Page 1 Determining changes in extratropical storm conditions - homogeneity and representativity Hans von Storch Institute of Coastal Research, GKSS Research Center, Geesthacht and CLISAP KlimaCampus, Hamburg University, Germany

Page 2 Message a)The major problem for statisticians applying their (ancient or modern) tools is the sampling assumption. Most data (both observational or derived = analyzed) are inhomogeneous. b)Simulation data are usually “better” – namely homogenous are available for long even very long times. But the degree of realism is always an issue. Advice: 1) When dealing with “real” data have contact to meteorologists, oceanographers etc. 2) When dealing with simulated data, have contact to modellers. 3) When testing new method, begin with simulation data.

Page 3 Challenge Storminess best represented by wind statistics, possibly derived quantities such as stream function, vorticity, but wind time series are almost always inhomogeneous too short

Page 4 10-yearly sum of events with winds stronger than 7 Bft in Hamburg

Page 5 Wind speed measurements  SYNOP Measuring net (DWD)  Coastal stations at the German Bight  Observation period: Representativity of near surface wind speed measurements This and the next 3 transparencies: Janna Lindenberg, GKSS

Page 6 Representativity of near surface wind speed measurements 1.25 m/s

Page 7 Causes of inhomogenities: Changes in  Instruments  Sampling frequencies  Measuring units  Environments (e.g. trees, buildings)  Location Station relocations (Dotted lines) Representativity of near surface wind speed measurements

Page 8 Representativity of near surface wind speed measurements

Page 9 Difference of max storm surge heights Hamburg - Cuxhaven Storm surges intensified after 1962 event due to improving coastal defense and deepening shipping channel in the river.

Time series of frequency of stormy days in Kullaberg (south- western Sweden), number of days per year with wind speed V≥21 m/s.

Page 12 Counting storms in weather maps – steady increase of NE Atlantic storms since the 1930s ….

Page 13 “Great Miami”, 1926, Florida, Alamaba – damages of 2005 usage - in 2005 money: 139 b$ Katrina, 2005: 81 b$ Pielke, Jr., R.A., Gratz, J., Landsea, C.W., Collins, D., Saunders, M., and Musulin, R., Normalized Hurricane Damages in the United States: Natural Hazards Review The increase in damages related to extreme weather conditions is massive – but is it because the weather is getting worse? Losses from Atlantic Hurricanes

Page 14 T The major problem for applying statistical analysis (ancient and modern) is related - to the assumptions on the sampling process, and - to the assumption of representativity The local data change their statistics by changing observational practices and conditions; also their representativity for a larger area is often compromised.

Page 15 For assessing changing storm conditions, principally two approaches are possible: a)Use of proxies, such a air pressure readings. b)Simulation by empirical or dynamical downscaling of large scale information. Usage of weather analyses, incl. re-analyses and proxies such as damages are not suitable.

Page 16 Pressure proxies

Page 17 Pressure based proxies Air pressure readings are mostly homogenous Annual/seasonal percentiles of geostrophic wind derived from triangles of pressure readings (e.g., 95 or 99%iles); such percentiles of geostrophic wind and of “real” wind are linearly related. Annual frequency of events with geostrophic wind equal or larger than 25 m/s Annual frequency of 24 hourly local pressure change of 16 hPa in a year Annual frequency of pressure readings less than 980 hPa in a year

Page 18 Geostrophic Wind The tropospheric wind is to first order approximation proportional to atmospheric pressure gradient The scaled pressure gradient is called “ geostrophic wind”. The geostrophic wind flows parallel to isobars The geostrophic wind is a proxy for real wind. At the surface, the geostrophic wind is stronger than the real wind. Oliver Krüger

Page 19 Geostrophic Wind - why is it useful? good approximation for synoptic scale airflow surface air pressure measurements can be used to calculate the geostrophic wind Pressure readings are usually homogenous Geostrophic winds may be derived for a long period.

Page 20 An example is given by Schmidt and von Storch, They calculated daily geostrophic wind speed from surface air pressure measurements at three synoptic stations around the German Bight. Annual distributions were obtained for the period of Upper percentiles of wind speed were derived and used as a proxy for storm activity over the German Bight. Upper percentiles are the values of wind speed above which a certain percent of observations fall. Assessing these upper percentiles gives information about changes in the marine storm climate.

Page 21 Other storm proxies Variance of local water levels relative to annual mean (high tide) water level. Repair costs of dikes in historical times. Sailing times of ships on historical routes.

Page 22 Proxies: N Europe

Page 23 Geostropic wind stats N Europe 99%iles of annual geostrophic wind speeds for a series of station triangles in the North Sea regions and in the Baltic Sea region. Alexandersson et al., 2002

Page 24 Local pressure stats since 1800 Time series of pressure-based storminess indices derived from pressure readings in Lund (blue) and Stockholm (red). From top to bottom: Annual number of pressure observations below 980 hPa (N p980 ), annual number of absolute pressure differences exceeding 16 hPa/12 h (N Dp/Dt ), Intra-annual 95-percentile and 99-percentile of the pressure differences (P 95 and P 99 ) in units of hPa. From Bärring and von Storch, 2005: see also BACC Stockholm Lund

Page 25 Regional development of storminess and temperature Unchanging extratropcial storm conditions is not contradicting the fact that temperature is rising.

Page 26 Dynamical Dowscaling N Europe Downscaling NCEP/NCAR large-scale analysis of weather

Page 27 Downscaling cascade Globale development (NCEP) Dynamical Downscaling REMO or CLM Simulation with barotropic model of North Sea Empirical Downscaling Pegel St. Pauli Cooperation with a variety of governmental agencies and with a number of private companies Mixed dynamical/empirical. downscaling cascade for constructing variable regional and local marine weather statistics

Page 28 Baroclinic storms in N Europe Problem solved for synoptic systems in N Europe in using RCM spectrally nudged to NCEP - retrospective analysis good skill with respect to statistics, but not all details are recovered. Weisse, R., H. von Storch and F. Feser, 2005: Northeast Atlantic and North Sea storminess as simulated by a regional climate model and comparison with observations. J. Climate 18,

Page 29 Stormcount Weisse et al., J. Climate, 2005 Change of # Bft 8/year t ≤ T t ≥ T

Page 30 Oct Polar Lows

c=0,72 Comparison with satellite data Count of Polar Lows per month. – downscaling - satellite data (Blechschmidt, 2008)

Page 32 Downscaling re-analysis

Page 34 Conclusion: Usage of proxies 1.Monitoring extra-tropical storminess may be based on air pressure proxies. 2.This allows assessments for 100 and more years. 3.Decades long upward and downwards trends have been detected in recent years. 4.These trends are not sustained and have show recent reversals in all considered regions. 5.Recent trends are not beyond the range of natural variations, as given by the historical past, but are more of intermittent character. Regional temperatures rose significantly at the same time.

Page 35 Conclusion: Usage of dynamical downscaling 1.Dynamical downscaling for describing synoptic and mesoscale variability is doable. 2.Meso-scale Polar Lows are also described, but the depth of the cyclones is not as deep as found in reality; also the winds are too weak. 3.Validation difficult, because homogeneous long term observed data do not exist. 4.Analysis of 60 year simulations point to strong year-to-year variability, to less decade-to-decade variability and no noteworthy trend.

Page 36 Message a)The major problem for statisticians applying their (ancient or modern) tools is the sampling assumption. Most data (both observational or derived = analyzed) are inhomogeneous. b)Simulation data are usually “better” – namely homogenous are available for long even very long times. But the degree of realism is always an issue. Advice: 1) When dealing with “real” data have contact to meteorologists, oceanographers etc. 2) When dealing with simulated data, have contact to modellers. 3) When testing new method, begin with simulation data.