How a state-of-the-art w ind atlas is m ade: The exam ple of the W ind Atlas for South Africa Andrea N. Hahmann (ahah@dtu.dk), Jake Badger, Patrick Volker, Jens Carsten Hansen, Niels G. Mortensen Department of Wind Energy, Technical University of Denmark, Risø Campus, Roskilde, Denmark Chris Lennard and Brendan Argent Climate System Analysis Group University of Cape Town, Cape Town, South Africa Final WASA Workshop Windaba, Cape Town 2 November 2016
Outline Mesoscale modeling within the WASA project Why do we need downscaling? How was the downscaling done for the WASA project • Generalization procedure • Validation and comparisons of wind climate Validation of seasonal and diurnal cycles Microscale downscaling and validation Available products Conclusions 03/ 11/ 2016 2
W hat is a w ind atlas? A wind atlas is much more than a simple map containing mean wind speed (or kinetic energy flux) for a region of the Earth
The atm osphere is in constant m otion… Shown is the wind speed at 10 m AGL, January 1998 snapshots every 6 hours The w ind clim atology is a summary of all these motions Andrea Hahmann, 12 Nov 2015
Num erical W ind Atlas m ethodology Downscaling from global reanalysis data + verification Mesoscale m odelling Microscale m odelling Global wind data Local surface wind Regional wind climate Global wind resources 2 0 0 km × 2 0 0 km 3 km × 3 km 1 – 1 0 m Verification Local surface wind Microscale m odelling Measurements
W hat kind of processes does the m esoscale m odel sim ulates? valley circulations coastal jet upslope and download winds sea breezes Andrea Hahmann, 12 Nov 2015
W RF Sim ulations for the W ASA project Simulating 8 years of the wind (on a 3 km x 3 km grid) of South Africa took ~ 6 weeks in our cluster Andrea Hahmann, 12 Nov 2015
Once w e run the m odel for a ( pretty) long period of tim e… W e get A very big collection of numbers How ever … Andrea Hahmann, 12 Nov 2015
I m portance of resolution Wind resource (power density) calculated at different resolutions meso only meso + micro 2.5 km 0.1 km 505 W/m 2 323 W/m 2 641 W/m 2 378 W/m 2 average over the light blue to green bits (50% percentile) of the image. Badger et al (2011) mean power density of total area mean power density for windiest 50% of area Andrea Hahmann, 12 Nov 2015
W hy do w e need generalization? GENERALI ZATI ON Nature Mesoscale Model W AsP “lib” files Andrea Hahmann, 12 Nov 2015
From m esoscale m odel to site conditions direct mesoscale model output site conditions micro corrections only Microscale modelling meso & micro corrections Microscale Mesoscale generalisation modelling Numerical Wind Atlas Badger, J., H. Frank, A. N. Hahmann and G. Giebel, 2014: Wind climate estimation based on mesoscale and microscale modeling: statistical-dynamical downscaling for wind energy applications. J. Applied Meteorology and Climatology, 5 3 , 1901-1919.
W RF-based sim ulations Steps towards the new research-based new numerical wind atlas Determine optimal model configuration (some learned from previous wind atlases), others are new to WASA project Run simulations (18 days on a almost fully dedicated cluster; a total of 293 runs; each 6 hours, on 8 nodes) Data processing – output from simulations are 8Tb! Generalization and validation Generation of data products 03/ 11/ 2016 16
Weather, Research and Forecast (WRF) model Complex model with m any options that need to chosen by the user Best configuration not found by chance: Extensive set of year-long simulations were performed to optimize domain size and location and various parameterizations. Mesoscale & Microscale Meteorology Division / NCAR 11/ 3/ 2016 17
Sensitivity Experim ents One year-long (Oct 2010 – Sep 2011) simulations (5 km x 5 km grid) Compare mean annual wind speed (m/ s) at 100 meters Forcing reanalysis Boundary layer scheme Radiative param | dU| < 0.5 m/ s Land surface model Land use class Convective param 03/ 11/ 2016 18
Results from the various sensitivity experim ents 5 km x 5 km grid spacing Error= (U model -U obs )/ U obs , U= year-long mean generalized wind speed 20.0% 15.0% 10.0% ERA 5.0% CFSR ERA YSU 0.0% ERA ULCC WM02 WM04 WM06 WM08 WM10 ERA RRTMG -5.0% WM01 WM03 WM05 WM07 WM09 MAE ERA YSU RRTMG ERA PLX (var Z0) -10.0% -15.0% -20.0% -25.0% Error reduction by using high-resolution 03/ 11/ 2016 19 WASA Final Wind Seminar
New research w ind atlas: W RF Model Configuration Very large (309 x 435) inner grid (3km x 3 km grid spacing) Changes to standard WRF land Simulations: use and roughness 8 years for (27/ 9/ 3 km) – 2005-2013; High- resolution SSTs; ERA-Interim forcing, 1/ 12 24 years (27/ 9 km) – 1990-2013; degree SSTs; MYJ PBL; 41 vertical levels (further details in incoming report) 03/ 11/ 2016 20
Validation after generalization GENERALI ZATI ON Nature Mesoscale Model W AsP “lib” files 22
Mesoscale generalization procedure Similar generalization procedure for KAMM and WRF simulations. Wind speeds and directions from WRF simulations are binned according to wind direction, wind speed, and stability (1/ L). Each binned wind class is then generalized and aggregated using their frequency of occurrence Neutral or non-neutral assumption was tested Term modified to account for non- neutral conditions. Hahmann, A. N., Pena Diaz, A., & Hansen, J. C. (2016). WRF Mesoscale Pre-Run for the Wind Atlas of 03/ 11/ 2016 23 Mexico. DTU Wind Energy. (DTU Wind Energy E, Vol. E-0126).
Num erical w ind atlas – W RF 3 km sim ulation Generalized wind speed, h= 100 m, z0= 0.03 m 03/ 11/ 2016 26
Microscale m odelling at the 1 0 W ASA m asts Som e background Wind-climatological inputs Three-years-worth of wind data Five levels of anemometry Topographical inputs Elevation maps (SRTM 3 data) Simple land cover maps (SWBD + Google Earth); water + land Preliminary results Microscale modelling verification • Site and station inspection • Simple land cover classification • Adapted heat flux values Wind atlas data sets from 10 sites Analysis show prevalence This data was used to verify the numerical of non-neutral conditions at wind atlas, but not to create them the sites.
Verification of the w ind atlas by m easurem ents So we can compare the numerical wind atlas GWC that is closest to each mast with the GWC derived from the mast data Please note: • Both sets of GWCs must have the same attributes i.e. • Same height a.g.l. • Flat terrain • Uniform roughness (The NWA data was also adjusted so that it was NWA GWC WM10 GWC representative over the same period the met mast measurements were taken.) SDC Cour
The verified num erical w ind atlas A state-of-the-art wind atlas is verified by measurements The Wind Atlas project is designed from the beginning to include high quality measurements against which the numerical wind atlas could be checked This produces a “Verified Numerical Wind Atlas” So, alongside the mesoscale modelling, the project has a second, parallel, activity: Numerical wind VERIFICATION Mesoscale Generalised climatological wind modelling climates @ grid points atlas High quality Microscale Generalised wind climates @ modelling mast locations measurements
Com parison at specific sites WM01 Observed versus numerical wind atlas at 3 sites h= 100 meters, z0= 0.03 m October 2010-September 2013 03/ 11/ 2016 30
Exam ple: W ASA site 1 , far northw est Observed wind atlas Weighted (solid) Re-fit (dashed) Numerical wind atlas WRF 03/ 11/ 2016 31
Com parison at specific sites Observed versus numerical wind atlas at 3 sites h= 100 meters, z0= 0.03 m WM05 October 2010-September 2013 03/ 11/ 2016 32
Exam ple: W M0 5 , southern coast Observed wind atlas Numerical wind atlas WRF 03/ 11/ 2016 33
Com parison at specific sites WM10 Observed versus numerical wind atlas at 3 sites h= 100 meters, z0= 0.03 m 03/ 11/ 2016 34
Exam ple: W M1 0 , Eastern cape Observed wind atlas Numerical wind atlas WRF 03/ 11/ 2016 35
Observational W ind Atlas Wind speed at 80 m above ground level WAsP resource grids from Observational Wind Atlas •10 x 10 km 2 grid •100 meter grid spacing
Andrea Hahmann, 12 Nov 2015 W ASA w ind resource @ 1 0 0 m – w ind speed
W ind Atlas for South Africa – verification of num erical w ind atlas Modelling versus m easurem ents @ 6 2 m wind speed yield Wind speed Energy yield • Slope: 102% • Slope: 105% • Spread: 5.9% • Spread: 12% 38
Seasonal and diurnal cycles in the observations and the W RF sim ulations Chris Lennard and Brendan Argent 03/ 11/ 2016 Univ. Cape Town 39
Recommend
More recommend
