonal generation potential aggregations reveal macro signals that drive zonal basis spreads. This signal is completely obscured in actual (post-curtailed) generation data. 
• For grid planners: The data serves as a high-precision input for production cost modeling, helping grid planners identify the exact transmission infrastructure needed to realize specific volumes of weather-driven energy potential.

Sunairio ONE represents a shift from reactive grid observation to predictive, physics-based intelligence. For more information, please visit sunairio.com or email info@sunairio.com.

###

About Sunairio
Founded in 2020, Sunairio is the pioneer of award-winning, next-generation grid forecasting software that’s the first to provide integrated energy, weather, and climate insights. Sunairio helps energy traders, grid operators, utility-scale asset developers, and VPP and demand response aggregators make better commercial decisions in the face of increasing grid variability and extreme event risks. Sunairio and Sunairio ONE have received recognition from the NSF, ACP, and EPRI. For more information, please visit sunairio.com.

Media Contact
Logan Varsano, Inflection Point Agency for Sunairio
logan@inflectionpointagency.com

Generation Potential: The Pure Meteorological Signal

Forecasting the weather-driven energy production of wind and solar farms is critical for understanding the physical impact renewable generators have on the grid: variable power flow dynamics, regional supply-demand balances, and ultimately economic price risk. Yet the renewable generation data that’s typically been available to forecasters and grid modelers has a fundamental flaw: it’s reported after curtailment. That is, when historical renewable generation data is provided from public sources, those data sets report as-generated values that include losses due to curtailment (caused by transmission congestion that limits renewable output). 

This means that forecasts designed to predict actual renewable generation are missing the most important piece of the story: the potential generation that could have been produced if not for the congestion – i.e., the true meteorological signal of renewable generation risk. 

Launching Sunairio Asset-Level Generation Potential Forecasts

Today Sunairio is excited to fill this gap by launching hourly Asset-Level Generation Potential Forecasts for every utility-scale wind and solar farm across major ISO power markets, beginning with ERCOT. By combining high-resolution Sunairio ONE weather forecast ensembles with high-fidelity wind and solar farm Sunairio Digital Twins, Asset-level Generation Potential Forecasts provide the purest signal of variable generation-driven price volatility on the grid – from individual generator nodes all the way up to ISO-level aggregates and trading hubs. 

Figure 1 provides a sneak peak of what this feature looks like in our platform.

Figure 1. (Left) Map of solar and wind assets in the Sunairio Maps module; (Right) Example wind asset forecast shown in Sunairio Ensemble Explorer module.

Why it Matters: The Unseen MWs that Actually Drive Congestion

The as-generated energy vs. generation potential nuance creates a vexxing problem for anyone trying to understand power market fundamentals. First, it means that we can’t try to replicate past market outcomes by simply using as-generated wind and solar values as inputs to engineering models of the grid – because that actual generation was potentially a curtailed solution of the balancing authority’s dispatch model, not an input to it. Moreover, it means we can’t easily distinguish between A) an hour that had low renewable generation because the wind/solar resource was low and B) an hour that had low renewable generation because the wind/solar resource was so high that it was curtailed due to congestion. 

Trying to triangulate the curtailment signal from LMP or raw weather is messy and difficult given the often non-linear relationship between weather and renewable generation. The fundamental signal we want is straightforward: weather-driven hourly renewable generation before curtailment. 

How We Do It: Sunairio ONE + Sunairio Digital Twins

Creating Generation Potential starts with building a Sunairio Digital Twin – a high-fidelity physics-based model – of each asset. Sunairio Digital Twins incorporate plant-level technical characteristics (turbine coordinates, hub heights, power curves, PV tracking, PV panel efficiency, etc.) in addition to hyperlocal terrain modeling. For wind, this includes running a 100m-resolution CFD model over the farm footprint to translate mesoscale (3km) wind speed averages into 100m-resolution turbine-scale wind speed fields, and then factoring in turbine-by-turbine waking. For solar, we leverage the best-in-class PVLIB solar modeling suite to replicate asset-specific generation characteristics. These Sunairio Digital Twins already support our existing Historical Generation Potential dataset – covering more than 1,400 utility-scale wind farms and 7,500 utility-scale solar farms – and leverage our market-leading Sunairio High-Resolution Earth Data (SHED) as the source for historical weather.

With the development of Sunairio ONE, we’re now able to provide Generation Potential forecasts, in addition to historical Generation Potential, for all of these renewable sites (Figure 2). By isolating the weather-driven potential from the grid-enforced transmission limits, we enable our users to quantify the volume of energy at risk of curtailment before it happens – and identify high-conviction nodal arbitrage opportunities that post-curtailment forecasts overlook.

Figure 2. Process of creating Generation Potential given weather inputs and Sunairio Digital Twin model of a wind or solar asset.

Performance: Accuracy Where It Matters

We validated our Asset-Level Generation Potential Forecasts against data from IPP partners to confirm how they perform in real-world conditions. For this analysis, we limited our scope to sites that could provide a complete generation picture: hourly availability, hourly curtailment, and hourly net generation. By accounting for availability and adding back curtailment MW we’re able to validate our generation potential forecasts on an apples-to-apples basis.

On a portfolio of three solar sites and six wind sites, Sunairio’s forecast of Generation Potential using same-day and day-ahead weather forecasts achieved scores similar to that of using historical actual weather (Figure 3). 

Figure 3. Forecast validation results for a solar portfolio and wind portfolio over three months. 

For Nodal Traders: The "Electric Peninsula" Case Study

An analysis of the coastal wind farm cluster near the Ajo substation in ERCOT South demonstrates the impact of renewable generation on transmission congestion and highlights the value of Sunairio Generation Potential forecasts for nodal trading. 

As shown in Figure 5, five wind farms totaling more than 1,000 MW compete for transmission capacity on the lone path off the coast: the Nelson Sharpe - Rio Hondo line. The Nelson Sharpe - Rio Hondo Generic Transmission Constraint (or NELRIO GTC) in Kennedy County, TX is a notorious bottleneck that ranked as the third most frequently binding constraint in 2025 (Potomac Economics) (Figure 5). 

Figure 5. Map of five wind farms and the transmission lines associated with the Nelson Sharp- Rio Hondo Generic Transmission Constraint in ERCOT.

Using Sunairio’s Historical Generation Potential and aligning with three years of historical node and hub LMP data, we can explore the relationship between Generation Potential for this cluster of wind farms and the average node to South hub basis (Figure 6, left). (Note that from 2022 to 2025, the four nodes associated with this cluster of wind farms experience the same real-time LMP in 99.99% of hours.) As the left panel shows, as Generation Potential for that wind farm cluster increases, there is a non-linear shift in the occurrence of more negative local basis. 

Traders can visualize this conditional probability of congestion as a surface with Generation Potential (Figure 6, right). Given a Generation Potential of ~900 MW, there is a 90% chance of the LMP basis being less than or equal to -$5, and a 85% chance of it being less than or equal to -$10. For virtuals traders trading into the day ahead market to take on congestion exposure, Sunairio asset-level Generation Potential forecasts are a powerful tool for building optimal bid/offer ladders and accurately quantifying the risk of the resulting trading strategy.

Figure 6. (Left) Sunairio Generation Potential vs. average node basis to South Hub of the five wind farms (2022 to 2025). (Right) Conditional probability of real-time congestion of the nodes associated with the five wind farms.

For Hub Traders: Generation Potential Zonal Aggregations Reveal Macro Signals  

Asset-level generation potential forecasts aren’t just useful for nodal trading. When aggregated up to load zone or ISO-aggregates, they provide a powerful signal of the regional and market-wide curtailment that’s necessary to balance the grid.  For example, Figure 7 shows a strong correlation between total ERCOT West Zone Generation Potential and actual West Zone curtailment, which isn’t visible when looking at Actual (post-curtailed) Generation. Predicting this dynamic via generation potential forecasts is critical for managing zonal basis risk.

Figure 7. Total ERCOT West Sunairio Generation Potential provides a strong signal of actual total renewable curtailment in the region while actual (post-curtailed) generation completely obscured the relationship

For Grid Planners: Precision for Planning and Production Cost Modeling

Beyond trading, this granularity is the new baseline for production cost modeling (PCM) and Grid Planning. Our hourly Asset-Level Generation Potential is ideally suited as an input for every individual renewable generation node of SCED models. 

Further, Sunairio’s Generation Potential data enable planners to precisely quantify the incremental export capacity required to mitigate regional congestion and validate the specific transmission infrastructure necessary to support it. This shift moves the analytical focus from generalized capacity expansions toward targeted infrastructure investments designed to realize a specific volume of weather-driven energy potential.

See the Signal, Find the Missing MWs

The increasing penetration of intermittent renewables and the rising occurrence of unprecedented weather have rendered historical grid performance a diminishingly reliable proxy for future risk. The integration of Sunairio ONE-powered Generation Potential forecasts at the asset level provides the high-fidelity visibility required to navigate a fundamentally decentralized grid.

By reconciling high-resolution weather modeling – including terrain-induced fluid dynamics and advanced irradiance physics – with Digital Twin asset specifications, Sunairio replaces regional approximations with nodal-specific signals. This granularity is essential for quantifying the supply-side volatility that drives congestion, curtailment, and basis risk in modern power markets.Sunairio ONE represents a shift from reactive grid observation to predictive, physics-based intelligence. Want to see the Generation Potential for your most congested nodes? Contact us to run your portfolio through Sunairio ONE.

Potomac Economics. “ERCOT Wholesale Electricity Market Monthly Report.”
https://www.potomaceconomics.com, Potomac Economics, 10 12 2025,
https://www.potomaceconomics.com/wp-content/uploads/2025/12/2025-11_Nodal_Monthly_Report.pdf. Accessed 16 3 2026.

ERCOT’s 85 GW load forecast for Winter Storm Fern wasn't just high—it was indefensible

We’re one month removed from the havoc that was Winter Storm Fern. Focusing on its effects in ERCOT, we dug into the weather, grid, and market fundamentals to provide six strategic insights:

1. Five days out, temperature forecasts disagreed about timing and intensity but were reasonably correct about the event low/peak heating period on 1/26. While the peak heating event was relatively extreme, in this particular case it wasn’t much of a short-term surprise. In terms of long-run climatology though, the peak heating event on 1/26 was a P99 event for that calendar day, the week was a P97 and the month was P65.

2. ERCOT appears to have deliberately biased their load forecasts high (incorrectly). Assuming ERCOT was using weather forecast consensus (which was relatively accurate) as the input, we find their 85 GW forecast for 1/26 and 80 GW forecast for 1/27 to be essentially indefensible, falling well outside the range of recent weather-load relationships.

3. ERCOT appears to have deliberately biased wind generation forecasts low (correctly). We know this because their STWPP forecast (representing a P50 level) was below their WGRPP forecast (representing a P20 level). They appear to have manually adjusted the STWPP low but left the WGRPP alone, creating a mathematically impossible scenario.

4. Load in West Texas was significantly affected by icing of oil and gas infrastructure. A large share of load growth in West Texas has been driven by the electrification of oil and gas drilling and processing. When these wells/pipes froze, the associated electrical compression didn’t operate, resulting in exceptionally low load.

5. The market result: a high day-ahead (DA) clear driven by ERCOT’s posturing and then a real-time (RT) fail due to reality and load destruction. ERCOT’s posturing appeared to have the intended effect, supporting a high DA clear with ample reserves to blunt the risk of RT market capacity shortfalls — even in the face of lower than expected wind.

6. Sunairio’s price forecast explained the actual realization and the advanced market fear premium. Looking at our forecast of the 1/26 5x16 North Hub locational marginal pricing (LMP) realizations from the week before, we can see that our expected value was almost spot on, while the long tail to the right explains why some were willing to pay $600+ (because there was a small chance of clearing over $1,000). Note: the mean of our price distribution was at the 80th percentile — far from the median — because of the extreme right skew (risk of high prices).

Over the past few weeks, we have explored what makes Sunairio ONE a "next-generation" forecast. In Part 1, we discussed the necessity of a calibrated ensemble that accurately captures extremes. In Part 2, we demonstrated why high spatial and temporal resolution is critical for modeling modern renewable assets like wind and solar. In this final installment, we show that Sunairio ONE provides a unified, seamless outlook from hours to years, eliminating the fragmentation issues that the industry faces today. 

The Current Landscape: A Patchwork of Compromises

Today, energy traders, grid operators, and other energy professionals have access to a growing collection of public weather forecasts that are each published with differing outlook horizons, temporal resolutions, and refresh schedules. For example, NOAA’s HRRR provides hourly forecasts for the next two days. Other forecasts, such as the GFS or ECMWF’s IFS, stretch to about 2 weeks, but with lower temporal resolution further in the outlook period (see Figure 1, top panel). Seasonal-range forecasts such as the CFS or IFS SEAS look many months into the future, but are low temporal resolution (see Figure 1, bottom panel) and in the case of the SEAS, published only once per month, letting forecasts go stale quickly. While most weather forecasts are updated just four times per day (e.g., 00Z, 06Z, 12Z, and 18Z) or fewer, Sunairio ONE is refreshed each hour using the latest information. 

To look beyond 9 months, one must turn to climate models (the current iteration of models are known as CMIP6) instead of weather forecasts, which can be significantly biased, typically provide only vague daily averages, and obscure intraday volatility. 

Figure 1. (Top panel) Even within a short 16-day outlook, alternative forecasts provide sparse data with low temporal resolution across days while Sunairio ONE provides dense hourly data without gaps; (Bottom panel) For seasonal or longer outlook periods, only Sunairio ONE provides hourly data. 

Thus, in order to build a complete picture of future weather risk– both in the short-term and in long-term planning– energy traders and asset managers need to stitch together information across multiple sources in a patchwork manner as illustrated in Figure 2. This fragmented approach requires building multiple complex data pipelines, and perhaps more importantly, it makes it challenging to synthesize insights for key operational and planning decisions. Sunairio ONE provides the seamless solution that the industry needs. 

Figure 2. The standard industry approach involves stitching together different models with varying resolutions, creating data "seams" and leaving a massive void for long-term planning. Sunairio ONE provides a seamless long-term outlook. 

The Sunairio ONE Advantage: The Unbroken Line

Sunairio ONE was designed to eliminate the seams that exist when moving across timescales, providing full continuity and high resolution. 

How is it possible to generate a credible hourly ensemble forecast a decade in advance?

It requires bridging the gap between traditional numerical weather prediction (NWP) and long-term climate modeling. Sunairio ONE is not just "extending" a standard weather model until it falls apart. It utilizes a proprietary blend of physics-based modeling and AI-driven calibration.

As detailed in our overview of ENSO-informed climate simulations, our long-range ensembles are constrained by large-scale climate signals, such as El Niño and La Niña cycles. This ensures that the weather patterns generated in years 5, 10, or 15 are physically consistent with the broader climate realities expected during those periods, while still providing the hourly volatility required for asset modeling.

The Business Impact:  Strategic Consistency

Moving from a fragmented patchwork to a seamless solution offers more than just technical convenience; it solves fundamental business problems:

• Eliminating "Model Basis Risk": When a short-term trading desk uses one model and the term traders use another, the firm risks making operational decisions that contradict a strategic outlook. A unified model ensures everyone is reading from the same book.
• Aligning Analytics with Market Contracts: Power markets operate on financial timelines (e.g., balance-of-month (balmo), seasonal (summer/winter), and annual contracts) that frequently collide with the rigid boundaries of standard weather forecasts. Sunairio ONE solves this misalignment by providing an unbroken, 15-year hourly stream. 
• Streamlined Data Pipelines: Data engineers no longer need to build complex intake systems to parse GRIB files from four different government agencies, normalize the data, and try to stitch it together. Sunairio ONE offers a single API feed and consistent timeseries format.
• Accurate Long-Term Valuation: To value an energy asset over its lifecycle, modeling assessments need realistic forward-looking weather assumptions that replicate trends and  volatility instead of simple historical averages. Sunairio ONE’s 15-year hourly ensemble captures trends, replicates extremes, and provides the necessary fidelity for accurate long-term financial modeling.

The Future of the Grid is Unified

Over this three-part blog series, we have outlined why the energy transition demands a new class of forecast technology. The grid of the future cannot run on forecasts that fail to see extremes, lack necessary resolution, or fragment after two weeks.

Sunairio ONE delivers unprecedented fidelity at all time scales. It is calibrated, sharp, high-resolution, and, crucially, seamless. It’s time to stop stitching together forecast models and start solving energy challenges with a unified view of the future.To see the difference seamless data can make for your organization, contact us today for a demonstration of the Sunairio ONE 15-year hourly ensemble.

This is the second blog post in our three-part series that explores key areas where Sunairio’s Omniscale Next-generation Ensemble (ONE) forecast model outperforms traditional solutions. Our first blog post demonstrated that Sunairio ONE was more calibrated, sharp, and extremes-conspicuous than legacy ensemble methods. Here, we show that Sunairio ONE’s high spatial resolution captures local variability of wind speeds better than existing models and demonstrate how that wind speed gradient can translate to a large variation in expected power output.

Introduction 

Variable renewable energy assets like utility-scale wind and solar farms experience meaningful fluctuations in power output as local weather conditions change. However, most weather forecasts aren’t generated at a spatial or temporal granularity that’s sufficient to accurately anticipate those dynamics.  In the case of wind, for example, detailed topographical features and their resulting phenomena (e.g., terrain-induced waking or speed-up effects) are lost at coarse resolutions¹. Sunairio ONE addresses these challenges by providing a high resolution weather forecast, which captures small-scale variability, enabling best-in-class asset-level generation potential forecasts.

Wind farm forecasting best practices   

Asset-level wind energy forecasting requires both highly-accurate weather forecasts of mesoscale wind (on the order of a few km) and sophisticated models of wind field dynamics within the wind farm footprint, which are affected by many factors including terrain, vegetation cover, and turbine waking.

In fact, we find that wind energy modeling best practices dictate separating this problem into three stages: 1) the mesoscale free-field wind² forecast, 2) the microscale wind field within the wind farm footprint (less than 1km effects usually influenced by terrain), and 3) turbine-by-turbine waking (Figure 1).

Figure 1. Diagram showing three mains steps involved in accurate modeling wind energy at a wind farm. 

Sunairio has already pioneered novel solutions to Steps 2 and 3, previously reporting on our use of highly-efficient CFD models to model turbine-scale (100m) wind fields and waking for all operational wind farms in the U.S.

We are now excited to leverage Sunairio ONE, our high-resolution ensemble weather forecast, to improve mesoscale wind forecasting for asset-level generation.

Existing weather forecasts are too coarse, or fine-but with limited outlook    

Surveying the landscape of publicly available weather models that can provide mesoscale free-field wind forecasts shows that none offer truly high spatial and temporal resolution output that extend beyond a few hours (Table 1). Furthermore, Figure 2 helps provide some scale to the current problem, comparing the spatial resolution of Sunairio ONE (at approximately 2km) and the IFS (at approximately 10km) against actual wind farm footprints in ERCOT.  Mesoscale forecasting at 10km (like the IFS) vs. 2km may average out important wind gradients that vary over a wind farm and lead to exponential wind generation errors (more on this below). 

Additionally, Sunairio ONE forecasts are generated at hourly temporal resolution–critical for energy forecasting applications–for the full forecast period; no other major publicly available forecast model does this. For example, the IFS starts at hourly resolution for the first three to four days but then drops to 3-hourly and 6-hourly steps as it gets further out.

Forecast Model	Outlook Period	Temporal Resolution	Spatial Resolution
HRRR (NOAA)	48 hours3	1-hourly	~	3km
GFS/GEFS (NOAA)	16 days	0-120 at 1-hourly 123-384 at 3-hourly	0.25°	~28km
IFS/IFS ENS (ECMWF)	15 days	0-90 at 1-hourly 93-144 at 3-hourly 150-360 at 6-hourly	0.1°	~10km
AIFS/AIFS ENS (ECMWF)	15 days	0-360 at 6-hourly	0.25°	~28km
Sunairio ONE	15 years	1-hourly	0.02°	~2km

Table 1. Comparison of forecast models by outlook period, temporal resolution, and spatial resolution. 

Figure 2. Zoomed area of map of wind farms in Texas (shaded bounding boxes show footprints of wind turbines at wind farms). 

Wind is highly variable across wind farm scales

To visualize just how important it is to capture variations in free-field wind forecasts, we analyzed a day of wind forecast performance (three days out) between Sunairio ONE and the IFS across a major ERCOT wind farm, Capricorn Ridge. As Figure 3 shows, the difference between wind speeds at the north end of the farm compared to the south end (turbines marked in red in left panel) were as great as 3 m/s in some hours. While Sunairio ONE successfully forecasted this wind field gradient (right panel), the lower-resolution IFS could not resolve the proper dynamics, instead seeing a fairly uniform wind field.  

Figure 3. (Left panel): Map of Sunairio ONE forecasted wind speed across a wind farm in Texas over a 24-hour period; (Right panel): Line plot tracking the wind speed gradient (i.e., difference in wind speeds) from the north side to the south side of the farm compared toSunairio ONE 3-day ahead forecast and IFS ENS 3-day ahead forecast.

Why it all matters: Even seemingly minor weather variations can have outsized impacts in power output

Does a 3 m/s gradient in free-field wind speed over a wind farm really matter? Depending on the absolute wind speed levels, it can matter immensely. Below their rated capacity, wind turbine power output scales cubically with wind speed which causes even seemingly minor variations in wind speed to lead to large differences in generation potential. Turning to Figure 4, we plot an example power curve of a 1.5 MW turbine, which represents the majority of turbines at Capricorn Ridge. A difference of 3 m/s in wind speed can translate to up to a 959 kW delta in power output, or 64% of rated capacity!

Figure 4. Power curve for the 1.5 MW wind turbines that comprise the Capricorn Ridge Farm. 3 m/s wind speed gradients can translate to power differences as large as 959 kW.

Over a wind farm with hundreds of turbines, the impact of modeling wind speeds at a lower resolution on generation potential forecast performance can quickly add up.

Conclusion

Accurately forecasting local wind speeds is essential for creating reliable forecasts of asset-level generation. Sunairio ONE provides 2km spatial resolution, hourly temporal resolution forecasts that are ideally suited to serve as the critical free-field wind forecast step in wind farm generation modeling pipelines.

1. Energy Systems Integration Group. 2023. Weather Dataset Needs for Planning and Analyzing Modern Power Systems (Full Report). A Report of the Weather Datasets Project Team. Reston, VA. https://www.esig.energy/weather-data-for-power-system-planning ↩︎
2. Free-field wind speeds are wind speeds expected in open, unobstructed areas free of turbulence caused by structures such as wind turbines. ↩︎
3. The HRRR goes out to 48 hours only for the 0Z, 6Z, 12Z, and 18Z initializations. All other initializations go out 18 hours. ↩︎

Late last month, on June 24, 2025, PJM experienced a record-setting heatwave that sent heat indexes soaring above 100^oF throughout the Mid-Atlantic, hotter than anything observed in late June since at least 1950. This extreme weather event sent regional power demand and prices spiking — though not at the same time. 

The price spike occurred in the two hours after demand peaked that day — a (perhaps unexpected) consequence of the growing reliance on renewables in PJM. Here's how solar fundamentally changed PJM's risk profile on that sweltering June day.

How net load drives grid stress and power prices

Net load — not native load — drives today’s price spikes. PJM native load peaked in hour ending (HE) 18, but Western Hub LMP spiked to $1,500/MWh+ in subsequent HE 19 and 20, as the plot in Figure 1 shows. 

Why? Because those are the hours in which net load peaked. As we’ve discussed before, net load (native load minus renewables) is the primary driver of grid stress in power markets with a significant share of renewables. And when grid stress rises, so do prices.

We can now officially count PJM as a market in which renewables can’t be ignored.

As Figure 2 shows, there was approximately 10 GW of solar generation in PJM on June 24 during hours 10–17, which significantly reduced net load — and therefore grid stress — throughout the day. Even in the peak load hour (HE 18) there was still 9 GW of solar generation. 

But this solar output dropped precipitously as the sun set across the ISO footprint to just 3 GW in HE 20 — causing net load (and overall grid stress) to continue to rise even as native load was falling. The resulting net load ramp required PJM to quickly dispatch expensive units to stabilize the grid.

The end result? The highest prices we’ve seen in PJM since 2022 occurred much later in the day than would have been historically expected for this period in June.  

Solar is shifting PJM fundamentals

Peak net loads have been creeping later in the day for some time. The highest net load hour in June has shifted from HE 17 in 2018 and 2019 to HE 20 this year, as Figure 3 shows  for each June’s highest net load day since 2018.

This pattern extends to less extreme days as well: 70% of daily net load peaks occurred at hour ending 19 or later in 2025 compared to 0% of peaks occurring that late in June 2018, as Figure 4 demonstrates across all June days since 2018.

This intraday net load pattern shift mirrors the massive growth of installed solar capacity in PJM. Utility-scale solar capacity has increased roughly 500% since 2020, from 2 GW to more than 12 GW as of June 2025, as Figure 5 illustrates.

As solar penetration increases in PJM, it drives greater price volatility and concentrates the highest price risks later into the evening hours during summer months. 

Fundamentally, it creates a grid management challenge as the quickly setting sun necessitates a steep ramp for other generators to be brought online. This highlights the need for accurate net demand (load, solar, and wind generation) forecasting and fast-ramping resources to successfully enable renewables integration.

Even with the headwinds that new solar projects face today, we still expect a sizable capacity addition over the next few years as projects that are under construction reach completion and the pipeline of projects that qualify for remaining tax incentives break ground. 

The bottom line: add PJM to the list of power markets whose fundamental outcomes are increasingly controlled by the intraday volatility and intermittency of weather-driven generation resources.

In modern electric grids, grid operators must continually balance power demand with power generation to maintain steady grid frequency and ensure overall grid stability. This entails anticipating, in real time, short-term weather-driven changes in both customer load and renewables output to match the remaining “net demand” with dispatchable generation. Grid planners, on the other hand, look far into the future to ensure long-term “resource adequacy” — i.e., that future generation resources will be sufficient to cover demand under all possible weather scenarios and grid conditions.

Satisfying resource adequacy through long-term grid planning processes is a critical component of grid reliability because generation resources and transmission infrastructure can’t be built in the few days ahead of a forecasted heat wave or winter storm. 

To study long-term resource adequacy, grid planners:

1. Construct a set of future weather scenarios
2. Make some assumptions about asset-level availability
3. Calculate the resulting hourly net demand given anticipated customer load projections and renewable energy capacity buildout
4. Quantify the likelihood of capacity shortfalls that would necessitate shedding load to maintain grid stability (i.e., brownouts or blackouts)  

Resource adequacy studies, in other words, rely on an accurate characterization of future weather variability. In practice, grid planners typically use historical weather as a proxy for future weather — tacitly assuming that a) future weather will be similar to historical weather and b) historical weather is a large enough sample to assess the risk of extreme events.

Unfortunately, both of these assumptions are wrong. Future weather is different from historical weather. Average temperatures, wind speeds, and irradiance are all expected to shift over time from climate change effects, to varying degrees and in different directions depending on what region is being studied. Moreover, the shape of weather distributions will also change. For example, climate modeling shows that extreme cold risks may actually become more frequent in the future as climate change weakens the jet stream — even though average temperatures will rise. That means that the bottom tail of a temperature distribution gets longer while the rest of the distribution shifts right.

Regarding the second assumption — that we have enough historical weather data to assess extremes — the historical record is much smaller than you might think.

First, we can only use serially-complete historical datasets that have values for every hour of the year because we can’t make reliable statistics from partial-year data. This generally excludes airport station temperature data before 1980.

Next, we have to find high-quality weather data to model wind and solar generation. Traditionally, that has meant relying on satellite-based irradiance data (available since 1998) and the NREL WIND Toolkit, which is a high-resolution wind speed dataset calculated for 2007-2013.

As a result, a resource adequacy study that includes correlated temperature-driven load, irradiance-driven solar generation, and wind-driven wind generation would only be able to call upon 7 historical year samples — drastically limiting the view of weather extremes that the grid could face. 

Sadly, flawed grid planning due to insufficient weather assumptions can lead to catastrophe, with major power outages occurring recently in California (2020), Texas (2021), and North Carolina (2022). An NREL post-mortem of all three events found modern electric grids, which increasingly rely on intermittent weather-dependent renewable energy generation, to be increasingly impacted by extreme weather conditions — and that “weather in recent years has exceeded the bounds of anticipated conditions,” highlighting the need for improved planning processes that accurately account for jointly correlated extremes of weather, power generation, and generation outages.

To fill this gap, Sunairio generates a 1,000-path ensemble of future hourly weather to give grid planners robust, complete, and actionable distributions of weather, grid conditions, and asset-level variability. This ensemble is climate-change adjusted, trained on the longest possible series of weather data, and numerous enough to gain intelligence into future extreme events — in whatever form they may come.

For plants, spring is a season of growth. But for power market participants and grid operators, it’s a time of surprises and volatility.

While typically mild spring weather results in low expected grid demand, extreme spring weather — through a confluence of factors — can drive real-world hourly grid balances to levels that approach or exceed emergency conditions. This contrast between mostly moderate days and acute periods of serious grid stress makes spring just as challenging to navigate as the traditional “peak” seasons in summer and winter.

Properly anticipating this inherently stochastic risk requires both a nuanced understanding of the underlying fundamentals and a probabilistic framework to quantify low-probability but high-impact economic and reliability outcomes. In this blog post, we examine actual weather, grid, and market events from this spring in ERCOT within a probabilistic context. Looking ahead to spring 2026, we then explore how rapidly changing grid fundamentals may alter next year’s spring grid risk profile.

Late season cold and an early heat scare in Texas

This year has not been an exception to the rule that spring weather varies wildly. Temperatures on April 7 dropped to 40ºF at DFW and into the 30s around Dallas (colder than all of March) while an early season heat wave drove forecast highs over 100ºF last week (levels not typically seen until July).

Figure 1 plots this spring’s weekly average, minimum, and maximum temperatures (green lines) against ranges derived from Sunairio probabilistic climatology (box and whisker plots) at DFW. As the plot shows, most hourly temperatures throughout spring are relatively mild — between 50ºF and 70ºF — but extremes dip into freezing territory and extend into severe heat.

An epic forecast fail

Temperatures could have been even more extreme this year if the weather forecasts for the May 12 week hadn’t flopped. As we see in Figure 2, the forecast for the May 14 afternoon high at DFW was 102ºF just 12 hours out — yet realized 8 degrees lower at 94ºF. That 8-degree temperature forecast error likely reduced ERCOT RTO load by approximately 9 GW versus higher temp load expectations, leading to a $40 Day-ahead/Real-time spread (DA higher than RT) — highlighting the difficulties of navigating grids amid temperature variability and forecast uncertainty.

Generation outages peak

In spring, dispatchable resources are often not available on purpose. Given that the majority of Texas spring weather is mild and that the season immediately precedes the peak demand summer months, generators schedule the bulk of their maintenance outages during this time. As we see in Figure 3, nonrenewable (thermal) generation outages usually peak in early April (coinciding with the mildest expected temperatures and lowest expected load), though unscheduled outages can cause significant variability. Ironically, such high levels of dispatchable unit outages can tip the grid from normal conditions into capacity shortfalls. Scheduling generation outages in spring makes sense, until it doesn’t.

Renewables ramp up

Temperatures and unit outages aren’t the only thing rising in ERCOT during spring. Wind generation typically maxes out in early April and solar generation typically increases by 38% from March 1 to May 31. Further complicating grid dynamics are intraday generation patterns, where wind generation peaks overnight and solar generation is dictated by the diurnal sunup-to-sundown cadence. Figure 4 plots both these seasonal (left panel) and intraday (right panel) trends.

Implications for power markets and reliability

Understanding grid risk these days is much more than understanding a simple temperature -> load relationship. It’s a multidimensional puzzle whose solution is driven by correlated variability between load, wind generation, solar generation, and the availability of dispatchable resources. It’s sensitive to temperature extremes, shifts in renewable generation patterns, and unit outages.

At Sunairio, we combine these fundamentals into one metric that provides a reliable indicator of overall grid stress — the Grid Stress Index (GSI), which measures the ratio of Net Demand (load minus renewables) to Available Dispatchable Capacity (read more about the construction of GSI here).

As Figure 5 shows, this metric anticipates hourly price volatility, with conditions in ERCOT (measured by delivered spark spreads) being relatively tame until GSI surpasses 60% — and becoming extremely volatile as GSI rises above 70%.

To understand grid risks on spring days, we use Sunairio’s historical simulations in Figure 6 to plot expected (P50) and extreme (P99) hourly GSI in both early (March 15) and late (May 15) spring 2025. For clarity, we shade the background darker above GSI = 60% to reflect increasing price risk. As the plot shows, while the majority of hourly conditions in spring fall below the 60% level corresponding to high prices/capacity shortages, extremes in several hours present significant risks.

In particular, cold mornings and low renewables can drive hours ending (HE) 7–9 to the danger zone in early spring (left panel, Figure 6), while afternoon heat and decreasing solar in the evening drives HE 20–22 even further beyond in late spring (right panel, Figure 6).

Indeed, on the exceptionally cold early spring ERCOT morning this year (April 7), we saw prices spiking close to the price cap in ERCOT and averaging $1,552/MWh in an hour (Figure 7) due to high GSI resulting from exactly this combination of low temperatures (dipping into the 30ºFs surrounding Dallas, low renewables (minimal solar at HE7), and high generation outages (25+ GW).

That one hour on April 7 added approximately $4/MWh to the monthly peak price, meaning that just 0.2% of hours in the April peak contract (1/352) accounted for 10% of the entire contract value.

Looking ahead to 2026

With the spring drama almost behind us for 2025, what should we expect next year? Accounting for load growth and new unit additions/retirements, we find that conditions should generally be calmer next year across all spring hours. In other words, generation additions will outpace load growth — though the risk of acute price spikes remains, especially in the evening. As Figure 8 shows, P50 and P99 hourly GSI in both early spring (March 15) and late spring (May 15) are expected to be lower in 2026 compared to 2025, representing downside risk to ERCOT market prices.

Navigating spring volatility

Spring volatility highlights the challenges of managing grid and energy markets risk without a robust framework for evaluating the jointly correlated, multidimensional, granular, and skewed nature of the electric grids. To effectively navigate this period, grid planners and energy traders need to understand the complicated interplay between averages and extremes — to expect not just the seasonal trends but also to expect extreme events that drastically alter reliability and economic outcomes. It’s not enough to know that ERCOT grid balances for HE 7 typically don’t present much of a risk. We need to quantify the risk — via, for example, stochastic simulations — that weather and grid conditions conspire to drive capacity reserves low and market prices to the moon (as they did during HE 7 on April 7, 2025).

Introduction

“Prediction of instantaneous weather patterns at sufficiently long range is impossible.” This statement from Edward Lorenz remains as true today as when it was first written by the chaos theory founder in 1982, with current numerical weather prediction (NWP) model skill rapidly decreasing to zero over the course of a couple of weeks (Figure 1, also see Zhang 2018).

Although attempts to make deterministic predictions of localized and instantaneous weather devolve into chaos within a two week forecast horizon, it is possible to meaningfully predict large-scale atmospheric phenomena on seasonal time scales (several months to a year). These model predictions, however, need to be carefully processed before use–as the ECMWF states, “use of the raw numerical forecast products without interpretation is not recommended” (ECMWF 2021).

In this blog post, we (1) discuss one of the most impactful large-scale climate phenomenon (the El Niño - Southern Oscillation), (2) describe how Sunairio performs statistical processing and model assimilation to incorporate the latest El Niño intelligence into our climate simulations, and (3) evaluate how the resulting ENSO-informed climate simulations compare to both historical patterns and NOAA’s seasonal-timescale Climate Forecast System (CFS).

El Niño - Southern Oscillation

The El Niño - Southern Oscillation (ENSO) cycle refers to periodic changes in sea level air pressure and sea surface temperature (SST) in the southern Pacific Ocean. It is typically measured by the mean sea surface temperature anomaly of the “Niño 3.4” region shown in Figure 2, left panel (known as the Niño 3.4 index).

In the United States, especially high Niño 3.4 values (an “El Niño event”) are associated with warmer and drier winter weather in the northern US as well as cooler and wetter winter weather in the southern US (Figure 2, right panel, Halpert 2014). On the other hand, especially low Niño 3.4 values (a “La Niña event”) are associated with cooler and wetter winter weather in the northern US as well as warmer and drier weather in the southern US.

Figure 2: Regions of ENSO sea surface temperature indices (left) and the typical weather pattern expected in winter due to an El Nino event (right) (Source NOAA)

 

As the ENSO cycle is the “strongest interannual signal in the global climate system” (Tang 2018), it is heavily studied and forecasted, with current models showing positive predictive skill 6-12 months out (Barnston 2012). Figure 3 shows historical values of the Niño 3.4 index (left) and predictions from the CFS (right).

Figure 3: Historical Nino 3.4 region sea surface temperature anomaly (left) and current forecast of Nino 3.4 SST anomalies from the NOAA CFS (right)

 

Creating ENSO-aware Sunairio Simulations

Surveying the predictive power of current weather science, we find that high-resolution NWP models are predictive for at most 15 days, medium-resolution predictions of the ENSO cycle are predictive for 6-12 months, and low-resolution models of long-term climate trends can be predictive for longer time periods.

Sunairio’s climate simulations aim to therefore combine the best-available intelligence at each time horizon: long-term climate trends from the latest climate models (CMIP6), medium-term ENSO predictions from NOAA, and simulated hourly anomalies from Sunairio’s proprietary stochastic simulation generator.

Concretely:

1. Sunairio simulates 1,000 probabilistic hourly climate-trend aware weather paths from 15 days to 15 years (picking up where the NWP models lose skill) as correlated anomalies from climatological means that reflect historical weather patterns and adjust for CMIP6 climate trends.
2. Sunairio generates 1,000 probabilistic ENSO paths over a 12-month time horizon by extrapolating a range of outcomes from the latest 40 CFS model runs (4 runs per day times 10 days) (Figure 5, left panel).
3. Sunairio derives, from historical data, the correlation and effect of Niño 3.4 Index values on local weather variables throughout the year (Figure 5, middle panel).
4. Sunairio links hourly weather simulations (1) with ENSO paths (2)–and adjusts the simulated weather according to the corresponding Niño 3.4 index effect (3)(Figure 5, right panel).

 

Inspecting the January 2025 ENSO-informed Sunairio climate simulations in Figure 5, for example, we find that the median ENSO forecast is approximately -1σ (left panel), that the corresponding ENSO temperature effect at DFW is approximately +0.5^oC (middle panel), and that the ENSO adjusted Sunairio simulations at DFW are indeed approximately +0.5^oC warmer than January 2025 climatology (right panel).

Do Sunairio’s ENSO-adjusted Weather Simulations Reproduce Expected ENSO Effects Across Multiple Geographies and Weather Variables?

In this section, we verify that Sunairio simulations appropriately reproduce the historical relationship between the Niño 3.4 index over CONUS for major weather variables (2m temperature, 100m wind speed, and irradiance).

At 191 weather stations across the contiguous United States, we performed a backtest in which we simulated local hourly weather adjusted with historical Niño 3.4 Index values between the years 1997-2022.

Dividing calendar years into seasons (Dec - Feb, Mar - May, Jun - Aug, and Sep - Nov), we then calculated the impact of a +1σ Niño 3.4 value with monthly simulated weather averages at each site (Figure 7, right)–and compared these patterns to the historical (1950-2023) relationship between Niño 3.4 deviations and monthly weather averages (Figure 6, left). As we can see in Figure 6, Sunairio simulations accurately reproduce the historical ENSO relationship. (Click the weather variable headings to select different plots.)

Figure 6: Historical (left) and simulated (right) relationships between weather variables and Niño 3.4 index values. A positive +1σ Niño 3.4 index in winter, for example, tends to cause a 1 ^oC increase in monthly 2m temperature in Minnesota. Sunairio simulations faithfully reproduce historical relationships.

 

Predictive ENSO Signal + Stochastic Climate Simulation = High-fidelity Seasonal Forecast

As Sunairio simulations combine skillful ENSO predictions with high-fidelity climate simulations, medium-term Sunairio simulations are actually more predictive of future local weather than the CFS.

We validated this result by comparing historical monthly mean temperature measurements at the same 191 weather stations to both CFS predicted means and Sunairio simulated means. We found Sunairio simulations to be more predictive than CFS counterparts at 60% of the weather stations with an average skill score improvement of 2.1%. 

Incorporating high-fidelity climate modeling, in other words, makes Sunairio simulation averages more predictive than the CFS.

Conclusions

Reviewing the sections above, we find that:

• The El Niño – Southern Oscillation (ENSO) cycle is strongly correlated with global weather patterns.
• While deterministic models cannot predict instantaneous weather beyond a 15-day time frame, the Niño 3.4 Index can be forecast with some skill over a seasonal (6-12 month) time frame.
• Sunairio incorporates the latest intelligence on climate trends (CMIP6) and Niño 3.4 (CFS) predictions into its weather simulations.
• Sunairio simulations accurately reproduce the historical relationship between ENSO and local weather.
• Sunairio simulations, by combining high-fidelity climate simulation with a predictive ENSO signal, are more predictive of weather means than seasonal weather models.

Finally, we note two additional advantages of high-resolution Sunairio simulations over the CFS. First, Sunairio simulations can be generated at arbitrary locations, while CFS predictions must be interpolated from a 56km horizontal-resolution grid. Second, while seasonal forecast models like the CFS only offer one possible view of future weather in the medium term, Sunairio simulations offer 1000 paths of future weather for the next 15 years. As such, Sunairio weather simulations can be used to derive probabilistic risk distributions for commercial applications.

References

Barnston, Anthony G., Michael K. Tippett, Michelle L. L'Heureux, Shuhua Li, and David G. DeWitt. "Skill of Real-Time Seasonal ENSO Model Predictions during 2002–11: Is Our Capability Increasing?". Bulletin of the American Meteorological Society 93.5 (2012): 631-651.

ECMWF. “ECMWF System 4 user guide.” 1.1 (2011): 1-41. 

ECMWF “SEAS5 user guide.” 1.2 (2021): 1-44.

Halpert, M. “United States El Niño Impacts.” NOAA (2014).

Lorenz, Edward N. "Atmospheric predictability experiments with a large numerical model." Tellus 34.6 (1982): 505-513.

Saha, Suranjana, Shrinivas Moorthi, Xingren Wu, Jiande Wang, Sudhir Nadiga, Patrick Tripp, David Behringer, Yu-Tai Hou, Hui-ya Chuang, Mark Iredell, Michael Ek, Jesse Meng, Rongqian Yang, Malaquías Peña Mendez, Huug van den Dool, Qin Zhang, Wanqiu Wang, Mingyue Chen, and Emily Becker. "The NCEP Climate Forecast System Version 2". Journal of Climate 27.6 (2014): 2185-2208.

Tang, Youmin, Rong-Hua Zhang, Ting Liu, Wansuo Duan, Dejian Yang, Fei Zheng, Hongli Ren, Tao Lian, Chuan Gao, Dake Chen, Mu Mu. “Progress in ENSO prediction and predictability study.” National Science Review 5.6 (2018): 826–839.

Zhang, Fuqing, Y. Qiang Sun, Linus Magnusson, Roberto Buizza, Shian-Jiann Lin, Jan-Huey Chen, and Kerry Emanuel. "What Is the Predictability Limit of Midlatitude Weather?". Journal of the Atmospheric Sciences 76.4 (2019): 1077-1091.