Presentation on theme: "Matthew Hendrickson, and Pascal Storck"— Presentation transcript:
1 Matthew Hendrickson, and Pascal Storck Downscaling Global Reanalyses with WRF for Wind Energy Resource AssessmentMark Stoelinga,Matthew Hendrickson, and Pascal Storck3TIER, Inc.
2 Wind Resource Assessment What is the long-term average wind resource at each turbine location within a proposed wind farm?
3 Wind Resource Assessment Install “met towers” for a period of ≥ 1 year.
4 Wind Resource Assessment Need to extend the observed information, bothspatially (around proposed windfarm) andtemporally (to estimate long-term mean from 1 year of measurements)
5 Estimating Temporal Variability of Wind Resource How can we extend the short (1-year) record into a long-term mean?Conventional approachIdentify a nearby, long-term, routine 10m wind observation (“reference station”) that correlates well with the 1-year tower measurement. Use linear regression to craft a relationship between reference site and tower, and then predict long-term mean at tower -> MCP
6 Estimating Temporal Variability of Wind Resource First-Generation Reanalysis Data Sets(NCAR/NCEP “R1”, ERA-40): Can potentially provide a “synthetic long-term reference station”, but with potential pitfallsCoarse resolution of underlying model ( deg)Flaws/limitations in DA methodChanges in observations over 50 yearsGrids available only every 6 h (hourly is preferred)
7 Estimating Temporal Variability of Wind Resource Downscaling of Reanalysis Data Sets with a Mesoscale ModelFoundation: a mesoscale model can produce good climatology of local surface wind if provided with appropriate large-scale flow conditions.Model can “fill in” at hourly frequencyModel can also provide multiple predictors to inform a statistical relationship between observations and the synthetic long-term reference (e.g., MOS)
8 2nd-Generation Reanalyses (CFSR, ERA-Interim, MERRA) 33-year record, entirely during satellite erahigh-resolution (~0.5 degrees)modern DA methodologies (4DVAR, or much better 3DVAR)Direct assimilation of satellite radiances
9 2nd-Generation Reanalyses (CFSR, ERA-Interim, MERRA) Questions:Do these new reanalysis data sets result in more accurate downscaled retrospective simulations?Are the reanalyses so good that we don’t need to downscale?Will look at:global mapsvalidation of regional runs at individual met towers
10 Global 80-m long-term mean wind maps NCAR/NCEP “R1” ReanalysisR1 w/ WRF downscaling3TIER “FirstLook” data setCompleted 2008, 5-km / 10-year global land coverage, WRF 2.2, YSU PBL, simple land surfaceCFSRERA-InterimMERRA
19 Regional Runs at Tower Sites 4.5-km WRF runs, V3.0PBL: YSU or MYJ; LSM: simple or Noah; grid nudging3-day runs strung together continuously for multiple years16949112
20 Regional Runs at Tower Sites Towers provide hourly data for periods ranging from 1 – 8 years.Wind speed error metrics R2 and MAE were calculated for WRF time series at the tower sites at hourly, daily, monthly, and yearly time scales
21 Wind Speed R2 for downscaled CFSR vs. NCAR/NCEP “R1” DailyMonthlyCFSR R2CFSR R2N. AmerS.AmerEuropeAfricaIndiaAustr.R1 R2R1 R2
22 Wind Speed R2 for downscaled ERA-Int vs. NCAR/NCEP “R1” DailyMonthlyERA-Int R2ERA-Int R2N. AmerS.AmerEuropeAfricaIndiaAustr.R1 R2R1 R2
23 Wind Speed R2 for downscaled CFSR vs. raw CFSR DailyMonthlyDownscaled CFSR R2Downscaled CFSR R2N. AmerS.AmerEuropeAfricaIndiaAustr.Raw CFSR R2Raw CFSR R2
24 ConclusionsSeveral new 33+ -year reanalysis data sets with ~0.5° resolution have recently become available for general useNew reanalyses show improved performance when used to drive downscaled WRF retrospective simulations for wind energy assessmentAlthough resolution and DA have been improved compared to 1st-generation reanalyses, considerable value is still added with WRF downscaling
25 Caveats about new reanalyses ERA-Interim and MERRA lag real time by a few monthsMostly “WRF-ready”, though MERRA requires some work (HDF4 file format)Freely availableCFSR not consistently produced after Jan 2011