Consistency of 20th century sea level pressure trends as simulated by a coupled and uncoupled GCM

Authors

  • Ioana Colfescu,

    Corresponding author
    1. Department of Atmospheric, Oceanic and Earth Sciences, George Mason University, Fairfax, Virginia, USA
    • Corresponding author: I. Colfescu, Department of Atmospheric, Oceanic and Earth Sciences, George Mason University, Fairfax, VA 22030, USA. (colfescu@cola.iges.org)

    Search for more papers by this author
  • Edwin K. Schneider,

    1. Department of Atmospheric, Oceanic and Earth Sciences, George Mason University, Fairfax, Virginia, USA
    2. Center for Ocean-Land-Atmosphere Studies, Calverton, Maryland, USA
    Search for more papers by this author
  • Hua Chen

    1. Department of Atmospheric, Oceanic and Earth Sciences, George Mason University, Fairfax, Virginia, USA
    Search for more papers by this author

Abstract

[1] A coupled (ocean-atmosphere) general circulation model (CGCM) and an uncoupled atmospheric general circulation model forced with the SST and external forcing of the coupled model simulate similar 2 m air temperature (TS) trends and also similar sea level pressure (SLP) trends for the latter half of the 20th century. This suggests that the inability of atmospheric models forced by observed SST and external forcing to reproduce observed SLP trends in the Indian Ocean could be due to model bias rather than lack of coupling. The internally generated TS trend in the CGCM is found to be small in comparison to the externally forced component. Intrinsic atmospheric noise explains most of the CGCM's internally generated high-latitude SLP trend, while in low latitudes, the response of the SLP trend to the internally generated SST trend is important.

1 Introduction

[2] Sea level pressure (SLP) trends, as simulated by atmospheric general circulation models (AGCMs) forced by observed SST and estimated external forcing for the latter half of the 20th century, do not agree with observed trends, particularly in the Indian Ocean. Using the HadAM3 AGCM, Copsey et al. [2006] found that the simulated SLP trend in the Indian Ocean was negative, while the observed trend was positive [Vecchi et al., 2006]. The explanations they suggested for this result were that either the SST-forced AGCMs have different responses in comparison to coupled models (even with the same SST) or the AGCM has a biased response to SST forcing. Although Deser et al. [2012] found better agreement between the observed DJF tropical SLP trend and the corresponding trend in the CAM3 AGCM, differences from the observed trend appear to have same sign and similar structure as in Copsey et al. [2006]. Meng et al. [2012] found a similar inconsistency from forcing the ECHAM5 AGCM with observed SST, although they found no inconsistency in a perfect model comparison (AGCM forced by the SST simulated by the coupled (ocean-atmosphere) general circulation model, CGCM) using the MPI CGCM and the ECHAM5 AGCM. The impact of observed SST trends in the latter half of 20th century on the trends of 500 hPa height (Z500) in Northern Hemisphere winter was analyzed by Schneider et al. [2003] using an ensemble of simulations made with the COLA AGCM. The study found that intraensemble Z500 trend variability, attributable to intrinsic atmospheric noise, was comparable to the SST-forced trend in high latitudes. The SST-forced Z500 trend was attributed primarily to forcing by the tropical SST trend.

[3] The mechanism of the internal variability of the SLP trend in CGCM simulations has been addressed by Deser et al. [2012] using Community Climate System Model version 3 (CCSM3). The externally forced trend and its internal variability were evaluated from an ensemble of future climate simulations. The main source of intraensemble variability in the CGCM-simulated SLP trends depended on the region of study—in middle and high latitudes, the structure of the coupled variability was similar to that of the intrinsic atmospheric variability, while in the tropics, the structures in the coupled and uncoupled simulations differed, demonstrating an important role for ocean-atmosphere coupling.

[4] Recently, Chen et al. [2013] examined the similarity between the SST-forced response in a CGCM current climate control simulation (constant external forcing), and the AGCM component of the CGCM, forced by the CGCM-generated SST in the Community Climate System Model version 3 (CCSM3). The differences they found between the CGCM and AGCM fields were attributed to forcing of SST by intrinsic atmospheric noise in the CGCM, but not the AGCM, essentially as described by the simple model of Barsugli and Battisti [1998].

[5] If coupled and uncoupled models have different responses with/to the same SST, as suggested by Copsey et al. [2006] and if the SST-forced AGCM does not correctly simulate the atmospheric response due to lack of coupling, conclusions drawn using SST-forced AGCM simulations could be in error. Our study examines the issue of whether atmosphere-ocean coupling is essential to simulate the externally forced SLP trends and identifies the role of intrinsic atmospheric noise in the coupled internal variability. The experiments use a perfect model framework, eliminating differences between CGCM and AGCM due to model bias and isolating the role of coupling.

2 Models and Methods

[6] The experimental design and analysis extend these used by Chen et al. [2013] to include 20th century external forcing and to examine trends in the latter half of the 20th century. The SLP trends in a CGCM ensemble and in an AGCM ensemble forced by the CGCM SST, sea ice, and external forcing are compared in a situation where uncertainty due to model bias is eliminated through the experimental design. Additionally, the design allows attribution of the trends in the CGCM to external forcing and internally generated variability, and furthermore, for the internally generated variability to be decomposed into two components—a component where coupling to the SST internal variability is important and a noise component. The models used were the coupled model, the Community Climate System Model (CCSM3) [Collins et al., 2006a] and its atmospheric component, the Community Atmosphere Model (CAM3) [Collins et al., 2006b]. The atmospheric model for both the coupled and uncoupled simulations had T42 spectral resolution in the horizontal and 26 levels. The ocean model configuration was an approximately 1° × 1° horizontal grid and 40 levels.

[7] A CGCM simulation (CONTROL) for the period 1870–1998 with prescribed 20th Century historical forcing represents the observations. Extending the approach of Chen et al. [2013], the results for field V in the CGCM, VCGCM, are decomposed into externally forced, inline image, and internally generated variability, inline image.

display math(1)

[8] For this study, V represents the SLP or surface air temperature (TS) trends evaluated over the 1950–1996, the period chosen for comparison with the results of Copsey et al. [2006].

[9] Similarly, the internally generated atmospheric variability is separated into the atmospheric response to the internally generated coupled SST variability, inline image, and the intrinsic atmospheric noise, inline image:

display math(2)

[10] In order to estimate the externally forced trends, an ensemble of four additional CCSM3 simulations with prescribed 20th Century historical external forcing was performed for the period 1870–1998. The differing initial conditions of the coupled ensemble members were obtained by choosing arbitrarily from a 500 year preindustrial control run with external forcing fixed at 1870 levels (archived as run b30.043 in the Community Earth System Model database at the National Center for Atmospheric Research). inline image is determined by

display math(3)

where the subscript refers to the ith ensemble member (including CONTROL), and the brackets represent ensemble averaging. inline image is then found using equations (1) and (3) by

display math(4)

[11] An ensemble of six AGCM simulations forced by the CONTROL SST and external forcing was made in order to decompose inline image into inline image and inline image, as well as for comparison with the CONTROL trend. We improved the consistency of the SST forcing for the AGCM ensemble compared to Chen et al. [2013] by evaluating it from the CGCM monthly ocean model output rather than from the atmospheric model output. Taking the AGCM ensemble mean, inline image, as the SST and externally forced trend, inline image is found by

display math(5)

[12] Then, inline image is calculated from equations (2) and (5):

display math(6)

[13] Two statistical tests are applied to the trends. One is a two-sided t test of the significance of the linear trend with respect to the yearly residuals from the trend (47 degrees of freedom if the lag-1 autocorrelation of the residuals is small, a condition satisfied for the SLP). The other is a two-sided t test of the reproducibility of the trend in the ensemble members, where the residuals are trends of the N individual ensemble members minus the ensemble mean trend (N − 1 degrees of freedom) as in Deser et al. [2012]. This test will be called the “reproducibility” of the trend to distinguish it from the first test.

3 Results

[14] inline image (Figure 1a) shows reproducible positive trends in the Pacific, over India, and in the midlatitude North Atlantic, and reproducible negative trends in high latitudes of both hemispheres. The tropical Indian and Atlantic Oceans have large regions of negative SLP trend, but these are reproducible only in the Indian Ocean near 15°S and in the eastern tropical North Atlantic. inline image (Figure 1c) has significant positive values over most of the globe. Compared with the observed SST trend between ±40° latitude shown by Copsey et al. [2006], the externally forced trend in CCSM3 is similar in the Indian and Atlantic Oceans but more uniform in the Pacific and weaker in the eastern Pacific.

Figure 1.

(a) inline image and (c) inline image for 1950–1996. Standard deviation of the trends in the CGCM ensemble for (b) SLP and (d) TS. Units are hPa per 100 years in Figures 1a and 1b, and K per 100 years in Figures 1c and 1d. Areas in Figures 1a and 1c where the reproducibility is significant at the 10% level are shaded, and other areas are plotted with contours only.

[15] The internally generated variabilities in the CGCM ensemble, taken as the unbiased standard deviations (SD) of inline image and inline image (Figures 1b and 1d, respectively) are small in the tropics and increase with latitude. The increase is more pronounced in the case of the SLP. The increase of SD of inline image with latitude is probably related in part to the internal atmospheric variability of NAO-like or annular modes of trend variability [Schneider et al., 2003; Deser et al., 2012]. However, these modes do not appear to be responsible for the lack of reproducibility of the SLP trends in the middle latitudes of both hemispheres because the annular mode structures are clearly seen in the externally forced response (Figure 1a), and also because the irreproducible regions occur near the nodal surfaces of the annular modes.

[16] SLPCGCM and TSCGCM (Figure 2) have areas of significant trends substantially smaller than the areas of reproducible trends shown in Figure 1. The AGCM ensemble mean SLP trend, inline image (Figure 3a), has a spatial structure that is similar to SLPCGCM (area-weighted correlation 0.69 globally, 0.84 between ±30° latitude). inline image is reproducible in regions where SLPCGCM trend is significant, but also in additional regions in the North Atlantic, North Pacific, and Indian Oceans and Indian subcontinent. Over land and sea ice, inline image (Figure 3c) increases toward higher latitudes. TSCGCM (Figure 2b) and inline image are similar even in regions where the significance of the trend is low, as long as the trends are reproducible. inline image is reproducible everywhere, of course, since the SST is identical in all of the AGCM ensemble members.

Figure 2.

CONTROL coupled run linear trends for (a) SLP (hPa century−1) and (b) TS (°C century−1). Areas where the trend is significant at the 10% level are shaded, and other areas are plotted with contours only.

Figure 3.

(a) inline image, (b) SD of inline image, (c) inline image, (d) SD of inline image for the six-member AGCM ensemble forced by the CONTROL coupled run SST. Units and significance shading as in Figure 1 for the respective panels.

[17] The variability of the intrinsic atmospheric noise trend in the AGCM ensemble is calculated as the SD of inline image and inline image (Figures 3b and 3d). SD of inline image is smaller in the tropical and subtropical regions and increases with latitude. This increase is probably partly the result of internal atmospheric variability related to the NAO and annular modes. SD of inline image over land shows the highest values over the Euro-Asian continent and can be attributed to internal atmospheric variability, perhaps involving land surface feedbacks, since the SST and external forcing is the same in all of the AGCM ensemble members.

[18] SD of inline image (Figure 1b) and SD of inline image (Figure 3b) have similar structures. An f test shows that the ratio of the intraensemble SLP trend variability between the CGCM and AGCM is not different from one at the 10% level except for a few isolated, small regions over land.

[19] inline image (Figure 4a) and inline image (Figure 4b) have similar structures in middle and high latitudes, but large areas where the sign differs in low latitudes, for example, in the Indian Ocean. inline image (Figure 4c) is comparable to or larger in magnitude than inline image in low latitudes and resembles the phase of the Southern Oscillation associated with La Nina there. inline image (Figure 4d) has elements of a La Nina-like structure, with warming in the western equatorial Pacific and cooling in the east, consistent with a positive coupled feedback between inline image and inline image. Cold (warm) temperature centers in the North Pacific (North Atlantic) correspond to increased (decreased) westerlies inferred from inline image. The configuration is consistent with the internally generated SST trends in those regions being forced by the SLP noise.

Figure 4.

Trends in CONTROL: (a) inline image, (b) inline image (c) inline image, and (d) inline image. Units (hPa century−1) and upper color bar for Figures 4a–4c, (°C century−1) and lower color bar for Figure 4d.

4 Conclusions

[20] The role of atmosphere-ocean coupling in the simulation of 1950–1996 TS and SLP trends was investigated in a perfect model setting, using coupled and SST-forced simulations with specified 20th century external forcing. We found that the SLP trend of a coupled simulation was well reproduced over most of the globe by the mean trend of an AGCM ensemble forced by the SST from the coupled simulation. Our results show that the major characteristics of the trends are the response to the SST and external forcing in either the coupled or uncoupled models. While differences from the observed SLP trend over the Indian Ocean are seen both here and in the AGCM simulations of Copsey et al. [2006], our results explain the error as a bias in the model, seen both with and without coupling rather than an error due to lack of coupling alone. Attribution of the trend errors to atmospheric model bias unrelated to coupling is supported by the model dependence of earlier results, i.e., the apparently better success of CAM3 in simulating the observed tropical SLP trends when forced by observed SST [Deser et al., 2012].

[21] It is important to note that our simulations were made with a single relatively low-resolution model. The results might change in the model world or in comparison to the real world if a set of very much higher horizontal resolution climate models [e.g., Scaife et al., 2011] were used. In such a case, qualitative changes in the strength of AGCM and CGCM atmospheric-ocean interactions, especially in the extratropics, should be expected [Minobe et al., 2008]. We recommend that similar perfect model experiments to those presented here be undertaken using other models, and especially high resolution coupled models, in order to determine the model dependence of the results.

Acknowledgments

[22] The contributions of Colfescu et al. were supported by NSF grants ATM-0653123 and AGS-1137902. Schneider was also supported by NSF grants ATM-0830068 and ATM-0830062, NOAA grant NA09OAR4310058, and NASA grant NNX09AN50G. Computer resources were provided by NASA on the Pleiades computer, and by the NCAR CISL. Our work is part of the CLIVAR International Climate of the Twentieth Century Project (C20C). Data analyses and plotting were done using GrADS.

Advertisement