Embed Size (px)
Citation preview
1
Improved representation of tropical Pacific ocean-atmosphere dynamics in a coarse-1 resolution Earth system model 2
3
Ryan L. Sriver1, Axel Timmermann2, Michael E. Mann3,4,5, Klaus Keller4,5, Hugues 4 Goosse6 5 6 1Department of Atmospheric Sciences, University of Illinois at Urbana-Champaign, 7 Urbana, IL 61801, U.S.A 8 2Department of Oceanography, University of Hawaii, Honolulu, HI 96822, U.S.A 9 3Department of Meteorology, The Pennsylvania State University, University Park, PA 10 16802, U.S.A. 11 4Department of Geosciences, The Pennsylvania State University, University Park, PA 12 16802, U.S.A. 13 5Earth and Environmental Systems Institute, The Pennsylvania State University, 14 University Park, PA 16802, U.S.A. 15 6Georges Lemaitre Center for Earth and Climate Research, Earth and Life Institute, 16 Louvain-la-Neuve, Belgium 17
18
Corresponding Author: R. L Sriver ([email protected]) 19
20
In Preparation for: Journal of Climate 21
22
Draft: 11/27/2012 23
24
2
Abstract: A new anomaly coupling technique is introduced into a coarse-resolution 25
dynamic Earth system model (LOVECLIM), improving the model’s representation of 26
eastern equatorial Pacific sea surface temperature (SST) variability. The anomaly 27
coupling amplifies the surface diabatic atmospheric forcing within a Gaussian-shaped 28
patch applied in a specified region in the tropical Pacific Ocean. The scheme is 29
implemented with two other model enhancements: replacement of quasi-geostrophic 30
balance with a linear balance equation, and an improved predictive cloud scheme based 31
on empirical relationships between cloud cover and key state variables. Results are 32
presented from a perturbed physics ensemble systematically varying the parameters 33
controlling the anomaly coupling patch size, location, and the coupling amplitude. The 34
model’s optimal parameter combination is chosen through calibration against the 35
observed power spectrum of monthly mean SST anomalies in the Niño3 region during 36
the period 1960-2009. The calibrated model exhibits substantial improvement in 37
equatorial Pacific interannual SST variability and robustly reproduces El Niño-Southern 38
Oscillation (ENSO)-like variability, rivaling the skill of more comprehensive models 39
from the Coupled Model Intercomparison Project Phase 5 (CMIP5). The calibrated 40
version of LOVECLIM also simulates the discharge/recharge of the equatorial Pacific 41
ocean heat content, thus capturing key features of the underlying ENSO dynamics. 42
Because of the tractability of LOVECLIM and its consequent utility in exploring long-43
term climate variability and large ensemble perturbed physics experiments, improved 44
representation of tropical Pacific ocean-atmosphere dynamics in the model may more 45
readily allow for the investigation of the role of tropical Pacific ocean-atmosphere 46
dynamics in past climate changes. 47
48
3
1. Introduction 49
The El Niño/Southern Oscillation (ENSO) is the dominant mode of interannual 50
variability in Earth’s climate system. It affects temperature and precipitation patterns 51
worldwide, with global environmental and socioeconomic implications (McPhaden et al. 52
2006). A hallmark of ENSO is the anomalous associated sea surface temperature (SST) 53
pattern in the eastern tropical Pacific ocean, alternating between warm phases (El Niño) 54
and cold phases (La Niña). These temperature variations are caused by complex thermal 55
and dynamic ocean-atmosphere interactions and feedbacks. Correctly simulating ENSO 56
behavior in coupled Earth system models, including its internal variability and response 57
to external forcings, is of paramount importance to the climate modeling community, 58
particularly because it is currently unclear how this critical component of the Earth 59
system may be affected by anthropogenic climate change (e.g. Collins et al. 2010, Vecchi 60
and Wittenberg 2010). 61
Comprehensive coupled general circulation models, such as those used in the 62
Coupled Model Intercomparison Project (CMIP) Phases 3 and 5, have become powerful 63
tools for examining ENSO behavior and dynamics (Randall et al. 2007), as well as 64
potential changes in ENSO mean state and variability (e.g. Meehl et al. 2007). However, 65
challenges remain in simulating all of ENSO’s key statistical features, given biases in the 66
current generation of these models. For example, many models are unable to simulate a 67
realistic annual cycle of tropical Pacific SST (Jin et al. 2008), which is important for 68
understanding potential interactions between ENSO variability and the mean state (e.g. 69
Tziperman et al. 1997, Guilyardi 2006, Stein et al. 2011, McGregor et al. 2012). 70
4
CMIP5 models appear to show some improvements in simulating ENSO 71
characteristics compared to CMIP3. These improvements are primarily related to 72
decreased model spread in ENSO amplitude, resulting in an ensemble mean that is closer 73
to observations (Guilyardi et al. 2012, Kim and Yu 2012). Preliminary analysis suggests 74
power spectra of monthly SST anomalies in the Niño3 region (-5 to 5 degrees North, 210 75
to 270 degrees East) is also improved in the CMIP5 models. However, such ENSO 76
metrics should be treated with caution due to the relatively short observational record 77
used to gauge model performance. In other words, the observational record typically used 78
to evaluate model performance may not be of sufficient length to capture ENSO statistics 79
robustly (Wittenberg 2009). 80
Given the high computational cost of running comprehensive, state-of-the-art coupled 81
Earth system models, such as those developed by international modeling groups 82
participating in the CMIP3 and CMIP5 projects, it is difficult to utilize these models to 83
examine possible sensitivity of ENSO behavior to systematical variations in the relevant 84
model parameters over their physically plausible ranges. These efforts would require 85
large ensemble experiments run for multiple centuries, or even millennia when 86
considering possible interactions with parameters controlling deep ocean processes (such 87
as vertical diffusion which requires multiple millennia spin-ups to achieve approximate 88
dynamic equilibrium of the full ocean). Such efforts are currently not possible using 89
higher-resolution (e.g. 1x1 degree) fully-coupled modeling frameworks. In recent 90
decades, more simple minimum physics ENSO models have produced major insights into 91
fundamental ENSO dynamics (e.g. Karspeck et al. 2007, Bejarano and Jin 2008, 92
McGregor et al. 2012). In particular, past theoretical and numerical modeling approaches 93
5
have shown that capturing a realistic representation of tropical Pacific SST variability 94
hinges on correctly simulating: 1) the recharge/discharge of ocean heat along the equator 95
(e.g. Wyrtki 1975, Jin 1997); 2) the Bjerknes feedback (e.g. Bjerknes 1969), which 96
requires an accurate representation of tropical atmospheric dynamics and the Walker 97
circulation; and 3) oceanic Kelvin waves (e.g. McPhaden 1999). A reduced complexity 98
coupled Earth system model that is capable of capturing the essential physical processes 99
responsible for ENSO may provide an advantage over more comprehensive CMIP 100
models in examining characteristics of ENSO variability both when long-run (e.g. 101
millennial) simulations are required and where large state spaces (e.g. multi-dimensional 102
perturbed physics ensembles) are explored. 103
Here we present findings from a perturbed physics ensemble to optimize the 104
representation of ENSO variability using an Earth system model of intermediate 105
complexity (LOVECLIM). Our version of the LOVECLIM model includes several key 106
enhancements, including a new dynamic cloud scheme, an updated linear balance 107
equation, and implementation of an anomaly coupling technique to improve the 108
representation of ocean-atmosphere coupling in the equatorial Pacific. These 109
improvements lead to more realistic ENSO-like variability in the model. The motivation 110
of this work is to highlight the use of an efficient and tractable reduced complexity Earth 111
system model, which can robustly simulate ENSO-like variability. Such a model may 112
provide a useful tool for examining past and potential future changes in ENSO using 113
multi-parameter perturbed physics ensemble approaches, potentially in concert with 114
assimilation of observational and proxy climate data (e.g. Steinman et al. 2012). 115
6
The paper is organized as follows: section 2 describes the version of LOVECLIM 116
used in this study, including details about modifications and the anomaly coupling 117
parameterization; section 3 outlines the model experiments and the perturbed physics 118
ensemble design; section 4 highlights the main results and interpretations; section 5 119
provides caveats and open questions; and section 6 summarizes our main findings and 120
implications. 121
122
2. LOVECLIM 123
We use a modified version of the fully-coupled Earth system model LOVECLIM 1.2 124
(Goosse et al. 2010, Loutre et al. 2011). The LOVECLIM version used in this study 125
consists of a spectral T21, three-level quasi-geostrophic atmosphere component 126
(ECBilt2) which includes also parameterizations of the ageostrophic circulation, coupled 127
to a 3 by 3 degree ocean general circulation model with 20 vertical levels and includes a 128
thermodynamic-dynamic representation of sea ice (CLIO3). ECBilt-CLIO is coupled to a 129
land vegetation model (VECODE) that includes dynamic representation of trees, grasses, 130
and desert. LOVECLIM contains two additional model components, simulating ocean 131
biogeochemical cycles (LOCH) and ice sheet dynamics (AGISM), which are not used in 132
the present study. We introduce several notable modifications to LOVECLIM, including: 133
(i) replacement of the quasi-geostrophic balance equation in the atmosphere using a linear 134
balance equation, (ii) development of a new dynamic cloud scheme based on observed 135
relationships between cloud distribution and reanalyzed spatial fields of state variables, 136
and (iii) implementation of an anomaly coupling technique to improve the representation 137
7
of coupled ocean-atmosphere processes (and thus potentially the faithfulness of 138
interannual variability) in the equatorial Pacific. These modifications are described in the 139
following sections. 140
141
a. Balance Equation 142
We have replaced the classical quasi-geostrophic balance equation in ECBilt, which 143
diagnostically links streamfunction Ψk in layer k with the geopotential height in this 144
layer Φk, with an extended linear balance equation as presented in (Stevens et al. 1990). 145
The new linear balance equation captures a more realistic representation of tropical 146
atmospheric dynamics. The new balance equation reads: 147
148
The key innovation is that we included the effect of divergent winds in each layer (as 149
represented by the velocity potential χ on the streamfunction ψ (Ω is the angular velocity 150
of the earth’s rotation, R represents the earth’s radius, and φ and λ represent the latitude 151
and longitude, respectively). In the previous version of ECBilt the third term was omitted 152
leading to a misrepresentation of the tropical trade winds and a damping of the equatorial 153
Kelvin wave, as demonstrated in Stevens et al (1990). Model diagnostics indicate that the 154
implementation of the divergence term decreases the cold bias and intensifies the Walker 155
circulation and the divergent wind component on the equator, thus improving the physics 156
necessary to reproduce ENSO variability. 157
158
r2 (�k)�r (2⌦ sin � r k) +2⌦
R2
@�k
@�= 0
8
b. Dynamic Cloud Scheme 159
The standard version of LOVECLIM uses either a prescribed observed cloud cover field 160
based on present day climatology or a highly simplified diagnostic cloud scheme. In 161
order to address potential short-wave and long-wave cloud feedbacks, we incorporated a 162
new empirical cloud scheme derived from spatial, long-term mean fields of the European 163
Centre for Medium-Range Weather Forecasts (ECMWF) 40-Year Reanalysis Project 164
(ERA-40) (Uppala et al. 2005). 165
The new scheme predicts cloud cover in LOVECLIM based on observed 166
relationships between key state variables in ERA-40 and total cloud cover from the 167
International Cloud Climatology Project (ISCCP) satellite product (Rossow and Schiffer 168
1999), calculated using cloud fields between ~1983-2001. No differentiation into 169
different cloud levels was undertaken. We used the Alternating Conditional Expectation 170
(ACE) Value algorithm (Timmermann et al. 2001) to derive a nonparametric, multiple 171
regression: 172
173
where the observed long-term mean ISCCP total cloud coverage (y) at each grid point 174
depends on the following four long-term mean state variables at the same grid point: 175
vertical velocity at 400 hPa (x1), surface temperature (x2), precipitation (x3), and relative 176
humidity in the boundary layer (x4). By using long-term means, rather than daily or 177
monthly fields, we make the implicit assumption that short-term (synoptic, intraseasonal 178
to interannual) variability of the predictors does not rectify nonlinearly into the predicted 179
long-term mean cloud cover. The corresponding look-up tables for the general transforms 180
y = C1(x1) + C2(x2) + C3(x3) + C4(x4)
9
(Ci) are relatively smooth and are fitted using cubic polynomials (Figure 1). The 181
nonparametric fit of these 4 variables resolves more than 80% of the variance of the 182
observed annual mean clouds. We find that the global long-term mean ranges of vertical 183
velocity at 400 hPa between -0.08 to 0.06 Pa/s (Figure 1A) and of precipitation (Figure 1, 184
C) translate only into weak contributions to the total cloudiness (~0.1). Moreover, a 185
saturation of cloudiness can be found for precipitation values larger than 1 mm (per 3 186
hours). The dominant contributions to total long-term mean cloudiness are provided by 187
temperature and relative humidity. Higher relative humidity translates into larger cloud 188
fraction. The cloud-temperature relation is non-monotonous and peaks at temperatures of 189
around 280K. The reduction of cloud coverage for temperatures >280K is associated with 190
the relatively low values of cloud cover in subtropical regions. For annual mean 191
temperatures >300K, one observes again an increase in cloudiness, associated with the 192
onset of deep tropical convection. Note, that the cubic fit does not capture the increase in 193
clouds for high temperatures, which may lead to a slight underestimation of cloudiness in 194
the ITCZ regions, the Indian Ocean and Western Pacific Warm Pool in our implemented 195
scheme (Figure 2). 196
The cloud scheme is implemented in LOVECLIM by calculating cloud fraction at 197
each grid point in the model, based on the polynomial fits derived from the ACE 198
algorithm procedure, using the respective sub-daily atmospheric fields (xi) from 199
LOVECLIM. The results (Figures 2 and 3) for a pre-industrial control simulation show 200
the new dynamic cloud scheme generally reproduces the prescribed climatological cloud 201
fields used in the standard version of LOVECLIM, with overall better agreement with 202
ISCCP fields in the mid-to-high latitudes and eastern equatorial Pacific. 203
10
204
c Anomaly Coupling 205
Because of the coarse resolution of LOVECLIM, its T21 atmosphere is essentially blind 206
to relatively narrow ENSO-related SST anomalies on the equator. We introduce tropical 207
anomaly coupling to reduce this problem. The parameterization enhances the diabatic 208
forcing within a specified patch in the tropical Pacific that intensifies SST anomalies. In 209
other words, the SST variability within this patch, as seen by the atmosphere, is 210
intensified. The anomaly coupling parameterization is implemented using the following 211
equation: 212
213
where SSTa* is the perturbed atmosphere SST, SSTa is the unperturbed atmosphere SST 214
(the value obtained by the atmosphere from the coupling routine), and SSTc is the 215
climatological atmosphere SST corresponding to coupled model equilibrium (unforced) 216
conditions. SSTc is calculated from a 20-year reference of the equilibrium model 217
conditions, prior to activation of the anomaly coupling. The shape of the anomaly patch 218
is defined using a Gaussian function where the parameters P, L1, and L2, define the zonal 219
location on the equator, and the zonal and meridional length scales, respectively. Figure 4 220
shows a schematic of the anomaly coupling parameterization and parameters. The aim of 221
this parameterization is to amplify the atmospheric response to tropical SST variability 222
and hence to increase the atmosphere-ocean coupling strength. We thereby circumvent 223
the model’s inability to simulate a realistic tropical response to SST anomalies due to 224
SST ⇤a (�, �, t) =SSTa(�, �, t) + A · [SSTa(�, �, t)� SSTc(�, �)]
· exp
"�
✓�� P
L1
◆2
�✓
�
L2
◆2#
11
limited atmospheric resolution. The anomaly coupling amplifies only the SST passed to 225
the atmosphere component, so throughout the rest of the text any reference to the model’s 226
SST pertains specifically to surface temperature in the model’s atmosphere component 227
ECBilt (i.e. surface air temperature). 228
It has been noted previously that flux adjustment methods in the tropics with large 229
coupling amplitudes can lead to potential instabilities, resulting in bifurcations and 230
multiple ENSO equilibria (Neelin and Dijkstra 1995). As a cross-check, we analyzed the 231
model’s mean SST and thermocline depth in the equatorial Pacific over a wide range of 232
anomaly coupling parameter settings, and we found no instability issues for reasonable 233
values of the coupling amplitude parameter (A). The model’s mean state is relatively 234
insensitive to the anomaly coupling parameterization, exhibiting differences of less than 235
~0.1 C in mean tropical Pacific SST across the considered range in A (0.5 to 3.5). 236
Similarly, the range in zonal mean thermocline depth of the equatorial Pacific (averaged 237
from 120 to 270 degrees East) is less than ~3 meters over the considered range in A. In 238
other words, the anomaly coupling only affects tropical Pacific SST variability but not 239
the long-term mean state. We find no rectification effects of the enhanced variability on 240
the mean state of upper-ocean properties or on the annual cycle of tropical Pacific SST. 241
The mean state appears to be conserved for the range in coupling strength (A) considered 242
here. 243
244
3. Ensemble Design 245
12
We constructed two perturbed physics ensemble experiments to explore the sensitivity of 246
tropical Pacific variability to anomaly coupling parameters A, P, L1, and L2. In addition 247
to the anomaly coupling, the LOVECLIM version in both ensembles contains the 248
modifications discussed in section 2 (i.e. the use of a linear balance equation and a 249
diagnostic cloud scheme which uses the instantaneous fields of mid level vertical 250
velocity, precipitation, surface temperature and relative humidity). The main ensemble 251
consists of 49 members representing unique combinations of the anomaly coupling 252
amplitude parameter (A: 0.5, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5) and location parameter (P: 190, 253
200, 210, 220, 230, 240, 250 degrees East). The zonal and meridional length scale 254
parameters (L1 and L2) were optimized separately using a smaller preliminary ensemble. 255
We separated the ensembles into two distinct optimizations in order to keep the ensemble 256
sizes manageable as dictated by computational constraints. In the ensemble presented 257
here, the length scale parameters are held fixed at their optimal values L1=35 degrees 258
(zonal width) and L2=10 degrees (meridional width). All ensemble members are initiated 259
from a 2000-year spin-up simulation with anomaly coupling turned off. Each ensemble 260
member is re-equilibrated with the anomaly coupling turned on for an additional 2000 261
years. Atmospheric forcing conditions are held fixed at pre-industrial levels for the entire 262
ensemble experiment (both spin-up and anomaly coupling phases). 263
264
4. Results/Discussion 265
We perform a model calibration using observational ENSO statistics to determine the 266
optimal combination of anomaly coupling parameters (A and P) within the ensemble 267
13
described in section 3. The model parameters are calibrated with second moment 268
statistics using the observed spectrum of monthly mean SST anomalies averaged over the 269
Niño3 region (-5 to 5 degrees North, 210 to 270 degrees East), based on the European 270
Centre for Medium-Range Weather Forecasts (ECMWF) Ocean Re-Analysis (ORA-S3) 271
product (Balmaseda et al. 2008). Results remain generally the same for other SST 272
products (e.g. Tropical Atmosphere-Ocean (TAO) buoy array data). 273
The model’s monthly temperature anomalies are calculated relative to the coupled 274
model climatology of each ensemble member. The calibration technique emphasizes 275
model/data agreement of ENSO statistical properties, specifically the variance of the 276
Niño3 monthly temperature anomalies at frequencies typically associated with ENSO 277
variability. We compare power spectra calculated from 50-years of temperature output 278
from the equilibrated LOVECLIM ensemble with anomaly coupling to 50-years of ORA-279
S3 reanalysis (1960-2009) (Figure 5A). The optimal parameter settings are defined as the 280
ensemble member with the minimum root mean square error in the spectrum compared to 281
observed SSTs between the frequencies 0.1 and 1 year-1 (or period 1 to 10 years). 282
Optimal parameter values for the calibrated LOVECLIM are: P=240E, A=3.0 (Red curve 283
in Figure 5A). 284
Though the anomaly coupling ensemble contains ~2000 years of model output for 285
each member, we chose to analyze 50 years (after 2000 years of equilibration) so that the 286
lengths of model and data records would be consistent. As discussed in the introduction, 287
it is not clear whether the relatively short length of the instrumental record provides for a 288
robust characterization of ENSO statistics and behavior, and we preserve this uncertainty 289
in our parameter optimization—i.e. the methodology does not necessarily assume the 290
14
ENSO statistics derived from ocean reanalysis over the past 50 years are indicative of 291
millennial-scale ENSO behavior, given there is also significant internal model variability 292
of ENSO properties on decadal-to-centennial time scales. We discuss how changes in 293
ENSO behavior and spectral uncertainties affect our results and interpretations in section 294
5. 295
The spectral results of the perturbed physics ensemble show that the anomaly 296
coupling parameterization in LOVECLIM substantially improves the model’s 297
representation of interannual variability in Niño3 SST. The power spectrum for the 298
calibrated model and reanalysis show similar broad spectral peaks between 3 and 7 years, 299
indicating that LOVECLIM is capable of simulating the irregular nature of interannual El 300
Niño-La Niña cycles (Figures 5 and 6). We compare the spectral results of the calibrated 301
simulation with 35 more comprehensive coupled atmosphere-ocean general circulation 302
models (Figure 5B), comprising the majority of the models (and international modeling 303
groups) participating in the Coupled Model Intercomparison Project Phase 5 (CMIP5). 304
The CMIP5 models represent the most up-to-date and state-of-the-art coupled Earth 305
system models. Figure 5B shows that the ENSO-like variability in the calibrated version 306
of LOVECLIM rivals the performance of some of the CMIP5 models. Many of the 307
CMIP5 models overestimate the SST variance and/or simulate unrealistically short ENSO 308
cycles that are oftentimes too periodic. 309
Figure 5 highlights the improved LOVECLIM performance in simulating 310
interannual tropical Pacific SST variability. Key features of these results include: 1) the 311
calibrated version of LOVECLIM generally agrees with the observed spectrum of 312
monthly Niño3 SST anomalies, and 2) the calibrated model is comparable with more 313
15
comprehensive CMIP5 models in terms of correctly capturing the variance and frequency 314
characteristics of Niño3 SST variability. However, we also find an apparent bias in the 315
skewness of the Niño3 time series for the calibrated LOVECLIM, as shown in Figure 6. 316
The observed Niño3 SST time series is positively skewed, meaning the warm anomalies 317
associated with El Niño events are typically larger than the cold anomalies associated 318
with La Niña. The calibrated version of LOVECLIM predicts a negatively skewed time 319
series (i.e. stronger La Niña compared to El Niño). This bias is a robust feature of the 320
model and appears in all ensemble members, and the magnitude of the skewness bias 321
generally increases with the variance. The positive skewness in the observed Niño3 SST 322
time series is primarily due to large El Niño events, whose growth can be accelerated as a 323
result of westerly wind bursts (Jin et al. 2007) or nonlinear dynamical heating (Jin et al. 324
2003). We speculate that the simplified 3-layer atmosphere model in LOVECLIM does 325
not capture the necessary atmospheric dynamics to generate these large El Niño events. 326
The skewness of the calibrated LOVECLIM is ~ -0.6 C (see Figure 6). For reference, the 327
mean skewness of the 35 CMIP5 models analyzed in Figure 5 is ~0.06, with a standard 328
deviation of ~0.28. Even with the negatively skewed Niño3 time series, the calibrated 329
version of LOVECLIM reflects a major model improvement through enhanced 330
interannual variability of tropical Pacific SSTs compared to the standard model (Figure 331
6C). 332
In addition to improved time series statistics, the calibrated model also shows 333
improved patterns in SST variability within the Niño3 region compared to the standard 334
version (Figure 7). The standard model substantially underestimates the magnitude and 335
zonal extent of the equatorial Pacific temperature anomalies. While the zonal variability 336
16
is improved in the calibrated model, the model overestimates the magnitude of the SST 337
variability, primarily in the Niño3 region where the anomaly coupling is applied. 338
Further, the meridional width of the modeled SST anomalies is too large, but this bias is 339
likely due to the coarsely resolved atmospheric model grid (~ 5x5 degrees). Given these 340
limitations, the calibrated LOVECLIM is comparable to CMIP3 (Guilyardi et al. 2009) 341
and CMIP5 models (Kim and Yu 2012) in simulating spatial patterns of SST variability 342
in the Niño3 regions, and many of those models share the same biases noted here. 343
However, due to the fixed positioning of the anomaly coupling patch, this technique is 344
likely not capable of capturing potential westward progression of SST anomalies, e.g. 345
“central Pacific” flavors of El Niño (Ashok et al. 2007). The teleconnections and climate 346
implications of central Pacific El Niños can be much different than eastern Pacific events, 347
and recent evidence suggests that these events have become more frequent and more 348
intense in recent years (Lee and McPhaden 2010). Because the eastern Pacific is the 349
primary region of interest for Pacific SST variability, we focus on the Niño3 region here 350
for model/data comparison and calibration. 351
To explore LOVECLIM’s underlying ENSO dynamics, we examine the cross-352
correlation structure between Niño3 monthly SST anomalies and the variation in depth of 353
the 20-degree isotherm averaged over the entire equatorial Pacific ocean (-5 to 5 degrees 354
North, 120 to 270 degrees East). The 20-degree isotherm depth is a good indicator of the 355
ocean heat content, so its cross-correlation with Niño3 SST provides insight into the 356
model’s ability to capture the portion of the recharge-discharge of the equatorial ocean 357
heat content (Jin 1997, McPhaden et al. 2006). Comparisons of the lagged cross-358
correlation between Niño3 SST and the variation in the 20-degree isotherm are presented 359
17
in Figure 8. The result from the ORA-S3 reanalysis data for the period 1980-2009 360
generally agrees with previous observational analysis (Meinen and McPhaden 2000), 361
showing a peak cross-correlation at roughly 7 months lag (equatorial heat content leading 362
Niño3 SST). When analyzing the more complete ORA-S3 record (1960-2009), the 363
general shape of the observed lagged correlation curve is consistent with the latter portion 364
of the record (1980-2009), though the magnitude of the peak cross-correlation from 365
ORA-S3 is considerably lower (r~0.35) than for the 1980-2009 ORA-S3 record (r~0.5) or 366
the Tropical Ocean-Atmosphere (TAO) buoy data 1980-1999 (r~0.7) (Meinen and 367
McPhaden 2000). The main reason for this discrepancy is the fact that prior to 1976, 368
ENSO variability was characterized by the westward propagating SST mode that does not 369
invoke thermocline dynamics (Fedorov and Philander 2000). Only after 1976 the 370
thermocline recharge-discharge mode became more prevalent in ENSO dynamics, as 371
expressed in the higher correlations in the TAO dataset and ORA-S3 captured during this 372
period. 373
Figure 8 shows that both versions of LOVECLIM (calibrated and standard model) 374
generally reproduce some of the observed lagged correlation between monthly Niño3 375
SST and the spatial average depth of the 20-degree C isotherm over the equatorial Pacific 376
basin, analogous to the equatorial Pacific warm water volume. The calibrated version, 377
however, exhibits much improved agreement with observations, especially in the lag 378
corresponding to maximum correlation (~7 months) and the correlation structure at large 379
lags, corresponding to both discharge (negative lag) and recharge (positive lag) 380
mechanisms. At lags greater or less than ~25 months, the calibrated model exhibits better 381
agreement with observations for both reference periods (Figure 8A and 8B). This feature 382
18
is due to the improved interannual SST variability in the calibrated model, which results 383
in more pronounced ENSO-like cycles (see Figure 6). Overall, Figure 8 indicates that the 384
calibrated model is robustly reproducing the recharge of the ocean heat content, which is 385
an important mechanism for initiating El Niño events. 386
One of the key elements of the ENSO cycle is the equatorial Kelvin wave, which 387
is important in initiating the eastward expansion of water from the western Pacific warm 388
pool. Coarse resolution B-grid finite difference models (such as the CLIO3 ocean 389
component in LOVECLIM) have been shown to robustly simulate realistic oceanic 390
Kelvin waves (Ng and Hsieh 1994, Sriver et al. 2012). In analyzing the annual cycle of 391
equatorial 20 degree C isotherm depths (not shown), we find that both the calibrated and 392
standard versions of LOVECLIM produce Kelvin waves with realistic phase speeds, 393
consistent with previous results analyzing CMIP3 models (Timmermann et al. 2007). 394
To further illustrate the dynamics of the recharge-discharge phenomenon and the 395
role of equatorial Kelvin waves in our coarse-resolution model, we include a Hovmoller 396
(longitude-time) plot showing the spatially-resolved lagged correlations between monthly 397
SST anomalies averaged over the Niño3 region and the anomalies of the 20 degree C 398
isotherm depth at each model grid point (Figure 9). The figure is a spatial representation 399
of the lagged correlation structure in Figure 8. The calibrated and standard versions of 400
LOVECLIM both generally simulate the spatially-resolved lagged cross-correlation 401
patterns in accordance to observations. This response illustrates the time-lagged 402
relationship between eastern equatorial Pacific SST and thermocline depth in advance of 403
an El Niño event shown in Figure 8, in that anomalously warm ocean heat content 404
accumulates in the central-western equatorial Pacific and is redistributed eastward 405
19
through initiation of an oceanic Kelvin wave several months prior to El Niño onset. In 406
Figure 9, positive correlation between deepened thermocline depth and Niño3 SST 407
reflects the conditions prior to the onset of El Niño (negative lags), and the eastward 408
propagation of the positive anomaly reflects initiation of El Niño. Figure 9 shows that 409
both versions of LOVECLIM agree with reanalysis in generally simulating the spatial 410
structure of the lagged correlation pattern. The time evolution of the correlation patterns 411
is, however, substantially improved in the calibrated version of the model, especially for 412
large lags and inter-connections to west Pacific conditions (see also Figure 8). This result 413
provides further demonstration of the robustness of the calibrated LOVECLIM in 414
capturing the ocean-atmosphere dynamics underlying ENSO. 415
416
5. Caveats 417
Our methodology includes several simplifying assumptions. For example, the anomaly 418
coupling parameterization enhances the eastern Pacific SST variability as seen by the 419
atmosphere, but the underlying SST anomalies in the ocean model are strongly muted. 420
This muting emerges from the fact that the difference in surface air-temperature 421
anomalies (strongly affected by the anomaly coupling) and the ocean surface temperature 422
induces a damping heat flux. The major oceanic feedbacks on eastern equatorial Pacific 423
SST in the ocean model, such as thermocline depth, upwelling and zonal advective 424
feedbacks, are largely driven by atmospheric processes, including local and remote wind 425
forcing which are influenced by the anomaly coupling. Thus, the main factors 426
controlling eastern Pacific SST variability are related atmospheric effects due to the 427
20
anomaly coupling rather than local damping of surface temperature anomalies by 428
increased heat fluxes. 429
An additional caveat relates to uncertainties in our spectral and time series 430
analyses. In the results presented in section 4, we compare observed temperature records 431
from a 50-year period (1960-2009) with 50-years of model output corresponding to 432
equilibrium conditions using constant pre-industrial atmospheric forcings. The size of 433
this record is likely not of sufficient length to ensure a robust calibration, due to internal 434
modulation of ENSO in the model (e.g. Wittenberg 2009, Ault et al. In Press) and 435
interdecadal/intercentennial variability of ENSO behavior (e.g. Cane 2005, Mann et al. 436
2005). Indeed, our optimal combination of anomaly coupling parameters is slightly 437
sensitive to the size and model start date of the reference period used in the calibration 438
technique. This sensitivity is highlighted in Figure 10, which shows the spectral range in 439
monthly Niño3 SST anomalies for the calibrated version of LOVECLIM calculated from 440
different 50-year periods of model output of the equilibrium simulation. While the 441
location of the spectral peak is robust, the variance is sensitive to the choice of the 442
averaging interval within the run. In other words, the data-model calibration may result 443
in a different optimal combination of anomaly coupling parameters when analyzing 444
different 50-year intervals. The parameter values are selected only for the purposes of 445
the analysis shown here, and the primary goal of this paper is to highlight the improved 446
representation of ENSO-like variability in a reduced complexity Earth system model. 447
But, the parameter optimization may become physically meaningful in a formal data-448
model assimilation experiment, in which case a careful consideration of spectral 449
uncertainties would be necessary. These uncertainties could be reduced by: 1) using a 450
21
longer time series for calibration, 2) considering the range of spectra from multiple 451
averaging periods (e.g. as in Figure 10) during the data assimilation process through the 452
use of a probabilistic representation of the observational constraints, and/or 3) using 453
transient forcing conditions that are consistent between the model and observations. 454
455
6. Conclusions 456
We present a modified version of the LOVECLIM Earth system model that 457
improves the representation of tropical Pacific coupled ocean-atmosphere dynamics and, 458
accordingly, the performance of the resulting interannual variability in tropical Pacific 459
SST. These modifications include a new linear balance equation, an empirical diagnostic 460
cloud scheme, and an anomaly coupling technique that amplifies diabatic atmospheric 461
forcing at the surface in the eastern tropical Pacific. We find the modifications improve 462
the representation of the large-scale temperature patterns and modes of variability in the 463
tropical Pacific. When the anomaly coupling is calibrated to observations, the model 464
generally reproduces the characteristics of the observed spectrum of Niño3 SST 465
variability (frequency range and variance). Furthermore, implementation of the anomaly 466
coupling preserves LOVECLIM’s tropical climatology and improves the correlation 467
structure between Niño3 SST and upper ocean heat content of the equatorial Pacific, a 468
strong indicator that the recharge-discharge mechanism known to be important for real-469
world ENSO dynamics is at work in the model. Overall, the calibrated LOVECLIM 470
robustly reproduces observed ENSO characteristics, and its performance compares well 471
with more comprehensive state-of-the-art CMIP5 models. Given LOVECLIM’s reduced 472
22
complexity and tractability, the results shown here highlight the usefulness of the model 473
for research efforts focusing on assimilation of tropical Pacific paleo-information, 474
parameter estimation, and uncertainty quantification. 475
476
Acknowledgements 477
M.E.M., K.K., and A.T acknowledge support for this work from the ATM program of the 478
National Science Foundation (grant ATM-0902133). H.G. is Senior Research Associate 479
with the Fonds National de la Recherche Scientifique (F.R.S.-‐ FNRS-‐Belgium). The 480
authors acknowledge the World Climate Research Programme's Working Group on 481
Coupled Modelling, which is responsible for CMIP, and we thank all the climate 482
modeling groups for producing and making available their model output. For CMIP the 483
U.S. Department of Energy's Program for Climate Model Diagnosis and Intercomparison 484
provides coordinating support and led development of software infrastructure in 485
partnership with the Global Organization for Earth System Science Portals. CMIP5 486
model output was downloaded via the KNMI climate explorer (http://climexp.knmi.nl). 487
We gratefully acknowledge the European Centre for Medium-Range Weather Forecasts 488
in creating the ocean (ORA-S3) and atmosphere (ERA-40) reanalysis products. The 489
ISCCP D2 data were obtained from the International Satellite Cloud Climatology Project 490
web site http://isccp.giss.nasa.gov, maintained by the ISCCP research group at the NASA 491
Goddard Institute for Space Studies, New York, NY. Reanalysis and satellite products 492
were downloaded via the Asia-Pacific Data-Research Center (APDRC) of the 493
International Pacific Research Center (IPRC) (http://apdrc.soest.hawaii.edu). The 494
23
authors acknowledge the use of the NCAR Command Language (NCL) in the data 495
analysis and visualization herein. We are grateful to Sonya Miller and Audine Laurian 496
for additional LOVECLIM code support. 497
498
24
References 499
Ashok, K., S. K. Behera, S. A. Rao, H. Y. Weng, and T. Yamagata, 2007: El Nino 500
Modoki and its possible teleconnection. J. Geophys. Res.-Oceans, 112. 501
Ault, T. R., C. Deser, M. Newman, and J. Emile-Gaey, 2012: Characterizing internal and 502
forced low-frequency temperature variability in the equatorial Pacific during the last 503
millennium, Geophys. Res. Lett., (In Press). 504
Balmaseda, M. A., A. Vidard, and D. L. T. Anderson, 2008: The ECMWF ocean analysis 505
system: ORA-S3. Mon. Weather Rev., 136, 3018-3034. 506
Bejarano, L., and F. F. Jin, 2008: Coexistence of equatorial coupled modes of ENSO. J. 507
Clim., 21, 3051-3067. 508
Bjerknes, J., 1969: ATMOSPHERIC TELECONNECTIONS FROM EQUATORIAL 509
PACIFIC. Mon. Weather Rev., 97, 163-&. 510
Cane, M. A., 2005: The evolution of El Nino, past and future. Earth Planet. Sci. Lett., 511
230, 227-240. 512
Collins, M., and Coauthors, 2010: The impact of global warming on the tropical Pacific 513
ocean and El Nino. Nat. Geosci., 3, 391-397. 514
Dijkstra, H. A., 2006: The ENSO phenomenon: theory and mechanisms. Advances in 515
Geosciences, 6, 3-15. 516
Fedorov, A. V., and S. G. Philander, 2000: Is El Nino changing? Science, 288, 1997-517
2002. 518
Goosse, H., and Coauthors, 2010: Description of the Earth system model of intermediate 519
complexity LOVECLIM version 1.2. Geosci. Model Dev., 3, 603-633. 520
Guilyardi, E., 2006: El Nino-mean state-seasonal cycle interactions in a multi-model 521
25
ensemble. Clim. Dyn., 26, 329-348. 522
Guilyardi, E., and Coauthors, 2009: UNDERSTANDING EL NINO IN OCEAN-523
ATMOSPHERE GENERAL CIRCULATION MODELS Progress and Challenges. 524
Bull. Amer. Meteorol. Soc., 90, 325-+. 525
Guilyardi, E., and Coauthors, 2012: NEW STRATEGIES FOR EVALUATING ENSO 526
PROCESSES IN CLIMATE MODELS. Bull. Amer. Meteorol. Soc., 93, 235-238. 527
Jin, F. F., 1997: An equatorial ocean recharge paradigm for ENSO .1. Conceptual model. 528
J. Atmos. Sci., 54, 811-829. 529
Jin, F. F., S. I. An, A. Timmermann, and J. X. Zhao, 2003: Strong El Nino events and 530
nonlinear dynamical heating. Geophys. Res. Lett., 30. 531
Jin, F. F., L. Lin, A. Timmermann, and J. Zhao, 2007: Ensemble-mean dynamics of the 532
ENSO recharge oscillator under state-dependent stochastic forcing. Geophys. Res. 533
Lett., 34. 534
Jin, E. K., and Coauthors, 2008: Current status of ENSO prediction skill in coupled 535
ocean-atmosphere models. Clim. Dyn., 31, 647-664. 536
Karspeck, A. R., and J. L. Anderson, 2007: Experimental implementation of an ensemble 537
adjustment filter for an intermediate ENSO model. J. Clim., 20, 4638-4658. 538
Kim, S. T., and J. Y. Yu, 2012: The two types of ENSO in CMIP5 models. Geophys. Res. 539
Lett., 39. 540
Lee, T., and M. J. McPhaden, 2010: Increasing intensity of El Nino in the central-541
equatorial Pacific. Geophys. Res. Lett., 37. 542
Loutre, M. F., A. Mouchet, T. Fichefet, H. Goosse, H. Goelzer, and P. Huybrechts, 2011: 543
Evaluating climate model performance with various parameter sets using observations 544
26
over the recent past. Clim. Past., 7, 511-526. 545
Mann, M.E., Cane, M.A., Zebiak, S.E., Clement, A., Volcanic and Solar Forcing of the 546
Tropical Pacific Over the Past 1000 Years, Journal of Climate, 18, 447-456, 2005. 547
McGregor, S., A. Timmermann, N. Schneider, M. Stuecker, and M. England, 2012: The 548
effect of the South Pacific Convergence Zone on the termination of El Niño events 549
and the meridional asymmetry of ENSO, Journal of Climate, 25, 5566–5586, 550
doi:10.1175/JCLI-D-11-00332.1 551
McPhaden, M. J., 1999: Genesis and evolution of the 1997-98 El Nino. Science, 283, 552
950-954. 553
McPhaden, M. J., S. E. Zebiak, and M. H. Glantz, 2006: ENSO as an integrating concept 554
in Earth science. Science, 314, 1740-1745. 555
Meehl, G. A., and Coauthors, 2007: Global Climate Projections. Cambridge Univ Press, 556
747-845 pp. 557
Meinen, C. S., and M. J. McPhaden, 2000: Observations of warm water volume changes 558
in the equatorial Pacific and their relationship to El Nino and La Nina. J. Clim., 13, 559
3551-3559. 560
Neelin, J. D., and H. A. Dijkstra, 1995: OCEAN-ATMOSPHERE INTERACTION AND 561
THE TROPICAL CLIMATOLOGY .1. THE DANGERS OF FLUX CORRECTION. 562
J. Clim., 8, 1325-1342. 563
Ng, M. K. F., and W. W. Hsieh, 1994: THE EQUATORIAL KELVIN WAVE IN 564
FINITE-DIFFERENCE MODELS. J. Geophys. Res.-Oceans, 99, 14173-14185. 565
Randall, D. A., and Coauthors, 2007: Climate Models and Their Evaluation. Cambridge 566
Univ Press, 589-662 pp. 567
27
Rossow, W. B., and R. A. Schiffer, 1999: Advances in understanding clouds from 568
ISCCP. Bull. Amer. Meteorol. Soc., 80, 2261-2287. 569
Sriver, R. L., M. Huber, and L. Chafik, 2012: Excitation of equatorial Kelvin and Yanai 570
waves by tropical cyclones in an ocean general circulation model. Earth System 571
Dynamics Discussions, 3, 999-1020. 572
Stein, K., A. Timmermann, and N. Schneider, 2011: Phase Synchronization of the El 573
Nino-Southern Oscillation with the Annual Cycle. Phys. Rev. Lett., 107. 574
Steinman, B. A., M. B. Abbott, M. E. Mann, N. D. Stansell, and B. P. Finney, 2012: 575
1,500 year quantitative reconstruction of winter precipitation in the Pacific Northwest. 576
Proc. Natl. Acad. Sci. U. S. A., 109, 11619-11623. 577
Stevens, D. E., H. C. Kuo, W. H. Schubert, and P. E. Ciesielski, 1990: QUASI-578
BALANCED DYNAMICS IN THE TROPICS. J. Atmos. Sci., 47, 2262-2273. 579
Timmermann, A., H. U. Voss, and R. Pasmanter, 2001: Empirical dynamical system 580
modeling of ENSO using nonlinear inverse techniques. J. Phys. Oceanogr., 31, 1579-581
1598. 582
Timmermann, A., and Coauthors, 2007: The influence of a weakening of the Atlantic 583
meridional overturning circulation on ENSO. J. Clim., 20, 4899-4919. 584
Tziperman, E., S. E. Zebiak, and M. A. Cane, 1997: Mechanisms of seasonal - ENSO 585
interaction. J. Atmos. Sci., 54, 61-71. 586
Uppala, S. M., and Coauthors, 2005: The ERA-40 re-analysis. Q. J. R. Meteorol. Soc., 587
131, 2961-3012. 588
Vecchi, G. A., and A. T. Wittenberg, 2010: El Nino and our future climate: where do we 589
stand? Wiley Interdiscip. Rev.-Clim. Chang., 1, 260-270. 590
28
Wittenberg, A. T., 2009: Are historical records sufficient to constrain ENSO simulations? 591
Geophys. Res. Lett., 36. 592
Wyrtki, K., 1975: EL NINO - DYNAMIC-RESPONSE OF EQUATORIAL PACIFIC 593
OCEAN TO ATMOSPHERIC FORCING. J. Phys. Oceanogr., 5, 572-584. 594
595
29
Figure Captions: 596
Figure 1. From top to bottom: C1(x1), C2(x2), C3(x3), and C4(x4) (blue look-up table) 597
calculated using the ACE algorithm with ERA-40 state variables (vertical velocity at 400 598
hPa height, surface temperature, precipitation and relative humidity), and the observed 599
ISCCP cloud climatology data. Green lines represent cubic polynomials. 600
601
Figure 2. Mean total cloudiness for: A) long-term ISCCP satellite product, B) 602
LOVECLIM equilibrium simulation with the empirical ACE cloud scheme driven by 603
model fields of vertical wind velocity at 400 hPa, precipitation, surface air temperature 604
and relative humidity, and C) LOVECLIM equilibrium simulation with the standard 605
prescribed climatological clouds. Observed mean cloudiness is ~0.68, and simulated 606
mean cloudiness (in both the new diagnostic cloud scheme and the standard prescribed 607
cloud climatology) is ~0.6. LOVECLIM equilibrium simulations represent constant pre-608
industrial atmospheric forcing conditions. 609
610
Figure 3. Zonally averaged long-term mean cloud cover for ISCCP satellite product 611
(black curve), LOVECLIM equilibrium simulation with the empirical ACE cloud scheme 612
driven by model fields of vertical wind velocity at 400 hPa, precipitation, surface air 613
temperature and relative humidity (red curve), and LOVECLIM equilibrium simulation 614
with the standard prescribed climatological clouds (blue curve). LOVECLIM 615
equilibrium simulations represent constant pre-industrial atmospheric forcing conditions. 616
30
617
Figure 4. Schematic of the anomaly coupling parameterization. The parameter P 618
specifies the center location of the Gaussian-shaped patch. The zonal and meridional e-619
folding length scales of the patch size are controlled by the parameters L1 and L2, 620
respectively. An additional parameter (A) modulates the coupling strength at the air-sea 621
interface. 622
623
Figure 5. Power spectra (~10 degrees of freedom) of monthly mean surface temperature 624
anomalies averaged over the Niño3 region (-5 to 5 degrees North, 210 to 270 degrees 625
East), plotted as the normalized variance on the y-axis, as in (Dijkstra, 2006). A. The 626
observed spectrum (black curve) is derived from ERA’s ORA-S3 ocean reanalysis for the 627
period 1960-2009. Gray curves show spectra for the individual equilibrium LOVECLIM 628
ensemble members varying anomaly coupling strength and zonal patch location, using 50 629
years periods of model output. The red curve denotes the ensemble member exhibiting 630
the minimum root mean square error referenced to the observations between frequencies 631
of 0.1 and 1 (dashed black lines). For reference, we also highlight the standard 632
LOVECLIM model with no anomaly coupling and the standard cloud scheme based on 633
climatological cloud cover (blue curve). B. Comparison between the observed Niño3 634
spectrum (black curve), the calibrated LOVECLIM (red curve), and the range of spectra 635
from 35 CMIP5 model simulations (gray curves). Observed and calibrated LOVECLIM 636
spectra are plotted as in Figure 5A, and CMIP5 models are analyzed corresponding to 637
historic forcing years 1960-2009. 638
31
639
Figure 6. Time series of monthly mean surface temperature anomalies averaged over the 640
Niño3 region (-5 to 5 degrees North, 210 to 270 degrees East) for: (A) ERA’s ocean 641
reanalysis from 1960-2009, (B) the LOVECLIM model calibrated to the observed power 642
spectrum (see Figure 4), and (C) the out-of-box version of LOVECLIM. All time series 643
are smoothed using a 5-month running mean. The shading represents El Niño (red) and 644
La Niña (blue) events. The variance/skewness of each smoothed time series is (A) 645
0.64/0.98, (B) 0.65/-0.65, (C) 0.16/-0.64. 646
647
Figure 7. Standard deviations of monthly modeled and observed surface temperature for 648
(A) ORA-S3 ocean reanalysis (1960-2009), (B) the LOVECLIM model calibrated to the 649
observed power spectrum (see Figure 5), and (C) the out-of-box version of LOVECLIM. 650
651
Figure 8. Lagged cross correlation between monthly mean surface temperature 652
anomalies averaged over the Niño3 region (-5 to 5 degrees North, 210 to 270 degrees 653
East) and monthly mean anomalies of the 20 degree Celsius isotherm depth averaged 654
over the equatorial Pacific (-5 to 5 degrees North, 120 to 270 degrees East). The black 655
curve shows the observed relationship derived from ORA-S3 ocean reanalysis during the 656
periods (A) 1960-2009 and (B) 1980-2009. Red and blue curves in both (A) and (B) 657
represent the calibrated and out-of-box version of the LOVECLIM model. Analyzed 658
time periods for the model are the same in both (A) and (B) and consistent with 659
32
description in Figure 5. Negative lags indicate that equatorial Pacific thermocline depth 660
anomalies are leading Niño3 surface temperature anomalies. 661
662
Figure 9. Longitude-lag plots of cross correlation between monthly mean surface 663
temperature anomalies averaged over the Niño3 region (-5 to 5 degrees North, 210 to 270 664
degrees East) and monthly mean anomalies of the 20 degree Celsius isotherm depth at 665
each longitude (averaged between -5 to 5 degrees North), for: A) ORA-S3 ocean 666
reanalysis (1960-2009), B) the calibrated version of LOVECLIM, and C) the out-of-box 667
version of LOVECLIM model. Negative lags indicate that equatorial Pacific thermocline 668
depth anomalies are leading Niño3 surface temperature anomalies. 669
670
Figure 10: Power spectra of monthly mean surface temperature anomalies averaged over 671
the Niño3 region. The black curve represents observational product using ORA-S3 672
reanalysis (1960-2009), as shown in Figure 4A. The gray shading highlights the range of 673
the spectra calculated from 15 different 50-year intervals of the calibrated LOVECLIM 674
model. We use a 50-year moving average (with 10-year offset), centered on the data-675
model calibration period. The red curve denotes the mean of the 15 different model 676
spectra. 677
678
679
33
Figures 680
681
Figure 1. From top to bottom: C1(x1), C2(x2), C3(x3), and C4(x4) (blue look-up table) 682
calculated using the ACE algorithm with ERA-40 state variables (vertical velocity at 400 683
hPa height, surface temperature, precipitation and relative humidity), and the observed 684
ISCCP cloud climatology data. Green lines represent cubic polynomials. 685
686
−0.1 −0.08 −0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.080.55
0.6
0.65
0.7
t400 hPa
c1
220 230 240 250 260 270 280 290 300
−0.2
−0.1
0
0.1
TS [K]
c2
0 1 2 3 4 5x 10−3
−0.1
0
0.1
Precipitation
c3
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
−0.2
−0.1
0
relative humidity
c4A.
B.
C.
D.
34
687
Figure 2. Mean total cloudiness for: A) long-term ISCCP satellite product, B) 688
LOVECLIM equilibrium simulation with the empirical ACE cloud scheme driven by 689
model fields of vertical wind velocity at 400 hPa, precipitation, surface air temperature 690
A.
B.
C.
35
and relative humidity, and C) LOVECLIM equilibrium simulation with the standard 691
prescribed climatological clouds. Observed mean cloudiness is ~0.68, and simulated 692
mean cloudiness (in both the new diagnostic cloud scheme and the standard prescribed 693
cloud climatology) is ~0.6. LOVECLIM equilibrium simulations represent constant pre-694
industrial atmospheric forcing conditions. 695
696
697
36
698
Figure 3. Zonally averaged long-term mean cloud cover for ISCCP satellite product 699
(black curve), LOVECLIM equilibrium simulation with the empirical ACE cloud scheme 700
driven by model fields of vertical wind velocity at 400 hPa, precipitation, surface air 701
temperature and relative humidity (red curve), and LOVECLIM equilibrium simulation 702
with the standard prescribed climatological clouds (blue curve). LOVECLIM 703
equilibrium simulations represent constant pre-industrial atmospheric forcing conditions. 704
705
37
706
707
Figure 4. Schematic of the anomaly coupling parameterization. The parameter P 708
specifies the center location of the Gaussian-shaped patch. The zonal and meridional e-709
folding length scales of the patch size are controlled by the parameters L1 and L2, 710
respectively. An additional parameter (A) modulates the coupling strength at the air-sea 711
interface. 712
713
L2
L1
PEquator
38
714
Figure 5. Power spectra (~10 degrees of freedom) of monthly mean surface temperature 715
anomalies averaged over the Niño3 region (-5 to 5 degrees North, 210 to 270 degrees 716
East), plotted as the normalized variance on the y-axis, as in (Dijkstra, 2006). A. The 717
observed spectrum (black curve) is derived from ERA’s ORA-S3 ocean reanalysis for the 718
period 1960-2009. Gray curves show spectra for the individual equilibrium LOVECLIM 719
ensemble members varying anomaly coupling strength and zonal patch location, using 50 720
A.
B.
39
years periods of model output. The red curve denotes the ensemble member exhibiting 721
the minimum root mean square error referenced to the observations between frequencies 722
of 0.1 and 1 (dashed black lines). For reference, we also highlight the standard 723
LOVECLIM model with no anomaly coupling and the standard cloud scheme based on 724
climatological cloud cover (blue curve). B. Comparison between the observed Niño3 725
spectrum (black curve), the calibrated LOVECLIM (red curve), and the range of spectra 726
from 35 CMIP5 model simulations (gray curves). Observed and calibrated LOVECLIM 727
spectra are plotted as in Figure 5A, and CMIP5 models are analyzed corresponding to 728
historic forcing years 1960-2009. 729
730
40
731
Figure 6. Time series of monthly mean surface temperature anomalies averaged over the 732
Niño3 region (-5 to 5 degrees North, 210 to 270 degrees East) for: (A) ERA’s ocean 733
reanalysis from 1960-2009, (B) the LOVECLIM model calibrated to the observed power 734
spectrum (see Figure 4), and (C) the out-of-box version of LOVECLIM. All time series 735
A.
B.
C.
41
are smoothed using a 5-month running mean. The shading represents El Niño (red) and 736
La Niña (blue) events. The variance/skewness of each smoothed time series is (A) 737
0.64/0.98, (B) 0.65/-0.65, (C) 0.16/-0.64. 738
42
739
740
Figure 7. Standard deviations of monthly modeled and observed surface temperature for 741
(A) ORA-S3 ocean reanalysis (1960-2009), (B) the LOVECLIM model calibrated to the 742
observed power spectrum (see Figure 5), and (C) the out-of-box version of LOVECLIM. 743
744
A.
B.
C.
43
745
Figure 8. Lagged cross correlation between monthly mean surface temperature 746
anomalies averaged over the Niño3 region (-5 to 5 degrees North, 210 to 270 degrees 747
East) and monthly mean anomalies of the 20 degree Celsius isotherm depth averaged 748
over the equatorial Pacific (-5 to 5 degrees North, 120 to 270 degrees East). The black 749
curve shows the observed relationship derived from ORA-S3 ocean reanalysis during the 750
periods (A) 1960-2009 and (B) 1980-2009. Red and blue curves in both (A) and (B) 751
represent the calibrated and out-of-box version of the LOVECLIM model. Analyzed 752
A.
B.
44
time periods for the model are the same in both (A) and (B) and consistent with 753
description in Figure 5. Negative lags indicate that equatorial Pacific thermocline depth 754
anomalies are leading Niño3 surface temperature anomalies. 755
756
45
757
758
Figure 9. Longitude-lag plots of cross correlation between monthly mean surface 759
temperature anomalies averaged over the Niño3 region (-5 to 5 degrees North, 210 to 270 760
degrees East) and monthly mean anomalies of the 20 degree Celsius isotherm depth at 761
each longitude (averaged between -5 to 5 degrees North), for: A) ORA-S3 ocean 762
reanalysis (1960-2009), B) the calibrated version of LOVECLIM, and C) the out-of-box 763
version of LOVECLIM model. Negative lags indicate that equatorial Pacific thermocline 764
depth anomalies are leading Niño3 surface temperature anomalies. 765
A. B. C.
46
766
767
Figure 10: Power spectra of monthly mean surface temperature anomalies averaged over 768
the Niño3 region. The black curve represents observational product using ORA-S3 769
reanalysis (1960-2009), as shown in Figure 4A. The gray shading highlights the range of 770
the spectra calculated from 15 different 50-year intervals of the calibrated LOVECLIM 771
model. We use a 50-year moving average (with 10-year offset), centered on the data-772
model calibration period. The red curve denotes the mean of the 15 different model 773
spectra. 774
