1. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 1 1 Reconstruction of Near-Global Precipitation Variations Based on Gauges and Correlations with SST and SLP Thomas Smith 1 Phillip Arkin 2 1. NOAA/NESDIS/STAR SCSB and CICS, College Park, Maryland 2. CICS/ESSIC/University of Maryland, College Park, Maryland Cooperative Institute for Climate and Satellites (CICS) Earth System Science Interdisciplinary Center (ESSIC)

2. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 2 2 Outline All analyses are of Precipitation Anomalies Base Satellite Data –IR (from 1979), MW (from mid 1980s) –Need global satellite analyses for reconstruction statistics Direct Reconstructions: fitting data to Empirical Orthogonal Functions (REOF) –EOF (or PC) analysis, for covariance maps –Fit available gauge-station data to a set of covariance maps –Monthly gauge-based 5-degree analyses available beginning 1900 Indirect Reconstructions: using Canonical Correlation Analysis (RCCA) –Correlate fields of sea-surface temperature (SST) and sea-level pressure (SLP) with fields of precipitation –Both SST and SLP analyzed for the 20 th century Merged Direct & Indirect Reconstructions –Direct Recons for over land and interannual and shorter variations over oceans –Indirect Recons more reliable for multi-decadal variations over oceans

3. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 3 3 Satellite-Based Analyses Monthly analyses, 1979-Present Several analyses available –Global precipitation climatology project (GPCP), multiple inputs, begins 1979, developed for climate studies –CAMS/OPI, several inputs, begins 1979, developed for interannual studies –Optimum Interpolation (OI) of MW and ERA-40 reanalysis, begins 1987, data problems over land before 1992 New OI analyses being tested in an attempt to obtain longer record using OI methods

4. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 4 4 Anomaly S.D. CAMS/OPI ocean S.D. more concentrated in tropics –OIP best for convective precipitation Both GPCP and OI S.D. have more extra-tropical variations –GPCP uses mix of satellite estimates + gauges –OI uses only microwave estimates, stronger variations than GPCP

5. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 5 5 Examples of spatial modes: Joint EOFs of OI and GPCP Anomalies EOF (PC) analysis jointly of 2 fields Shows common variations by mode Mode 1: Main ENSO

6. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 6 6 Joint EOF 2: Secondary ENSO Mode Mostly variance associated with the very strong 1997-1998 episode –Lags mode 1 Both analyses have similar ocean variations Tested using different analyses, use GPCP for final REOF

7. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 7 7 Reconstruction Based on EOFs (REOF) EOF spatial covariance modes –3 regions: 80S-20S, 30S-30N, 20N-80N –Separate so tropics do not dominate –Only large-scale modes used In each region, fit available gauge anomalies to the set of modes –3 areas merged with smoothing at boundaries Cross-validation testing to find the best set of modes for each region –For S.H. 5 EOFs, for Tropics 15 EOFs, for N.H. 10 EOFs

8. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 8 8 Anomaly Distribution Relative frequency distribution for common months and locations (1992-2001) Both anomalies approximately normal –reconstruction methods should work

9. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 9 9 Gauge Sampling of 5-deg Regions Gauge-based analyses, annual averages of monthly % sampling –Global Historical Climate Network (GHCN) –Global Precipitation Climatology Center (GPCC) –Climate Research Unit (CRU) CRU gives best sampling of 5-deg areas in historical period –Differences due to data processing & how many stations needed to form a 5- deg area –Test all & use CRU based analysis

10. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 10 REOF Methods Space points, x, and times, t Data: Anomalies, a, minus a first guess –Can sometimes estimate a first guess; otherwise can use 0 first guess Anomaly reconstruction: First guess plus weighted sum of the modes

11. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 11 Best Fit Weights for the Modes Reconstruction fit, F, using modes –Modes function of space –Weights function of time Squared error of the fit over all areas with sampling

12. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 12 System of Equations For Weights Differentiate error with each weight and set to 0 for system of equations to minimize error Equations to minimize error –Weights affected by data and sampling –System of equations solved each time for the set of weights Note: with complete data and orthogonal modes, computing the weights is simplified –EOF (or PC) time series

13. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 13 Sampling and Screening Modes Data must be sufficient to sample a mode’s variations –Mode variance proportional to the square of the mode value Compute the fraction of sampled variance of each mode Use only modes with a critical fraction of variance sampled –Use cross-validation testing to find the critical sampling fraction

14. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 14 Cross Validation Simulate historical periods –Modes computed from data excluding the time to be analyzed, and enough surrounding time to be independent –Data sub-sampled to simulate historical sampling Reconstruct using simulated historical conditions and compare to full data to find errors Used for tuning the reconstruction and for finding its errors

15. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 15 Cross-Validation For Tuning and Skill Evaluation Independent modes –Delete base data for each test analysis year Station sampling for 3 periods –Here GHCN sampling used Anomaly correlation with base full data –S.H. extra-tropics: low skill –Tropics: highest skill –N.H. extra-tropics some skill over oceans Ocean Avgs 1912-21 1952-61 1992-01 0.37 0.40 0.39

16. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 16 REOF Spatial Statistics Global spatial standard deviation (upper) –Similar interannual changes –GHCN low before 1940 (low sampling) –CRU strong most of record –Filtered GPCP strong at the end of record Global spatial correlation between analyses –High GHCN, CRU for high- sampling period, lower values before 1940

17. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 17 Regression Against SOI SOI represents ENSO interannual variability (annual averages) Shows typical ENSO precipitation patterns GHCN-based recon gives slightly weaker regression

18. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 18 Regression Against NAO Dec-March Regression Similar patterns, especially in Northern Hemisphere GPCC tropical Pacific different from the others

19. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 19 REOF Conclusions Satellite-based analysis are needed for historical reconstruction statistics Major interannual variations can be reconstructed for the 20 th century REOF oceanic multi-decadal variations may be less reliable –Different base data & analysis data change multi-decadal results much more than interannual results –REOF multi-decadal inconsistent with theoretical modeling results Improvements may be obtained from improved satellite base data Ocean-area sampling too sparse to use more EOF modes

20. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 20 Reconstructions Based on CCA (RCCA), and Comparisons to REOFs Canonical Correlation Analysis (CCA) –Fields of predictors correlated with a predictand field –Data smoothed and condensed using EOFs before CCA computed Training Data: GPCP, SST, SLP (1979-2004) –Annual average anomalies, GPCP satellite based –SST and SLP give ocean observations correlated with precipitation on long time scales Analysis Data: SST, SLP (1900-2004) –SST & SLP global analyses available

21. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 21 Canonical Correlation Analysis (CCA) Compress anomaly predictor (SST, SLP) and predictand (GPCP) fields using separate EOFs –All fields normalized –One joint predictor EOF, separate predictand EOF Predictor CCA equations: to relate compressed predictor fields to compressed predictand field –CCA modes developed –CCA eigenvalue used to weight analysis –For reconstruction, predictor data used to find weights for predictand field CCA predicting annual precipitation from annual SST and SLP –8 modes used, most oceanic variations from the first 3 modes CCA is described in detail by Barnett and Preisendorfer (1987, Mon. Wea. Rev., 115, 1825-1850)

22. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 22 Correlations Between Analyses Computed for period when all are available (1900-1998) –Averages over oceans, land, & areas with CRU gauge sampling –Annual-spatial averages correlated Each individual reconstruction correlates well with CRU gauges –ENSO and other major modes allow interannual variations to be resolved –REOF(Blend) has the best fit over land, but the nearly-independent RCCA is almost as good –AR4 ensemble averages out interannual variations, leaving in multi-decadal variations RCCA has same oceanic multi-decadal tendency as AR4, REOF has opposite tendency Correlations between averages over the given areas Oceans Land Gauge-sampling CRU, REOF(Blend) ---- ---- 0.88 CRU, RCCA ---- ---- 0.74 REOF(Blend), RCCA 0.64 0.81 0.83 REOF(Blend), AR4 -0.06 -0.01 -0.07 RCCA, AR4 0.32 -0.02 0.00

23. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 23 Land Comparisons RCCA & REOF & CRU data land averages, filtered –REOF(Blend) from REOF(CRU) and REOF(GPCP) RCCA & REOF similar for most of period RCCA & REOF(GPCP) similar for the GPCP period

24. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 24 Ocean Comparisons RCCA & REOF ocean averages, filtered RCCA & REOF differ before 1980 –1970s climate shift in RCCA –REOF does not resolve trend in RCCA & in AR4 ensemble RCCA & REOF(GPCP) similar –REOF(GPCP) can be used for updates

25. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 25 Trends Computed for period when all are available (1900-1998) –Averages over oceans, land, & land areas with CRU gauge sampling –Annual and low-pass filtered (as in figures) In each individual reconstruction, opposite trends over ocean & land –May be from use of ENSO modes to analyze ENSO-like multi-decadal –ENSO has opposite land-sea anomalies –Gauge data make land trends positive for REOF, no gauge data in RCCA Trends in mm/mon per 100 years for averages over the given areas Oceans Land Gauge-sampling CRU Gauges ---- ---- 1.2 REOF(Blend) -0.4 0.4 0.4 RCCA 1.6 -0.5 -1.1 AR4 0.7 -0.1 -0.5

26. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 26 Spatial Standard Deviation of Recons RCCA underestimates interannual signals REOFs give consistent level of signal over analysis period GPCP resolves variations filtered by REOF modes

27. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 27 RCCA Conclusions Multi-decadal variations over oceans can be reconstructed from SST and SLP using RCCA Over land, RCCA large-scale multi-decadal & interannual variations are consistent with independent observations over the 20 th century Over oceans, RCCA large-scale multi-decadal & interannual variations are consistent with model variations 20 th century

28. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 28 Merged Reconstructions REOF reliable over land where gauges are available Interannual REOF reliable over oceans, but multi-decadal REOF less reliable over oceans Multi-decadal RCCA appears to be more reliable over oceans Merge by replacing ocean multi-decadal REOF with ocean multi-decadal from RCCA For recent period, use REOF(GPCP)

29. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 29 Merged Recon Averages Filtered Recons for All Areas and Ocean Areas Ocean average changes most Including land removes the 1970s climate shift and most interannual variations

30. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 30 Recon Trends Ocean tropical trend greatest Land trends weaker & tend to be opposite to ocean trends –Similar to ENSO land-sea differences, suggests ENSO-like processes

31. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 31 Normalized Joint EOF Merged Recon and AR4 Ensemble Both annual averaged and filtered before JEOF First mode indicates joint trend-like variations –Tropical ENSO-like increase –Mid-latitude decrease –High-latitude increase –Pattern differences may reflect model biases

32. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 32 Possible Uses of Reconstructed Precipitation Climate-dynamics studies of global precipitation on interannual to multi-decadal time scales –Developing ocean-land changes over the 20 th century can be better understood and diagnosed –Oceanic influence on dry and wet regimes can be more clearly shown To validate and improve coupled-climate model precipitation –Both improvement of the models and statistical adjustment of model output possible with global reconstructions Data available at http://cics.umd.edu/~tsmith/recpr/

33. Cooperative Research Programs (CoRP) Satellite Climate Studies Branch (SCSB) 33 Summary & Conclusions EOF-based reconstructions resolve oceanic interannual variations through the 20 th century –Direct reconstruction using the available gauge data –Over land REOF does best for all variations CCA-based reconstructions resolve oceanic multi-decadal variations through the 20 th century –Indirect method using correlations with better sampled variables Merged analysis takes advantage of the best qualities of both Future improvements possible with new data or refined reconstruction methods –Extended reanalyses may yield independent precipitation information Data available at http://cics.umd.edu/~tsmith/recpr/