15 NOVEMBER 2014 PENDERGRASS AND HARTMANN 8357

Two Modes of Change of the Distribution of Rain*

ANGELINE G. PENDERGRASS AND DENNIS L. HARTMANN

Department of Atmospheric Sciences, University of Washington, Seattle, Washington

(Manuscript received 28 February 2014, in final form 23 July 2014)

ABSTRACT

The frequency and intensity of rainfall determine its character and may change with climate. A method-
ology for characterizing the frequency and amount of rainfall as functions of the rain rate is developed. Two
modes of response are defined, one in which the distribution of rainfall increases in equal fraction at all rain
rates and one in which the rainfall shifts to higher or lower rain rates without a change in mean rainfall.
This description of change is applied to the tropical distribution of daily rainfall over ENSO phases in
models and observations. The description fits observations and most models well, although some models
also have an extreme mode in which the frequency increases at extremely high rain rates. The multimodel
mean from phase 5 of the Coupled Model Intercomparison Project (CMIP5) agrees with observations in
showing a very large shift of 14%–15% K21, indicating large increases in the heaviest rain rates associated
with El Niño. Models with an extreme mode response to global warming do not agree as well with ob-
servations of the rainfall response to El Niño.

1. Introduction structure, and wet-day threshold. Some potential tem-

poral resolutions include daily rainfall accumulation and
The frequency and intensity of rain are expected to
instantaneous, hourly, 6-hourly, 5-day, or monthly data.
change due to global warming (Trenberth 1999). There
Stephens et al. (2010) argue that instantaneous rain rate
are also indications that they change on shorter time
is the best variable to consider, while Trenberth et al.
scales. What is the present day distribution of rain fre-
(2003) recommend hourly. Liu and Allan (2012) argue,
quency and intensity, and how do the frequency and
however, that averaging of observational data is neces-
intensity of rain change in response to warming?
sary for purposes of intercomparison. They show that
Rainfall and its changes are often considered from the
the distributions of precipitation in satellite data and
perspective of the mean or the extremes (e.g., Allen and
gridded analyses agree better when averaged over 5 days
Ingram 2002). We approach the problem by instead
than for daily or instantaneous data. Daily satellite ob-
considering the whole distribution and how it changes,
servations are the average of only two snapshots, while
which includes the mean and extremes but also their
the daily merged, gridded products have implicit aver-
context. In preparation for understanding the distribu-
aging. Other recent studies use monthly-mean data to
tion of rain, we consider carefully the best way to
look at the distribution of rain (Liu et al. 2012; Lau et al.
quantify it.
2013). This removes all variability on less than monthly
A variety of approaches to quantifying the distribu-
time scales, which is not desirable if one is interested in
tion of rain can be found in the scientific literature, in-
possible changes in extremes.
cluding different spatial and temporal resolution, bin
The choice of rain-rate bin structure is fundamental to
all rainfall distribution calculations. Rain rate, even ac-
* Supplemental information related to this paper is available at
cumulated over a day, varies by over orders of magni-
the Journals Online website: [Link] tude from day to day and from one location to another.
14-00182.s1. A wide variety of possible bin structure choices can be
found in the scientific literature. Some studies use line-
Corresponding author address: Angeline G. Pendergrass, NCAR, arly distributed rain-rate bins, often with 1 mm day21
P.O. Box 3000, Boulder, CO 80307. spacing (e.g., Chou et al. 2012). Linear spacing of bins
E-mail: apgrass@[Link] has the benefit that the bins are related in an obvious

DOI: 10.1175/JCLI-D-14-00182.1

Ó 2014 American Meteorological Society

Brought to you by NATIONAL ATMOSPHERIC RESEARCH LAB | Unauthenticated | Downloaded 07/24/24 11:51 AM UTC
8358 JOURNAL OF CLIMATE VOLUME 27

way, but a substantial drawback is that the bins do not we explore the response to global warming in climate
have reasonable sampling at high rain rates. Other model simulations more extensively and discuss the role
studies use irregular bin spacing to improve sampling that the modes of change play in the energetic and
and capture the entire distribution, but this may make it moisture constraints on precipitation.
intractable to work with the distribution mathemati- In the next three sections, we introduce the GCM and
cally. Watterson and Dix (2003) use logarithmically observational datasets we will use, present our meth-
distributed rain-rate bins. This coordinate makes odology for calculating the rainfall distribution, and
mathematical operations tractable and captures the consider the climatological rainfall distributions. In
entire distribution from light to heavy rain with ade- section 5, we discuss the two modes of change. Then, in
quate sampling. section 6, we fit the modes of change to the rain amount
Gridded observations and model data represent response to ENSO phases in observations and models
a spatial average rather than a single point in space, so from phase 5 of the Coupled Model Intercomparison
that an objective threshold for when it is raining is not Project (CMIP5) and make comparisons with the re-
obvious. Some studies and guidelines define wet days as sponse to increasing CO2 documented in Pendergrass
having a minimum accumulation of 1 mm (Sun et al. and Hartmann (2014b).
2006; Klein Tank et al. 2009). Other studies extend the
dry threshold to 0.1 mm day21 (e.g., Chou et al. 2012).
2. Daily rainfall data
The area over which averages are taken also influences
the interpretation of rainfall statistics. Some studies (e.g., In this section, we document the GCM simulations
Pall et al. 2007; O’Gorman and Schneider 2009) take and observational datasets. We analyze daily rainfall
a zonal-mean perspective, in which expected shifts of the accumulation from models and observations. Daily
ITCZ and storm tracks are prominent. Watterson and gridded data strike a compromise by providing some
Dix (2003) look at the changes in distributions at in- temporal averaging but not more than is necessary to
dividual locations. For changes in precipitation at many make (gridded) observations comparable with models.
locations (and thus for impacts of precipitation change), For observational distributions of rainfall and its
geographical shifts in precipitation systems may domi- response to ENSO phases, we use two gridded ob-
nate over thermodynamic effects. servational datasets that merge data from satellite and
Spatial resolution of precipitation data has been iden- rain gauges. Global Precipitation Climatology Project
tified as a contributing factor to disagreements among (GPCP) One-Degree Daily (1DD) data (Huffman et al.
observational datasets and between models and obser- 2001) have global coverage at 18 resolution in latitude
vations (Chen and Knutson 2008; Kopparla et al. 2013). and longitude. We use data from 1997 to 2012. Tropical
Both daily gridded data and model data have implicit Rainfall Measuring Mission (TRMM) 3B42 version 7
spatial averaging, so in this sense they are fundamentally (Huffman et al. 2007) covers 508S–508N at 0.258 resolu-
different from station data, which measure a particular tion from 1998 to 2012.
location. To determine ENSO phase, we obtain the monthly
Extreme precipitation is typically quantified by the time series of SSTs in the Niño-3.4 region (58N–58S, 1708–
rain rate for some cumulative percentile of the rain 1208W) from the National Oceanic and Atmospheric/
frequency distribution (e.g., Allen and Ingram 2002; Pall Administration Earth System Research Laboratory/
et al. 2007; O’Gorman and Schneider 2009). There are Physical Sciences Division (NOAA/ESRL/PSD;
large differences in the extreme precipitation response [Link]
to interannual variability and longer-time-scale climate and calculate a Niño-3.4 index following Trenberth
changes in models (O’Gorman 2012; Allan et al. 2013). (1997). For the warm phase composite, we choose the
The goal of this study is to quantify ways the distribu- months with the highest Niño-3.4 index. We use 50
tion of rain could change over time. To accomplish this, months for observations and 100 months for Atmo-
we introduce two modes of coherent change of the rain spheric Model Intercomparison Project (AMIP) simu-
amount distribution: an increase in rain amount at all rain lations (to make use of the longer model simulations).
rates (the increase mode) and a shift of the rain amount Because we cannot remove the seasonal cycle when
distribution to higher rain rates (the shift mode). The calculating the rainfall distribution, it is important to
increase mode changes the total amount of rain, while make sure we use the same months of the year in the
the shift mode does not. We apply these two modes to warm and cold phases. To enforce this requirement, we
the responses of modeled and observed distributions over choose cold months so that the distribution of months is
phases of El Niño–Southern Oscillation (ENSO). In a the same as for the warm months. For example, five of
companion paper, Pendergrass and Hartmann (2014b), the warm months over the GPCP period are January, so

Brought to you by NATIONAL ATMOSPHERIC RESEARCH LAB | Unauthenticated | Downloaded 07/24/24 11:51 AM UTC
15 NOVEMBER 2014 PENDERGRASS AND HARTMANN 8359

we choose the five coldest Januaries for the cold phase. has a width of 0.0025 mm day21, and the largest bin needed
We use Hadley Centre/Climatic Research Unit surface for CMIP5 model data is centered at 663 mm day21 and has
temperature anomalies, version 4 (HadCRUT4; Jones a width of 49.0 mm day21. Bins include all rain rates in the
et al. 2012) to calculate the observed change in tropical dataset greater than 0.03 mm day21. Because of the loga-
(308S–308N) surface temperature from cold to warm rithmic structure, movement along the rain-rate axis
ENSO phase (0.35 K for the GPCP period). is expressed as a percentage change in rain rate. A 7% in-
To examine the modeled response to ENSO, we use the crease in rain rate is simply a movement of one rain-rate bin.
AMIP experiments from the CMIP5 archive (Taylor et al. Another specification that must be chosen is the
2012). (The models used and their resolutions are listed in threshold used to define dry days. We use a dry threshold
Table S1 of the supplemental material.) In AMIP experi- of 0.0321 mm day21. While most dry days have zero
ments, sea surface temperature (SST) is prescribed from rainfall, in GPCP 3% of days have rainfall that is non-
observations over the historical period, along with pre- zero but below the dry threshold. We miss a negligible
scribed historical forcings. These experiments allow us to amount of the total rainfall (0.014% in GPCP) by ig-
examine the modeled response to ENSO SST forcing de- noring the rain amount below the dry threshold.
spite that not all models have good ENSO responses when Focusing on the global scale allows us to integrate
run in coupled mode. We use one ensemble member from over spatial shifts in precipitating systems, like the ITCZ
each of the 27 models with archived daily precipitation and storm tracks, which occur with climate change. To
data from 1979 to 2007. This time period overlaps with but address differing spatial resolution of datasets, when
is longer than the observational time period. We normalize comparing two distributions, we coarsen the higher
by each model’s tropical mean surface air temperature spatial-resolution precipitation data using an averaging
change from cold to warm ENSO phase. We choose warm scheme that conserves the total amount of rain (Jones
and cold ENSO months based on the same Niño-3.4 1999) before calculating the distribution. We coarsen
dataset index we use for observational data, which is pos- the 0.258 TRMM data to the 18 GPCP grid. Model res-
sible because the SSTs are prescribed in the models. olution ranges from 0.758 to 2.88 (for models analyzed
To calculate the model precipitation distributions and from the CO2-increase experiment). We coarsen the
their response to CO2 doubling, we use climate model GPCP data to model grids when the model resolution is
simulations from the CMIP5 archive. We use simulations coarser than 18 and coarsen neither GPCP nor the model
from two experiments: 1% yr21 CO2 increase (1pctCO2) for the two models with resolution finer than 18.
and representative concentration pathway (RCP) with ra- With these decisions about coordinate system, dry
diative forcing reaching 8.5 W m22 near 2100 (RCP8.5). threshold, and spatial averaging made, we calculate the
The fully coupled 1pctCO2 scenario provides the response distribution of rain frequency and amount. At each grid
to CO2 doubling, initialized from a preindustrial base state. point, we calculate a rain histogram and normalize by
We compare years 1–10 (the start of the simulation) and the number of days to form the rain frequency distri-
years 61–70 (CO2 doubles at year 70). We use one ensemble bution. We also tabulate the total amount of rain that
member from each of the 22 models with archived daily falls in each rain-rate bin to form the rain amount dis-
rainfall accumulation and surface air temperature. From the tribution. Then we take the area-weighted average of
RCP8.5 simulations, which are initialized from historical the distributions over the globe to obtain global-mean
forcing in 2005, we compare 2006–15 with 2090–99. distributions. A mathematical description of how we
calculate the distributions is presented in the appendix.
We calculate rain amount and rain frequency distri-
3. Calculating rain amount and rain frequency
butions separately for the first 10 yr of the CO2-doubling
Next, we discuss the methodology for calculating model simulations and the 10 yr immediately prior to
distributions of rain amount and frequency and their doubling (years 61–70) as well as for the warm and cold
changes. We use logarithmically distributed rain-rate phases of ENSO. We report the absolute changes in rain
bins following Watterson and Dix (2003). In log co- frequency and rain amount. Some studies report the per-
ordinates, each rain-rate bin is a fixed percentage wider centage change of rain frequency and amount distributions
than the previous one and has a bin center the same (e.g., Sun et al. 2007; Lau et al. 2013), but this has some
percentage larger. We choose bin spacing of 7% to undesirable properties. The maximum rain rate in-
balance resolution and sampling. For example, one rain- creases with warming, so for the highest rain rates
rate bin is centered at 10.6 mm day21 and has a width of the percent change in the distribution is not defined. The
0.78 mm day21, while the next bin is centered at percentage change will also depend very strongly on the
11.4 mm day21 and has a width of 0.84 mm day21. The bin structure used. For these reasons, we look at absolute
smallest nonzero bin is centered at 0.03 mm day21 and changes in the rain frequency and amount distributions.

Brought to you by NATIONAL ATMOSPHERIC RESEARCH LAB | Unauthenticated | Downloaded 07/24/24 11:51 AM UTC
8360 JOURNAL OF CLIMATE VOLUME 27

Extreme precipitation is typically quantified by the rain

rate for some cumulative percentile of the rain frequency
distribution (e.g., Allen and Ingram 2002; Pall et al. 2007;
O’Gorman and Schneider 2009). Using our global dis-
tribution of rain frequency, we form the cumulative
probability distribution of rainfall. We include both wet
and dry days, following other studies (e.g., O’Gorman
and Schneider 2009). We then interpolate the rain rates
from this cumulative distribution onto a logarithmic
percentile axis to compare the rate rain as a function of
percentile between an initial distribution and a later one.
We report the percent change in rain rate at a given
percentile of the distribution. The large range of rain
rates makes relative change a convenient metric.
Last, we put error bounds on our calculated distributions.
To estimate the error in model simulations, we use the
spread across models as a measure of variability and de-
termine the 95% confidence interval using Student’s t test
assuming each model is independent. This method will
underestimate the true uncertainty since climate models
are not independent (Masson and Knutti 2011; Pennell and
Reichler 2011). To estimate the error of our observed
composites, we calculate the distribution for each month of
the 50 months of data separately and then determine the
95% confidence interval using the difference-of-means test.

4. The rain amount distribution

In this section, we look at similarities and differences
among modeled and observed rain amount and rain
frequency distributions. The global-mean distributions
of rain amount and rain frequency for both GPCP and
the CMIP5 multimodel mean are shown in Fig. 1. The
multimodel mean is taken from the first 10 yr of the
RCP8.5 simulation so that the time period is similar to
(though not exactly the same as) the GPCP time period.
To address potential inconsistency between models and
observations due to their differing resolution, the GPCP
distribution is calculated as follows: The GPCP data are
coarsened to each model’s grid, the distribution of the
coarsened GPCP data at each grid point is calculated,
each distribution of coarsened data is averaged globally,
and then the global distributions are averaged together
to obtain the distributions shown in Fig. 1. The GPCP FIG. 1. Global (a) rain amount (mm day21) and (b) rain fre-
and model rain amount distributions agree broadly. quency distribution (%) climatologies in CMIP5 models and GPCP
They both peak around 10 mm day21 and are in agree- observations of daily rainfall. Dry-day frequency is noted at the top
left of (b), with models (normal) and GPCP (thick). GPCP gridded
ment for moderate rain rates. However, the model rain observations are coarsened before distributions are calculated.
amount is slightly higher at the highest and lowest rain Gray lines show the 95% confidence interval on the multimodel
rates. The GPCP rain amount distribution is narrower mean according to Student’s t test across the model distributions.
than the multimodel mean. The area under the rain
amount curve is the global-mean rainfall, which is lower
in GPCP, at 2.7 mm day21, than in the models, at
2.9 mm day21. Stephens et al. (2012) argued that GPCP

Brought to you by NATIONAL ATMOSPHERIC RESEARCH LAB | Unauthenticated | Downloaded 07/24/24 11:51 AM UTC
15 NOVEMBER 2014 PENDERGRASS AND HARTMANN 8361

underestimates the total rainfall compared to what is Two modes of change: Shift and increase of rain
needed to balance observed atmospheric radiative cooling. amount
The multimodel-mean rain frequency distribu-
In this section, we will formulate some simple modes
tion agrees relatively well with GPCP above about
of change of the rain amount distribution. Before con-
10 mm day21 (Fig. 1b). At lighter rain rates, between
sidering changes in rain, we must define some terms. The
about 0.1 and 5 mm day21, models have a much higher
rain rate r is in millimeters per day, though we will work
frequency of rain than GPCP. Modeled rain frequency
in coordinates of lnr. The cumulative rain amount P(r) is
peaks at 1.1 mm day21, where it is nearly twice as fre-
quent as GPCP, while the GPCP rain frequency peaks at ð lnr
10.6 mm day21. Models and GPCP disagree drastically P(r) 5 _ d lnr_ ,
p(lnr) (1)
lnrmin
on the dry-day frequency. The dry-day frequency is 22%
in the multimodel mean but 41% in GPCP. Extending
where p(lnr) is the rain amount distribution. Dots in-
the precipitation bins below 1 mm day21 heightens the
dicate placeholder variables over which the integral is
discrepancy compared to studies that include this light
taken. The units of P are millimeters per day, and
rain in the dry-day count.
p 5 dP/d lnr. Then the total rainfall (the global-mean
Most models have a shape similar to GPCP (see Fig. S1
precipitation) is P 5 P(‘) 5 P(rmax ). The dry-day threshold
in the supplemental material; see also Pendergrass 2013),
is rmin , and the maximum daily rainfall in our dataset is
with a peak somewhere slightly above 10 mm day21.
rmax . We assume P(rmin ) 5 0. The cumulative frequency
Some models have an extra feature on the high end of
distribution of rain F(r) is
the distribution, while others have an extra bump at light
rain rates. Many models have distributions that closely ð lnr ð lnr
resemble GPCP, including the MPI-ESM models F(r) 5 _ lnr_ 5 Fd 1
f (lnr)d _ lnr_ ,
f (lnr)d (2)
2‘ lnrmin
(Mauritsen et al. 2012).
TRMM and GPCP datasets agree on the general shape
of the distribution but disagree on many details (Fig. S2 where f (lnr) is the rain frequency distribution, F(‘) 5
in the supplemental material). TRMM has a broader F(rmax ) 5 1, and Fd 5 F(rmin ) is the dry-day frequency.
distribution of rainfall than GPCP, with more rain at low The rain amount p and rain frequency f distributions are
and high rain rates, particularly over ocean. The datasets related by
agree closely on the total amount of precipitation. p(lnr) 5 rf (lnr). (3)
Overall, GPCP and TRMM show better agreement on
rain amount than rain frequency, just as GPCP and We have d lnr 5 dr/r, and as long as changes are small,
models agree better on total amount than frequency. we can approximate the infinitesimal changes with fi-
The rain amount distribution emphasizes higher rain nite ones (D lnr 5 Dr/r), expressed as a percentage.
rates more than the rain frequency distribution does. Changed distributions will be indicated by primes ( f 0 ,
Models and observations agree better at higher rain rates p0 , and Fd0 ).
than lower ones, so the modeled rain amount distribution How might the distribution of rain amount and rain
is more similar to GPCP than the frequency distribution. frequency respond to changes in climate? One way rain
The similarity in rain amount despite differences in rain could change is that it could become more frequent. The
frequency could be anticipated because rain amount is an shape of the distribution of rain when it is raining would
important factor in the energy budget, and models are stay the same, but it would rain more often, with more rain
roughly in energetic balance. On other side of this coin, falling at each rain rate. We will call this mode the increase
having many or few light rain events makes only a small mode. If the rain increases by a fraction a at each rain rate,
difference in terms of latent heating and so are not
strongly constrained by the energy balance. p0 (lnr) 5 (1 1 a)p(lnr), which also requires (4)

f 0 (lnr) 5 (1 1 a)f (lnr) . (5)

5. The shift and increase modes of change
We focus our analysis on the change in each distri- The increase mode results in a decreased dry-day frequency,
bution relative to its own base state to account for Fd0 5 1 2 (1 1 a)(1 2 Fd ), and an increase in total rainfall
the differences in climatological distributions among P0 5 (1 1 a)P. The increase or decrease in rain must be
models and observations. There is no need to assume an balanced by a corresponding increase or decrease in total
underlying functional form (e.g., gamma or lognormal) atmospheric cooling. We can also have negative a, where
of the distribution for our analysis. the total precipitation and frequency of rainfall decrease.

Brought to you by NATIONAL ATMOSPHERIC RESEARCH LAB | Unauthenticated | Downloaded 07/24/24 11:51 AM UTC
8362 JOURNAL OF CLIMATE VOLUME 27

FIG. 2. Schematic of the modes of change of the rain distribution introduced here: (a),(c) the
increase (purple) and (b),(d) the shift (green) in (a),(b) the rain amount distributions and (c),(d)
the accompanying rain frequency distributions. Initial distributions from GPCP are shown in black,
and the new distributions resulting from a large (30%) shift or increase are shown in color.

A response consisting of the increase mode alone is un- neutral, not requiring a change in atmospheric radiative
likely for several reasons. For example, it does not allow for cooling. If the distribution shifts to the right by b 5 Dr/r,
an increase in the maximum rain rate. We expect that the then this shift mode is
rainfall intensity will increase where convergence occurs or
where air flows over topography by at least 7% K21 be- p0 (lnr) 5 p(lnr 2 b) . (6)
cause the specific humidity will increase with temperature.
However, the magnitude of global-mean precipitation in- If the same amount of rain falls at higher rain rates, then
crease is limited by energetic constraints to be 1%–2% K21 less time is spent raining. So, we should expect the rain
(e.g., Pendergrass and Hartmann 2014a). For the increase frequency to decrease and dry-day frequency to increase
in global-mean precipitation to occur at a different rate with the shift mode. With some algebra, we obtain the
than events driven by convergence and upward motion, new frequency distribution,
which are generally the events with the highest rain rates, f 0 (lnr) 5 e2b f (lnr 2 b) and (7)
the shape of the precipitation distribution must change.
Another possible response is that the rain amount 1 2 Fd
distribution could shift to higher rain rates. Then, the Fd0 5 1 2 . (8)
eb
same amount of rain would fall, but at higher rain rates.
If the distribution stays the same shape, enclosing the The total rainfall does not change, so P0 5 P. A sche-
same total area, then this change will be energetically matic of the increase and shift modes is shown in Fig. 2.

Brought to you by NATIONAL ATMOSPHERIC RESEARCH LAB | Unauthenticated | Downloaded 07/24/24 11:51 AM UTC
15 NOVEMBER 2014 PENDERGRASS AND HARTMANN 8363

The modes we have chosen have convenient mathe-

matical forms and focus on key energetic concepts, but
others are possible. For example, one can imagine
a mode where the shape of the rain frequency distribu-
tion and the total frequency of rain do not change, but
rain amount increases when it is raining. This would be
a shift of the rain frequency distribution to higher rain
rates, f 0 (lnr) 5 f (lnr 2 D lnr), analogous to the shift of
the rain amount distribution in the shift mode. In fact,
this movement of the rain frequency distribution with no
change in shape is equivalent to the shift and increase
modes simultaneously changing by the same amount,
a 5 b (for small changes a, b  1 when f and p can be
linearized). Because this mode is degenerate with the
sum of the shift mode and the increase mode, we will not
calculate it independently.
We can calculate the magnitudes of the shift and in-
crease modes that produce the closest possible fit to the
modeled or observed response to warming. We will call
their superposition the ‘‘shift plus increase.’’ To opti-
mize the shift plus increase, we make some assumptions. FIG. 3. CMIP5 multimodel-mean rain amount response to CO2
First, we assume that all changes are small, so changes in doubling normalized by the increase in global-mean surface air
p can be linearized, temperature (mm day21 K21). The model response (black), in-
crease (purple), shift (green), and shift plus increase (pink) are
dp shown. Gray lines show 95% confidence interval of the multimodel-
p(lnr 2 b) ’ p(lnr) 2 b . (9)
d lnr mean response.

The model distribution and its response are discrete

distributions, so we use a discrete approximation. Next, extensively in Pendergrass and Hartmann (2014b). The
we must choose an error metric. We choose the sum sum of squared differences will emphasize the largest
of the square of the difference between the shift plus differences, but we are more interested in how much of
increase and the observed or model response, E 5 the change in rainfall is captured by the response. We
S(Dp 2 Dpm )2 , where Dpm is the model response, Dp 5 report the error in terms of how much of the pre-
p0 2 p, and the sum is taken over all r bins. cipitation response it fails to capture,
To fit the magnitude of the increase and shift, a and b,
we find where E has a local minimum with respect to SjDp 2 Dpm j
Error 5 . (11)
each of a and b. This produces the following two- SjDpm j
variable linear set of equations,
2 3
dp 2 3
6
6
Sp 2 2Sp
d lnr 7  
7 a
pDpm
6 7 6 7 6. Response to ENSO phases in models and
6 7 54 5.
4   25 b dp
dp dp 2 Dp m observations
2Sp S d lnr
d lnr d lnr In this section, we ask the following: Do the shift and
(10) increase in the rain amount distribution provide a good
description of observed rainfall change? Gridded daily
The optimal shift plus increase for the multimodel- observational precipitation datasets only go back to
mean response to CO2 increase is shown in Fig. 3. The 1997, so at present we would not expect to see statisti-
magnitude of the shift mode is 3.3% K21 and the in- cally significant global trends in rain frequency or rain
crease mode is 0.9% K21. Between about 1 and amount in observations. Instead of looking at trends, we
50 mm day21, the shift plus increase falls within the range explore the response of tropical rain to ENSO. We look
of uncertainty in the model response. At the highest rain at changes in tropical (308S–308N) precipitation in the
rates and at light rain rates, the shift plus increase un- GPCP 1DD and TRMM 3B42v7 datasets (described in
derestimates the rain amount response, which we discuss section 2) composited over warm and cold phases of

Brought to you by NATIONAL ATMOSPHERIC RESEARCH LAB | Unauthenticated | Downloaded 07/24/24 11:51 AM UTC
8364 JOURNAL OF CLIMATE VOLUME 27

FIG. 4. Rain amount response in the tropics to ENSO phases in (a) GPCP, (b) TRMM, and (c) AMIP models.
Warm minus cold ENSO precipitation from GPCP 1DD over the tropics (308N–308S) is shown. Gray lines show 95%
confidence interval of the change according to the difference-of-means test on the observations.

ENSO. The response to ENSO can be studied in both qualitatively somewhat different responses. The sharper
observations and models. So we can also ask whether response seen in GPCP compared to TRMM is consis-
model changes are similar to the observed changes. tent with its steeper climatological distribution (Fig. S2).
Figure 4 shows the difference between warm and Table 1 lists the shift, increase, error, and total rainfall
cold ENSO rain amount distributions (along with the response for dataset and region. In all cases, the shift-
fitted shift and increase modes) for the two sets of ob- plus-increase fit the response to ENSO phase within the
servations as well as the multimodel mean. The dif- uncertainty at most rain rates (Fig. 4 and Figs. S3 and S4
ference in rain amount distribution crosses zero near in the supplemental material). Figures 5a,c show the
the peak of the climatological rain amount distribution, magnitude of the shift and increase modes over the
with a local maximum at high rain rates and minimum whole tropics, sea, and land. Over land, the shift and
at lower rain rates. This is consistent with Allan and increase closely fit the rain amount response in all cases
Soden (2008), though they looked at changes as a func- (Fig. S3). Both observational datasets and the AMIP
tion of percentile of the cumulative frequency distribu- models show a large negative increase mode of 25%–
tion. GPCP and TRMM observational datasets have 26% K21 over land, which is slightly larger than the

Brought to you by NATIONAL ATMOSPHERIC RESEARCH LAB | Unauthenticated | Downloaded 07/24/24 11:51 AM UTC
15 NOVEMBER 2014 PENDERGRASS AND HARTMANN 8365

TABLE 1. The shift and increase in response to ENSO phases for GPCP and TRMM observations and AMIP model simulations, along
with the error of the fit and the change in mean precipitation. For the AMIP simulations, both the fit to the multimodel mean (MMM) and
the average of the fit to each model are shown. Temperature changes are taken from tropical average HadCRUT4 for observational
datasets and surface air temperature for models.

Dataset Shift (% K21) Increase (% K21) Error DP (% K21)

Whole tropics
GPCP 14 6.8 0.31 6.1
TRMM 14 27.8 0.21 26.1
AMIP mean of models 15 24.7 0.51 21.8
AMIP MMM 15 25.2 0.49 21.8
Sea
GPCP 20 19 0.21 17
TRMM 18 20.85 0.17 20.92
AMIP mean of models 20 2.0 0.48 3.9
AMIP MMM 22 1.2 0.31 3.8
Land
GPCP 23.6 226 0.30 222
TRMM 25.7 225 0.24 222
AMIP mean of models 23.3 225 0.42 221
AMIP MMM 24.5 224 0.32 221

decrease in total rainfall. The datasets also agree on GPCP has an increase of 6.1% K21. AMIP models fall in
a shift of rain amount to lighter rain rates, ranging from between the two, with 1.8% K21 of decrease. The changes
23.3% to 25.7% K21. Most of the response to ENSO in atmospheric energy balance over ENSO are unlikely to
phase over land is a decrease in rain amount. be the same as in response to CO2 increase because both
Over the ocean, the observational datasets disagree on the nature of the forcing and the spatial structures of the
the response in total rainfall (Fig. 5c). Interannual vari- changes are different. Energy transport into and out of the
ability over the ocean in the previous version of TRMM, tropics plays an important role in the relationship between
version 6, was different from other observational datasets tropical surface temperature and precipitation (Su and
(both GPCP and satellite; Liu and Allan 2012). As a result Neelin 2003), unlike in the global average case.
of the difference in the total rainfall response, the obser- The observational datasets and AMIP models agree on
vational datasets disagree on the sign of the increase mode a shift between 18% and 22% K21 over ocean (Fig. 5a),
over ocean. TRMM has a decrease of 6.1% K21, while despite the disagreement in total rainfall response. The rain

FIG. 5. (a),(b) The shift and (c),(d) the increase modes fit to (a),(c) the tropical response to ENSO phases in the
CMIP5 multimodel mean and (b),(d) the global response to CO2 increase. Units are percent per kelvin. The total
(black), ocean (blue), and land (brown) are included.

Brought to you by NATIONAL ATMOSPHERIC RESEARCH LAB | Unauthenticated | Downloaded 07/24/24 11:51 AM UTC
8366 JOURNAL OF CLIMATE VOLUME 27

FIG. 6. Extreme precipitation response (% K21) to ENSO phases in (a) GPCP and (b) TRMM observations and
(c) AMIP models.

amount response to ENSO is dominated by a shift to higher and models. Next we consider the implications of the
rain rates in both observational datasets and models. shift and increase on the extreme events.
The response to ENSO phases differs from the response to
a. Extreme events
increasing CO2 [the response to CO2 increase is explored in
Pendergrass and Hartmann (2014b)]. Figures 5b,d compare Now we turn to the response of the extreme end of the
the magnitude of the shift-plus-increase fit to the modeled frequency distribution: How does observed tropical ex-
responses to CO2 increase and ENSO phases. The differ- treme precipitation respond to ENSO, and do models
ences between land and ocean in response to CO2 increase capture the observed changes? The response of extreme
are much smaller than the differences in response to ENSO precipitation to ENSO phases in both observational
phases. The magnitude of the shift is much bigger in response datasets is shown in Fig. 6. All datasets and the fitted
to ENSO (15% K21) than in response to CO2 increase shift plus increases show increases in extreme events
(3.3% K21) for the total area and over the ocean. Finally, the beyond the 99th percentile in the tropical mean. At the
increase mode has a larger magnitude in the model response 99.99th percentile, the increases are highest in AMIP
to ENSO phases than in response to CO2 increase. models (19% K21), followed by TRMM (12% K21) and
It is remarkable that the shift and increase fit the rain then GPCP (5% K21). At the 99th percentile, GPCP (at
amount response to ENSO phases in both observations 10% K21) has more increase than TRMM (5% K21),

Brought to you by NATIONAL ATMOSPHERIC RESEARCH LAB | Unauthenticated | Downloaded 07/24/24 11:51 AM UTC
15 NOVEMBER 2014 PENDERGRASS AND HARTMANN 8367

FIG. 7. Change in extremes over land in models in response to (a) CO2 doubling and (b) ENSO phases in AMIP experiments.

while AMIP is similar to GPCP (10% K21). Up to the than for increases expected over the next century in
99th percentile, the shift plus increase is within the un- models (O’Gorman 2012). Allan et al. (2013) also found
certainty in each observational dataset. At the 99.9th muted increase of extreme precipitation for climate
and 99.99th percentiles, the shift plus increase un- change relative to interannual variability.
derestimates the increase for AMIP models. Land and ocean could be expected to have different
The change in extreme precipitation over ocean (Fig. S5 responses to ENSO or CO2 increase for physical reasons
in the supplemental material) follows the same pattern as (Trenberth 2011). Figure 7a shows the multimodel-
the whole tropics. GPCP and TRMM responses over land mean response to CO2 doubling over land (at all lat-
(Fig. S6) are similar to the AMIP response (Fig. 7b). Un- itudes). It is very similar to the total response, except for
like for CO2 doubling, the response of extreme rain rate somewhat better agreement among models than for the
over land is very different from the total response (con- global response. The intermodel standard deviation of
sistent with Allan et al. 2013). The ‘‘shift plus decrease’’ 99.99th-percentile increase is 7.2% K21 over ocean and
predicts a decrease in rain rate at nearly all percentiles in 3.9% K21 over land.
both observational datasets and AMIP experiment. The
b. Sorting models by comparing their response to
only exception is that in GPCP, the most extreme per-
ENSO phases with observations
centiles do not show significant change. The shift plus in-
crease predicts a larger decrease in rain rate at all Can we discern whether some models are more con-
percentiles beyond the 99th in all datasets over land. The sistent with observations than others by comparing AMIP
models agree less over land than over ocean, unlike in re- experiments with observations? O’Gorman (2012)
sponse to CO2 doubling. The standard deviation of the showed that the responses of interannual variability and
99.99th-percentile rain-rate response is 16% K21 over long-term change in each model’s 99.9th-percentile pre-
land and 10.8% K21 over ocean. For the entire tropics, the cipitation response are related and used the modeled re-
standard deviation among models is lower than for either lationship between interannual variability and long-term
land or ocean, at just 9.0% K21. This is consistent with change to predict the long-term change based on ob-
shifting of rain (spatially) from land to ocean resulting in served variability. For the 14 models analyzed here with
some compensation across models. simulations for both RCP8.5 and AMIP experiments, the
The increase in extreme precipitation implied by the 99.99th-percentile response due to twenty-first-century
shift plus increase is tightly tied to the magnitude of the climate forcing and ENSO phase are positively corre-
shift mode. The finding that the shift is much greater for lated, significant at the 90% (but not 95%) confidence
changes during ENSO events than for the response to level. In these models, the shift mode in response to
CO2 doubling (by a factor of at least 4) is consistent with ENSO phase and RCP8.5 experiments is also correlated
the finding that the sensitivity of 99.9th-percentile pre- (significant at 95% confidence). If the observed response
cipitation is 2.5 times greater for interannual variability to ENSO phase clearly favored some models over others,

Brought to you by NATIONAL ATMOSPHERIC RESEARCH LAB | Unauthenticated | Downloaded 07/24/24 11:51 AM UTC
8368 JOURNAL OF CLIMATE VOLUME 27

TABLE 2. CMIP5 models sorted by whether or not their tropical response to ENSO phases in AMIP simulations are consistent with
GPCP and TRMM observed responses. To be considered consistent, a model’s shift must be within one standard deviation (6.5% K21) of
the observed shift, 14% K21, and its 99.99th-percentile rain-rate response (DR99:99 ) must also fall with the uncertainty range which spans
from 22% to 21% K21. Only models participating in AMIP as well as either CO2-doubling or RCP8.5 experiments are included. Global
responses are shown for CO2-doubling and RCP8.5 experiments.

AMIP 2 3 CO2 RCP8.5

Shift DR99:99 Shift DR99:99 Shift DR99:99
Model Expanded model name (% K21) (% K21) (% K21) (% K21) (% K21) (% K21)
Consistent
BNU-ESM Beijing Normal University–Earth 12 3.4 2.3 9.0 6 4.7
System Model
CMCC-CM Centro Euro-Mediterraneo per I 19 2.6 1.8 8.2 6 3.2
Cambiamenti Climatici Climate
Model
CSIRO Mk3.6.0* Commonwealth Scientific and 9.9 8.5 3.6 5.0 6 2.4
Industrial Research Organisation
Mark 3.6.0
CanAM4 Fourth Generation Canadian Coupled 15 14 2.6 8.9 6 2.3 1.9 7.5 6 3.6
Atmospheric Global Climate Model
FGOALS-g2 Flexible Global Ocean–Atmosphere– 14 8.3 0.41 1.1 6 1.9
Land System Model, gridpoint
version 2
INM-CM4.0 Institute of Numerical Mathematics 13 9.8 2.8 2.3 6 1.4
Coupled Model, version 4.0
IPSL-CM5B-LR* L’Institut Pierre-Simon Laplace Coupled 18 20 3.7 3.4 6 2.2 4.0 3.2 6 3.5
Model, version 5B, low resolution
MPI-ESM-LR Max Planck Institute Earth System 19 17 5.7 6.3 6 2.5 4.7 6.8 6 2.9
Model, low resolution
MPI-ESM-MR Max Planck Institute Earth System 17 9.8 5.6 6.9 6 2.5
Model, medium resolution
Not Consistent
ACCESS1.3 Australian Community Climate and 21 0.37 4.0 4.5 6 1.9
Earth-System Simulator,
version 1.3
BCC-CSM1.1 Beijing Climate Center, Climate System 30 7.9 4.8 5.8 6 0.84 5.2 8.1 6 1.3
Model, version 1.1
CCSM4 Community Climate System Model, 15 27 4.7 9.7 6 2.6 2.8 14 6 8.7
version 4
CNRM-CM5 Centre National de Recherches 5.9 21 2.0 7.7 6 3.5
Météorologiques Coupled Global
Climate Model, version 5
FGOALS-s2 Flexible Global Ocean–Atmosphere– 25 29 0.18 6.6 6 3.2
Land System Model, second spectral
version
GFDL CM3 Geophysical Fluid Dynamics Laboratory 21 30 2.5 9.4 6 5.5 2.8 13 6 7.6
Climate Model, version 3
IPSL-CM5A-LR L’Institut Pierre-Simon Laplace Coupled 4.9 17 2.7 12 6 2.8 2.1 21 6 4.2
Model, version 5A, low resolution
IPSL-CM5A-MR L’Institut Pierre-Simon Laplace Coupled 0.96 13 2.1 12 6 1.8
Model, version 5A, mid resolution
MIROC5 Model for Interdisciplinary Research on 6.8 11 20.16 4.8 6 2.6 0.77 8.0 6 3.2
Climate, version 5
MRI-CGCM3 Meteorological Research Institute 20 31 5.2 16 6 3.1
Coupled Atmosphere–Ocean General
Circulation Model, version 3
NorESM1-M Norwegian Earth System Model, 14 25 3.7 4.1 6 2.2 3.7 9.0 6 5.3
version 1 (intermediate resolution)

* Not consistent when rain-rate response criterion is applied at the 99.9th percentile.

Brought to you by NATIONAL ATMOSPHERIC RESEARCH LAB | Unauthenticated | Downloaded 07/24/24 11:51 AM UTC
15 NOVEMBER 2014 PENDERGRASS AND HARTMANN 8369

we might take that as an indication that those models produce a higher frequency of rain falling at light rain
were closer to reality and more heavily weight their rates compared to observations, but the rain amount
response to global warming. The companion paper, contributed at these light rates is small.
Pendergrass and Hartmann (2014b), focuses on the re- We introduce two modes of coherent change of the
sponse to CO2 increase in models. One of the findings rain amount distribution: an increase in rain amount at
there is that some models have an extreme mode, which all rain rates (the increase mode) and a shift of the rain
is an increase in rain rate at the highest percentiles of the amount distribution to higher rain rates (the shift
distribution, while other models’ responses are closely fit mode). The increase changes total rainfall, while the
by the shift plus increase. shift does not. The degree of shift is expressed as a per-
A challenge with this approach is that the two obser- centage change in rain rate, representing a movement of
vational datasets disagree about many aspects of the re- the rain amount distribution along an axis of the log of
sponse to ENSO. They disagree about the sign of the total rain rate. The increase is expressed as a percentage
change in rainfall. They disagree about whether the change in rain amount at all rain rates.
heaviest events should increase at rates greater (TRMM) We fit the shift and increase modes of the rain amount
or less (GPCP) than the shift-plus-increase prediction. distribution to the modeled and observed response to
However, they do agree that the magnitude of the shift warm and cold ENSO phases. Observational datasets
mode is 14% K21. They put a wide margin on the mag- disagree with each other about the change in mean rainfall
nitude of the increase at the 99.99th percentile, from over ocean in response to ENSO phases. Nonetheless, the
23% to 21% K21 (the bottom of the GPCP range shift plus increase fits much of the change in rain amount
through to the high end of TRMM). We use two mea- reasonably, reinforcing our confidence in the framework.
sures to sort the models: first, the 99.99th-percentile Despite disagreement about total rain in the observational
rain-rate response within the error bounds of either ob- datasets, AMIP model simulations and both observational
servational dataset (from 22% to 21% K21, a wide range of datasets produce a similar shift, between 14% and 15%
responses) and, second, a shift within one standard deviation K21 during ENSO events. Models and observations agree
(6.5% K21) of the observational datasets (8%–20% K21). on the differences between land and ocean responses to
The range of acceptable model responses is quite wide, ENSO events. The response of the rain frequency as
but half of models do not fall within it. Two more models a function of rain rate and extreme rain rate as a function
are inconsistent when the rain-rate response criterion is of percentile are also included in the framework.
applied at the 99.9th, rather than the 99.99th, percentile.
Table 2 lists the models that do and do not meet the cri- Acknowledgments. CMIP5 modeling groups gener-
teria for consistency. Only models that also participated ously shared their simulations, and the resulting data
in the RCP8.5 or CO2-increase experiments are shown. were distributed by PCMDI. GPCP 1DD and TRMM
Models that are consistent with observations according to 3B42 v7 groups generously made their data available.
this criterion have increases in 99.99th-percentile rain rate NOAA merged air land and SST anomalies data were
for RCP8.5 or CO2-doubling experiments that are at most provided by the NOAA/OAR/ESRL/PSD, Boulder,
9.0% K21. Furthermore, in Pendergrass and Hartmann (2014b) Colorado, from their website ([Link]
we show that one consistent model (MPI-ESM-LR) has a large psd/). Dargan Frierson participated in helpful discus-
initial convective fraction while one inconsistent model sions, and Kevin Trenberth and three anonymous
(IPSL-CM5A-LR) has a small initial convective fraction. reviewers provided helpful feedback. This work was
Models with large increases in extreme precipitation funded by NSF under Grant AGS-0960497.
participating in the AMIP experiment are inconsistent
with observations according to the criteria used here.
APPENDIX
7. Conclusions
Rain Amount and Rain Frequency Distribution
We have considered the rain amount and rain fre-
Calculations
quency distributions of daily precipitation data in
CMIP5 models and in GPCP 1DD and TRMM 3B42 We calculate the distribution of rain amount p and
version 7 observational datasets and their changes in rain frequency f for each dataset, using daily rain ac-
response to warm and cold phases of ENSO. CMIP5 cumulation r from model output or gridded observa-
models do a tolerable job of simulating the global dis- tions. The Rli and Rri are left and right bin edges, and
tribution of rain amount by rain rate when compared Rci 5 (Rli 1 Rri )/2 is bin centers, which we use in trans-
to the GPCP 1DD observational dataset. The models forming the distribution,

Brought to you by NATIONAL ATMOSPHERIC RESEARCH LAB | Unauthenticated | Downloaded 07/24/24 11:51 AM UTC
8370 JOURNAL OF CLIMATE VOLUME 27

ð lnRr 127, 2204–2210, doi:10.1175/1520-0493(1999)127,2204:

1 i
pi (Rci ) 5 p(lnr)d lnr [Link];2.
D lnR lnRli Klein Tank, A. M. G., F. W. Zwiers, and X. Zhang, 2009: Guide-
Agridpt lines on analysis of extremes in a changing climate in support
1
5
D lnR
 r(Rli , r , Rri )
Atotal
, (A1) of informed decisions for adaptation. World Climate Data and
gridpts Monitoring Programme Rep. 72 and WMO Tech. Doc. 1500,
56 pp.
ð lnRr Kopparla, P., E. Fischer, C. Hannay, and R. Knutti, 2013: Improved
1 i
fi (Rci ) 5 f (lnr)d lnr simulation of extreme precipitation in a high resolution at-
D lnR lnRli mosphere model. Geophys. Res. Lett., 40, 5803–5808,
doi:10.1002/2013GL057866.
1 Nd (Rli , r , Rri ) Agridpt
5
D lnR
 SNd Atotal
, and Lau, W. K.-M., H.-T. Wu, and K.-M. Kim, 2013: A canonical re-
sponse of precipitation characteristics to global warming from
gridpts
CMIP5 models. Geophys. Res. Lett., 40, 3163–3169,
(A2) doi:10.1002/grl.50420.
Liu, C., and R. P. Allan, 2012: Multisatellite observed responses
1 Agridpt of precipitation and its extremes to interannual climate
Fd 5
SNd
 Nd (r 5 0)
Atotal
, (A3) variability. J. Geophys. Res., 117, D03101, doi:10.1029/
gridpts 2011JD016568.
——, ——, and G. J. Huffman, 2012: Co-variation of temper-
where A represents areas of each grid point and the ature and precipitation in CMIP5 models and satellite
total area and Nd is the number of days. The distri- observations. Geophys. Res. Lett., 39, L13803, doi:10.1029/
2012GL052093.
butions are calculated a bin width D lnR 5
Masson, D., and R. Knutti, 2011: Climate model genealogy. Geo-
(Ri11 2 Ri )/Ri 5 7:67%, which gives reasonable sam- phys. Res. Lett., 38, L08703, doi:10.1029/2011GL046864.
pling across the distribution. Mauritsen, T., and Coauthors, 2012: Tuning the climate of a global
model. J. Adv. Model. Earth Syst., 4, M00A01, doi:10.1029/
2012MS000154.
REFERENCES
O’Gorman, P. A., 2012: Sensitivity of tropical precipitation ex-
Allan, R. P., and B. J. Soden, 2008: Atmospheric warming and the tremes to climate change. Nat. Geosci., 5, 697–700,
amplification of precipitation extremes. Science, 321, 1481– doi:10.1038/ngeo1568.
1484, doi:10.1126/science.1160787. ——, and T. Schneider, 2009: The physical basis for increases in
——, C. Liu, M. Zahn, D. A. Lavers, E. Koukouvagias, and precipitation extremes in simulations of 21st-century climate
A. Bodas-Salcedo, 2013: Physically consistent responses of change. Proc. Natl. Acad. Sci. USA, 106, 14 773–14 777,
the global atmospheric hydrological cycle in models and doi:10.1073/pnas.0907610106.
observations. Surv. Geophys., 35, 533–552, doi:10.1007/ Pall, P., M. Allen, and D. Stone, 2007: Testing the Clausius–
s10712-012-9213-z. Clapeyron constraint on changes in extreme precipitation
Allen, M. R., and W. J. Ingram, 2002: Constraints on future changes under CO2 warming. Climate Dyn., 28, 351–363, doi:10.1007/
in climate and the hydrologic cycle. Nature, 419, 224–232, s00382-006-0180-2.
doi:10.1038/nature01092. Pendergrass, A., 2013: The atmospheric energy constraint on pre-
Chen, C.-T., and T. Knutson, 2008: On the verification and com- cipitation change. Ph.D. dissertation, University of Wash-
parison of extreme rainfall indices from climate models. ington, 134 pp.
J. Climate, 21, 1605–1621, doi:10.1175/2007JCLI1494.1. ——, and D. Hartmann, 2014a: The atmospheric energy constraint
Chou, C., C.-A. Chen, P.-H. Tan, and K. T. Chen, 2012: Mech- on global-mean precipitation change. J. Climate, 27, 757–768,
anisms for global warming impacts on precipitation fre- doi:10.1175/JCLI-D-13-00163.1.
quency and intensity. J. Climate, 25, 3291–3306, doi:10.1175/ ——, and ——, 2014b: Changes in the distribution of rain fre-
JCLI-D-11-00239.1. quency and intensity in response to warming. J. Climate, 27,
Huffman, G. J., R. F. Adler, M. M. Morrissey, D. T. Bolvin, 8372–8383, doi:10.1175/JCLI-D-14-00183.1.
S. Curtis, R. Joyce, B. McGavock, and J. Susskind, 2001: Pennell, C., and T. Reichler, 2011: On the effective number of climate
Global precipitation at one-degree daily resolution from models. J. Climate, 24, 2358–2367, doi:10.1175/2010JCLI3814.1.
multisatellite observations. J. Hydrometeor., 2, 36–50, Stephens, G. L., and Coauthors, 2010: Dreary state of pre-
doi:10.1175/1525-7541(2001)002,0036:[Link];2. cipitation in global models. J. Geophys. Res., 115, D24211,
——, and Coauthors, 2007: The TRMM Multisatellite Precipitation doi:10.1029/2010JD014532.
Analysis (TMPA): Quasi-global, multiyear, combined-sensor ——, and Coauthors, 2012: An update on Earth’s energy balance in
precipitation estimates at fine scales. J. Hydrometeor., 8, 38– light of the latest global observations. Nat. Geosci., 5, 691–696,
55, doi:10.1175/JHM560.1. doi:10.1038/ngeo1580.
Jones, P. D., D. H. Lister, T. J. Osborn, C. Harpham, M. Salmon, Su, H., and J. D. Neelin, 2003: The scatter in tropical average
and C. P. Morice, 2012: Hemispheric and large-scale land- precipitation anomalies. J. Climate, 16, 3966–3977,
surface air temperature variations: An extensive revision and doi:10.1175/1520-0442(2003)016,3966:[Link];2.
an update to 2010. J. Geophys. Res., 117, D05127, doi:10.1029/ Sun, Y., S. Solomon, A. Dai, and R. W. Portmann, 2006: How often
2011JD017139. does it rain? J. Climate, 19, 916–934, doi:10.1175/JCLI3672.1.
Jones, P. W., 1999: First-and second-order conservative remapping ——, ——, ——, and ——, 2007: How often will it rain? J. Climate,
schemes for grids in spherical coordinates. Mon. Wea. Rev., 20, 4801–4818, doi:10.1175/JCLI4263.1.

Brought to you by NATIONAL ATMOSPHERIC RESEARCH LAB | Unauthenticated | Downloaded 07/24/24 11:51 AM UTC
15 NOVEMBER 2014 PENDERGRASS AND HARTMANN 8371

Taylor, K. E., R. J. Stouffer, and G. A. Meehl, 2012: An overview of ——, 2011: Changes in precipitation with climate change. Climate
CMIP5 and the experiment design. Bull. Amer. Meteor. Soc., Res., 47, 123–138, doi:10.3354/cr00953.
93, 485–498, doi:10.1175/BAMS-D-11-00094.1. ——, A. Dai, R. M. Rasmussen, and D. B. Parsons, 2003: The
Trenberth, K. E., 1997: The definition of El Niño. Bull. Amer. Meteor. changing character of precipitation. Bull. Amer. Meteor. Soc.,
Soc., 78, 2771–2777, doi:10.1175/1520-0477(1997)078,2771: 84, 1205–1217, doi:10.1175/BAMS-84-9-1205.
[Link];2. Watterson, I., and M. Dix, 2003: Simulated changes due to global
——, 1999: Conceptual framework for changes of extremes of the warming in daily precipitation means and extremes and their
hydrological cycle with climate change. Climatic Change, 42, interpretation using the gamma distribution. J. Geophys. Res.,
327–339, doi:10.1023/A:1005488920935. 108, 4379, doi:10.1029/2002JD002928.

Brought to you by NATIONAL ATMOSPHERIC RESEARCH LAB | Unauthenticated | Downloaded 07/24/24 11:51 AM UTC