Empirical/Statistical Downscaling Dependencies in application. Bruce Hewitson : University of Cape Town

Transcription

1 Empirical/Statistical Downscaling Dependencies in application Bruce Hewitson : University of Cape Town

2 Empirical/statistical downscaling: A pragmatic response to need by impacts community for useful scenarios Diverse methodology -- Wide range of predictors -- Broad range of applications Region Technique Predictor Predictand Time Author (s) Africa South Africa TF C P D Hewitson & Crane, 1996 America USA WT T Tmax, Tmin D Brown & Katz, 1995 USA WG C P D Zorita et al., 1995 USA WG, TF C, T, VOR P D Wilby & Wigley, 1997 USA TF C, Q P D Crane & Hewitson, 1998 USA WG, TF C, T, VOR T, P D Wilby et al., 1998a, b USA WG, WT C T, P D Mearns et al., 1999 USA TF C, T, RH, W T D Sailor & Li, 1999 USA WG P D Bellone et al., 1999 Mexico and USA TF C, TH, O P D Cavazos, 1997 Mexico and USA TF, WT C, TH, Q P D Cavazos, 1999 Central Argentina TF C, W T, Tmax, Tmin M Solman & Nuñez, 1999 Asia Japanese coast TF C Sea level M Cui et al., 1995, 1996 Chinese coast TF Sea level M Cui and Zorita, 1998 variability Oceania New Zealand WT C Tmax, Tmin, P D Kidson & Watterson, 1995 New Zealand TF C, TH, T, P D Kidson & Thompson, 1998 VOR, W Australia TF C Tmax, Tmin D Schubert &Henderson-Sellers, 1997 Australia TF C Tmax, Tmin D Schubert, 1998 Australia WT C, T P Timbal & McAvaney, 1999 Australia WT Schnur & Lettenmaier, 1999 Europe Europe WG VOR, W Conoway et al., 1996 Europe WG, TF C, P, Tmax, T, P D Semenov & Barrow, 1996 Tmin, O Europe TF C, W, VOR, T, P M Murphy, 1998a, b T, Q, O Europe TF C T, P, vapour D Weichert & Bürger, 1998 pressure Germany TF T Phenological event Maak &van Storch, 1997 Germany TF C Storm surge M Von Storch & Reichardt, 1997 Germany TF Salinity Heyen & Dippner, 1998 Germany WT Thunderstorms D Sept, 1998 Germany TF Ecological Krönke et al., 1998 variables Iberian Peninsula WG C P D Cubash et al., 1996 Iberian Peninsula TF C Tmax, Tmin D Trigo & Palutikof, 1998 Iberian Peninsula TF P, NST Boren et al., 1999 Iberian Peninsula TF P, NST Ribalaygua et al., 1999 Spain (and USA) TF C Tmax, Tmin D Palutikof et al., 1997 Spain (and USA) TF C Tmax, Tmin D Winkler et al., 1997 Spain WT D Goodess & Palutikof, 1998 Portugal TF C P M Corte-Real et al., 1995 Portugal WT C D Corte-Real et al., 1999 The Netherlands WT C, VOR, W T, P D,M Buishand & Brandsma, 1997 Norway TF C, O T, P and others M Benestad, 1999a, b Norway (glaciers) TF C, O Local weather D Reichert et al., 1999 Romania TF C P M Busuioc & von Storch, 1996 Romania TF C P M Busuioc et al, 1999 Switzerland TF P Buishand & Klein Tank, 1996 Switzerland TF P Brandsma & Buishand, 1997 Switzerland TF D Widmann & Schär, 1997 Switzerland WG C Local Weather H Gyalistras et al., 1997 Switzerland TF P Buishand & Brandsma, 1999 Poland TF C T, sea level, wave height, salinity, wind, run-off D,M Mietus, 1999 Alps WT Fuentes & Heimann, 1996 Alps TF C, T T, P M Fischlin & Gylistras, 1997 Alps WT C Snow Martin et al., 1997 Alps WT Fuentes et al., 1998 Alps TF C, T T, P, Gyalistras et al., 1998 Alps, TF C, T Snow cover Hantel et al., 1998 Alps WT C, T Landslide activity Dehn, 1999a, b Alps WT T, P D Heimann and Sept, 1999 Alps WT P D Fuentes & Heimann, 1999 Alps TF, WG C, T Weather statistics M Riedo et al., 1999 Alps TF C P M Burkhardt, 1999 Mediterranean TF C, P T Palutikof & Wigley, 1995 Mediterranean TF C P S Jacobeit, 1996 North Atlantic TF C Pressure M Kaas et al., 1996 tendencies North Atlantic TF C Wave height M WASA, 1998 North Sea TF Ecological Dippner, 1997a, b variables North Sea coast TF C Sea level M Langenberg et al., 1999 Baltic Sea TF SLP Sea level M Heyen et al., 1996 Region not specified WT Frey-Buness et al., 1995 WT C Matyasovszky & Bogardi, 1996 WT Enke & Spekat, 1997 TF C, VOR, W Kilsby et al., 1998 TF Ecological variables Heyen et al., 1998

3 Given the fundamental dependency on validity of AOGCM information as the starting point for the application of all regionalization techniques -- for RCMs and statistical methods. Attractions of empirical/statistical approaches: Computational efficiency Rapid application to multiple GCMs Tailoring to target variables (eg: storm surge) Applicability to broad range of temporal and spatial resolutions Accessibility beyond the modeling community Complementary to regional modeling Significant lack of systematic evaluation. More co-ordinated efforts are thus necessary to evaluate the different methodologies, inter-compare methods and models IPCC, TAR 2001 For empirical/statistical approach: what are the considerations for effective application?

4 Two broad categories for downscaling Transfer functions relating atmospheric forcing to target variable Stochastic functions, possibly conditioned on synoptic state Variance explained Synoptic Forcing Sub-grid scale forcings Synoptic scale Local scale For climate change: what proportion of response is due to sub-gcm grid scale structure in the regional climate change signal? Both transfer functions and stochastic functions vulnerable to nonstationarity of sub-grid scale forcing. Q: for a given location, which is dominant: local or synoptic forcing?

5 Example: From a RCM experiment High-res SST Precipitation as a function of three different SST fields with identical NCEP boundary forcing: Precipitation (primarily convective) is temporally consistent between SST fields. Implication of dominance by synoptic state. 1 SST Zonal SST

6 Given the caveats, what needs to be evaluaed for effective empirical/statistical downscaling? 1) Choice of predictor variables Most commonly used are circulation related variables

7 Given the caveats, considerations for effective empirical/statistical downscaling? 1) Choice of predictor variables 2) Predictor spatial representation Indices, EOFs, Synoptic classifications, Raw grid data Local versus remote (teleconnections) Surface versus upper air fields

8 Given the caveats, considerations for effective empirical/statistical downscaling? 1) Choice of predictor variables 2) Predictor spatial representation 3) Antecedent conditions Local forcing as a function of antecedent events

9 Given the caveats, considerations for effective empirical/statistical downscaling? 1) Choice of predictor variables 2) Predictor spatial representation 3) Antecedent conditions 4) Target (predictand) resolutions Station scale, impacts scale (scale of user community), RCM scale?

10 Given the caveats, considerations for effective empirical/statistical downscaling? 1) Choice of predictor variables 2) Predictor spatial representation 3) Antecedent conditions 4) Target (predictand) resolutions 5) Training data periods Observational data that sufficiently spans the relationship for training downscaling function

11 Given the caveats, considerations for effective empirical/statistical downscaling? 1) Choice of predictor variables 2) Predictor spatial representation 3) Antecedent conditions 4) Target (predictand) resolutions 5) Training data periods 6) Representing the climate change signal Predictors explaining significant variance may not be predictors sensitive to the climate change signal

12 Given the caveats, considerations for effective empirical/statistical downscaling? 1) Choice of predictor variables 2) Predictor spatial representation 3) Antecedent conditions 4) Target (predictand) resolutions 5) Training data periods 6) Representing the climate change signal 7) Stationarity of function / predictors Is climate change primarily characterized by changes in frequency of existing events? Are changes in local sub-grid-scale forcing small with respect to synoptic forcing? Are residuals in downscaling from GCM-resolution due to low predictor resolution, or sub-grid scale forcing?

13 Given the caveats, considerations for effective empirical/statistical downscaling? 1) Choice of predictor variables 2) Predictor spatial representation 3) Antecedent conditions 4) Target (predictand) resolutions 5) Training data periods 6) Representing the climate change signal 7) Stationarity of function / predictors A given downscaling implementation needs to take cognizance of, and evaluate, the dependencies

14 Exploring the dependencies. Transfer function based methodology: gives dominance to synoptic forcing Challenging case: continental summer convective daily precipitation NCEP derived predictors Station derived precipitation

15 Exploring the dependencies. Transfer function based methodology: gives dominance to synoptic forcing Problematic case: continental summer convective daily precipitation NCEP derived predictors Station derived precipitation Topography Regional context steep topography elevated inversions strong interannual variability

16 Dominance by semi-permanent high pressure systems with surface thermal trough Strong spatial gradients of precipitation strongly dependant on moisture transport January mean SLP January mean precip

17 Characteristic 7-day back trajectories into test region for downscaling (shading by specific humidity).

18 Downscaling methodology: Transfer function methodology - derives local response as function of synoptic forcing, excludes subgrid scale local forcing (useful for evaluation of dependencies). - Artificial Neural Nets (analogous to non-linear multiple regression) - derives non-linear transfer functions between NCEP (2.5 ) atmospheric variables and precipitation (0.25 ) 20 years of training data ( ) * Focus not on optimizing results, but a sensitivity study * Pre-1980 reanalysis data problematic for southern hemisphere

19 1: Evaluation of predictor variables (20 examples) Surface 700hPa 500hPa temperature temperature temperature divergence divergence divergence geopotential height geopotential height vertical velocity vertical velocity vertical velocity relative humidity specific humidity specific humidity u wind u wind u wind v wind v wind v wind Each predictor used independently to derive a transfer function to precipitation at Predictor temporal resolution: Predictor spatial resolution: 12 hourly 9 grid cells (7.5 by 7.5 ) centered on target location 48 hour antecedent predictor state included

20 Predictor variable R Results suggest: Dominant relationship is with mid and upper troposphere humidity and predictors related to vertical motion. Specific Humidity (500hPa) 0.56 Vertical Velocity (500hPa) 0.55 v wind (700hPa) 0.53 Relative Humidity (Surface) 0.53 Specific Humidity (700hPa) 0.49 Divergence (700hPa) 0.49 Temperature (Surface) 0.45 Geopotential height (700hPa) 0.44 v wind (500hPa) 0.44 Divergence (Surface) 0.41 Vertical Velocity (700hPa) 0.40 Divergence (500hPa) 0.35 u wind (Surface) 0.34 u wind (500hPa) 0.34 Vertical velocity (Surface) 0.34 u wind (700hPa) 0.34 v wind (Surface) 0.34 Temperature (500hPa) 0.30 Temperature (700hPa) 0.27 Geopotential height (500hPa) 0.19

21 Similar examination of other locations supports the above results. Suggests predictors should include mid-troposphere indicators of humidity and circulation dynamics. Place Arg Aus Bot Zam Bra Nin3 Cri Ban Mex Chi Atl Por Iow Ger Sib Latitude Season W W W W W D D D D D W W D D D Rank 1 q7 rh7 rh7 q5 rh7 th q7 rh7 v0 q7 z8 v0 q7 v0 rh7 2 z5 q7 q5 z7 q5 q5 rh7 q7 rh7 v0 z7 z8 rh7 rh7 z5 3 rh7 q5 q7 z5 q7 z2 q5 z5 q5 rh7 z5 rh7 z8 z7 z7 4 z7 z5 z7 z8 u8 q8 z7 z7 q7 q5 z2 z5 z5 z5 z2 5 q5 v0 z5 rh7 d8 rh7 z5 q5 z5 z5 v0 z7 q5 z8 d8 6 z8 z7 z2 q7 d8 z8 v0 z8 z8 rh7 q7 z2 q5 7 v0 z2 z8 th d2 th vo0 v8 z7 q7 z2 z7 z2 8 z2 u8 z2 vo0 v7 z2 v7 q5 q5 d2 q7 9 th slp v0 u5 d2 q8 u5 u5 th 10 d8 v8 d2 u5 q8 th slp 11 v7 u5 u8 d2 u8 slp vo0 12 u8 v7 v0 d8 d8 d2 v8 13 v8 d8 q8 d2 14 d2 u8 z2 u8 15 u5 u0 u0 16 vo0 v7 17 q8 slp

22 Based on the above, a set of predictors may be chosen. eg: Surface temperature, u and v winds 700hPa specific humidity and geopotential heights 500hPa specific humidity and geopotential heights Trained function results: r = 0.7 predicted mean precipitation: 4.2mm/day observed mean precipitation: 3.8mm/day mm * Observed Downscaled Days

23 Residuals From either: Lack of information in predictors (choice or predictor or resolution) Local sub-grid scale forcing unrelated to synoptic state May be stochastically modeled (stationarity issues) mm * Observed Downscaled Days

24 2: Predictor Spatial Resolution Test relationship of target variable to atmospheric predictors progressively further away from region of interest. ANN downscaling using mid-troposphere (700hPa) specific humidity and geopotential height Predictors drawn from progressively larger regions: a) single NCEP grid cell co-located with target b) 7.5 by 7.5 window centered on target c) 15 by 15 window centered in target d) 22.5 by 22.5 window centered on target e) 30 by 30 window centered on target Spatial resolution r Single cell x x x x Increase in predictor window size, once large enough to represent spatial gradient, has minimal improvement.

25 2: Predictor Spatial Resolution Downscaling a function of information content in predictors -- a function of resolution. eg: Station daily rainfall downscaled from MRF (1 ) atmospheric fields: Average error: 0.5 mm/day Observed and Predicted Rainfall Rainfall (mm/day) Day Predicted Observed Suggests variance from sub-grid scale forcing is minimal in this case

26 3: Predictor Antecedent State Test relationship of target variable to inclusion of the antecedent state of atmospheric predictors. Predictors used as: a) time coincident with target b) time coincident with target and increasing lag in 12 increments to 90 hours lag. Lags of at least 24 hours are very beneficial Correlation Lag (hours)

27 5: Training data period Test sensitivity of downscaled function to data used in training. Case 1 Train on 1980s Case 2 Train on 1990s Case 3 Train on 1982/83 For each case, test function on independent decades. Case 1: Trained on 1980s, predicted 1980s: r = 0.66 mean ppt: +7% Trained on 1980s, predicted 1990s: r = 0.59 mean ppt: -9% Case 2: Trained on 1990s, predicted 1990s: r = 0.78 mean ppt: +6% Trained on 1990s, predicted 1980s: r = 0.51 mean ppt: +18% Case 3: Trained on 82/82, predicted 1980s: r = 0.33 mean ppt: -34% Trained on 82/83, predicted 1990s: r = 0.11 mean ppt: -28% Where training data spans the variability, performance good

28 6: Representing the climate change signal Predictors that explain the most variance may not be the predictors that capture the climate change signal. Test: for each predictor, determine the climate change signal Train on the predictors, and predict from GCM control and future climate simulations Predictor variable Future - present downscaled % change Specific humidity (500hPa) 4.49 Specific humidity (700hPa) 4.71 Surface Temperature 2.43 Surface u-wind Surface v-wind hPa geopotential heights hPa geopotential heights Choice of predictor may change sign of downscaled response

29 6: Representing the climate change signal Downscaling using: Future - control: +2.1% Specific humidity (500hPa) Specific humidity (700hPa) Surface u-wind Surface v-wind 500hPa geopotential heights 700hPa geopotential heights Or excluding humidity: Future - control: -3.5% Surface u-wind Surface v-wind 500hPa geopotential heights 700hPa geopotential heights

30 Incorporating the issues discussed: Downscaled summer precipitation anomaly (future - present) Downscaled from CSM transient 1%/year simulation

31 7: Stationarity: Predictors: Do future synoptic events have present day representation Transfer function: Stability of relationship Sub-grid scale forcing: % contribution to local variance, feedbacks to atmosphere At a minimum, evaluate predictors...

32 7: Stationarity Consider distribution of 700hPa geopotential height fields in CSM control simulation 2001 ESIG/NCAR. Not for reproduction without written permission.

33 7: Stationarity Frequency of occurrence of each mode may be determined, and change under future climate calculated % change in frequency of occurrence from CSM future-control silumations:

34 7: Stationarity % change in frequency of occurrence from CSM future-control simulations: Similarity of future patterns to present day may be determined, and a measure of change in pattern calculated. % change in pattern from CSM future-control simulations: Where significant increases in frequency have occurred, variance of pattern modes has generally decreased. Hence: CSM 700hPa geopotential height fields under a future climate are spanned by events in present day simulation

35 Some conclusions: Empirical/statistical downscaling has pragmatic attractions. Appropriate implementation can produce downscaled results consistent to changes in synoptic forcing. Fundamental need for further systematic evaluation

36 Results offer useful downscaled scenarios?