Proxy reconstructions of Pacific decadal variability Michael N. Evans Laboratory of Tree-Ring Research/Geosciences/Atmospheric Sciences ATMO/GEOS 513 web pages: http://ic.ltrr.arizona.edu/ic/enso/ March 8, 2006 References: Gedalof and Sm ith, GRL 28(8):1515-1518 (2001); Evans et al. (2001), GRL 28(19): 36893692 (2001); optional: Evans et al. (2001) in Diaz and Markgraf, eds. (2001). 1) What is the Pacific decadal variability, and why does it matter? 2) How is PDV reconstructed from proxy observations? What are the primary uncertainties? 3) What have we learned about the mechanism and timescales of PDV? 4) What data would you like to have to resolve these questions?
You are the jargon police Mike, what did you mean by...? Image from the NYPD website, http://www.nypd.org/
What is the Pacific Decadal Oscillation?
What is the Pacific Decadal Oscillation? warm phase cold phase Source: N. Mantua, JISAO/Univ. Washington (http://tao.atmos.washington.edu/pdo/)
What is the Pacific Decadal Oscillation? Source: N. Mantua, JISAO/Univ. Washington (http://tao.atmos.washington.edu/pdo/)
How is the PDO similar to/different from ENSO? Spatial patterns: warm phases Time series Source: N. Mantua, JISAO/Univ. Washington (http://tao.atmos.washington.edu/pdo/)
Why does anyone care? Climate impacts Wind-driven ocean circulation, sea surface temperature, surface air temperature, precipitation, streamflow, snowpack, flood risk over much of the Pacific basin and over much of western and southern North America Ecological impacts Nutrient and phytoplankton distribution, sardine, anchovy, and salmon production in the CA coastal current, Pacific Northwest and Alaska Source: http://www.iphc.washington.edu/staff/hare/html/papers/jclimate/jclimate.html
Pacific climate variability on decadal timescales: Unresolved questions What is the underlying physical mechanism? Is there a characteristic timescale? Why might answers to these questions be useful?
Can statistical reconstruction of paleoclimate fields address our unresolved questions? We require reconstruction of climate fields to resolve spatial and temporal patterns of variability We require longer time series to resolve questions concerning quasi-decadal timescale behavior We require a mechanistic understanding of the processes which produce the observed PDO/impacts
Methodology of climate field reconstruction from high resolution proxy data networks Evans et al., 2001b
Methodology of climate field reconstruction from high resolution proxy data networks Characterization of the targeted climate field and the proxy data Calibration of the proxy data Reconstruction: least-squares estimation of fields and errors Verification: testing the reconstruction against independent estimates Interpretation: analysis of the results
Historical climate data contain signal and noise Uncertainties in their analysis are due to changes in Observational coverage Measurement method Source: COADS (left) and I-COADS (right) observatonal density over time, from Diaz et al. (2002): http://www.cdc.noaa.gov/coads/publications.html
Analysis of historical climate data sets Source: data from Kaplan et al. (1998) Identification of robust features by EOF analysis Decomposition into orthogonal patterns and time series Truncation (filtering) of patterns not resolved by observations Reconstitution of filtered fields
Analysis of proxy observations Evans et al. (2001b)
Methodology of climate field reconstruction from high resolution proxy data networks Characterization of the targeted climate field and the proxy data Calibration of the proxy data Reconstruction: least-squares estimation of fields and errors Verification: testing the reconstruction against independent estimates Interpretation: analysis of the results
Calibration of proxy observations Robust regression methods: leading pattern of covariance between tree-ring data and SST field (SVD) Result: correlation of first tree-ring principal component with SST, 1856-1990 How can tree ring data tell us about SST patterns? Evans et al. (2001b)
Methodology of climate field reconstruction from high resolution proxy data networks Characterization of the targeted climate field and the proxy data Calibration of the proxy data Reconstruction: least-squares estimation of fields and errors Verification: testing the reconstruction against independent estimates Interpretation: analysis of the results
Least-squares estimation of climate variables: Reduced space optimal interpolation field reconstruction from proxy data Objective: Obtain linear least-squares fit to available observations and a model of the largescale modes of spatial field variation. Statistical model of climate field variation: T=E Calibration of the proxy data: D=H patterns calibration (robust linear regression) Reduced space cost function: S( ) = (H -D)TR-1(H -D) + T -1 fit to observations and model within error Assumptions: unbiased estimates; uncorrelated errors; R,E,,H known produces Optimal reduced space estimate: T' = E ' = EPHTR-1D fields Error covariance in ': P = (HT R-1H + -1)-1 and errors. Evans et al. (2001b)
Methodology of climate field reconstruction from high resolution proxy data networks Characterization of the targeted climate field and the proxy data Calibration of the proxy data Reconstruction: least-squares estimation of fields and errors Verification: testing the reconstruction against independent estimates Interpretation: analysis of the results
Verification Comparison with observations withheld from the calibration Result: correlation (skill) of reconstruction for pre-calibration interval (1856-1990) Where does this reconstruction have skill? Where does this reconstruction not have skill? Evans et al. (2001b)
When things go wrong Or, opportunities for improved statistical reconstructions Source: www.perdixfire.org/ pics.html
Ideal conditions for paleoclimatic field reconstruction: Description of modern climate is complete and unbiased. Errors in modern observations are relatively small. Proxies are unbiased, constant, linear functions of the reconstruction target. There is no age model error in the proxy data.
Ideal conditions for paleoclimatic field reconstruction: Description of modern climate is complete and unbiased. Errors in modern observations are relatively small. Proxies are unbiased, constant, linear functions of the reconstruction target. There is no age model error in the proxy data. Reality: Proxies and historical climate data are biased. Small number of proxies = smaller dimension of reconstructed fields. Calibration (regression) is error-prone. There is age model error in all proxy data.
Proxy data fidelity is frequency-dependent Evans et al. (2002) Solutions: Prefilter proxy data Incorporate data-adaptive filter into calibration
Proxy data contains non-climatic information Calibration skill Verification skill Is there artificial calibration skill? (compare calibration to verification) Solutions: Use principal components to isolate robustly resolved variability Compare skill to benchmark reconstructions based on 'pseudoproxies' Evans et al. (2001b)
Proxy data are multivariate-dependent Solutions: Isolate a single controlling variable through proxy selection Exploit complementary multivariate dependencies to enhance signal Test interpretation of the proxy data using forward models Base proxy calibrations on more than empirical statistical relationships Vaganov-Shashkin model; Evans et al. (2006)
Proxy data intercomparison exercise http://ic.ltrr.arizona.edu/ic/enso/ex3.html
What are the robust interpretations of the proxy PDO/PDV reconstructions? Gedalof and Smith (2001)
What are the robust interpretations of the proxy PDO/PDV reconstructions? Evans et al. (2001)
Is the PDO centered in the North Pacific, or is it basin-wide? warm phase cold phase Source: N. Mantua, JISAO/Univ. Washington (http://tao.atmos.washington.edu/pdo/)
Is the PDO mechanism in the tropics or extratropics? Zhang et al., 1997: pattern Gu and Philander, 1997: mechanism
Is the PDO mechanism in the tropics or extratropics? Garreaud and Battisti, 1999: pattern Karspeck et al. (2002): mechanism
Is there a preferred PDO timescale from proxybased reconstructions? My $0.02: No.
What observations would you help you better address the timescale and mechanisms of Pacific decadal-scale climate variability?
Summary Predictability is a key goal of understanding PDV mechanism(s) and timescales(s). There are many uncertainties in reconstructing PDV from proxy data using robust forms of multiple linear regression (calibration, verification, data availability, error, shifting teleconnection patterns...) The mechanism underlying PDV remains unknown. Tropical dynamics and subtropical ventilation of the equatorial thermocline are leading candidates. We may need thousands of years of high resolution, wellreplicated proxy data from the right locations, plus a consistent OA model that includes the recent influence of anthropogenic forcing of the oceans and thermocline ventilation, to diagnose the mechanisms further and explore predictabity.