Long-term synthetic wind dataset

This page presents a synthetic wind dataset prepared by Meridian Energy, who have kindly made it available for public use.

The dataset was derived using two sources of information: Meridian wind monitoring mast data and NIWA station data available from the National Climate Database.

A synthetic wind dataset has been available for use in wind integration studies since 2009.

The NIWA dataset has a ten-minute time step and covers a five-year span, which means it is suitable for modelling wind variability over minutes to days. However, due to its short length, it is not suitable for modelling variability over months to years.

The new synthetic dataset provided by Meridian fills this niche. It consists of daily mean wind speeds at twelve different sites distributed around New Zealand, and covers a nineteen-year span, which means it can be used for analysis of longer-term wind variability.

A peer review carried out by NIWA supports the methodology used, compares the synthetic dataset with other available wind datasets (to the extent possible), and concludes that the dataset is fit for the purpose of gaining a high level understanding of long-term variability in wind generation output.

If you have any questions or comments on the dataset, please contact info@ea.govt.nz.
Methodology

The two key inputs were:

    wind speed records for twelve Meridian sites distributed around the country, using measurements from actual wind monitoring masts at either established or potential wind farms. This data has a ten minute time step and covers the 24 month period from Jan 2005 to Dec 2006;

    wind speed records for a large number of automatic weather stations (AWS) across the length and breadth of the country, as recorded in NIWA's National Climate Database. Wind records are available at most AWS locations from 1990 (or earlier, in some cases), generally with an hourly time step.

If NIWA's AWS were sited at potential or actual wind farm locations, then there would be no need for a synthetic data set - the AWS data could be used to analyse variability in wind farm output over long time frames. However, the AWS are not at wind farm sites and are not a good guide to the dynamics of wind generation. A more complex approach is needed.

The approach taken by Meridian was to use the two-year period of overlap between wind monitoring mast data and AWS data to establish a statistical relationship between the two. This relationship was then used to extrapolate wind speeds at monitoring sites over the much longer span of the AWS data.

There is a well-established linear relationship between mean wind speed and wind farm energy yield (as opposed to the theoretical cubic relationship between wind speed and the output of a single idealised turbine). Accordingly, the analysis was carried out with a daily resolution, yielding a synthetic dataset of daily mean wind speeds. It could have been carried out at an hourly resolution, but this would have added little value in terms of understanding monthly, quarterly or annual energy yields.

The steps carried out for each of the 12 sites were to:

    obtain wind speed data as recorded by a Meridian wind monitoring mast;

    obtain wind speed data at 65 AWS stations distributed throughout the country (including several outlying islands);

    convert both wind datasets into daily averages over Jan 2005 - Dec 2006;

    fit a linear regression model to the data, using the Meridian mast data as the y-variable and the three most informative AWS sites as the x-variables;

    generate the synthetic dataset by applying the fitted regression model to AWS daily averages over 1990 - 2008; and

    scale the results to a mean of 8 m/s.

The final scaling was carried out to make it impossible to distinguish the monitoring sites with higher wind speeds from those with lower wind speeds. This protects Meridian IP without reducing the usefulness of the data (given the linear relationship between mean wind speed and wind farm energy output, end users can legitimately rescale the synthetic wind data to whatever level they desire).

For some sites, one or two of the AWS data series used in the regression model did not extend back as far as 1990. In these cases, the earliest part of the synthetic datasets were filled using a reduced regression model with only two (or one) predictors.

The 12 sites have been labelled: Auckland, Canterbury, Central NI, Hawkes Bay, Northland, Otago, Southland 1, Southland 2, Taranaki, Tararua, Wairarapa and Wellington.