Sensitivity of tropical precipitation extremes to climate
change
The MIT Faculty has made this article openly available. Please share
how this access benefits you. Your story matters.
Citation
O’Gorman, Paul A. “Sensitivity of Tropical Precipitation Extremes
to Climate Change.” Nature Geoscience 5.10 (2012): 697–700.
CrossRef. Web.
As Published
http://dx.doi.org/10.1038/ngeo1568
Publisher
Nature Publishing Group
Version
Author's final manuscript
Accessed
Mon May 23 10:56:52 EDT 2016
Citable Link
http://hdl.handle.net/1721.1/77904
Terms of Use
Article is made available in accordance with the publisher's policy
and may be subject to US copyright law. Please refer to the
publisher's site for terms of use.
Detailed Terms
Sensitivity of tropical precipitation extremes to climate
change
Paul A. O’Gorman
1
Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA
Precipitation extremes increase in intensity over many regions of the globe in simulations of
a warming climate1–3 . The rate of increase of precipitation extremes in the extratropics is
consistent across global climate models, but the rate of increase in the tropics varies widely,
depending on the model used3 . The behaviour of tropical precipitation can, however, be constrained by observations of interannual variability in the current climate4–6 . Here I show
that, across state-of-the-art climate models, the response of tropical precipitation extremes
to interannual climate variability is strongly correlated with their response to longer-term
climate change, although these responses are different. I then use satellite observations to
estimate the response of tropical precipitation extremes to the interannual variability. Applying this observational constraint to the climate simulations and exploiting the relationship
between the simulated responses to interannual variability and climate change, I estimate a
sensitivity of the 99.9th percentile of daily tropical precipitation to climate change at 10%
per K of surface warming, with a 90% confidence interval of 6-14% K−1 . This tropical
sensitivity is higher than expectations for the extratropics3 of about 5% K−1 . The inferred
percentage increase in tropical precipitation extremes is similar when considering only land
regions, where the impacts of extreme precipitation can be severe.
1
Increases in precipitation extremes (defined here as high percentiles of daily precipitation) associated with climate change would have important impacts, such as on flooding, soil erosion, and
landslides7, 8 . Changes in the distribution of precipitation are expected in a warmer climate because
of the dependence of the saturation vapor pressure of water on temperature3, 9, 10 . Observations suggest that precipitation extremes may have increased in intensity as the climate warmed in recent
decades, at least regionally11, 12 . Extratropical precipitation extremes consistently increase at close
to the “thermodynamic” rate (∼ 6%K−1 ) in simulations with global climate models, corresponding
to little change in the magnitude of vertical winds associated with the extremes3 . The thermodynamic rate is similar in the tropics, but the simulated rate of increase of tropical precipitation extremes may be substantially lower or higher depending on the model used, with close to no change
in some models and rates of increase of up to 30%K−1 in others3 . This inter-model scatter likely
results from the strong dependence of tropical precipitation on moist-convective processes that
must be represented by subgrid parameterizations in global climate-change simulations13 . Recent
idealized studies of radiative-convective equilibrium using models that resolve convective-scale
dynamics found that intense precipitation increases with warming at close to the thermodynamic
rate14, 15 , but a different response could occur in the tropics because of convective organization and
large-scale dynamics that were not included in the idealized studies.
Given that climate models simulate robust large-scale patterns of temperature change in the
tropics16 but have difficulty in reliably simulating tropical precipitation extremes, it is reasonable to ask whether observations of temperature and precipitation may be used to help constrain
the expected response of tropical precipitation extremes to climate change. Studies of observed
2
variability within the current climate suggest stronger increases in certain types of precipitation
extremes with warming than given by the thermodynamic rate4, 5, 17 . But the sensitivity of precipitation extremes to temperature changes within a given climate cannot be assumed to be the same as
the sensitivity under climate change. For example, interannual variability in tropical precipitation
extremes is largely related to El Niño-Southern Oscillation (ENSO) which has distinct temperature
patterns and dynamics compared with global warming18, 19 .
Here, I show that the sensitivity of tropical precipitation extremes to temperature changes
associated with variability is in fact strongly correlated across models with the sensitivity to global
warming (although the sensitivities are not the same). I use this relationship between sensitivities to variability and climate change, together with observations of variability, to constrain the
climate-change sensitivity of tropical precipitation extremes. Similar approaches have previously
been used to constrain snow-albedo feedback20 and climate sensitivity21 using the observed seasonal cycle. An important feature of the approach presented here is that it is physically plausible
that the same subgrid parameterizations responsible for moist convection (and the division between convective and stratiform rainfall) cause the inter-model scatter in the response of tropical
precipitation extremes to both variability and climate change.
The default simulations used involve 18 climate models from the World Climate Research
Programme’s (WCRP’s) Coupled Model Intercomparison Project phase 3 (CMIP3) archive. Simulated variability is analyzed over the period 1981-2000 in the 20C3M simulation, and climate
change is calculated as the difference between this period and 2081-2100 in the SRES-A1B sce-
3
nario (slightly different time periods are used for some models; see Supplementary Information).
The analysis was also repeated for a subset of “good-ENSO” climate models that have been identified as having ENSO temperature variability similar to that found in observations22 , and for simulations drawn from the recently-available CMIP5 archive. The default precipitation observations are
taken from the Special Sensor Microwave Imager (SSM/I) using the dataset from Remote Sensing
Systems (RSS) over the period 1991-200823 , and four other observational precipitation datasets
are used for comparison. Surface temperatures are taken from the National Oceanic and Atmospheric Administration Merged Land-Ocean Surface Temperature Analysis24 . Climate change is
calculated over the whole tropics or over tropical land, while variability is generally analyzed over
the tropical oceans because this is found to give the strongest constraint on sensitivities to climate
change. Results are also reported using variability over the whole tropics.
Time series are first constructed of precipitation extremes and mean surface temperature
over the tropical oceans between 30S and 30N (Methods). The influence of ENSO on precipitation
extremes over the tropical oceans is clearly evident in observations, as shown in Fig. 1 for the
99.9th percentile of daily precipitation and consistent with results from previous studies4–6 . Positive anomalies in surface temperature tend to be associated with positive anomalies in precipitation
extremes; the calculated sensitivity to surface temperature (Methods) is 25%K−1 with a 90% confidence interval of 16 to 36%K−1 . A similar behavior is found in the climate-model simulations,
but with different time series of surface temperature because coupled models are considered, and
with very different sensitivities depending on the climate model used (Fig. 1 and Supplementary
Fig. S1).
4
Sensitivities to climate change are calculated over the whole tropics in the climate model
simulations and are normalized by changes in mean surface temperature (Methods). For the 99.9th
percentile of precipitation, the sensitivities to climate change are strongly correlated across models
with the sensitivities to variability (Fig. 2), with a correlation coefficient of 0.866. The relationship between sensitivities is further quantified using ordinary-least-squares regression (Table S1).
The regression line passes close to the origin, and the sensitivity to variability is greater than the
sensitivity to climate change by roughly a factor of 2.5.
The relationship between the sensitivities to variability and climate change, together with
the observed sensitivity to variability, yields an inferred sensitivity to climate change. For the
99.9th percentile of precipitation, the inferred sensitivity to climate change is 10%K−1 , which is
higher than what most of the models simulate (Fig. 2). Uncertainty is estimated by a bootstrapping procedure involving resampling of the models used and 12-month blocks in the observed and
simulated time series (Methods). The resulting 90% confidence interval of 6 to 14%K−1 is substantially narrower than the inter-model scatter of 2 to 23%K−1 , clearly illustrating the value of the
observational constraint.
The inferred sensitivity to climate change increases with percentile from the 98th to the
99.9th percentile and decreases slightly to the 99.95th percentile (Fig. 3a); it exceeds the multimodelmedian sensitivity (and by as much as 68%), although the associated 90% confidence interval does
not exceed the multimodel median for all percentiles. multimodel median continues to increase
with percentile. Both intermodel scatter and the strength of the relationship between sensitivities
5
for variability and climate change increase with percentile (Table S1), such that the observational
constraint is more useful for higher percentiles of precipitation.
The inferred sensitivities were also calculated to climate change over land only, with variability over the ocean as before. A strong relationship holds between climate change and variability for
the higher percentiles of precipitation considered (Fig. S2, Table S2), and the inferred sensitivities
to climate change over land approach the sensitivities over the whole tropics at these percentiles
(Fig. 3b). This similar response over land and the whole tropics occurs despite ∼60% greater
surface warming over land than ocean (all sensitivities to climate change are normalized by temperature changes over the whole tropics for ease of comparison). Indeed, the percentage changes in
precipitation extremes in the simulations of climate change are close to equal over land and ocean
across all the models (Fig. S3), which is likely related to the importance of oceanic water vapor
sources for precipitation over land and to decreases in land surface-air relative humidity under
global warming25 .
For the “good-ENSO” subset of models (Supplementary Information), the relationship between sensitivities to climate change and variability is very tight for the 99.9th percentile of precipitation (Fig. S4), with a correlation coefficient of 0.997, and the resulting inferred sensitivities
to climate change are similar to what is obtained using all the models (Table S1). This robustness
suggests that the inferred response to climate change is not strongly affected by the relativelypoor quality of simulated ENSO temperature variability in some of the model simulations. Similar
results are also obtained using the CMIP5 simulations; the relationship between sensitivities to
6
variability and climate change is less tight than in CMIP3 for the 99.9th percentile of precipitation
(Fig. S5 and Table S3), but the inferred sensitivities to climate change are only slightly higher at
11%K−1 versus 10%K−1 (Fig. 3c).
To help assess uncertainties related to observational estimation of precipitation (which are
not included in the estimates of uncertainty given above), the analysis was repeated for four alternative observational precipitation datasets: the Goddard Profiling Algorithm (GPROF)26 applied to
SSM/I radiances, a dataset from the Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (TMI), the 1-degree daily merged dataset from the Global Precipitation Climatology Project
(GPCP 1DD)27 , and the TRMM 3B42 merged daily dataset28 . Most of these alternative precipitation datasets cover a shorter time period than the default precipitation dataset, but they all give
similar inferred sensitivities to climate change (Table S4, Fig. S6). Similar results are also obtained whether variability is calculated over the whole tropics or over the tropical oceans (using
the SSM/I GPROF dataset because the default precipitation dataset does not include values over
land), despite different relationships between variability and climate change in each case (Table
S5).
The results presented show how the simulated response to climate change of an important aspect of the tropical hydrological cycle may be constrained using observed variability. The inferred
sensitivities of tropical precipitation extremes under climate change have ranges of uncertainty that
are considerably narrower than the intermodel scatter. The inferred sensitivity of 10%K−1 (with
a 90% confidence interval of 6 to 14%K−1 ) for the 99.9th percentile of tropical precipitation is
7
higher than what climate models simulate for the same percentile of extratropical precipitation (36%K−1 across models and 5%K−1 in the multimodel median3 ). A higher sensitivity in the tropics
than the extratropics is physically possible if, for example, the circulations associated with precipitation extremes strengthen with warming in the tropics while they remain roughly constant in the
extratropics3 . One caveat is that other sensitivities may apply at hourly timescales for extratropical convective storms17 . The similarity of the inferred response when the analysis is restricted to
climate change over tropical land regions only (for sufficiently high percentiles of precipitation)
is important for impacts of climate change, and it suggests that precipitation extremes over land
may be more strongly tied to changes in surface temperatures over ocean rather than land. The observational constraint provides additional motivation for monitoring of tropical precipitation and
efforts to better understand the associated observational uncertainties. Consistent estimates were
obtained from only a decade of observations for three of the datasets considered (Table S4) which
suggests that the analysis could be applied reasonably quickly to data from new observing platforms. Ongoing research continues to lead to improvements in the parameterization of moist convection in climate models, but precipitation extremes are particularly challenging for convective
parameterizations13 and observational constraints are expected to continue to be useful.
Methods
Calculation of sensitivities Details of the climate models and observational datasets used are
given in the Supplementary Information. All precipitation datasets are first conservatively interpolated to an equal-area grid with constant spacing in longitude of 3 degrees. The interpolation
8
method is 1st order and weights data according to the area of overlap between gridboxes in the
original and coarse grid,29 consistent with the treatment of precipitation as a flux30 . The use of
a conservative interpolation scheme and a relatively-coarse common grid helps to allow for a
fair comparison of precipitation extremes in observations and simulations with different native
resolutions30 and improves the robustness of the observational estimates. For the SSM/I and TMI
observational datasets, both ascending and descending passes are given equal weight in the interpolation when available.
Time series of surface temperature and precipitation extremes are constructed as follows. For
each month, daily precipitation rates are aggregated in time over the month and in space between
30S and 30N. Precipitation extremes are then calculated as high percentiles of the aggregated
precipitation rates (including rates equal to zero) to yield one value per month for each percentile.
Given a balance between the desire to study extreme precipitation and a sample size of roughly
70,000 precipitation rates per month or less, the 99.9th percentile is the primary focus, but results
for other percentiles are also reported. Monthly surface temperature is spatially averaged between
30S and 30N (again giving one value per month). Surface skin temperatures are consistently
used throughout the paper with the exception of the results in Table S5, for which the observed
temperatures over land are surface air temperatures.
In the case of sensitivities to variability, the time series of precipitation extremes and surface
temperature are calculated either over the whole tropics or over the tropical oceans (30S to 30N).
The time series are deseasonalized by subtracting the mean seasonal cycle as estimated from the
time series themselves. The time series are then detrended and filtered with a 6-month running
9
average, followed by ordinary-least-squares regression of precipitation extremes against surface
temperature. The calculated sensitivities (% K−1 ) are expressed as a fraction of the mean value of
the precipitation extremes over the time period. This simple sensitivity to variability is adequate
despite the spatially-heterogeneous response to ENSO because it is used as an observable that is
strongly correlated with the sensitivity for climate change and not to fully characterize the response
of precipitation to ENSO.
In the case of sensitivities to climate change, the time series for precipitation and surface
temperature described above are calculated either over the whole tropics or over tropical land
(30S to 30N). The time series are then averaged over the 20th and 21st century time periods, and
the climate-change sensitivity of precipitation extremes (% K−1 ) is expressed as the difference
in precipitation extremes normalized by their 20th century value and the difference in surface
temperature.
Relatively strict land and ocean masks are used in this study. The masks are specified such
that grid boxes with less than 90% ocean are excluded when considering ocean, and grid boxes with
less than 90% land are excluded when considering land. The masks are applied after interpolation
to the common grid in the case of precipitation. The use of a relatively strict mask is needed
for consistency between the observational datasets for ocean-only precipitation; the SSM/I RSS
dataset has missing values over land that would bias the sensitivities to variability otherwise. The
use of a relatively strict mask also helps to minimize the contribution of ocean precipitation when
calculating climate change over land in the climate-model simulations.
10
Estimation of uncertainty Uncertainty is estimated using a bootstrapping method. A total of
2000 bootstrap estimates of the inferred sensitivity to climate change are generated by resampling
both the models used and the time series for models and observations. Time series are resampled
using 12-month moving blocks because of autocorrelation in the time series; use of shorter blocks
results in smaller error estimates. In the case of the model time series, resampled time series are
used in the calculation of the sensitivities to both variability and climate change. The same block
resampling of time series is used in the calculation of confidence intervals of the sensitivities for
variability.
It is important to note that the estimates of uncertainty do not account for errors in the estimation of observed precipitation rates (cf. Fig. S6 and Table S4). Note also that the models used
are not truly independent. For example, models GFDL-CM2.0 and GFDL-CM2.1 are developed at
the same laboratory and share many of the same components. At a minimum, the simulations from
these models may be regarded as giving independent realizations with a similar physical model.
1. Kharin, V. V., Zwiers, F. W., Zhang, X. & Hegerl, G. C. Changes in temperature and precipitation extremes in the IPCC ensemble of global coupled model simulations. J. Climate 20,
1419–1444 (2007).
2. Sun, Y., Solomon, S., Dai, A. & Portmann, R. W. How often will it rain? J. Climate 20,
4801–4818 (2007).
3. O’Gorman, P. A. & Schneider, T. The physical basis for increases in precipitation extremes in
simulations of 21st-century climate change. Proc. Natl. Acad. Sci. 106, 14773–14777 (2009).
11
4. Allan, R. P. & Soden, B. J. Atmospheric warming and the amplification of precipitation
extremes. Science 321, 1481–1484 (2008).
5. Allan, R. P., Soden, B. J., John, V. O., Ingram, W. & Good, P. Current changes in tropical
precipitation. Environ. Res. Lett. 5, 025205 (2010).
6. Liu, C. & Allan, R. P. Multisatellite observed responses of precipitation and its extremes to
interannual climate variability. J. Geophys. Res. 117, D03101 (2012).
7. Parry, M. L. et al. Technical summary. In Parry, M. L. et al. (eds.) Climate Change 2007:
Impacts, Adaptation and Vulnerability, 23–78 (Cambridge Univ. Press, 2007).
8. Pall, P. et al. Anthropogenic greenhouse gas contribution to flood risk in England and Wales
in autumn 2000. Nature 470, 382–385 (2011).
9. Trenberth, K. E., Dai, A., Rasmussen, R. M. & Parsons, D. B. The changing character of
precipitation. Bull. Amer. Meteor. Soc. 84, 1205–1217 (2003).
10. O’Gorman, P. A. & Schneider, T. Scaling of precipitation extremes over a wide range of
climates simulated with an idealized GCM. J. Climate. 22, 5676–5685 (2009).
11. Groisman, P. Y. et al. Trends in intense precipitation in the climate record. J. Climate 18,
1326–1350 (2005).
12. Min, S. K., Zhang, X., Zwiers, F. W. & Hegerl, G. C. Human contribution to more-intense
precipitation extremes. Nature 470, 378–381 (2011).
12
13. Wilcox, E. M. & Donner, L. J. The frequency of extreme rain events in satellite rain-rate
estimates and an atmospheric general circulation model. J. Climate 20, 53–69 (2007).
14. Muller, C. J., O’Gorman, P. A. & Back, L. E. Intensification of precipitation extremes with
warming in a cloud resolving model. J. Climate 24, 2784–2800 (2011).
15. Romps, D. M. Response of tropical precipitation to global warming. J. Atmos. Sci. 68, 123–
138 (2011).
16. Sobel, A. H. & Camargo, S. J. Projected future seasonal changes in tropical summer climate.
J. Climate 24, 473–487 (2011).
17. Lenderink, G. & van Meijgaard, E. Increase in hourly precipitation extremes beyond expectations from temperature changes. Nature Geoscience 1, 511–514 (2008).
18. Lu, J., Chen, G. & Frierson, D. M. W. Response of the zonal mean atmospheric circulation to
El Niño versus global warming. J. Climate 21, 5835–5851 (2008).
19. Chou, C. & Tu, J. Y. Hemispherical asymmetry of tropical precipitation in ECHAM5/MPIOM during El Niño and under global warming. J. Climate 21, 1309–1332 (2008).
20. Hall, A. & Qu, X. Using the current seasonal cycle to constrain snow albedo feedback in
future climate change. Geophys. Res. Lett. 33, L03502 (2006).
21. Knutti, R., Meehl, G. A., Allen, M. R. & Stainforth, D. A. Constraining climate sensitivity
from the seasonal cycle in surface temperature. Journal of climate 19, 4224–4233 (2006).
13
22. Guilyardi, E. et al. Understanding El Niño in ocean–atmosphere general circulation models.
Bull. Amer. Meteor. Soc 90, 325–340 (2009).
23. Hilburn, K. A. & Wentz, F. J. Intercalibrated passive microwave rain products from the unified
microwave ocean retrieval algorithm (UMORA). J. Appl. Meteorol. 47, 778–794 (2008).
24. Smith, T. M., Reynolds, R. W., Peterson, T. C. & Lawrimore, J. Improvements to NOAA’s
historical merged land-ocean surface temperature analysis (1880-2006). J. Climate 21, 2283–
2296 (2008).
25. O’Gorman, P. A. & Muller, C. J. How closely do changes in surface and column water vapor follow Clausius-Clapeyron scaling in climate-change simulations? Environ. Res. Lett. 5,
025207 (2010).
26. Kummerow, C. et al. The evolution of the Goddard Profiling Algorithm (GPROF) for rainfall
estimation from passive microwave sensors. J. Appl. Meteorol. 40, 1801–1820 (2001).
27. Huffman, G. J. et al. Global precipitation at one-degree daily resolution from multisatellite
observations. J. Hydrometeor. 2, 36–50 (2001).
28. Huffman, G. J. et al. The TRMM multisatellite precipitation analysis (TMPA): Quasi-global,
multiyear, combined-sensor precipitation estimates at fine scales. J. Hydrometeorol. 8, 38–55
(2007).
29. Jones, P. W. First- and second-order conservative remapping schemes for grids in spherical
coordinates. Mon. Wea. Rev. 127, 2204–2210 (1999).
14
30. Chen, C. T. & Knutson, T. On the verification and comparison of extreme rainfall indices from
climate models. J. Climate 21, 1605–1621 (2008).
Supplementary Information is linked to the online version of the paper at www.nature.com/nature
Acknowledgements I am grateful to Christian Kummerow, Tapio Schneider, Martin Tingley, Richard
Allan, Kerry Emanuel, Carl Wunsch, Susan Solomon, and Timothy Merlis for helpful discussions. I acknowledge the World Climate Research Programme’s Working Group on Coupled Modelling, which is
responsible for CMIP, and I thank the climate modeling groups for producing and making available their
model output. For CMIP the U.S. Department of Energy’s Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership
with the Global Organization for Earth System Science Portals. SSM/I (V6) data were provided by Remote
Sensing Systems (www.remss.com) and sponsored by the NASA Earth Science MEaSUREs DISCOVER
Project. SSM/I and TMI GPROF (V10) data were downloaded from http://rain.atmos.colostate.edu/. GPCP
1DD (V1.1) data were downloaded from http://www1.ncdc.noaa.gov/pub/data/gpcp/. TRMM 3B42 (V7)
daily data were provided by the Goddard Earth Sciences Data and Information Services Center. NOAA
Merged Air Land and SST anomalies (V3.5.1) were provided by the NOAA/OAR/ESRL PSD from their
website at http://www.esrl.noaa.gov/psd/. I acknowledge support from NSF grant AGS-1148594 and NASA
grant NNX-11AO92G.
Competing Interests The author declares that he has no competing financial interests.
Correspondence Correspondence and requests for materials should be addressed to P.A.O’G. (email:
pog@mit.edu).
15
Observations
mm/day
10
P99.9
0
25%K−1
−10
1992
1996
2000
2004
2008
mm/day
Model: GFDL−CM2.0
P99.9
15
51%K−1
Implied
0
−15
1985
1990
1995
2000
Model: ECHAM5/MPI
mm/day
3
P99.9
0
11%K−1
Implied
−3
1985
1990
1995
2000
Figure 1: Time series of precipitation extremes and surface temperature over the tropical oceans
in observations and simulations (GFDL-CM2.0 and ECHAM5/MPI). Anomalies in the 99.9th percentile of precipitation (blue) and surface temperature rescaled by the sensitivity (%K−1 ) for variability in each case (green) are shown. Also shown (red) for the models are surface temperature
anomalies rescaled by the sensitivity to variability implied by the sensitivity to climate change
(over the whole tropics) and the regression relationship between sensitivities to variability and climate change for all the CMIP3 models (Table S1). Time series are filtered with a 6-month running
average.
16
Climate change (% K−1)
20
Inferred
10
Observed
0
0
20
40
Variability (% K−1)
60
BCCR BCM2.0
CGCM3.1 T47
CGCM3.1 T63
CNRM−CM3
CSIRO−Mk3.0
CSIRO−Mk3.5
GFDL−CM2.0
GFDL−CM2.1
FGOALS−g1.0
ECHAM4/INGV
ECHAM5/MPI
INM−CM3.0
IPSL−CM4
MIROC3.2−med
MIROC3.2−hi
MRI−CGCM2.32
NCAR−PCM1
NCAR−CCSM3.0
Figure 2: Sensitivities (%K−1 ) of the 99.9th percentile of precipitation for variability versus climate change in the CMIP3 simulations. The solid line shows the ordinary-least-squares best fit.
Histograms show estimates (with uncertainty) of the observed sensitivity to variability and the inferred sensitivity to climate change. Sensitivities to variability are over the tropical oceans and
sensitivities to climate change are over the whole tropics.
17
Sensitivity (% K−1)
(a) Default
Inferred
Model max, min
Model median
20
10
0
Sensitivity (% K−1)
20
(b) Land only
Default inferred
10
0
Sensitivity (% K−1)
(c) CMIP5
20
10
0
98
99
99.5
99.8
99.9
Percentile of precipitation
99.95
Figure 3 Inferred and simulated climate-change sensitivities (%K−1 ) for high percentiles of precipitation. Black lines with circles show inferred sensitivities, shading shows the associated 90%
confidence intervals, solid green lines show multimodel maxima and minima, and dashed green
lines shows multimodel medians. (a) CMIP3 models and the whole tropics. (b) As in (a) but
for climate change over land only and normalized by temperature changes over the whole tropics. (c) As in (a) but for CMIP5 models. Black dashed lines in (b) and (c) reproduce the inferred
sensitivities shown in (a).
18
Sensitivity of tropical precipitation extremes to climate
change
Supplementary Information
Paul A. O’Gorman
1
Climate-model simulations
The 18 CMIP3 models used are BCCR-BCM2.0, CGCM3.1 T47, CGCM3.1 T63, CNRM-CM3,
CSIRO-Mk3.0, CSIRO-Mk3.5, GFDL-CM2.0, GFDL-CM2.1, FGOALS-g1.0, ECHAM4/INGV,
ECHAM5/MPI, INM-CM3.0, IPSL-CM4, MIROC3.2-med, MIROC3.2-hi, MRI-CGCM2.32, NCARPCM1, and NCAR-CCSM3.0. The time periods used are 1981-2000 (20C3M) and 2081-2100
(SRES A1B), except for BCCR-BCM2.0 (1981-1998, 2081-2098), CNRM-CM3 (1981-1999, 20812099), FGOALS-g1.0 (1981-1999, 2081-2099), MIROC3.2-hi (1981-1999, 2081-2099), NCARPCM1 (1980-1999, 2080-2098), and NCAR-CCSM3.0 (1980-1999, 2080-2099). Models not included in the analysis were primarily excluded because of lack of availability of the necessary
data or because of corrupt data. The GISS-AOM model was excluded because it has extremely
weak ENSO temperature variabilityS1 (it also gives a negative sensitivity of precipitation extremes
to variability). Inspection of Fig. 2 suggests that the two GFDL models may be influential in
the regression of sensitivities; omitting the GFDL models from the analysis changes the inferred
sensitivity to climate change of the 99.9th percentile of precipitation from 10%K−1 to 8%K−1 .
(The inferred sensitivity using the CMIP5 ensemble remains at 11%K−1 if the GFDL models are
omitted.)
1
The CMIP3 models with relatively-good simulations of ENSO (the “good-ENSO” models)
are taken to be ECHAM5/MPI, GFDL-CM2.0, GFDL-CM2.1, IPSL-CM4, and MRI-CGCM2.32.
This subset of models follows the selection in a previous studyS1 , with the exception of the UKMOHadCM3 model for which the necessary daily data were not available.
The 15 CMIP5 models used are ACCESS1.0, BNU-ESM, CCSM4, CSIRO-Mk3.6.0, GFDLESM2G, GFDL-ESM2M, GFDL-CM3, HadGEM2-CC, HadGEM2-ES, IPSL-CM5A-MR, IPSLCM5B-LR, MIROC-ESM-CHEM, MIROC5, MRI-CGCM3, and NorESM1-M. The time periods
used are 1981-2000 (historical) and 2081-2100 (RCP 8.5), except for ACCESS1.0 (1980-1999,
2081-2100), GFDL-CM3 (1980-1999, 2081-2100), HadGEM2-CC (1981-2000, 2081-2099), HadGEM2ES (1981-2000, 2081-2099), IPSL-CM5A-MR (1980-1999, 2081-2100), MIROC5 (1980-1999,
2080-2099), MRI-CGCM3 (1980-1999, 2081-2100), and NorESM1-M (1980-1999, 2081-2100).
2
Observational precipitation datasets
The default precipitation observations used are based on passive-microwave retrievals and are taken
from the SSM/I (V6) datasetS2 of Remote Sensing Systems (RSS) for the period 1991-2008 using
the satellites F10 (1991-1995) and F13 (1996-2008). The time series could have been extended
three years further back in time by also using F08, but using only two satellites helps to minimize
uncertainties related to intercalibration. Results for four other observational datasets are presented
in Fig. S6 and Table S4, and the time periods used are specified in Table S4. SSM/I GPROF refers
to version 10 of the Goddard Profiling AlgorithmS3 using the same satellites and time periods as for
2
the default SSM/I dataset. The Tropical Rainfall Measuring Mission (TRMM) Microwave Imager
(TMI) dataset used is also based on the GPROF version 10 algorithm. The 1-degree daily merged
product V1.1 from the Global Precipitation Climatology Project (GPCP 1DD) includes inputs from
infrared, passive-microwave, and gauge measurementsS4 . The TRMM 3B42 V7 merged daily
dataset includes inputs from infrared, passive and active microwave, and gauge measurementsS5 .
Taken together, the observational datasets for precipitation include different satellites, types of
sensors, and retrieval algorithms, although they are not independent.
TRMM 3B42 has somewhat different variability from the other datasets in the case of mean
precipitation, but it gives similar results to the other datasets in the case of precipitation extremes
(Fig. S6). An earlier version (V6) was found to be inconsistent with the other datasets as regards
interannual variability of both mean and extreme precipitation, a discrepancy that is noted in previous papersS5, S6 .
Daily precipitation is accumulated in models but must be combined from estimates at discrete
times during the day in satellite observations. This may be expected to affect the absolute daily
precipitation rates, but not necessarily their fractional changes. Some confidence that this issue
does not strongly impact the final results comes from the similarity of inferred sensitivities (Table
S4) from SSM/I (at most one ascending and one descending pass per day at each location), TMI
(a similar number of passes but with different orbital characteristics from SSM/I), and the merged
datasets GPCP 1DD and TRMM 3B42.
3
3
Dependence on tropical cyclones and choice of domain
The high precipitation percentiles considered in this study include contributions from a range of
different types of tropical systems, including tropical cyclones. To assess the influence of tropical
cyclones on the results, the analysis was repeated over the latitude band 5S to 5N in which there
is little tropical-cyclone activity. Although the relationship between sensitivities to variability and
climate change is not as strong for this narrower latitude band, similar results are obtained for
the inferred sensitivity to climate change. For the 99.9th percentile of precipitation, the inferred
sensitivity is 11%K−1 when 5S to 5N is used compared with 10%K−1 when 30S to 30N is used.
More generally, the calculated sensitivity to variability depends on the domain chosen (e.g.,
which parts of the Pacific are included or how much land is included) because of the spatiallyheterogeneous response of precipitation to ENSO. But this does not imply that the inferred sensitivity to climate change depends strongly on the choice of domain because both modeled and
observed sensitivities to variability are affected by the choice of domain. For example, for the
99.9th percentile of precipitation and the SSM/I GPROF precipitation dataset, the sensitivity to
variability is 33%K−1 over ocean and 21%K−1 over the whole tropics, while the resulting inferred
sensitivities to climate change are 13%K−1 in both cases. However, it is important that land is
masked out in a consistent way in both the models and observations. Results based on variability
over land alone are not reported because this gives a sensitivity to variability that is not strongly
related to the response to climate change.
4
4
Dependence on method of calculation of precipitation extremes
The sensitivities of tropical precipitation extremes to climate change calculated here are similar but
not exactly the same as those in a previous studyS7 , the results of which are used for comparison.
In particular, the multimodel-median sensitivity in the tropics for the 99.9th percentile of precipitation is similar at 6%K−1 in this study and 5%K−1 in ref. S7, but the intermodel scatter is smaller
in this study. The differences arise because of slightly different sets of climate models used and
different methods of aggregation of precipitation rates, and because the precipitation rates used
here are interpolated to a common grid prior to calculation of percentiles. In ref. S7, precipitation
rates are aggregated at each latitude over the entire time period prior to calculating percentiles,
and percentage changes in the precipitation percentiles are then averaged in latitude over the tropics or extratropics. The inferred sensitivity for climate change of the 99.9th percentile of tropical
precipitation increases by only 0.1%K−1 when the analysis presented here is repeated using the
aggregation method of ref. S7 for climate-change sensitivities. If the precipitation percentiles are
also calculated on the native model grids when calculating climate-change sensitivities, the intermodel scatter increases to what was found in ref. S7 and the inferred sensitivity to climate change
increases from 10 to 12%K−1 , but with a wider 90% confidence interval of 6 to 17%K−1 . The
thermodynamic rates and simulated extratropical sensitivities are more robust than the simulated
tropical sensitivities so that it is reasonable to use the values calculated in ref. S7 as a point of
comparison for the inferred sensitivities calculated here.
5
Supplementary references
S1. Guilyardi, E. et al. Understanding El Niño in ocean–atmosphere general circulation models.
Bull. Amer. Meteor. Soc 90, 325–340 (2009).
S2. Hilburn, K. A. & Wentz, F. J. Intercalibrated passive microwave rain products from the unified
microwave ocean retrieval algorithm (UMORA). J. Appl. Meteorol. 47, 778–794 (2008).
S3. Kummerow, C. et al. The evolution of the Goddard Profiling Algorithm (GPROF) for rainfall
estimation from passive microwave sensors. J. Appl. Meteorol. 40, 1801–1820 (2001).
S4. Huffman, G. J. et al. Global precipitation at one-degree daily resolution from multisatellite
observations. J. Hydrometeor. 2, 36–50 (2001).
S5. Huffman, G. J. et al. The TRMM multisatellite precipitation analysis (TMPA): Quasi-global,
multiyear, combined-sensor precipitation estimates at fine scales. J. Hydrometeorol. 8, 38–55
(2007).
S6. Liu, C. & Allan, R. P. Multisatellite observed responses of precipitation and its extremes to
interannual climate variability. J. Geophys. Res. 117, D03101 (2012).
S7. O’Gorman, P. A. & Schneider, T. The physical basis for increases in precipitation extremes in
simulations of 21st-century climate change. Proc. Natl. Acad. Sci. 106, 14773–14777 (2009).
6
Observations
mm/day
10
P99.9
0
25%K−1
−10
1992
1996
2000
2004
2008
mm/day
Model: GFDL−CM2.0
P99.9
15
51%K−1
Implied
0
−15
mm/day
Model: GFDL−CM2.1
P99.9
15
42%K−1
Implied
0
−15
mm/day
Model: IPSL−CM4
5
P99.9
0
24%K−1
Implied
−5
Model: ECHAM5/MPI
mm/day
3
P99.9
0
11%K−1
Implied
−3
Model: MRI−CGCM2.32
mm/day
3
P99.9
0
8%K−1
Implied
−3
1985
1990
1995
2000
Fig. S1. As in Fig. 1 but showing time series for all of the “good-ENSO” subset of CMIP3 models.
7
Climate change (% K−1)
15
10
5
0
0
20
40
60
BCCR BCM2.0
CGCM3.1 T47
CGCM3.1 T63
CNRM−CM3
CSIRO−Mk3.0
CSIRO−Mk3.5
GFDL−CM2.0
GFDL−CM2.1
FGOALS−g1.0
ECHAM4/INGV
ECHAM5/MPI
INM−CM3.0
IPSL−CM4
MIROC3.2−med
MIROC3.2−hi
MRI−CGCM2.32
NCAR−PCM1
NCAR−CCSM3.0
Variability (% K−1)
Fig. S2. As in Fig. 2 but for climate change over tropical land only. The sensitivity to climate
change over land is normalized by the change in surface temperature over the whole tropics. Variability is calculated over the tropical oceans.
8
Land (% change)
40
30
20
10
0
0
10 20 30 40 50 60
Ocean (% change)
BCCR BCM2.0
CGCM3.1 T47
CGCM3.1 T63
CNRM−CM3
CSIRO−Mk3.0
CSIRO−Mk3.5
GFDL−CM2.0
GFDL−CM2.1
FGOALS−g1.0
ECHAM4/INGV
ECHAM5/MPI
INM−CM3.0
IPSL−CM4
MIROC3.2−med
MIROC3.2−hi
MRI−CGCM2.32
NCAR−PCM1
NCAR−CCSM3.0
Fig. S3. Percentage changes in the 99.9th percentile of precipitation over land versus ocean in the
CMIP3 simulations. The solid line corresponds to equal percentage changes over land and ocean.
The correlation coefficient of percentage changes over land and ocean is 0.897. Unlike in other
figures, the results shown are not normalized by changes in surface temperature.
9
Climate change (% K−1)
ECHAM5/MPI
GFDL−CM2.0
GFDL−CM2.1
IPSL−CM4
MRI−CGCM2.32
20
10
0
0
20
40
60
Variability (% K−1)
Fig. S4. As in Fig. 2 but based on the “good-ENSO” subset of CMIP3 models.
10
Climate change (% K−1)
ACCESS1.0
BNU−ESM
CCSM4
CSIRO−Mk3.6.0
GFDL−ESM2G
GFDL−ESM2M
GFDL−CM3
HadGEM2−CC
HadGEM2−ES
IPSL−CM5A−MR
IPSL−CM5B−LR
MIROC−ESM−CHEM
MIROC5
MRI−CGCM3
NorESM1−M
20
10
0
0
20
40
60
Variability (% K−1)
Fig. S5. As in Fig. 2 but based on the CMIP5 climate models.
11
99.9th percentile
SSM/I RSS
SSM/I GPROF
GPCP 1DD
3B42
TMI
10%
0
−10%
SSM/I RSS
SSM/I GPROF
GPCP 1DD
3B42
TMI
Mean
10%
0
−10%
1992
1996
2000
2004
2008
Fig. S6. Anomalies (%) in (top) the 99.9th percentile of precipitation and (bottom) mean precipitation over the tropical oceans in the default (SSM/I RSS) and four other observational datasets.
Time series are filtered with a 6-month running average. Time periods differ for the different
datasets and are longest for SSM/I RSS and GPROF (see Table S4).
12
Table S1. Inferred sensitivities to climate change with 90% confidence intervals (CIs) for different high percentiles of precipitation based on CMIP3 models and the “good-ENSO” subset of
CMIP3 models. The correlation coefficient (r) and regression coefficients (a,b) for the relationship
sc =a+bsv between the sensitivity to variability (sv ) and the sensitivity to climate change (sc ) are
also given. The sensitivity to variability is over the tropical oceans, and the sensitivity to climate
change is over the whole tropics.
Percentile
All Models
Inferred sc
r
%K−1 (90% CI)
“Good-ENSO” models
a
b
%K−1
Inferred sc
r
%K−1 (90% CI)
a
b
%K−1
98
3 (1, 5)
0.124
2.5
0.03
-3 (-10, 6)
-0.600
5.2
-0.26
99
6 (3, 7)
0.392
2.2
0.12
7 (-13, 15)
0.345
0.8
0.22
99.5
7 (4, 9)
0.522
2.2
0.18
10 (-8, 17)
0.753
-0.4
0.38
99.8
9 (6, 12)
0.735
2.2
0.28
10 (7, 16)
0.988
1.1
0.38
99.9
10 (6, 14)
0.866
1.4
0.37
11 (8, 17)
0.997
0.4
0.43
99.95
9 (5, 13)
0.882
1.3
0.38
9 (5, 16)
0.973
0.6
0.43
13
Table S2. As in Table S1 but for the response of precipitation extremes to climate change over
tropical land only. To facilitate comparison with Table S1, the inferred sensitivities to climate
change are normalized with respect to surface temperature change averaged over the whole tropics.
Variability is calculated over the tropical oceans.
Percentile
All Models
Inferred sc
r
%K−1 (90% CI)
“Good-ENSO” models
a
b
%K−1
Inferred sc
r
%K−1 (90% CI)
a
b
%K−1
98
0 (-3, 3)
-0.398
3.6
-0.13
-3 (-14, 5)
-0.453
5.0
-0.27
99
3 (0, 6)
-0.063
3.4
-0.02
2 (-10, 7)
-0.128
3.9
-0.06
99.5
5 (2, 8)
0.199
3.3
0.08
7 (0, 13)
0.509
3.6
0.12
99.8
9 (5, 12)
0.609
2.0
0.27
10 (8, 16)
0.908
3.3
0.29
99.9
9 (6, 12)
0.735
2.1
0.28
11 (8, 15)
0.927
4.4
0.26
99.95
8 (5, 11)
0.760
2.3
0.28
10 (7, 16)
0.876
5.5
0.24
14
Table S3. As in Table S1 but comparing results using the CMIP3 and CMIP5 models.
Percentile
CMIP3 (All Models)
Inferred sc
r
%K−1 (90% CI)
CMIP5
a
b
%K−1
Inferred sc
r
%K−1 (90% CI)
a
b
%K−1
98
3 (1, 5)
0.124
2.5
0.03
5 (2, 8)
0.407
1.4
0.13
99
6 (3, 7)
0.392
2.2
0.12
8 (4, 11)
0.528
0.9
0.24
99.5
7 (4, 9)
0.522
2.2
0.18
9 (5, 14)
0.604
1.3
0.29
99.8
9 (6, 12)
0.735
2.2
0.28
10 (7, 15)
0.658
2.9
0.30
99.9
10 (6, 14)
0.866
1.4
0.37
11 (8, 16)
0.741
3.3
0.33
99.95
9 (5, 13)
0.882
1.3
0.38
10 (6, 15)
0.743
3.9
0.30
15
Table S4. Sensitivities of the 99.9th percentile of precipitation based on the default and alternative
observational precipitation datasets and the CMIP3 models. Sensitivities to variability over the
tropical oceans (sv ) and inferred sensitivities to climate change over the whole tropics (sc ) are
shown.
Dataset
Time period
sv
Inferred sc
%K−1 (90% CI)
%K−1 (90% CI)
SSM/I RSS (default)
1991-2008
25 (16, 36)
10 (6, 14)
SSM/I GPROF
1991-2008
33 (22, 49)
13 (7, 18)
TMI
1998-2008
33 (22, 40)
13 (7, 16)
GPCP 1DD
1997-2008
21 (14, 31)
9 (6, 12)
3B42
1998-2008
21 (12, 31)
9 (6, 12)
16
Table S5. As in Table S1 but comparing results using variability over ocean only or variability
over the whole tropics. Climate change is over the whole tropics in both cases. The results in this
table are based on the SSM/I GPROF precipitation dataset because the default precipitation dataset
does not include values over land. The full set of CMIP3 models is used.
Percentile
Ocean variability
Inferred sc
r
%K−1 (90% CI)
Land+ocean variability
a
b
%K−1
Inferred sc
r
%K−1 (90% CI)
a
b
%K−1
98
3 (2, 4)
0.124
2.5
0.03
3 (2, 4)
0.182
2.3
0.08
99
6 (3, 7)
0.392
2.2
0.12
6 (3, 7)
0.478
1.8
0.25
99.5
8 (4, 10)
0.522
2.2
0.18
8 (4, 10)
0.555
2.0
0.30
99.8
10 (6, 14)
0.735
2.2
0.28
10 (6, 15)
0.660
2.0
0.44
99.9
13 (7, 18)
0.866
1.4
0.37
13 (7, 20)
0.793
1.6
0.56
99.95
13 (7, 19)
0.882
1.3
0.38
13 (7, 20)
0.820
1.6
0.55
17