Our blogs Blogs home
Energy in Transition

Energy in Transition

Wind

Sources of long-term reference wind data in Ireland and southwest of Scotland: A consistency analysis

When undertaking an analysis of the future energy output of a wind farm, it is paramount to use sources of wind speed reference data which are consistent throughout the historical period considered. Inconsistent and/or short reference data sets can jeopardise and impact the expected financial returns of a project.

In the past, when undertaking energy assessments in the UK and Ireland, DNV GL would always look at a range of potential sources of long-term reference data, such as Reanalysis data sets and ground-based Meteorological Stations, or indeed indices composed of several of these sources. In recent studies, the informed adjustment from Meteorological Stations and Reanalysis data sets have shown an unsettling divergence, particularly in the northern part of Ireland and southwest Scotland.

Following these findings, DNV GL has been undertaking a thorough investigation into the consistency of the different data-sets. We have used as reference a series of long site wind data measurement campaigns, which have been proven consistent and are located across Ireland and the southwest of Scotland. Ten sites have been considered, with an average measurement period of 4.5 years.

In previous publications (see 1), DNV GL has demonstrated MERRA-2 to be in agreement with MERRA across Ireland and Great Britain, and therefore, both Reanalysis data sets have been included in this study. As for Meteorological Stations, DNV GL has been considering only four Meteorological Stations in Ireland, as consistent for long-term reference use (Aldergrove, Knock Airport, Casement Aerodrome and Cork Airport). These, along with three UK Met Office Meteorological Stations in the southwest coast of Scotland (Machrihanish, Islay Port Ellen and Prestwick Airport), have been looked at.

Investigations show that MERRA-2 exhibits better correlation quality with DNV GL’s sample of site measurements when compared to long-term ground-based sources and regional indices of stations, see Figure 1 1.

 

Figure 1 – Correlation quality for the considered sites in Ireland based on the coefficient of determination R2. MERRA and MERRA-2 values are calculated using the closest grid cell to each site.

In addition, we have also compared wind speed predictions using MERRA-2, ground-based Meteorological Stations and regional wind indices against the measured consistent site data sets. This was undertaken by dividing the site data and using only half of the data set to predict the wind speed for the whole period. Using this method, wind speed predictions have been compared to actual measured wind speed. The average wind speed prediction errors are presented in the following figure.

 Figure 2 – Average error in wind speed prediction for the sites analysed in this study using MERRA-2, ground-based Meteorological Stations and regional windiness indices.

As seen in Figure 2, Meteorological Stations present the bigger average error in the prediction of wind speed. This indicates that one or several of these stations are likely to have one or more inconsistencies throughout the period of measurements under consideration. The periods under consideration span across the last 15 years and are, therefore, crucial for any energy assessment long-term analysis.

For this reason, it is DNV GL’s recommendation that Reanalysis data sets, such as MERRA-2, are considered as primary sources of long-term data in Ireland and southwest Scotland, and that data from Irish Meteorological Stations is considered with extreme care.

Additionally, MERRA-2 Reanalysis data sets have the advantage of representing, by nature, a complete data set with full coverage and a longer measurement period than ground stations, which contributes to the decrease in long-term uncertainties for projects.

 

DNV GL in the UK and Ireland

DNV GL has analysed more than 200 GWs of projects globally throughout the last 30 years, with a team of over 100 wind and solar analysts. Due to its vast experience in the region and the continuous improvements of its global and regional methodologies, DNV GL can help you with your projects, from the measurement campaign to production. DNV GL’s recognised excellence and independence makes us the preferred consultancy of most financial institutions.

References

1 – Ebsworth, G. et al, “A review of the performance of MERRA-2: the next era of global long-term reference data”, DNV GL, Wind Europe Conference, Edinburgh, March 2017.

1 – In Figure 1, comparisons of quality of correlation are presented for three macro-regions. The results are shown in this fashion to better represent the locations of the selected measurements used in this study. The separation between the macro-regions is not to be intended as a precise separation.

1 Comments Add your comment
David says:

Thanks for this study – the results are certainly very interesting and it is a very relevant topic – accuracy of reanalysis data has been one of our focus areas over recent years. I’ve got a couple of questions related to what you’ve written above – sorry they go on a bit.
– you say: “In the past, when undertaking energy assessments in the UK and Ireland, DNV GL would always look at a range of potential sources of long-term reference data” does this mean that you are no longer looking at a range of sources?
– how does distance to the site impact on the correlations reported – the MERRA nodes are always going to be close to the site, but the distance to the nearest met station will vary significantly. I wonder whether the results above whilst representing an average could be somewhat skewed for met stations as they may be a long way from the test sites. For a nearby ground station (say within 50km) of the test site it would be good to know whether it correlates better than the nearby reanalyses.
– When testing the reanalysis you divided the data in half to predict the other half – again I’d be curious if 1-year was used to predict the other years – as this may be more representative of a lot of early assessments where limited data for correlation is available.
– Lastly – the decrease in uncertainty with long-term periods of >15 years, is this a true effect or just theoretical. Research historically has suggested that for long-term periods >10-years you may not see any improvement in uncertainty in reanalysis data. Have you been able to test this assumption?
Keep up the good work looking forward to reading any more similar studies

Reply with your comment

Your email address will not be published. Required fields are marked *