How to use reanalyses for representativity/quality estimation Andrea K. Kaiser-Weiss, Vera Heene, Michael Borsche, and Frank Kaspar
Monthly wind speed correlations ERA-20CEra-InterimCOSMO-REA6 Andrea Kaiser-WeissEMS & ECAC, Sofia
Monthly mean anomalies Wind speed anomaly [m/sec]
Added value of COSMO-REA6 Not nudged (independent stations) Nudged stations Andrea Kaiser-WeissEMS & ECAC, Sofia
Added value from COSMO-REA6 is shown: possibly a mix of benefit from higher resolution of the model, added value from data assimilation (nudging), and from temporal resolution of output Other regional reanalysis output (from UERRA) would be of interest Long term: only ERA-20C is available 5 Lessons learned? Andrea Kaiser-WeissEMS & ECAC, Sofia 2015
Outline 1. Station data and reanalyses 2. Comparing (surface winds, temperatures) -> frequency distributions, correlations, long-term changes -> Lessons learned 3. Summary: how to use reanalyses for quality and representativity estimation Andrea Kaiser-WeissEMS & ECAC, Sofia
The nicest example I found station Geissenheim 5yr smoothing No worries about the bias Andrea Kaiser-WeissEMS & ECAC, Sofia
A more typical example station Hamburg-Fuhlsbüttel worries about smoothed inhomogeneities Andrea Kaiser-WeissEMS & ECAC, Sofia
Another typical example 1991 change of instrument height above ground from 14 m to 16 m 1992 change to AWS Pearson corr ERA-20C ERA-Int HErZ NA NA NA NA Station Braunlage
Anecdotical evidence Station Marienberg Andrea Kaiser-WeissEMS & ECAC, Sofia
Monthly difference between stations (averaged from hourly measurements) and ERA20C, deseasonalized, filtered 5 years with gaussian filter, cutoff=2 Any trend / long-term change estimate would heavily depend on the subsample of stations used Inhomogeneities > long-term change Andrea Kaiser-WeissEMS & ECAC, Sofia
Lessons learned - wind In the recent years, the monthly means of station winds and reanalysis are highly correlated Hideous station inhomogenieties in the past (some documented in the meta-data) Station inhomogeneities are much larger than any „stilling effect“ over Germany Hundreds of wind stations would be needed to smooth historic inhomogeneities out - > stringent quality control or homogenization required before application Problem obscured by smaller time-scale variability, comparison with reanalysis can illuminate Maybe better consider using wind speed from reanalysis Andrea Kaiser-WeissEMS & ECAC, Sofia
Outline 1. Station data and reanalyses 2. Comparing (surface winds, temperatures) -> frequency distributions, correlations, long-term changes -> Lessons learned 3. Summary: how to use reanalyses for quality and representativity estimation Andrea Kaiser-WeissEMS & ECAC, Sofia
Biases are expected Biases from: -difference between local topography and model topography, smaller scale processes Andrea Kaiser-WeissEMS & ECAC, Sofia
Now a typical example for temperature No worries about the bias again station Geissenheim Andrea Kaiser-WeissEMS & ECAC, Sofia
De-seasonalized differences station Karlsruhe, now filter length = 2yrs Andrea Kaiser-WeissEMS & ECAC, Sofia
Lessons learned - temperature All hourly time series from stations over Germany (aggregated to monthly) are highly correlated with ERA- 20C. Some stations (e.g., Geissenheim, Hamburg- Fuhlsbüttel) match nearly perfectly the ERA-20C surface temperature trend. Possible exception: ERA-20C temperatures. Next: To include: more daily, monthly station data Topographic bias to be resolved by post-processing boring ? re-assuring less boring ? Andrea Kaiser-WeissEMS & ECAC, Sofia
Outline 1. Station data and reanalyses 2. Comparing (surface winds, temperatures) -> frequency distributions, correlations, long-term changes -> Lessons learned 3. Summary: how to use reanalyses for representativity/quality estimation Andrea Kaiser-WeissEMS & ECAC, Sofia
Summary (I) 1. For DWD station data, reanalyses were good enough to identify larger scale representative stations and suspicious stations (instrument changes, local changes, bugs) 2. Over Germany, all used reanalysis had better long-term stability than probably all wind station series, and long-term stability of temperature was comparable to station series, with the possible exception of (where there might be some reanalysis fault). Andrea Kaiser-WeissEMS & ECAC, Sofia
Lessons learned for >> AREA of GERMANY 10m winds: rather go for any reanalysis COMSO- REA6 (HErZ) with 6km resolution gives better correlations with stations than ERA-Interim. For temperature, with smoothing > 5 yrs you are pretty safe to use even the coarsest reanalysis (ERA-20C) for homogeneity tests. Take special care with deriving trends (see KNMI work). CORE-CLIMAX ECV Capability Workshop, January 2014, Darmstadt 20
Summary (II) & What’s Next 3. Regional reanalysis COSMO-REA6 showed added value for wind speed. Kaiser-Weiss et al.: Comparison of reanalysis near-surface winds with station observations, Adv. Sci Res. 1, 1-12, Next: * Move to higher spatial resolution / ensembles -> UERRA regional reanalyses, post-processing * Understand and attribute differences between station measurements at smaller timescales How far can we push resolution? Andrea Kaiser-WeissEMS & ECAC, Sofia
Frequency distribution quality control ERA-20C grid cell matching station location Daily wind speed for decades , and more than 10% calm days. Windmessanlage Junkalor (not heated) After 1990 instrument change to WMG 201 (heated) Effect found at Stations: Meiningen, Genthin, Obersdorf, Garmisch-Partenkirchen