How a hybrid forecasting arrangement in New Zealand aims to improve the accuracy of variable generation forecasts

At a glance

• Hybrid model: 7-day centralised forecasts, updated every 30 minutes
• Economic case: Forecast errors cost up to $273m annually; 15-year NPV of $151.5m–$326.5m
• Early results: Solar RMSE improved from 15.5% to 6.17%; wind from 13.9% to 12.5%
• Rigorous trial: 8-week assessment across 6 sites; DNV selected as centralised forecaster
• Why it matters: 80% of new generation will be wind and solar in the next 5–10 years

The Electricity Authority of New Zealand (Te Mana Hiko) is the regulator responsible for ensuring generators comply with industry rules and for procuring forecast services.

Prior to changes implemented by the Authority in July 2025 variable generators were responsible for procuring their own forecasts to inform their offers.

"Previously, variable generators had to submit an initial offer at least 71 trading periods (about 36 hours) before real time. They were not required to revise this offer until 4 trading periods (2 hours) prior to real time. As a result, updated wind and solar forecasts often weren’t reflected in offers until very close to real time. This was challenging from an operational perspective."‍‍
‍- Seb Hart, Electricity Authority New Zealand (Te Mana Hiko)

Forecast accuracy was inconsistent, with significant variation in accuracy across generators, including for near real-time forecasts (eg. 30 minutes ahead).

“Over-forecasting variable generation can depress price signals, leading other generators to offer less output. This creates a supply gap that must be filled by expensive firming generation at short notice, increasing costs to the market. Conversely, under-forecasting also causes inefficiencies as firming generators may start costly units based on price signals, only to not be dispatched when cheaper variable generation unexpectedly becomes available. This wastes resources and raises overall system costs”
- Seb Hart, Electricity Authority New Zealand (Te Mana Hiko)

The economic case for accurate forecasting

As part of the policy development, the Electricity Authority conducted a comprehensive cost-benefit analysis to understand the market impact of inaccurate variable generation forecasts.

An analysis of the trading periods between 1 November 2019 and 31 October 2022 revealed significant price impacts from forecast errors:

Under-forecasting of wind generation (occurring 32.5% of the time) resulted in an average spot price impact of -$6.90/MWh, equivalent to $94 million annually. Over-forecasting (occurring 67.5% of the time) resulted in an average spot price impact of $3.77/MWh, equivalent to $107 million annually.

When considering demand and forecast prices 12 hours before real time, these impacts increased substantially - under-forecasting impacts rose to $133 million annually, while over-forecasting impacts reached $273 million annually. The Authority calculated that wind forecast errors alone contribute approximately $960,000 per annum in deadweight loss and $2.2 million per annum in productive efficiency costs.

While this analysis focused on wind generation - the dominant form of intermittent generation at the time - similar dynamics apply to solar forecasting as the sector grows.

The business case for the hybrid forecasting arrangement was compelling. The five-year net present value was assessed at between $15.4 million and $33.9 million, with the 15-year net present value ranging from $151.5 million to $326.5 million - reflecting the expected substantial growth in intermittent generation over the coming years.

A hybrid centralised approach to forecasting

The Electricity Authority undertook a comprehensive policy process and looked at global approaches, including systems in:

• Australia
• ERCOT (Texas, USA)
• Alberta, Canada
• EU, UK, and Ireland

They decided to implement a hybrid forecasting arrangement where a centrally procured forecast will be provided for each variable generation site. Variable generators can use their own forecasts provided they can demonstrate these meet minimum performance standards.

The centralised forecaster (DNV) provides 7-day ahead forecasts and updates forecasts every 30 minutes. Variable generators must ensure their offers reflect the latest forecast.

Selecting a partner to be their official forecast provider

Once the Electricity Authority confirmed it would implement a hybrid forecasting arrangement, it carried out a competitive selection process for a centralised forecast provider.

Trial overview

Following publication of an RFP, shortlisted respondents were invited to take part in an eight-week trial to assess the accuracy of their wind and solar forecasts at six sites across New Zealand. The process was designed to ensure methodological fairness and to support the authority’s mandate for market reliability and consumer benefit.

Participants and scope

The three shortlisted vendors participated along with a benchmark which were all compared to the actual power generation from actual sites.

The trial focused on four wind farms (~290MW installed capacity) and two solar farms (~71MW installed capacity) which were selected because they are in different parts of New Zealand. This ensures forecast accuracy could be assessed across multiple areas of the country

Forecasting performance was assessed over eight different forecasting windows: 30-min, 1, 2, 4, 6, 12, 36, and 72-hour ahead forecasts. However, a heavier weight was put on the 4, 6, and 12-hours due to their importance for planning in the startup of slow-starting thermal plants and other flexible resources.

Assessment approach

All participants were assessed using a comprehensive suite of industry-standard metrics:

• Root Mean Squared Error (RMSE): To measure the magnitude of forecast errors, with lower values indicating higher accuracy.
• Mean Bias Error (MBE): To assess systematic over- or under-forecasting tendencies.
• Standard Deviation (STD): To quantify the variability of forecast errors.
• P10/P90 Interval Accuracy: To evaluate the reliability of forecast confidence intervals.
• P10/P90 Interval Width: To assess the sharpness of forecast intervals.
• Skewness: To measure the asymmetry of error distributions.
• Correlation: To indicate the strength of alignment between forecasted and actual generation.

Performance on RMSE, MBE and Correlation were of particular focus for the Electricity Authority.

Process notes

The Electricity Authority led a transparent and scientifically robust process:

• All participants were provided identical datasets and evaluation criteria.
• Data gaps were managed equitably to ensure fairness.
• Mid-trial feedback was incorporated, allowing participants to refine their approaches.

“During the trial – at the halfway point – we gave all the shortlisted providers a report on how they were doing. DNV did really well in the second half of the trial, which demonstrated the agility of their modelling team and technology to be able to improve so quickly.”
- Seb Hart, Electricity Authority New Zealand (Te Mana Hiko)

Selection & Outcome

During the trial, DNV emerged as the most consistent and reliable shortlisted forecasting provider, demonstrating strong overall performance across both wind and solar generation and exhibiting the least bias among all shortlisted respondents.

Implementation

Once DNV was confirmed as the centralised forecaster for the Electricity Authority, they began the process of onboarding the system operator Transpower, Energy Market Services (who provides real-time data to DNV) and their 7 variable energy generators (20 sites) - ensuring a smooth integration into the new hybrid forecasting arrangement.

“Our previous work in similar jurisdictions around the world ensured we were well-prepared to meet the challenges of integrating into the New Zealand market and its wide range of stakeholders and participants. In our experience, it’s also a credit to the Electricity Authority’s robust hybrid centralised forecast model that such a diverse group could be smoothly integrated, all relying on these forecasts to support market operations.”
- Mike Ferian, Principal Solutions Specialist, DNV

Variable generators may use their own forecasts if they demonstrate compliance with minimum performance standards, which is determined following an eight-week trial.

Results & benefits

After reviewing the first five months of the new forecasting system, DNV and the Electricity Authority were encouraged by the strong early results and positive market impact.

Compared to the same five month period in the previous year:

Wind forecasts (12-hour ahead):

• RMSE (averaged across all 14 wind farms) improved from 13.7% in 2024 to 12.8% in 2025.
• MBE shifted from +0.7% (over-forecast) in 2024 to -0.2% (under-forecast) in 2025.

Solar forecasts (12-hour ahead):

• RMSE (averaged across the two solar farms operating in both years) improved from 15.36% in 2024 to 6.38% in 2025.
• MBE improved from 15.19% in 2024 to 1.37% in 2025.

This extended forecast horizon now allows solar and wind generators to maximise their assets performance and returns, while allowing other generators in the market to plan ahead with confidence - especially coal which can take up to 3 days to ramp up if required.

In the New Zealand market, wholesale energy prices can be as low as $0.01 per MWh when there’s plenty of wind or hydro available. However, when wind or hydro generation is low and thermal generators set the price, costs can rise to $500 or more per MWh. More accurate forecasts from variable generators to improve system reliability and efficiency, which can reduce overall system costs.

“We look forward to continuing our work with DNV to ensure their wind and solar forecasts deliver the highest possible accuracy, helping variable generators make more informed offers. We’re confident that the new hybrid forecasting model, with DNV as the centralised forecaster, will improve variable generation forecast accuracy and offers - enhancing system reliability and efficiency, and reducing costs for consumers.”
- Seb Hart, Electricity Authority New Zealand (Te Mana Hiko)

With approximately 80% of new generation expected to come from wind and solar in the next 5-10 years, accurate forecasting is critical to New Zealand’s energy future.

‍

Visit: Electricity Authority New Zealand (Te Mana Hiko)

See: DNV Forecaster

‍