SUPPLEMENTARY INFORMATION

Transcription

1 SUPPLEMENTARY INFORMATION DOI:.8/NCLIMATE76 Supplementary information for Changes in South Pacific rainfall bands in a warming climate Matthew J. Widlansky, Axel Timmermann,, Karl Stein, Shayne McGregor, Niklas Schneider,, Matthew H. England, Matthieu Lengaigne 4, and Wenju Cai 5 International Pacific Research Center, University of Hawaii at Manoa, 68 East-West Road, Honolulu, Hawaii 968, USA Department of Oceanography, University of Hawaii at Manoa, Pope Road, Honolulu, Hawaii 968, USA Climate Change Research Centre, University of New South Wales, Sydney, NSW 5, Australia 4 Laboratoire d Océanographie et du Climat: Expérimentation et Approches Numériques (LOCEAN), IRD/UPMC/CNRS/MNHN, Paris, France 5 CSIRO Marine and Atmospheric Research-Aspendale, 7- Station Street, Aspendale, VIC 95, Australia NATURE CLIMATE CHANGE Macmillan Publishers Limited. All rights reserved.

2 SUPPLEMENTARY METHODS Data description. SST data from a continuous AB experiment are available for CMIP models (4 of these models use some form of SST flux correction). Rainfall data is available for two additional CMIP models and is shown in Supplementary Fig. S for comparison with CMIP5 (Fig. ). In cases where multiple model realizations are available, the ensemble for each model is averaged prior to calculating the multi-model mean. While preparing this manuscript, some CMIP5 SST ( models) and rainfall ( models) data became available. Additionally, some modeling groups participating in CMIP5 have forced the atmospheric components of their models with the SST trend from the CMIP 4xCO experiment ( AMIP-future shown in Supplementary Fig. S6; 5 models). Considering available CMIP5 models, the Pacific SST th century cold tongue bias and st century warming pattern from the RCP 4.5 W m - experiment are mostly similar to, respectively, the CMIP historical experiment and AB emissions scenario. Note that the magnitude of greenhouse warming is larger in AB. In CMIP5, at least for the model data released at the time of writing this paper, rainfall biases are similar to CMIP and the inter-model standard deviation of rainfall projections continues to be larger than the st century multi-model trend in most of the SPCZ (Fig. and Supplementary Fig. S). Each atmospheric model ( ½ layer troposphere model, ICTP GCM, and CAM GCM) in the experimental hierarchy utilized here is initialized from an atmosphere at rest and forced by a fixed annual cycle of either monthly climatological SST (Control), or by this SST climatology perturbed either by the CMIP bias, by the st century projected SST trend, or by the 4xCO SST trend used to force the AMIP-future experiment. CMIP average model biases are Page of 9 Macmillan Publishers Limited. All rights reserved.

3 calculated by subtracting the observed SST (NOAA Extended Reconstruction version b; refs.,) month varying climatology from the SSTs obtained from the CMIP th century historical experiment. The averaging period is chosen to achieve a balance between warm and cold ENSO events in the observed SST field. For comparison, SST and rainfall (GPCP Merged Monthly precipitation version ; ref. 4) observations are shown in Supplementary Fig. S. CMIP SST trends are calculated by subtracting the th century historical period (98-99) from the AB projection (9-99). All data are interpolated onto a x latitudelongitude grid prior to computing biases or trends for forcing atmospheric GCM experiments, except for the ICTP experiments, for which data are first interpolated onto the native model grid (total wavenumber, T: approximately.75 horizontal resolution). For the ICTP experiments, SST from the European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis (ERA5; ref. 5) is used for the Control instead of NOAA SST to best match previous experiments investigating the SPCZ (ref. 6). SST patterns are consistent regardless of grid resolution or choice of observational data source. To avoid perturbing sea-ice distributions (particular to the ICTP and CAM models), SST anomalies poleward of approximately 4.5 N/S are tapered to zero over three latitudinal grid spaces. No tapering is done for the 4xCO SST experiments as tapering the larger SST anomalies would lead to spurious gradients in the midlatitudes. Model specifics. The ½ layer troposphere model 7,8 is an intermediate complexity model of only the tropical atmosphere (-N) with x4 latitude-longitude horizontal resolution which has previously been shown to simulate the salient features of the tropical SPCZ (ref. 9). We integrate the model for years, of which only the last DJF seasons are shown. Page of 9 Macmillan Publishers Limited. All rights reserved.

4 Principally, the ½ layer model couples a first baroclinic mode Gill-type free troposphere with a Lindzen-Nigam type boundary layer. Long-wave radiative forcing and boundary layer thermodynamics 7 are also included. Rainfall is a function of the underlying surface temperature and accordingly we remove the spatial average st century SST warming trend (. C, calculated between 4.5 N/S) to preserve the observed threshold for convection in the tropics, allowing comparison of the simple troposphere model with higher complexity GCM experiments. For the more complex GCMs used in this study, the SST threshold for convection adjusts dynamically to greenhouse warming conditions and as a result of near-homogenous upper tropospheric warming. The simplified GCM we use is an idealized atmospheric model from the International Centre for Theoretical Physics (ICTP; refs. 4-7) which is based on a hydrostatic spectral dynamical core 8 and governed by many physical parameterizations including short- and long-wave radiation, large-scale condensation, convection, surface fluxes of momentum, heat and moisture, and vertical diffusion. Convection is parameterized by a mass-flux scheme that is activated where conditional instability is present. The ICTP GCM is configured with eight vertical (sigma) levels and T spectral resolution. Coupling is done to a slab land model, allowing land temperatures to adjust to modified atmospheric conditions. In polar regions, coupling to a thermodynamical seaice model is performed. SST anomalies are tapered to zero poleward of 4 N/S and the sea-ice distribution does not seem to affect rainfall in the tropical SPCZ. For the ICTP model, the first years of a -year integration are discarded and the remaining years are used to compile DJF seasonal averages. Page 4 of 9 Macmillan Publishers Limited. All rights reserved.

5 The most complex and highest resolution atmospheric GCM used in this study is the National Center for Atmospheric Research (NCAR) Community Atmosphere Model version (CAM.; ref. 9). CAM is configured with 6 vertical levels and T4 spectral resolution. Compared to the ICTP GCM, the CAM dynamical core is more sophisticated and the physical parameterization suite is more comprehensive. The atmosphere is coupled to sophisticated land and sea-ice models, which are more complex than in the ICTP GCM. Averages of the last DJF seasons from -year simulations are shown for the CAM st century experiments (for the 4xCO pattern SST experiments, -season averages from a 4-year simulation are shown to more closely match the AMIP-future experiment). All CAM experiments are compared to the last DJF seasons from a 4-year Control simulation. To test the effects of SST biases on the st century SST change and on the SPCZ s response to greenhouse gas increases, we conducted fully coupled GCM experiments with doubled CO concentrations with and without SST bias correction. A flux coupler is used to merge CAM with a full ocean, sea ice, and land surface model, the NCAR Community Climate System Model version (CCSM; ref. ). For our experiments, the atmospheric model (CAM) is configured with the same resolution as above. Coupling is made to the Parallel Ocean Program (POP) with a nominal horizontal resolution of and 4 vertical levels. The full sea ice and land models have horizontal resolution similar to the ocean. We run four CCSM experiments for 9 years each to reach approximate SST equilibrium and account for internal variability. For two Control experiments, the CO volume mixing ratio is held constant at 99 values (55 ppm; ref. ). CO is increased % per year to 7 ppm in two Ramping experiments. For one Control and one Ramping experiment, we correct SST biases internal to CCSM using a shortwave radiative flux Page 5 of 9 Macmillan Publishers Limited. All rights reserved.

6 adjustment (W m - ) applied to oceans in the global channel between 5N/S (Supplementary Fig. S) and tapered to zero elsewhere. SST and rainfall projections from the last years of the Ramping experiments, with or without SST bias-correction, are compared to -year averages from the respective Control simulations (Supplementary Fig. S4). The experiment without bias correction projects a warming pattern (Supplementary Fig. S4a) similar to the CMIP AB experiment (Supplementary Fig. Sa), but with smaller magnitude presumably because other greenhouse gases are not increased. Unlike most CMIP models, CCSM projects drying in the model s SPCZ (Supplementary Fig. S4b). Drying anomalies extend east into a region of the central tropical Pacific with little observed rainfall. The zonal SPCZ bias and associated rainfall anomalies in CCSM are related to SST biases common to CMIP, namely a cold bias in the equatorial East Pacific and a warm bias to the south (Supplementary Fig. Sb). By applying a shortwave radiative flux adjustment to the global channel, SST biases in the East Pacific are diminished and as a result the double-itcz rainfall bias vanishes (contours in Supplementary Fig. S4b,d), generating a more diagonal SPCZ. The projected warming patterns in both CO doubling experiments (shading in Supplementary Fig. S4a,c) are very similar regardless of whether SST bias correction is applied, and resemble the CMIP (Supplementary Fig. Sa) and CMIP5 (Fig. a) multi-model average. Future drying still occurs throughout the SPCZ, but rainfall anomalies are shifted west compared to the uncorrected coupled experiment. Unlike the CCSM experiment without flux correction, which projected greatest drying in a region of the South Pacific that receives little observed rainfall, drying in the SST bias-corrected experiment occurs in the Southwest Pacific mostly collocated with the Page 6 of 9 Macmillan Publishers Limited. All rights reserved.

7 observed SPCZ. Most importantly, st century simulated SST changes are robust to improvements of the coupled model s SST climatology. Page 7 of 9 Macmillan Publishers Limited. All rights reserved.

8 N S 9 Observed climatology 4 E 5E 8W 5W W 9W (mm day - ) Supplementary Figure S Observed rainfall and SST during DJF (98-). GPCP rainfall (mm day -, shading) and NOAA SST ( C, contours and labels). SST contour interval: C; starting at.5 C. Red contours depict the 7.5 C isotherm. Blue contours depict the 5 mm day - average rainfall. Page 8 of 9 Macmillan Publishers Limited. All rights reserved.

9 a N S. SST trend E 5E 8W 5W W 9W.5.5 ( C) d N S Rainfall trend E 5E 8W 5W W 9W (mm day - ) b N S Inter-model standard deviation of SST trend E 5E 8W 5W W 9W ( C) e N S Inter-model standard deviation of rainfall trend E 5E 8W 5W W 9W (mm day - ) c N S th century SST bias f th century rainfall bias E 5E 8W 5W W 9W ( C) E 5E 8W 5W W 9W Supplementary Figure S CMIP st century projections, inter-model variability, and th century biases during DJF. Multi-model ( for SST and for rainfall) mean projections (8-99 minus ) for the AB emissions scenario and biases (98-999) for the historical experiment with respect to observations. (a) SST projection ( C, shading) and th century multi-model average 7.5 C SST contour (green line). (b) Inter-model standard deviation of SST projections ( C, shading, tropical mean st century SST trend removed). (c) SST bias ( C, shading) and 7.5 C contour for observations (dashed) and CMIP (solid). (d) Rainfall projection (mm day -, shading) and th century multi-model average (mm day -, black contours). Average rainfall contour interval: mm day - ; starting at mm day -. Blue contours depict the 5 mm day - multi-model mean rainfall for the th century. (e) Inter-model standard deviation of rainfall projections (mm day -, shading). (f) Rainfall bias (mm day -, shading) and observations (mm day -, black contour interval as for d). Blue contours depict the 5 mm day - rainfall for observations (dashed) and CMIP (solid). N S (mm day - ) Page 9 of 9 Macmillan Publishers Limited. All rights reserved.

10 a Shortwave flux correction (W m - ) 5N 5 4N 4 N N S 4 5 E 6E E 8W W 6W E b N S SST correction E 5E 8W 5W W 9W Supplementary Figure S Shortwave flux correction applied to a CGCM (NCAR CCSM). (a) Shortwave flux correction (W m - ) applied to oceans in the global channel between 5N/S and tapered to zero elsewhere. (b) SST bias-correction anomalies ( C, shading) during DJF (-year average) and 7.5 C SST contour for the two experiments (no flux correction: red; flux correction: green). SST anomalies are the difference between the Control experiment with or without flux correction. ( C) Page of 9 Macmillan Publishers Limited. All rights reserved.

11 a N S No flux correction E 5E 8W 5W W 9W ( C) c N S Flux correction E 5E 8W 5W W 9W ( C) b N S E 5E 8W 5W W 9W (mm day - ) d E 5E 8W 5W W 9W Supplementary Figure S4 st century projected SST and rainfall anomalies during DJF (-year average) for a xco scenario using a CGCM (NCAR CCSM) either without SST bias-correction (left) or with radiative flux correction (right). (a and c) SST projection ( C, shading) and th century 7.5 C SST contour (green line) for the two experiments. (b and d) Rainfall projection (mm day -, shading) and th century Control average (mm day -, black contours). Rainfall contour interval: mm day - ; starting at mm day -. Blue contours depict the 5 mm day - Control average rainfall. N S (mm day - ) Page of 9 Macmillan Publishers Limited. All rights reserved.

12 a N b S st century trend (tropical mean removed) E 5E 8W 5W W 9W N S -. ½ layer model E 5E 8W 5W W 9W.6 -. ( C) (mm day - ) e N S st century trend E 5E 8W 5W W 9W ( C) c N S ICTP model E 5E 8W 5W W 9W (mm day - ) f N S ICTP model E 5E 8W 5W W 9W (mm day - ) d N S CAM model E 5E 8W 5W W 9W (mm day - ) g N S CAM model E 5E 8W 5W W 9W (mm day - ) Supplementary Figure S5 st century projected rainfall anomalies during DJF (9-99) for atmospheric models forced with SSTs modified from the AB emissions scenario. Experiments with the spatially averaged SST trend between 4.5 N/S removed (left column) are compared to those with the SST trend included (right column). (a and e) SST forcing anomalies ( C, shading) and 7.5 C contour for the Control (solid) and Experiments (dotted). (b-d) Rainfall anomaly (mm day -, shading) and Control average (mm day -, black contours) for the ½ layer, ICTP, and CAM models, respectively. (f-g) Same, but for only the ICTP and CAM models forced with the tropical mean SST trend included. For the ½ layer model (b), only the tropical channel domain is shown. Control average rainfall contour interval: mm day - ; starting at mm day -. Blue contours depict the 5 mm day - Control rainfall for each model. Page of 9 Macmillan Publishers Limited. All rights reserved.

13 a N b S N S 4. SST trend E 5E 8W 5W W 9W AMIP future ensemble E 5E 8W 5W W 9W ( C) (mm day - ) c N S CAM model E 5E 8W 5W W 9W (mm day - ) Supplementary Figure S6 Projected rainfall anomalies during DJF for atmospheric models forced with SSTs modified from the CMIP 4xCO emissions scenario. (a) SST forcing anomaly ( C, shading) and 7.5 C contour for the Control (solid) and Experiments (dotted). (b-c) Rainfall anomaly (mm day -, shading) and Control average (mm day -, black contours) for the AMIP-future ensemble and the CAM model, respectively. Control average rainfall contour interval: mm day - ; starting at mm day -. Blue contours depict the 5 mm day - Control rainfall for each model Page of 9 Macmillan Publishers Limited. All rights reserved.

14 Net moisture flux convergence (% th century observations) in the SPCZ 5 4 CMIP5 (- (u'q)) + (- (uq')) (RCP 4.5 W m - ) CMIP5 (RCP 6. W m - ) CMIP5 (RCP 8.5 W m - ) AMIP future (4xCO SST) CAM experiments Projected SST trend ( C) in the SPCZ Supplementary Figure S7 Multi-experiment projection estimate of the net moisture flux convergence in the SPCZ. Shaded shapes (see legend and Supplementary Table S) represent the sum of each respective experiment s near-surface convergence of anomalous moisture advected by the mean winds (- (uq')) and convergence of mean moisture advected by anomalous winds (- (u'q)) as in the sum of the green and brown shapes in Fig. 5, but shown here as a percentage of th century ERA4 observations. The solid curve represents the ensemble best fit (i.e., the multi-experiment projection), using a nd order polynomial constrained to pass through the origin. Dashed curves represent the functional bounds (95% confidence interval) for the projection estimate assuming 7 degrees of freedom (df=(n-)-k, where n=76 is the number of experiments and k= for a quadratic polynomial). Maximum drying is projected to be -6% of the observed mean at.5 C with projection bounds of the quadratic fit of ±4%. Net moistening is projected for warming greater than.c, although uncertainty increases to approximately ±% for the highest warming amount. Note that the functional bounds for the projection estimate (dashed curves) are narrower than the inter-model variance (blue error bars) because the functional bounds account only for sampling variations in the parameters of the multi-experiment projection. Error bars are centered about three warming bins inclusive of the 76 member ensemble (each interval is approximately C) and measure ± standard deviations about the inter-model mean of each bin. Page 4 of 9 Macmillan Publishers Limited. All rights reserved.

15 Supplementary Table S. List of atmospheric modeling experiments with prescribed SST anomalies and the CGCM experiment with radiative flux corrected SST. Experiment th century bias *Fig. st century trend *Fig. and Supplementary Fig. S5 Model hierarchy SST anomaly CAM CMIP (98:99) OBS (98:99) ½ layer ICTP CAM MME NOAA CMIPMME(9: 99) CMIPMME(98:99) (tropical mean warming removed) st century trend *Supplementary Fig. S5 4xCO SST trend *Fig. and Supplementary Fig. S6 st century trend (CGCM) *Supplementary Fig. S4 ICTP CAM CAM Non flux corrected- CCSM Flux corrected- CCSM CMIPMME(9: 99) CMIPMME(98:99) (tropical mean warming included) OBS NOAA (979: 8) plus patterned SST anomaly (CMIP 4xCO, tropical mean warming removed) OBS NOAA (979: 8) plus patterned SST anomaly (CMIP 4xCO, tropical mean warming included) [CO :7 ppm]- 99 CO levels [55 ppm] [CO :7 ppm]- 99 CO levels [55 ppm] Page 5 of 9 Macmillan Publishers Limited. All rights reserved.

16 Supplementary Table S. Description of the climate model hierarchy. Model name Type Resolution SST Climatology ½ layer ICTP CAM CCSM Tropical troposphere Atmospheric GCM Atmospheric GCM Ocean- Atmosphere GCM latitude x 4 longitude,.5 vertical levels T spectral, 8 vertical levels T4 spectral, 6 vertical levels Atmosphere (CAM) Ocean (POP:, 4 vertical levels) (98:99) OBS NOAA (98:99) OBS ERA5 OBS NOAA (979: 8) Non-flux corrected SST 99 CO levels [55 ppm] Flux corrected SST 99 CO levels [55 ppm] Page 6 of 9 Macmillan Publishers Limited. All rights reserved.

17 Supplementary Table S. List of models for each greenhouse warming experiment analyzed in Fig. 5 and Supplementary Fig. S7. In cases where multiple model realizations are available, only the first run for each model is analyzed. Greenhouse warming scenario CMIP5 (RCP 4.5 W m - ) 8 models CMIP5 (RCP 6. W m - ) 7 models CMIP5 (RCP 8.5 W m - ) 4 models AMIP future (CMIP 4xCO SST) 5 models Model names alphabetical order ACCESS-, ACCESS-, bcc-csm-, bcc-csm--m, BNU-ESM, CanESM, CCSM4, CNRM-CM5, CSIRO-Mk-6-, GFDL-ESMG, GFDL-ESMM, GISS-E-H, GISS-E-H-CC, GISS-E-R, GISS-E-R-CC, HadGEM-AO, HadGEM-CC, HadGEM-ES, inmcm4, IPSL-CM5A-LR, IPSL-CM5A-MR, IPSL-CM5B-LR, MIROC-ESM, MIROC-ESM-CHEM, MIROC5, MRI-CGCM, NorESM-M, NorESM-ME bcc-csm-, CCSM4, CSIRO-Mk-6-, FGOALS-s, GFDL-ESMG, GFDL-ESMM, GISS-E-H, GISS-E-R, HadGEM-AO, HadGEM-ES, IPSL-CM5A-LR, IPSL-CM5A-MR, MIROC-ESM, MIROC-ESM-CHEM, MIROC5, MRI-CGCM, NorESM-M ACCESS-, ACCESS-, bcc-csm-, bcc-csm--m, BNU-ESM, CanESM, CCSM4, CNRM-CM5, CSIRO-Mk-6-, FGOALS-s, GFDL-ESMG, GFDL-ESMM, GISS-E-R, HadGEM-AO, HadGEM-CC, HadGEM-ES, inmcm4, IPSL-CM5A-LR, IPSL-CM5A-MR, MIROC-ESM, MIROC-ESM-CHEM, MIROC5, MRI-CGCM, NorESM-M CanAM4, CNRM-CM5, HadGEM-A, MIROC5, MRI-CGCM *In addition, two CAM experiments forced with SSTs modified from either the CMIP AB or 4xCO emissions scenario are analyzed. Page 7 of 9 Macmillan Publishers Limited. All rights reserved.

18 Supplementary References Nebojsa Nakicenovic et al. in IPCC, : Emissions Scenarios (eds Nebojsa Nakicenovic & Rob Swart) Ch. 4, (Cambridge Univ. Press, ). Xue, Y., Smith, T. M. & Reynolds, R. W. Interdecadal changes of -yr SST normals during 87-. J. Climate 6, 6-6 (). Smith, T. M., Reynolds, R. W., Peterson, T. C. & Lawrimore, J. Improvements to NOAA's historical merged land-ocean surface temperature analysis (88-6). J. Climate, 8-96, doi:.75/7jcli. (8). 4 Adler, R. F. et al. The version- Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (979-present). J. Hydrometeor. 4, (). 5 Gibson, J. K. et al. ECMWF re-analysis. Project report series.. ERA description., (European Centre for Medium-Range Weather Forecasts, Reading, UK, 997). 6 Widlansky, M. J., Webster, P. J. & Hoyos, C. D. On the location and orientation of the South Pacific Convergence Zone. Climate Dyn. 6, (). 7 Fu, X. & Wang, B. The role of longwave radiation and boundary layer thermodynamics in forcing tropical surface winds. J. Climate, (999). 8 Wang, B. & Li, T. A simple tropical atmosphere model of relevance to short-term climate variations. J. Atmos. Sci. 5, 6-84 (99). 9 McGregor, S., Timmermann, A., Schneider, N., Stuecker, M. F. & England, M. H. The effect of the South Pacific Convergence Zone on the termination of El Niño events and the meridional asymmetry of ENSO. J. Climate, doi:.75/jcli-d-- (). Gill, A. Some simple solutions for heat-induced tropical circulation. Quart. J. Roy. Meteor. Soc. 6, (98). Lindzen, R. S. & Nigam, S. On the role of sea-surface temperature-gradients in forcing low level winds and convergence in the tropics. J. Atmos. Sci. 44, (987). Graham, N. E. & Barnett, T. P. Sea surface temperature, surface wind divergence, and convection over tropical cceans. Science 8, (987). Johnson, N. C. & Xie, S.-P. Changes in the sea surface temperature threshold for tropical convection. Nature Geoscience, , doi:.8/ngeo8 (). 4 Molteni, F. Atmospheric simulations using a GCM with simplified physical parametrizations. I: model climatology and variability in multi-decadal experiments. Climate Dyn., 75-9 (). 5 Bracco, A., Kucharski, F., Kallummal, R. & Molteni, F. Internal variability, external forcing and climate trends in multi-decadal AGCM ensembles. Climate Dyn., (4). 6 Kucharski, F., Molteni, F. & Bracco, A. Decadal interactions between the western tropical Pacific and the North Atlantic Oscillation. Climate Dyn. 6, 79-9 (6). 7 Kucharski, F., Bracco, A., Yoo, J. & Coauthors. A Gill-Matsuno-type mechanism explains the tropical Atlantic influence on African and Indian monsoon rainfall. Quart. J. Roy. Meteor. Soc. 5, (9). 8 Held, I. M. & Suarez, M. J. A proposal for the intercomparison of dynamical cores of atmospheric general circulation models. Bull. Amer. Meteor. Soc. 75, 85-8 (994). 9 Collins, W. D. et al. Description of the NCAR Community Atmosphere Model (CAM.). 6 pp (National Center for Atmospheric Research, Boulder, Colorado, 4). Page 8 of 9 Macmillan Publishers Limited. All rights reserved.

19 Hurrell, J. W., Hack, J. J., Phillips, A. S., Caron, J. & Yin, J. The dynamical simulation of the Community Atmosphere Model version (CAM). J. Climate 9, 6-8, doi:.75/jcli76. (6). Collins, W. D. et al. The Community Climate System Model Version (CCSM). J. Climate 9, -4 (6). Wilks, D. S. Statistical methods in the atmospheric sciences. Second edn, (Elsevier, 6). Page 9 of 9 Macmillan Publishers Limited. All rights reserved.