Abstract
This document reports on the methods used to create and the results of the WRF-based numerical wind atlases developed for the Wind Atlas for South Africa Phase 3 (WASA3) project.
The report is divided into four main parts. In the first part, we document the method used to run the mesoscale simulations and to select the best suited WRF model configuration using the measurements from the WASA masts. In the second part, we describe the method used to generalise and downscale the WRF model wind climate. We compare the results from the downscaled numerical wind atlas against the observed wind statistics from the 19 WASA masts in the third part. In the last part, we present the new wind resource maps and their long-term climatology.
In WASA3, there have been many updates to the configuration of the 2018 WASA2 simulations documented in
Hahmann et al (2018). Among the most important:
• We ran thirteen two-year simulations covering the period most observed in all the WASA sites to find the WRF model configuration most suited to the simulation of the wind climatology over South Africa.
• We used a new method of generalisation and downscaling of the WRF-derived wind climate that uses the PyWAsP engine and was demonstrated more accurate than the previous approach.
• We produced the most extensive to date wind climatology for South Africa, 30 years (1990–2019) simulation covering all South Africa at 3.33 km x 3.33 km spatial resolution and 30 minutes time output.
The final error statistics of the WASA3 wind atlas show that the WRF+PyWAsP method has a MAPE of 14.2 % and 4.3 % for the long-term power density and wind speed, respectively. This is improved from the same validation in WASA2. When ignoring the two more complex masts, WM09 and WM11, the WRF and WRF+PyWAsP downscaling significantly narrows the error distributions for both long-term wind speed and power density.
The report is divided into four main parts. In the first part, we document the method used to run the mesoscale simulations and to select the best suited WRF model configuration using the measurements from the WASA masts. In the second part, we describe the method used to generalise and downscale the WRF model wind climate. We compare the results from the downscaled numerical wind atlas against the observed wind statistics from the 19 WASA masts in the third part. In the last part, we present the new wind resource maps and their long-term climatology.
In WASA3, there have been many updates to the configuration of the 2018 WASA2 simulations documented in
Hahmann et al (2018). Among the most important:
• We ran thirteen two-year simulations covering the period most observed in all the WASA sites to find the WRF model configuration most suited to the simulation of the wind climatology over South Africa.
• We used a new method of generalisation and downscaling of the WRF-derived wind climate that uses the PyWAsP engine and was demonstrated more accurate than the previous approach.
• We produced the most extensive to date wind climatology for South Africa, 30 years (1990–2019) simulation covering all South Africa at 3.33 km x 3.33 km spatial resolution and 30 minutes time output.
The final error statistics of the WASA3 wind atlas show that the WRF+PyWAsP method has a MAPE of 14.2 % and 4.3 % for the long-term power density and wind speed, respectively. This is improved from the same validation in WASA2. When ignoring the two more complex masts, WM09 and WM11, the WRF and WRF+PyWAsP downscaling significantly narrows the error distributions for both long-term wind speed and power density.
| Original language | English |
|---|
| Place of Publication | Risø, Roskilde, Denmark |
|---|---|
| Publisher | DTU Wind Energy |
| Number of pages | 101 |
| ISBN (Electronic) | 978-87-93549-86-9 |
| Publication status | Published - 2021 |
| Series | DTU Wind Energy E |
|---|---|
| Number | E-0218 |