Metrics used to measure climate extremes

Metrics used to measure climate extremes A cautionary note: Caveats and uncertainties Sebastian Sippel 1 Milan Flach 2 1 Norwegian Institute of Bioeconomy Research, Ås, Norway 2 Max Planck Institute for Biogeochemistry, Jena, Germany 04.10.2017 S. Sippel Metrics used to measure climate extremes 04.10.2017 1 / 15

Outline 1. Introduction: Climate extremes and impacts 2. Statistical quantification of extremes: Precipitation and temperature 3. Developing benchmarking datasets for uni-/multivariate extreme detection metrics 4. Conclusions & Outlook S. Sippel Metrics used to measure climate extremes 04.10.2017 2 / 15

1. Introduction: Climate extremes and impacts S. Sippel Metrics used to measure climate extremes 04.10.2017 2 / 15

Points of departure 1. Several types of weather & climate extremes are changing in intensity and frequency S. Sippel Metrics used to measure climate extremes 04.10.2017 3 / 15

Points of departure 1. Several types of weather & climate extremes are changing in intensity and frequency 2. Climate extremes propagate to impacts, but in complex ways: Example: Forest and peat fires of Russian heat wave (Aug 5, 2010) Carbon losses through (indirect) wild fire effects (-256 Tg CO2, Yoshida et al. 2017, Env. Pol.) were about three times as large as direct carbon losses through reduced photosynthesis (-90 Tg CO2, Bastos et al. 2014, Biogeosciences). S. Sippel Metrics used to measure climate extremes 04.10.2017 3 / 15

Points of departure 1. Several types of weather & climate extremes are changing in intensity and frequency 2. Climate extremes propagate to impacts, but in complex ways: 3. Data (Impact & Climate), methodologies, and dialogue crucially needed to link climate extremes and impacts: S. Sippel Metrics used to measure climate extremes 04.10.2017 3 / 15

Approaches to deal with extremes: Image source: NOAA S. Sippel Metrics used to measure climate extremes 04.10.2017 4 / 15

2. Statistical quantification of extremes: Precipitation and temperature P(env novel ) P(env ref ) Building on a reference of normality S. Sippel Metrics used to measure climate extremes 04.10.2017 4 / 15

2. Statistical quantification of extremes: Precipitation and temperature P(env novel ) P(env ref ) NOAA Building on a reference of normality Spatial aggregation to improve S/N ratio S. Sippel Metrics used to measure climate extremes 04.10.2017 4 / 15

NGE IN TEMPERATURE VARIABILITY? Temperature and precipitation extremes Temperature extremes: Precipitation extremes: Z i = X i X ref s(x ref ) Donat et al. 2016, Nat. Clim. Change Hansen et al. 2012, PNAS ansen et al. (2012) PNAS Area covered by temperature anomalies in the Precipitation normalization: ategories Temperature definednormalization: as hot ( > 0.43 ), very hot (> 2 ), and extremely hot > 3 ), with analogous divisions for cold anomalies. Anomalies are relative o 1951-1980 base period, z = X with X ref also from 1951-1980 data. Lowest row is s(x ref ). z = X X ref outhern Hemisphere summer. S. Sippel Metrics used to measure climate extremes 04.10.2017 5 / 15

Increasing dry region precipitation extremes? LETTERS NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE2 a 1.30 1.20 Observations dprcptot = 0.063 Slope = 0.020 p < 0.001 Wet and dry 30% of HadEX2 grid cells Dry b CMIP5 Observations 1.30 dprcptot = 0.037 dprcptot = 0.008 Slope = 0.013 Slope = 0.003 p < 0.001 1.20 p = 0.064 Wet CMIP5 dprcptot = 0.010 Slope = 0.004 p < 0.001 PRCPTOT 1.10 1.00 1.10 1.00 0.90 0.90 1.12 1.08 drx1day = 0.048 Slope = 0.016 p < 0.001 drx1day = 0.031 Slope = 0.011 p < 0.001 1.12 1.08 drx1day = 0.021 Slope = 0.007 p = 0.001 drx1day = 0.028 Slope = 0.010 p < 0.001 Rx1day 1.04 1.00 0.96 1.04 1.00 0.96 0.92 1960 1980 2000 1960 Year 0.92 1980 2000 1960 1980 2000 1960 Year 1980 2000 c Extreme daily precipitation averaged over both dry and wet regimesprcptot shows robust increases in both observations Rx1day and climate models over the past six decades (Donat et al., 2016, Nature Climate Change) Conventional Precipitation Normalisation: z = X X ref Dry/wet (n = 299) Area fraction = 0.552 Sippel et al., 2017, HESS Dry/wet (n = 132) Area fraction = 0.264 S. Sippel Figure 1 Precipitation Metrics changes used to in the measure dry and climate wet regions extremes identified from observations. a,b, Time series of PRCPTOT (annual 04.10.2017 precipitation totals) 6 / 15and

Artificial example: simulate spatio-temporal dataset of extremes (X GEV ): 60 artificial years simulated time series are stationary S. Sippel Metrics used to measure climate extremes 04.10.2017 7 / 15

Artificial example: simulate spatio-temporal dataset of extremes (X GEV ): 60 artificial years simulated time series are stationary (a) 0.6 0.5 Original PDF Ref. period PDF Out of base PDF (b) 1.06 1.04 Original time series Ref. period time series Normalized time series Density 0.4 0.3 0.2 Mean of GEV 1.02 1.00 0.1 0.98 0.0 0.96 1 0 1 2 3 4 0 10 20 30 40 50 60 Time step Conventional Precipitation Normalisation: z = X X ref Sippel et al., 2017, HESS S. Sippel Metrics used to measure climate extremes 04.10.2017 7 / 15

Increasing dry region precipitation extremes? (a) 1.15 Rx1Day, Dry Regions 1.10 1.05 1.00 0.95 Ref. period 1951 1980 Ref. period 1981 2010 Ref. period 1951 2010 0.90 1950 1960 1970 1980 1990 2000 2010 Extreme daily precipitation averaged over both dry and wet regimes shows robust increases in both observations and climate models over the past six decades (Donat et al., 2016, Nature Climate Change) Year Conventional Precipitation Normalisation: z = X X ref Sippel et al., 2017, HESS S. Sippel Metrics used to measure climate extremes 04.10.2017 8 / 15

Increasing dry region precipitation extremes? (a) 1.15 Rx1Day, Dry Regions 1.10 1.05 1.00 0.95 Ref. period 1951 1980 Ref. period 1981 2010 Ref. period 1951 2010 0.90 1950 1960 1970 1980 1990 2000 2010 Extreme daily precipitation averaged over both dry and wet regimes shows robust increases in both observations and climate models over the past six decades (Donat et al., 2016, Nature Climate Change) Year Conventional Precipitation Normalisation: Here: Slopes reduced by 36 40%. z = X = bias σ 2. X ref µ 2 n ref Sippel et al., 2017, HESS S. Sippel Metrics used to measure climate extremes 04.10.2017 9 / 15

What about temperature extremes? Z i = X i X ref s(x ref ) sen et theal. emergence (2012) of [...] PNAS summertime Area extremely covered hot outliers, by temperature anomalies in the more than three standard deviations (3σ) warmer than [...] the gories 1951 1980 defined base period as hot [...] now ( > typically 0.43 covers ), very 10% of the hot (> 2 ), and extremely hot land area. ), with analogous divisions for cold anomalies. Anomalies are relative 951-1980 basehansen period, et al., 2012, with PNASalso 109(37), from E2415 2423. 1951-1980 data. Lowest row is thern Hemisphere summer. S. Sippel Metrics used to measure climate extremes 04.10.2017 10 / 15

What about temperature extremes? Z i = X i X ref s(x ref ) Conventional Normalisation: z = X X ref s(x ref ) sen et theal. emergence (2012) of [...] PNAS summertime Area extremely covered hot outliers, by temperature anomalies in the more than three standard deviations (3σ) warmer than [...] the gories 1951 1980 defined base period as hot [...] now ( > typically 0.43 covers ), very 10% of the hot (> 2 ), and extremely hot land area. ), with analogous divisions for cold anomalies. Anomalies are relative 951-1980 basehansen period, et al., 2012, with PNASalso 109(37), from E2415 2423. 1951-1980 data. Lowest row is thern Hemisphere summer. S. Sippel Metrics used to measure climate extremes 04.10.2017 10 / 15

The normalisation issue: Conceptual scenario simulate for many grid cells (n = 10.000) each 60 random Gaussian variables (i.e. 60 years ) a. count 2-sigma extremes across all grid cells for each time step a 2 sigma extremes 350 300 250 200 150 100 0 10 20 30 40 50 60 time steps Sippel et al., 2015, Geophys. Res. Lett. 42(22), 9990 9998 (Chapter 2). S. Sippel Metrics used to measure climate extremes 04.10.2017 11 / 15

The normalisation issue: Conceptual scenario simulate for many grid cells (n = 10.000) each 60 random Gaussian variables (i.e. 60 years ) a. count 2-sigma extremes across all grid cells for each time step b. normalise with X 1951 1980 and s(x 1951 1980), and count 2-sigma extremes across all grid cells for each time step a 2 sigma extremes 350 300 250 200 150 100 Gaussian, i.i.d. variables Normalized extremes 0 10 20 30 40 50 60 time steps Sippel et al., 2015, Geophys. Res. Lett. 42(22), 9990 9998 (Chapter 2). S. Sippel Metrics used to measure climate extremes 04.10.2017 11 / 15

The normalisation issue: Conceptual scenario simulate for many grid cells (n = 10.000) each 60 random Gaussian variables (i.e. 60 years ) a. count 2-sigma extremes across all grid cells for each time step b. normalise with X 1951 1980 and s(x 1951 1980), and count 2-sigma extremes across all grid cells for each time step a 2 sigma extremes 350 300 250 200 150 100 Gaussian, i.i.d. variables Normalized extremes 0 10 20 30 40 50 60 time steps 48.2% overestimation of 2-sigma extremes Sippel et al., 2015, Geophys. Res. Lett. 42(22), 9990 9998 (Chapter 2). S. Sippel Metrics used to measure climate extremes 04.10.2017 11 / 15

3. Developing benchmarking datasets for uni-/multivariate extreme detection metrics S. Sippel Metrics used to measure climate extremes 04.10.2017 11 / 15

Generation of Artificial data farm, 10 observables from 3 intrinsic dimensions: Evaluation of extreme detection algorithms and feature extraction methods: Flach et al., 2017, Earth Syst. Dyn. 8, 677 696. S. Sippel Metrics used to measure climate extremes 04.10.2017 12 / 15

Workflow for evaluation experiments: Flach et al., 2017, Earth Syst. Dyn. 8, 677 696. S. Sippel Metrics used to measure climate extremes 04.10.2017 13 / 15

Results of algorithm evaluation experiment: Data processing (feature extraction) methods are critical for event detection Flach et al., 2017, Earth Syst. Dyn. 8, 677 696. S. Sippel Metrics used to measure climate extremes 04.10.2017 14 / 15

Conclusion: Quantification of extremes Conventional reference period standardisation approaches are systematically biased. This includes any data processing based on reference period statistics! An analytical understanding and correction has been developed Temperature extremes a 2 sigma extremes 500 450 400 350 300 250 200 150 Gaussian, i.i.d. variables Normalized extremes Correction, out of base extremes Correction, ref. period extremes 0 10 20 30 40 50 60 time steps Hansen et al., 2012, PNAS 109(37), E2415 2423. Sippel et al., 2015, Geophys. Res. Lett. 42(22), 9990 9998. S. Sippel Metrics used to measure climate extremes 04.10.2017 15 / 15

Conclusion: Quantification of extremes Conventional reference period standardisation approaches are systematically biased. This includes any data processing based on reference period statistics! An analytical understanding and correction has been developed Temperature extremes Precipitation extremes (a) 1.15 Rx1Day, Dry Regions 1.10 1.05 1.00 0.95 Ref. period 1951 1980 Ref. period 1981 2010 Ref. period 1951 2010 0.90 1950 1960 1970 1980 1990 2000 2010 Year Hansen et al., 2012, PNAS 109(37), E2415 2423. Donat et al., 2016, Nat. Clim. Chang. 6, 508 513. S. Sippel Metrics used to measure climate extremes 04.10.2017 15 / 15

Conclusion: Quantification of extremes Conventional reference period standardisation approaches are systematically biased. This includes any data processing based on reference period statistics! An analytical understanding and correction has been developed Temperature extremes Precipitation extremes (a) 1.15 Rx1Day, Dry Regions 1.10 1.05 1.00 0.95 Ref. period 1951 1980 Ref. period 1981 2010 Ref. period 1951 2010 0.90 1950 1960 1970 1980 1990 2000 2010 Year Sippel et al., 2017, Hydrol. Earth Syst. Sc. 21, 441 458. Hansen et al., 2012, PNAS 109(37), E2415 2423. Donat et al., 2016, Nat. Clim. Chang. 6, 508 513. S. Sippel Metrics used to measure climate extremes 04.10.2017 15 / 15

Conclusion: Quantification of extremes Conventional reference period standardisation approaches are systematically biased. This includes any data processing based on reference period statistics! An analytical understanding and correction has been developed Temperature extremes Precipitation extremes Spatio-temporal variance or asymmetry of temperature Hansen et al., 2012, PNAS 109(37), E2415 2423. Donat et al., 2016, Nat. Clim. Chang. 6, 508 513. Huntingford et al., 2013, Nature 500, 327 330. Kodra and Ganguly, 2014, Sci. Rep. 4, 5884. S. Sippel Metrics used to measure climate extremes 04.10.2017 15 / 15

Conclusion: Quantification of extremes Conventional reference period standardisation approaches are systematically biased. This includes any data processing based on reference period statistics! An analytical understanding and correction has been developed Scrutinizing data analytical tools is crucial Statistical bias correction (often based on ref. period in Obs.) Generation of anomaly-based gridded observations (based on fixed ref. periods) Multivariate indicators for quantifying extremes S. Sippel Metrics used to measure climate extremes 04.10.2017 15 / 15

Conclusion: Quantification of extremes Conventional reference period standardisation approaches are systematically biased. This includes any data processing based on reference period statistics! An analytical understanding and correction has been developed Scrutinizing data analytical tools is crucial Data processing (feature extraction) can be more important than the choice of detection method. Flach et al., 2017, Earth Syst. Dyn. 8, 677 696. S. Sippel Metrics used to measure climate extremes 04.10.2017 15 / 15

Outlook Scrutinising data analytic tools and development of impact metrics using large, high-resolution model ensembles: Figure Courtesy ECMWF and Michael Wehner. S. Sippel Metrics used to measure climate extremes 04.10.2017 16 / 15

Thanks for the attention! S. Sippel Metrics used to measure climate extremes 04.10.2017 16 / 15

The underlying problem Artificial experiment yields the following pdf s: c Standardization of anomalies Original Gaussian PDF Out of base PDF Reference period PDF 6 4 2 0 2 4 6 z scores variability and extremes show a distinct change S. Sippel Metrics used to measure climate extremes 04.10.2017 16 / 15

The underlying problem Artificial experiment yields the following pdf s: c Standardization of anomalies Original Gaussian PDF Out of base PDF Reference period PDF Analytical understanding of biases:... what is meant to be done: z = X µ σ... and what is really being done: z = X ˆµ ref ˆσ ref Out-of-base period: Independent normalisation Reference period: Dependent normalisation 6 4 2 0 2 4 6 z scores variability and extremes show a distinct change Sippel et al., 2015, Geophys. Res. Lett. 42(22), 9990 9998 (Chapter 2). S. Sippel Metrics used to measure climate extremes 04.10.2017 15 / 15