1 MARCH 2021 HAGOS ET AL.

1587

The Relationship between Precipitation and Precipitable Water in CMIP6 Simulations

and Implications for Tropical Climatology and Change

SAMSON M. HAGOS,a L. RUBY LEUNG,a OLUWAYEMI A. GARUBA,a CHARLOTTE DEMOTT,b BRYCE HARROP,a
JIAN LU,a AND MIN-SEOP AHNc
a
Pacific Northwest National Laboratory, Richland, Washington
b
Colorado State University, Fort Collins, Colorado
c
University of Washington, Seattle, Washington

(Manuscript received 24 March 2020, in final form 25 November 2020)

ABSTRACT: It is well documented that over the tropical oceans, column-integrated precipitable water (pw) and pre-
cipitation (P) have a nonlinear relationship. In this study moisture budget analysis is used to examine this P–pw relationship
in a normalized precipitable water framework. It is shown that the parameters of the nonlinear relationship depend on the
vertical structure of moisture convergence. Specifically, the precipitable water values at which precipitation is balanced
independently by evaporation versus by moisture convergence define a critical normalized precipitable water, pwnc. This is a
measure of convective inhibition that separates tropical precipitation into two regimes: a local evaporation-controlled
regime with widespread drizzle and a precipitable water–controlled regime. Most of the 17 CMIP6 historical simulations
examined here have higher pwnc compared to ERA5, and more frequently they operate in the drizzle regime. When
compared to observations, they overestimate precipitation over the high-evaporation oceanic regions off the equator,
thereby producing a ‘‘double ITCZ’’ feature, while underestimating precipitation over the large tropical landmasses and
over the climatologically moist oceanic regions near the equator. The responses to warming under the SSP585 scenario are
also examined using the normalized precipitable water framework. It is shown that the critical normalized precipitable
water value at which evaporation versus moisture convergence balance precipitation decreases as a result of the competing
dynamic and thermodynamic responses to warming, resulting in an increase in drizzle and total precipitation. Statistically
significant historical trends corresponding to the thermodynamic and dynamic changes are detected in ERA5 and in low-
intensity drizzle precipitation in the PERSIANN precipitation dataset.
KEYWORDS: Climate change; Drizzle; Convective parameterization; Coupled models

1. Introduction much less clear, particularly over the tropics, where spatial and
temporal shifts related to dynamic and thermodynamic re-
Understanding and quantifying the effects of global warm-
sponses to warming can influence regional hydrology. This
ing on regional hydrological cycles is one of the most important
challenge is further exacerbated by the persistent precipitation
problems in climate science because of the societal implica-
distribution biases in generations of climate models, under-
tions. At global scale, atmospheric moisture increases with
mining our confidence in model projections of future changes
temperature under global warming at a rate that follows the
in precipitation. The biases that are well documented in the
Clausius–Clapeyron relationship of ;7% K21, while global
simulations from phases 3 and 5 of the Coupled Model
precipitation increases at a much slower rate of ;2% K21
Intercomparison Project (CMIP3 and CMIP5, respectively)
(Held and Soden 2006). This difference between the responses
include excessive precipitation over oceanic regions off of the
of precipitation and atmospheric moisture content is related to
equator (Hirota and Takayabu 2013; Fiedler et al. 2020), weak
the constraint imposed by the atmospheric radiative cooling
South Asian monsoon rainfall (Hagos et al. 2019), and weak
due to the increased temperature and humidity, which limits
Amazon precipitation (Yin et al. 2013). Understanding the
the precipitation change through the global energy and water
origin of these biases and how they relate to regional-scale
balances (Allen and Ingram 2002; Pendergrass and Hartmann
projections of precipitation changes is critical for building
2014). Regional differences of the precipitation response are
confidence in the projections.
Several studies suggest that model precipitation biases are
related to the representation of convection. For example,
Denotes content that is immediately available upon publica-
precipitation biases in coupled climate simulations can be
tion as open access.
reproduced in uncoupled Atmosphere Model Intercomparison
Project (AMIP)-style simulations (Zhang et al. 2007; Chikira
Corresponding author: Samson Hagos, samson.hagos@pnnl.gov 2010) while some modifications to the parameterization of

DOI: 10.1175/JCLI-D-20-0211.1
Ó 2021 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright
Policy (www.ametsoc.org/PUBSReuseLicenses).
Unauthenticated | Downloaded 04/29/23 12:21 PM UTC
1588 JOURNAL OF CLIMATE VOLUME 34

convection, specifically the sensitivity of convection to envi- the historical simulations and 2086–2100 for the future
ronmental humidity, appear to mitigate the biases (Song and simulations.
Zhang 2009; Hirota et al. 2011; Emori et al. 2001). The non- In Fig. 1, we compare the mean precipitation from the 17
linear relationship between convection and environmental CMIP6 models for the historical period to the mean observed
humidity over the tropical oceans has been well documented precipitation, computed as the average of TRMM and PERSIANN
(Bretherton et al. 2004; Rushley et al. 2018; Ahmed and precipitation. Compared to observations, the CMIP6 multi-
Schumacher 2015, and references therein). Several studies model mean shows excessive precipitation over the tropical
empirically approximate the relationship using an exponential oceans off the equator and drier conditions over the tropical
function with a ‘‘pick-up’’ precipitable water value at which landmasses. Figure 2 shows the climatological mean of the
precipitation starts to increase rapidly with precipitable water daily low-intensity precipitation, or drizzle, and evaporation.
(Igel et al. 2017; Sahany et al. 2014). Conceptually this rela- For reasons that will be apparent in the next section, drizzle is
tionship provides a measure of the effectiveness of environ- defined as precipitation rates less than 0.13 mm h21 (3.12 mm day21).
mental air in diluting rising plumes in moist convection (Peters The models have particularly excessive low-intensity precipi-
et al. 2009; Holloway and Neelin 2009, etc.) and the moistening tation, which has been well documented for previous genera-
effect of convection on the environment. Analyzing the non- tions of models (Sillmann et al. 2013; Liu et al. 2014; Stephens
linear relationship between precipitation and precipitable et al. 2010; Dai 2006). The corresponding evaporation fields are
water in the South Asian monsoon region and the equatorial shown in the right column of Fig. 2. Excess evaporation is also a
Indian Ocean, Hagos et al. (2019) identified the normalized common bias in previous generation of models. In the tropics,
precipitable water in the equatorial Indian Ocean as an im- this bias is driven primarily by wind biases (e.g., Small et al.
portant metric for understanding model biases in simulating 2019), but also possibly related to biases in downwelling
South Asian monsoon precipitation and the intermodel spread shortwave radiation and model resolution (Demory et al.
in future precipitation projections. 2014). The spatial similarities between the biases in total pre-
Extending the geographic focus from the South Asian mon- cipitation (Fig. 1c), drizzle (Fig. 2c), and evaporation (Fig. 2f)
soon region to the entire tropics, we aim to address two questions are striking and will be discussed later.
in this study: What are the origins of biases in CMIP6 precipi- The effects of tropical oceanic rainfall and 850-hPa wind
tation climatology over the tropics? What are the implications of biases on precipitable water over tropical landmasses are
the biases for uncertainties in the projected changes? To this end, shown in Fig. 3. Excessive CMIP6 rainfall over the tropical
we first derive the precipitation–precipitable water relationship oceans diverts moisture away from the tropical lands, weak-
discussed above from the moisture budget equation and examine ening the moisture transport to South Asia, Africa, and the
its representation in the historical and future climate simulations Amazon, resulting in dry biases over land regions. As the cir-
of the newest generation of models that participated in CMIP6 culation biases may be induced by precipitation biases over the
and in a global reanalysis. Then the relationship is normalized oceans through diabatic heating, we hypothesize that better
such that key parameters that control the model behavior in both representation of the climatology of precipitation over the
historical and projection simulations are identified and their oceans could reduce the perennial dry biases over land in
physical meanings in the context of model representation of generations of models (Sperber et al. 2013). This hypothesis
convection are discussed. will be examined in the next section.

3. The relationship between precipitation and

2. Tropical precipitation climatology in CMIP6 precipitable water
simulations
a. Derivation
We analyzed 17 historical simulations and 10 future simu-
lations following the Shared Socioeconomics Pathway (SSP585) To investigate the origins of model precipitation biases, we
from the CMIP6 model archive. The models were selected based first examine the processes behind the observed relationship
on availability of daily precipitation, precipitable water, and between precipitation and precipitable water. Consider the
evaporation output, as required by the analysis. Where it is vertically integrated moisture budget equation:
available vertically integrated moisture convergence is directly ð
›(pw) 1 ps
used; otherwise, it is estimated from the above listed variables 5E2P2 =
(vq) dp , (1)
›t g pt
using the moisture budget equation. We also analyzed data from
the ERA5 dataset (C3S climate services 2017), two observational where
precipitation datasets, and an observational evaporation dataset.
They are Precipitation Estimation from Remotely Sensed ð ps
1
Information using Artificial Neural Networks (PERSIANN; pw 5 q dp (2)
g pt
Ashouri et al. 2015) and Tropical Rainfall Measuring Mission
(TRMM 3B42; Huffman et al. 2010) and the Woods Hole is the column-integrated water vapor or precipitable water;
Objectively Analyzed Flux Project daily evaporation data E and P are evaporation and precipitation respectively, while
(OAFlux; Yu et al. 2008). All data are remapped to uniform ps and pt are the pressures at the surface and top of the at-
28 grid spacing. The study is focused on the tropics between mosphere, respectively. To obtain a relationship between
208S and 208N. We analyzed two 15-yr periods: 2000–14 for precipitation and precipitable water, the moisture convergence

Unauthenticated | Downloaded 04/29/23 12:21 PM UTC

1 MARCH 2021 HAGOS ET AL. 1589

Over the tropical oceans tropospheric temperature gradients are

weak and diabatic heating is primarily balanced by adiabatic
cooling (Sobel and Bretherton 2000; Sobel et al. 2001), leading to a
relationship among vertical velocity, static stability, and diabatic
heating (see the appendix). Using conservation of moisture, mass,
and energy, the relationship between moisture flux convergence
and precipitation was obtained in Hagos et al. (2019). Here we
focus the discussion on the physical interpretations and implica-
tions for subsequent analysis. As is apparent from (4), NMFC
plays an important role in moisture variability so its physical
interpretation deserves some discussion. Figure 4a shows the re-
lationship between wind divergence (contours) and moisture
normalized by the column integrated precipitable water [i.e.,
(q/pw)(dp/g); shading] over the tropical oceans between 208S
and 208N based on ERA5. For precipitation less than about
0.13 mm h21, there are strong upper-level convergence and near-
surface divergence indicating subsidence and convective inhibi-
tion. For precipitation greater than about 0.13 mm h21, near-
surface convergence (dashed green contours) increases and
deepens, and the height of maximum divergence (Fig. 4a, red
contours) increases. The profile of water vapor mixing ratio (i.e.,
the moisture mixing ratio normalized by precipitable water) is
displayed in shadings. This profile shows little change with in-
creasing precipitation. Increases in the height level of maximum
divergence as well as an increase in low-level moisture conver-
gence mean that a smaller fraction of moisture diverges with in-
creasing precipitation. Therefore, NMFC represents the strength
of convection not only in the sense of the magnitude of vertical
velocity but also in its effectiveness in converging moisture by
virtue of its vertical profile. The relationship between the diver-
gence profile and precipitation implies that NMFC increases with
precipitation, and the relationship is approximately linear as
demonstrated in Fig. 4b. Figure 4c shows the linear relationship
derived by calculating the mean NMFC values within equally
sized precipitation bins to find the intercept (0.13 mm h21) at
which, in a mean sense, evaporation balances precipitation [Eq.
(4)]. The value of pw at which this balance occurs, pw0, which is an
FIG. 1. 15-yr (2000–14) mean precipitation from (a) TRMM- important parameter in the subsequent discussion, is also calcu-
3B42 and PERSIANN (mm day21) and (b) multimodel mean of lated in a similar manner (Fig. 4d). Defining the inverse of the
the 17 CMIP6 historical simulations, and (c) the difference of slope of the regression line in Fig. 4c as the precipitable water limit
(b) minus (a). pwlim and the intercept as E0/pwlim,

P 2 E0
term has to be represented as a separable function of P and pw. NMFC 5 : (5)
pwlim
To do that we introduce a variable, normalized moisture flux
convergence (NMFC), as Before moving the analysis further, a brief discussion of NMFC in
ð ps the context of previous work is warranted. The normalization
1 introduced in Eq. (3) is also relevant to the concept of gross moist
2 =
(vq) dp
g pt stability (GMS) put forward by Neelin and Held (1987) and fur-
NMFC 5 (3)
pw ther developed by Raymond et al. (2009) and many others. The
relationship between these two concepts is made more explicit in
such that the moisture budget equation can be written as
the appendix. As a measure of the depth of the moisture con-
vergence field, NMFC is also related to temperature and satura-
›(pw)
5 E 2 P 1 (NMFC)pw: (4) tion. For example, if the circulation is shallow, moisture can
›t
converge and diverge close to the surface with little saturation and
As can be seen from its definition, NMFC is related to the precipitation as the near-surface temperature is warmer than the
vertical structure of wind divergence in relation to moisture temperature at upper levels, but the circulation could also be deep
profile. Therefore, by continuity it is related to vertical velocity. enough to allow significant saturation and condensation.

Unauthenticated | Downloaded 04/29/23 12:21 PM UTC

1590 JOURNAL OF CLIMATE VOLUME 34

FIG. 2. (left) 15-yr (2000–14) mean drizzle precipitation (precipitation , 3.12 mm day21) from (a) TRMM-3B42 and
PERSIANN (mm day21) and (b) multimodel mean of the 17 CMIP6 historical simulations, and (c) the difference of
(b) minus (a). (right) 15-yr (2000–14) mean evaporation from (d) Woods Hole OA Flux Project evaporation (mm day21)
and (e) multimodel mean of the 17 CMIP6 historical simulations, and (f) the difference of (e) minus (d).

Substituting Eq. (5) into Eq. (4), the moisture budget convergence. Thus, pwlim represents an asymptotic limit for
equation can be written as pw in the P–pw relationship at which precipitation is balanced
    by moisture convergence. On the other hand, E0 represents
›(pw) pwE0 pw the evaporation at grid points where there is drizzle but no
’ E2 2P 12 : (6)
›t pwlim pwlim moisture convergence. It is considered constant over the
domain and the value is given for each model (legend
To understand the physical meanings of pwlim and E0, consider
in Fig. 6).
Eq. (6) as pw approaches pwlim. In that case pw becomes less
The above derivations and analyses show that the mois-
and less sensitive to precipitation. On the other hand, com-
ture budget can be used to understand the relative control of
bining Eq. (3) and Eq. (5) yields the following:
moisture convergence and evaporation on precipitation in
ð ps   the low-intensity and high-intensity regimes. Here we will
1 pw
2 =
(vq) dp 5 (P 2 E0 ) : (7) demonstrate that the P–pw relationship can be derived from
g pt pwlim
the moisture budget using the physical quantities pwlim, pw0,
As pw approaches pwlim, the loss of moisture by the net and E0 defined above. Under steady state Eq. (6) can be
precipitation is compensated for by the gain through moisture written as

Unauthenticated | Downloaded 04/29/23 12:21 PM UTC

1 MARCH 2021 HAGOS ET AL. 1591

FIG. 3. Mean local summer precipitable water and 850-hPa circulation for three tropical landmasses from (a),(d),(g) ERA5 and (b),(e),(h)
CMIP6 MMM, and (c),(f),(i) their differences.

pw E0
E2 E P5  . (9)
pwlim 0 pw 2 pw0
P5 pw : (8) 12
12 pwlim 2 pw0
pwlim
It should be noted that (9) is not prognostic; rather, it should be
Equation (8) provides a P–pw relationship but it is not unique interpreted as a state defined by precipitation drying, precipi-
since it depends on evaporation. Of all the P–pw curves that arise table water, and the vertical structure of moisture convergence
from Eq. (8), the one that goes through (pw 5 pw0, E 5 E0) is of about which P and pw fluctuate under the prescribed evapo-
most interest to us as it represents the mean P–pw relationship ration E0.
when variability in evaporation and precipitable water ten- The analysis of Fig. 4 is repeated with all 17 CMIP6 historical
dencies are neglected. Therefore, the numerator in (8) be- simulations (Fig. 5) and the 10 SSP585 simulations (not
comes E0(1 2 pw0/pwlim). After collecting (pwlim 2 pw0) and shown), showing that the linear relationship in Fig. 4b based on
(pw 2 pw0) from the numerator and denominator respectively, ERA5 holds for the CMIP6 models. The model results are
Eq. (8) can be written as shown in blue, green, and red based on the value of pwlim, the

Unauthenticated | Downloaded 04/29/23 12:21 PM UTC

1592 JOURNAL OF CLIMATE VOLUME 34

FIG. 4. (a) The relationship of water vapor profile (shading) and wind divergence (105 s21) with precipitation in
ERA5. The blue line represents the approximate precipitation value where the NMFC is zero. (b) Moisture flux
convergence profile [NMFC; Eq. (3)] vs precipitation. The term pwlim is defined as the inverse of the slope of the
regression line. (c) The relationship between precipitation and normalized moisture flux convergence (NMFC) for
ERA5. (d) As in (c), but against precipitable water the points at which (NMFC) is zero. Values of E0 and pw0
are shown.

asymptotic pw limit, falling in the lower, middle, and higher averaging the corresponding daily precipitation in each bin.
terciles, respectively, of the multimodel ensemble. This color Figure 7b is the idealized relationship reconstructed from
scheme is used in all subsequent plots. The parameter pwlim Eq. (9) using values of pwlim, pw0, and E0 determined for each
varies widely among models. Similarly, the x-intercept value E0 model. The idealized relationship captures the main features of
for which evaporation at NMFC equals zero [Eq. (5)] is shown the P–pw relationship and the differences among the models,
in Fig. 6a for each CMIP6 historical simulation. Note that E0 as demonstrated by the correspondence between the color
has a range of values between 0.15 to 0.19 mm h21, all higher associated with each model relative to ERA5 in the two panels.
than that for ERA5, which is 0.13mm h21. We set the Our analysis shows that, to first order, the relationship between
threshold for drizzle precipitation as the value of E0 derived precipitation and precipitable water 1) is governed by moisture
from ERA5 (E0 5 0.13 mm h21 or 3.12 mm day21). Recall conservation and 2) can be uniquely defined by two pw values
that E0 corresponds to the evaporation (and precipitation) pwlim and pw0. When pw approaches pwlim, precipitation is
rate when NMFC is equal to zero. Rain rates below this E0 balanced by moisture convergence, but when pw approaches
threshold, defined as low-intensity precipitation, are pri- pw0 precipitation is balanced by evaporation. Each critical pw
marily controlled by evaporation rather than by moisture con- value is set by the vertical structure of moisture convergence and
vergence. Note that this threshold is smaller than 0.3 mm h21 its relation to precipitation (Fig. 4). In many ways, pw0 and pwlim
typically used as the drizzle threshold in an operational context together play similar roles as wc and b of Neelin et al. (2009) in
(e.g., AMS Glossary). that they determine the ‘‘pickup point’’ and the growth rate of
Equation (9) implies that one can construct the relationship precipitation with pw. As the atmosphere warms up, they both
between P and pw from pwlim, pw0, and E0, which are already increase such that the P–pw curves shift and stretch to the right, as
obtained for each model from the above analysis. Figure 7a one would expect from the Clausius–Clapeyron relation. The
shows the actual relationship obtained by partitioning the 15 idealized P–pw relationship will be used to analyze model diver-
years of daily pw (tropical ocean points only) into 80 bins and sity and biases from the CMIP6 simulations next.

Unauthenticated | Downloaded 04/29/23 12:21 PM UTC

1 MARCH 2021 HAGOS ET AL. 1593

FIG. 5. As in Fig. 4b, but for 16 of the CMIP6 model historical simulations. The blue, green, and red colors correspond to models with low,
medium, and high historical pwlim, respectively.

b. Normalization of precipitable water pw0

pwnc 5 (11)
pwlim 2 pw0
Here we examine the origins of precipitation biases and inter-
model spread. Since the P–pw curves for the diverse models are of is defined as the normalized critical precipitable water.
the same form, appropriate normalization could yield a general Equation (10) implies that precipitation can be estimated using
relationship for a straightforward interpretation. pwn 2 pwnc and E0. The raw and normalized forms of the P–pw
Equation (9) can be written more concisely by defining a relationship are plotted in Figs. 8a and 8c, while the actual and
normalized precipitable water as pwn 5 pw/(pwlim 2 pw0), estimated P frequency distributions are plotted in Figs. 8b and 8d.
such that As predicted by Eq. (10) normalization reduces the spread in the
E0 P–pw relationships and in the pw frequency distributions. Note,
P5 , (10) however, that the normalized P–pw curves diverge near pwn 2
1 2 (pwn 2 pwnc )
pwnc 5 0.5 (Fig. 8c); because of the nonlinearity, small differences
where at higher PW lead to larger spread in precipitation than at low pw.

Unauthenticated | Downloaded 04/29/23 12:21 PM UTC

1594 JOURNAL OF CLIMATE VOLUME 34

FIG. 7. (a) The relationship between precipitation and precipi-

FIG. 6. (a) The relationship between precipitation and normal- table water from the models and ERA5, and (b) that derived from
ized moisture flux convergence (NMFC) for all the CMIP6 his- Eq. (9) using the respective E0, pwlim, and pw0 values for each
torical simulations. (b) As in (a), but against precipitable water. E0 CMIP6 model historical simulation and for ERA5. The blue, green,
and pw0 are provided in the legend. The blue, green, and red colors and red colors correspond to models with low, medium, and high
correspond to models with low, medium, and high historical pwlim, pwlim, respectively.
respectively.

pw / pwlim, P/E0 becomes very large (Figs. 8a,c) and is much

more sensitive to the precipitable water. Thus pwnc can be
To understand the physical implications of Eq. (10), con- thought of as a form of convective inhibition (e.g., the strength
sider the normalized P–pw relationship at its limits. At one of upper-level convergence and subsidence in Fig. 4a) that a
extreme with pwn  pwnc, P ’ E0/(1 1 pwnc). That is, pre- moist column must overcome to transition to deep convection.
cipitation is largely controlled by surface evaporation and is Before discussing the implications of this for model biases and
essentially independent of the local precipitable water. This projections the next section, there is one more point worth
regime corresponds to the widespread drizzle over high evap- mentioning. Equation (10) is of the form y 5 1/(1 2 x), which at
oration areas. In Fig. 4a, pwn  pwnc corresponds to strong first glance appears different from exponential form used in
subsidence (left of the blue line) where wind divergence is several previous studies to empirically fit observed and mod-
confined to shallow levels, efficiently transporting moisture out eled P–pw relationships. Note however that for small x, 1/(1 2
of the column. At the other extreme with pwn / 1 1 pwnc or x) can be approximated by 1 1 x, which is also true for ex. That

Unauthenticated | Downloaded 04/29/23 12:21 PM UTC

1 MARCH 2021 HAGOS ET AL. 1595

FIG. 8. (a) The relationship between precipitation and precipitable water in historical and ssp585 simulations and
(b) the frequency distributions of precipitable water. (c),(d) As in (a) and (b), but both pw and P are normalized
(see text). The blue, green, and red colors correspond to models with low, medium, and high historical pwlim,
respectively.

appears to us to be a mathematical coincidence. The above The difference in mean drizzle precipitation (daily precipitation ,
analysis suggests that the former is a physically sound rela- 3.12 mm day21) between the five models with lowest and highest
tionship since it follows from conservation of moisture, mass, pwnc is shown in Fig. 10. As predicted by the analysis, the low pwnc
and energy under weak temperature gradient. models have much less drizzle. These models also simulate heavier
total precipitation over Africa, South Asia, and the Maritime
4. Model biases and projections
a. Characterizing model biases and spread
The convergence of the P–pw relationship (Fig. 8c) and the
frequency distribution of pw after normalization (Fig. 8d) suggests
that pwnc, a measure of convective inhibition, and E0, the evap-
oration under neutral NMFC (i.e., NMFC 5 0), may be useful
metrics for characterizing rainfall biases in models. Figure 9
shows a scatterplot of these two parameters. In comparison to
the parameters derived from ERA5, all models have higher E0,
implying that the models produce more rain in the form of drizzle
than ERA5 in the absence of moisture convergence (Figs. 2 and
6a). Most models also have higher pwnc than ERA5, implying
stronger convective inhibition than in observations, which means
it takes more moisture to transition from drizzle to intense con-
vection in the models. The fact that most models overestimate
pwnc implies they have a greater-than-observed frequency of days
with pwn smaller than pwnc (or pw , pw0; Fig. 8d). Most of the
models have frequency distribution with respect to pwn 2 pwnc FIG. 9. (a) Scatterplot of E0 vs pwnc for the 17 CMIP6 models and
that is extended to the left compared to ERA5 (Fig. 8d). ERA5 dataset.

Unauthenticated | Downloaded 04/29/23 12:21 PM UTC

1596 JOURNAL OF CLIMATE VOLUME 34

pwnc is a function of pw0 and pwlim [Eq. (11)], its future change
represents the combined effect of both thermodynamic change
(increase in moisture represented here by increase in pwlim) and
the dynamic change (increase in inhibition or stability represented
by increase in pw0). Figure 11c shows the changes in both pwlim
and pw0. Both changes are also depicted in the rightward shift of
the P–pw curve and the pw frequency under warming (Figs. 8a,b,
dashed lines). The dominance of thermodynamic change is re-
flected in the stronger response of pwlim to warming (and there-
fore the decrease in pwnc). Both changes in pwlim and pw0 in the
models as responses to warming are related to the equilibrium
climate sensitivity (ECS) of the models (Fig. 11d). Models with
larger climate sensitivity have a larger change in pwlim as one
would expect from the Clausius–Clapeyron relation and a corre-
sponding larger increase in pw0 as stability increases as well.
c. Historical trends
In the last subsection we show that the change in precipita-
tion in response to warming can be represented by the decrease
in pwnc, that is, the net decrease in convective inhibition re-
sulting from the competing effects of increase in moisture
(pwlim) and increase in pw0 (Fig. 11c). This raises the question
of whether trends in pwlim and pw0 can be detected in the
global reanalysis and the observed precipitation. To address
this question the trends in 40 years of pwlim and pw0 values are
calculated from ERA5. Because of the nonlinear nature of the
P–pw relationship, a slight decrease in pwnc leads to a shift of
many points from the evaporation-controlled drizzle regime to
FIG. 10. (a) The difference in mean precipitation between the top
the pw-controlled regime. Thus we expect a rapid increase in
and bottom five models according to their critical normalized
precipitable water (pwnc). (b) As in (a), but for precipitation precipitation near pwnc. To verify this expectation, we com-
, 3.12 mm day21. puted the historical precipitation trends using 36 years of
PERSIANN daily drizzle precipitation (precipitation ,
3.12 mm day21 as defined above). Figure 12 shows the trends
Continent. These results support the hypothesis put forward in the normalized by their corresponding historical mean values. The
first section that the dry bias over tropical land may be related to the statistical significances of the trend in ERA5 pwlim and pw0 as
excess low-intensity precipitation over the high evaporation regions well as that of the drizzle precipitation trend from PERSIANN
of the tropical ocean that limits the moisture transported to the are tested using the Kendall’s tau (KT) test (Kornbrot 2014)
landmasses through geostrophic adjustment. and the p values shown in the figure indicate that the trends are
statistically significant. Even though there are also increasing
b. Projected change in precipitation trends in the critical normalized precipitable water from ERA5
and total precipitation from PERSIANN (not shown), they are
Finally, the response to warming is examined in the nor-
not found to be statistically significant according to the KT test.
malized precipitable water framework. To understand the
This is consistent with the competing effects of the changes in
implication of the normalization of precipitation for projected
pwlim versus pw0 on the changes in pwnc, rendering larger un-
change in precipitation, we consider the partial derivative of
certainty in estimating the changes in pwnc and the total pre-
Eq. (9) with respect to E0 and pwnc:
cipitation compared to the changes in drizzle precipitation,
 
DP DE0 P which are determined by changes in pw0 alone.
’ 2 Dpwnc : (12)
P E0 E0
5. Conclusions
Note that the term in the square brackets is nearly model in-
dependent and does not change much with warming as implied Several previous studies have documented the excess pre-
by the collapse of the P–pw curves into one after normalization cipitation over tropical oceans and dry biases over tropical
(Fig. 8c). Figure 11 shows the end of the twenty-first-century landmasses in multiple generations of CMIP models. Motivated
fractional change in precipitation under the SSP585 scenario by these persistent biases and the uncertainties in the projected
versus the corresponding change in pwnc and E0. The change in changes, we developed a new framework to study the moisture
precipitation has a stronger relationship with Dpwnc under budgets of the ERA5 dataset and CMIP6 simulations. This
warming than with DE0. Six of the 10 models show a decrease in framework reveals that the regional distribution of precipitation,
Dpwnc and they have larger fractional increase in precipitation. As specifically whether precipitation occurs preferentially over high

Unauthenticated | Downloaded 04/29/23 12:21 PM UTC

1 MARCH 2021 HAGOS ET AL. 1597

FIG. 11. (a) The percentage difference in P vs in pwnc between historical (2000–14) and SSP585 (2086–2100)
simulations. (b) As in (a), but for change in P and in E0. (c) SSP585 pw0 and pwlim (mm) vs their historical values. (d)
Change in pw0 and pwlim (mm) under the SSP585 scenario vs equilibrium climate sensitivity (ECS).

evaporation areas (which leads to ‘‘double ITCZ’’) or high pre- There is increased recognition of the role of representation
cipitable water areas (Fig. 10), is related to the vertical structure of of entrainment and detrainment processes in defining the P–pw
moisture flux convergence and how it relates to precipitation in- relationship, not just in the inhibition or promotion of con-
tensity. It is shown that the well-documented relationship between vection by environmental moisture at diurnal and subdiurnal
precipitation (P) and column integrated precipitable water (pw) time scales (Derbyshire et al. 2004; Tompkins 2001; Ahmed
can be uniquely defined by two critical pw values: the pw where and Neelin 2018) but also through the impact of convection on
precipitation balances evaporation, and the pw where precipita- the moisture budget of the column itself at longer time scales
tion balances moisture flux convergence (Fig. 6). When the raw (Kuo et al. 2017; Singh et al. 2019; Emanuel 2019). In the
pw is scaled by the difference of these two values, one can define a framework presented in this study, we use daily averages of P
normalized critical precipitable water, pwnc, which is a measure of and pw, and the analytical solutions are derived from a steady-
convective inhibition that separates tropical precipitation into two state form of conservation of moisture; thus, they represent a
regimes: a local evaporation-controlled regime with widespread steady-state relationship about which P and pw fluctuate. From
drizzle, and a precipitable water–controlled regime with heavy the short time scale perspective, the critical normalized pre-
rainfall. Compared to ERA5, most of the 17 CMIP6 model his- cipitable water derived in this study represents a measure of
torical simulations examined have both higher pwnc (convective inhibition that the normalized moisture must exceed for deep
inhibition) and E0 (evaporation and precipitation balance point), convection to take place. That inhibition depends on the how
and too much drizzle (Fig. 9). It is also shown that the magnitude rapidly the updraft deepens with increasing precipitation and
of the response to warming is also linearly related to the change in therefore to what extent the column-integrated moisture con-
pwnc. A robust increasing trend in both the pw values that define vergence compensates for the moisture loss by precipitation.
the P–pw relationship are detected in ERA5 and the implied Thus, the normalized precipitable water framework (specifi-
trend of low-intensity precipitation is also observed in the cally the critical normalized precipitable water parameter)
PERSIANN precipitation dataset (Fig. 12). could be a theoretical tool for interpreting the combined

Unauthenticated | Downloaded 04/29/23 12:21 PM UTC

1598 JOURNAL OF CLIMATE VOLUME 34

APPENDIX

Relationship between Normalized Moisture Flux

Convergence and Gross Moist Stability
Using the steady-state form of the dry static energy and
moisture equations [Eq. (2.3) in Raymond et al. 2009], nor-
malized gross moist stability can be written as
ð ps
TR =
(ys) dp
pt
GR 5 2 ð ps , (A1)
L =
(yq) dp
pt

where TR 5 300 K is a reference temperature, L is the latent

heat of condensation, y is horizontal wind, and s is specific
moist entropy. Combined with Eq. (3), (A1) can be rewritten as
FIG. 12. (a) The trends in pwlim and pw0 from ERA5 data as well
ð ps
as in the amount of drizzle precipitation (precipitation rate
, 3.12 mm day21; see text) averaged over the tropical oceans. All TR =
(ys) dp
pt
three trends are plotted as deviations from their respective long- NMFC 5 2 : (A2)
LpwGR
term means and pass the Kendall’s tau statistical significance test at
the level indicated in the legend.
Using Eq. (2.1) of Raymond et al. (2009), again for steady state,
(A2) can be written as
TR (Fs 2 R)
2GR NMFC 5 , (A3)
Lpw
effects of promotion of convection by environmental moisture
as well as the moistening effect of convection in parameteri- where Fs is the moist entropy flux due to surface fluxes of heat
zations of entrainment and detrainment in climate models. and moisture and R is the pressure integral of the entropy sink
Furthermore, although the normalized precipitable water due to radiative cooling. Equation (A3) states that the strength
framework is inherently nonspatial, it can provide insights into of convection (NMFC) is related to the diabatic heating and
model biases in the spatial distribution of rainfall, particularly stability. This is analogous to the weak temperature gradient
over tropical landmasses versus ocean. representation of the energy conservation (Holton 1992):

2Sp v 5 J/Cp , (A4)

Acknowledgments. This is work is supported by U.S.
Department of Energy Office of Science Biological and where Sp is static stability, v is pressure vertical velocity, and J
Environmental Research as part of Global and Regional is diabatic heating. Hence it is not surprising that the linear
Modeling and Analysis Program. The contribution of YG is relationship between NMFC with precipitation follows from
supported by National Oceanic and Atmospheric Administration (A4) (Hagos et al. 2019).
(NOAA) Oceanic and Atmospheric Research, Program Climate
Program Office (CPO), under NOAA Grant NA17OAR4310263. REFERENCES
Computing resources for the analysis are provided by the National
Ahmed, F., and C. Schumacher, 2015: Convective and stratiform
Energy Research Scientific Computing Center (NERSC). We
components of the moisture–precipitation relationship. Geophys.
acknowledge the World Climate Research Programme, which,
Res. Lett., 42, 10 453–10 462, https://doi.org/10.1002/2015GL066957.
through its Working Group on Coupled Modelling, coordinated ——, and J. D. Neelin, 2018: Reverse engineering the tropical
and promoted CMIP6. We thank the climate modeling groups for precipitation–buoyancy relationship. J. Atmos. Sci., 75, 1587–
producing and making available their model output, the Earth 1608, https://doi.org/10.1175/JAS-D-17-0333.1.
System Grid Federation (ESGF) for archiving the data and pro- Allen, M., and W. J. Ingram, 2002: Constraints on future changes in
viding access, and the multiple funding agencies who support climate and the hydrologic cycle. Nature, 419, 228–232, https://
CMIP6 and ESGF. We thank DOE’s RGMA program area, the doi.org/10.1038/nature01092.
Data Management program, and NERSC for making this coor- Ashouri, H., K.-L. Hsu, S. Sorooshian, D. K. Braithwaite, K. R.
dinated CMIP6 analysis activity possible. Pacific Northwest Knapp, L. D. Cecil, B. R. Nelson, and O. P. Prat, 2015:
PERSIANN-CDR: Daily precipitation climate data record
National Laboratory is operated by Battelle for the U.S.
from multisatellite observations for hydrological and climate
Department of Energy under Contract DE-AC05-76RLO1830.
studies. Bull. Amer. Meteor. Soc., 96, 69–83, https://doi.org/
The global ocean heat flux and evaporation products were pro- 10.1175/BAMS-D-13-00068.1.
vided by the WHOI OAFlux project (http://oaflux.whoi.edu) Bretherton, C. S., M. E. Peters, and L. E. Back, 2004: Relationships
funded by the NOAA Climate Observations and Monitoring between water vapor path and precipitation over the tropical
(COM) program. We also would like to thank Daehyun Kim for oceans. J. Climate, 17, 1517–1528, https://doi.org/10.1175/1520-
insightful feedback on the early versions of the manuscript. 0442(2004)017,1517:RBWVPA.2.0.CO;2.

Unauthenticated | Downloaded 04/29/23 12:21 PM UTC

1 MARCH 2021 HAGOS ET AL. 1599

Chikira, M., 2010: A cumulus parameterization with state- Liu, Z., A. Mehran, T. J. Phillips, and A. AghaKouchak, 2014:
dependent entrainment rate. Part II: Impact on climatology Seasonal and regional biases in CMIP5 precipitation sim-
in a general circulation model. J. Atmos. Sci., 67, 2194–2211, ulations. Climate Res., 60, 35–50, https://doi.org/10.3354/
https://doi.org/10.1175/2010JAS3317.1. cr01221.
Dai, A., 2006: Precipitation characteristics in eighteen coupled Neelin, J. D., and I. M. Held, 1987: Modeling tropical convergence
climate models. J. Climate, 19, 4605–4630, https://doi.org/ based on the moist static energy budget. Mon. Wea. Rev., 115, 3–12,
10.1175/JCLI3884.1. https://doi.org/10.1175/1520-0493(1987)115,0003:MTCBOT.2.0.
Demory, M. E., P. L. Vidale, M. J. Roberts, P. Berrisford, CO;2.
J. Strachan, R. Schiemann, and M. S. Mizielinski, 2014: The ——, O. Peters, and K. Hales, 2009: The transition to strong con-
role of horizontal resolution in simulating drivers of the global vection. J. Atmos. Sci., 66, 2367–2384, https://doi.org/10.1175/
hydrological cycle. Climate Dyn., 42, 2201–2225, https:// 2009JAS2962.1.
doi.org/10.1007/s00382-013-1924-4. Pendergrass, A. G., and D. L. Hartmann, 2014: The atmospheric en-
Derbyshire, S. H., I. Beau, P. Bechtold, J.-Y. Grandpeix, J.-M. ergy constraint on global-mean precipitation change. J. Climate,
Piriou, J.-L. Redelsperger, and P. M. M. Soares, 2004: Sensitivity 27, 757–768, https://doi.org/10.1175/JCLI-D-13-00163.1.
of moist convection to environmental humidity. Quart. J. Roy. Peters, O., J. D. Neelin, and S. W. Nesbitt, 2009: Mesoscale con-
Meteor. Soc., 130, 3055–3079, https://doi.org/10.1256/qj.03.130. vective systems and critical clusters. J. Atmos. Sci., 66, 2913–
Emanuel, K., 2019: Inferences from simple models of slow, con- 2924, https://doi.org/10.1175/2008JAS2761.1.
vectively coupled processes. J. Atmos. Sci., 76, 195–208, Raymond, D. J., S. L. Sessions, A. H. Sobel, and Z. Fuchs, 2009:
https://doi.org/10.1175/JAS-D-18-0090.1. The mechanics of gross moist stability. J. Adv. Model. Earth
Emori, S., T. Nozawa, A. Numaguti, and I. Uno, 2001: Importance Syst., 1, 9, https://doi.org/10.3894/JAMES.2009.1.9.
of cumulus parameterization for precipitation simulation over Rushley, S. S., D. Kim, C. S. Bretherton, and M.-S. Ahn, 2018:
East Asia in June. J. Meteor. Soc. Japan, 79, 939–947, https:// Reexamining the nonlinear moisture–precipitation relation-
doi.org/10.2151/jmsj.79.939. ship over the tropical oceans. Geophys. Res. Lett., 45, 1133–
Fiedler, S., and Coauthors, 2020: Simulated tropical precipitation 1140, https://doi.org/10.1002/2017GL076296.
assessed across three major phases of the Coupled Model Sahany, S., J. D. Neelin, K. Hales, and R. B. Neale, 2014: Deep
Intercomparison Project (CMIP). Mon. Wea. Rev., 148, 3653– convective transition characteristics in the Community
3680, https://doi.org/10.1175/MWR-D-19-0404.1. Climate System Model and changes under global warming.
Hagos, S., L. R. Leung, M. Ashfaq, and K. Balaguru, 2019: South Asian J. Climate, 27, 9214–9232, https://doi.org/10.1175/JCLI-D-13-
monsoon precipitation in CMIP5: A link between inter-model 00747.1.
spread and the representations of tropical convection. Climate Dyn., Sillmann, J., V. V. Kharin, X. Zhang, F. W. Zwiers, and D. Bronaugh,
52, 1049–1061, https://doi.org/10.1007/s00382-018-4177-4. 2013: Climate extremes indices in the CMIP5 multimodel en-
Held, I. M., and B. J. Soden, 2006: Robust responses of the hy- semble: Part I. Model evaluation in the present climate.
drologic cycle to global warming. J. Climate, 19, 5686–5699, J. Geophys. Res. Atmos., 118, 1716–1733, https://doi.org/
https://doi.org/10.1175/JCLI3990.1. 10.1002/jgrd.50203.
Hirota, N., and Y. N. Takayabu, 2013: Reproducibility of precipi- Singh, M. S., R. A. Warren, and C. Jakob, 2019: A steady-state
tation distribution over the tropical oceans in CMIP5 multi- model for the relationship between humidity, instability, and
climate models compared to CMIP3. Climate Dyn., 41, 2909– precipitation in the tropics. J. Adv. Model. Earth Syst., 11,
2920, https://doi.org/10.1007/s00382-013-1839-0. 3973–3994, https://doi.org/10.1029/2019MS001686.
——, ——, M. Watanabe, and M. Kimoto, 2011: Precipitation re- Small, R. J., F. O. Bryan, S. P. Bishop, and R. A. Tomas, 2019: Air–
producibility over tropical oceans and its relationship to the sea turbulent heat fluxes in climate models and observational
double ITCZ problem in CMIP3 and MIROC5 climate analyses: What drives their variability? J. Climate, 32, 2397–
models. J. Climate, 24, 4859–4873, https://doi.org/10.1175/ 2421, https://doi.org/10.1175/JCLI-D-18-0576.1.
2011JCLI4156.1. Sobel, A. H., and C. S. Bretherton, 2000: Modeling tropical
Holloway, C. E., and J. D. Neelin, 2009: Moisture vertical structure, precipitation in a single column. J. Climate, 13, 4378–4392,
column water vapor, and tropical deep convection. J. Atmos. https://doi.org/10.1175/1520-0442(2000)013,4378:MTPIAS.2.0.
Sci., 66, 1665–1683, https://doi.org/10.1175/2008JAS2806.1. CO;2.
Holton, J. R., 1992: An Introduction to Dynamic Meteorology. 3rd ——, J. Nilsson, and L. M. Polvani, 2001: The weak temperature
ed., Academic Press, 511 pp. gradient approximation and balanced tropical moisture waves.
Huffman, G. J., R. F. Adler, D. T. Bolvin, and E. J. Nelkin, 2010: J. Atmos. Sci., 58, 3650–3665, https://doi.org/10.1175/1520-
The TRMM Multi-Satellite Precipitation Analysis (TMPA). 0469(2001)058,3650:TWTGAA.2.0.CO;2.
Satellite Rainfall Applications for Surface Hydrology, F. Song, X., and G. Zhang, 2009: Convection parameterization,
Hossain and M. Gebremichael, Eds., Springer Verlag, 3–22. tropical Pacific double ITCZ, and upper-ocean biases in the
Igel, M. R., S. R. Herbener, and S. M. Saleeby, 2017: The tropical NCAR CCSM3. Part I: Climatology and atmospheric feed-
precipitation pickup threshold and clouds in a radiative convec- back. J. Climate, 22, 4299–4315, https://doi.org/10.1175/
tive equilibrium model: 1. Column moisture. J. Geophys. Res. 2009JCLI2642.1.
Atmos., 122, 6453–6468, https://doi.org/10.1002/2016JD025907. Sperber, K. R., H. Annamalai, I.-S. Kang, A. Kitoh, A. Moise,
Kornbrot, D., 2014: Kendall’s tau: Basic. Wiley StatsRef: Statistics A. Turner, B. Wang, and T. Zhou, 2013: The Asian summer
Reference Online, N. Balakrishnan et al., Eds., https://doi.org/ monsoon: An intercomparison of CMIP5 vs. CMIP3 simula-
10.1002/9781118445112.stat06566. tions of the late 20th century. Climate Dyn., 41, 2711–2744,
Kuo, Y.-H., J. D. Neelin, and C. R. Mechoso, 2017: Tropical con- https://doi.org/10.1007/s00382-012-1607-6.
vective transition statistics and causality in the water vapor– Stephens, G. L., and Coauthors, 2010: Dreary state of precipitation
precipitation relation. J. Atmos. Sci., 74, 915–931, https:// in global models. J. Geophys. Res., 115, D24211, https://
doi.org/10.1175/JAS-D-16-0182.1. doi.org/10.1029/2010JD014532.

Unauthenticated | Downloaded 04/29/23 12:21 PM UTC

1600 JOURNAL OF CLIMATE VOLUME 34

Tompkins, A. M., 2001: Organization of tropical convection in low vertical (OAFlux) Project: Latent and sensible heat fluxes, ocean
wind shears: The role of water vapor. J. Atmos. Sci., 58, 529–545, https:// evaporation, and related surface meteorological variables.
doi.org/10.1175/1520-0469(2001)058,0529:OOTCIL.2.0.CO;2. Woods Hole Oceanographic Institution, OAFlux Project
Yin, L., R. Fu, E. Shevliakova, and R. E. Dickinson, 2013: How well Tech. Rep. OA-2008-01, 64 pp..
can CMIP5 simulate precipitation and its controlling processes Zhang, X., W. Lin, and M. Zhang, 2007: Toward understanding
over tropical South America? Climate Dyn., 41, 3127–3143, the double Intertropical Convergence Zone pathology in
https://doi.org/10.1007/s00382-012-1582-y. coupled ocean–atmosphere general circulation models.
Yu, L., X. Jin, and R. A. Weller, 2008: Multidecade Global Flux J. Geophys. Res., 112, D12102, https://doi.org/10.1029/
Datasets from the Objectively Analyzed Air-sea Fluxes 2006JD007878.

Unauthenticated | Downloaded 04/29/23 12:21 PM UTC