RAMOS DA SILVA AND AVISSAR / JOURNAL OF HYDROMETEOROLOGY

The Hydrometeorology of a Deforested Region of the Amazon Basin

RENATO RAMOS DA SILVA AND RONI AVISSAR*

Department of Civil and Environmental Engineering, Duke University, Durham, NC

(Manuscript received 21 March 2005, in final form January 2006)

ABSTRACT

A series of numerical simulations were performed to evaluate the capability of the

Regional Atmospheric Modeling System (RAMS) to simulate the evolution of convection in a
partly deforested region of the Amazon basin during the rainy season, and to elucidate some of
the complex land-atmosphere interactions taking place in that region. Overall, we demonstrate
that RAMS can simulate properly the domain-average accumulated rainfall in Rondônia when
provided with reliable initial profiles of atmospheric relative humidity and soil moisture. It is
also capable of simulating important feedbacks involving the energy partition at the ground
surface and the formation of convection. In general, more water in the soil and/or the atmosphere
produces more rainfall. But these conditions affect the onset of rainfall in opposite ways; while
higher atmospheric relative humidity leads to early rainfall, higher soil moisture delays its
formation. As compared to stratiform clouds, which tend to cover a large area, convective clouds
are localized and they let relatively more solar radiation reach the ground surface. As a result, a
stronger sensible heat flux is released at the ground surface, which enhances the atmospheric
instability and reinforces convection. Simulations using horizontal grid elements 2 and 4 km in
size show a delay and decrease of rainfall as compared to simulations with high-resolution grids
whose elements are not larger than 1 km and, as a result, afflict RAMS performance. We
conclude that RAMS can be used as a reliable tool to simulate the various hydrometeorological
processes involved in land-cover changes as a result of deforestation in this region.

-------------------------------------------------------------

1. Introduction discharge of over a trillion cubic meters of

Amazonia is a hydrographic basin of water into the Atlantic Ocean (Vital and
great hydrologic importance. It covers an Stattegger 2000). Recent human activity in
area of 6 million km2 of tropical forest the area has caused a gross rate of
(Goulding et al. 2003), receiving a total deforestation on the order of 18000 km2 yr-1
average rainfall of 2500 mm per year (INPE 2002). The hydroclimatic changes
(Richey et al. 1991), generating an annual resulting from this replacement of natural
*
Corresponding author address: Dr Roni Avissar, Department of Civil and Environmental
Engineering, Duke University, 123 Hudson Hall, PO Box 90287, Durham, NC, 27708-0287; Tel:
919 660-5458, Fax: 919 660-5459; E-mail: avissar@duke.edu
 2

forest by degraded vegetation has yet to be this region, an intensive observation

fully understood and quantified; estimating campaign was carried out in the rainy season
its impact is very important not only for the between January and February of 1999 as
region but also for the global water cycle part of the Large Scale Biosphere-
(Avissar and Nobre 2002; Avissar and Atmosphere Experiment in Amazonia
Werth 2005). (LBA) (Avissar and Nobre 2002). A large
Data analysis show that most of the number of meteorological observations with
moisture into the Amazon River basin radars, radiosondes, micrometeorological
comes from the Tropical Atlantic (Rao et al. towers, rain gauges, weather stations, and
1996). Also, long-term observations suggest airplanes were collected during this field
that the convergence of moisture over the study in Rondônia, which is referred to as
region has increased in the last 40 years, the Wet Atmospheric Mesoscale Campaign
producing more precipitation mainly during (WetAMC) . These observations show that
the rainy season (Henderson-Sellers et al. the wind regime and the landscape (i.e., land
2002). Two possible reasons have been cover and topography) have important
suggested for this increase of rainfall: impacts on convection and precipitation
greenhouse gas-induced hydroclimate (Halverson et al. 2002). But even when
change or locally-induced increase of carried out extensively, observations alone
convection (Henderson-Sellers and Pitman are not sufficient to explain all the processes
2002). involved in this complex
Governmental projects for the hydrometeorological system. Indeed, most
development of the Amazon (such as the measurements are made only at a few
one in Rondônia) have resulted in a patchy locations for limited periods, and often lack
landscape of forest and pasture that has a temporal and spatial resolution. In addition,
characteristic length scale of about 4 km instruments measure only a small number of
(Calvet et al. 1997). Satellite observations variables, which are not sufficient to explain
show an increase of cloudiness at the all the physical, biological and chemical
boundary of these patches during the dry mechanisms and feedbacks involved in these
season (Cutrim et al. 1995; Negri et al. processes. State-of-the-art models are useful
2004) and the rainy season (Durieux et al. tools to bridge the gap in our understanding.
2003). The possible effects of land-cover Several studies have been conducted with
change on convective clouds during the global and regional numerical models. But
rainy season are still not well understood. experiments performed with GCMs use grid
The impacts of land-cover change on size in the order of 2-5 degrees of latitude
rainfall may evolve in different ways. A and longitude (Nobre et al. 1991; Werth and
massive deforestation is likely to reduce Avissar 2002). At such coarse resolution,
rainfall due to a decrease in the transpiration mesoscale circulations are not resolved, and
and surface roughness and an increase in while a few attempts have been made to
albedo as has been shown in various studies develop an appropriate parameterization
with general circulation models (GCMs). (Lynn et al. 1995, Liu et al. 1999, Donner et
The current level of deforestation, however, al. 2001) important issues remain to be
may create mesoscale circulations that could solved and no GCM yet accounts for them
enhance convection and promote a transient through subgrid-scale parameterizations
increase of rainfall (Avissar et al. 2002). (Avissar et al. 2002). Current GCMs
To better understand the subgrid-scale parameterizations represent
hydrometeorological processes involved in convection and microphysical cloud
 3

processes based on the large-scale mean hydrometeorological processes in that region

state of the atmosphere, which leads to during the rainy season remains to be
errors in rainfall modeling (Molinari and demonstrated. Here, we show that modeling
Dudek 1992). moist convection accurately in this region
So far, it has not been shown that requires: (1) a reliable representation of the
atmospheric models are capable of current landscape heterogeneity; (2) a grid
simulating realistically precipitation during mesh with elements not larger than 1 km to
the wet season in the Amazon. Indeed, they explicitly resolve convection; (3) a grid
tend to produce too early rainfall in mesh that covers a large domain to correctly
Rondônia (Betts and Jakob 2002), and too represent relevant mesoscale dynamical
high precipitation east of the Andes (Horel processes; (4) several case studies with
et al. 1994; Tanajura 1996; Chou et al. 2000; various meteorological background
Chou et al. 2002; Druyan et al. 2002; Misra conditions during the wet season to have a
et al. 2002). Roads et al. (2003) performed robust evaluation; and (5) a comparison with
an inter-comparison of several models reliable observations.
applied to this region and concluded that The goal of the study summarized in this
they are overly dry over much of the domain article is twofold: (1) evaluate the
as compared to observations. The errors performance of a state-of-the-art numerical
seem to be associated with the models model based on recent observations
convective parameterization. Currently, the collected during the LBA field campaign
only known method to avoid this problem is together with vegetation characteristics
by performing high-resolution simulations. derived from modern satellites, and (2)
Thus, a careful evaluation of the capability improve our understanding of the effects of
of the models to represent weather events deforestation on the hydrometeorology of
associated with moist convection in the the Amazon basin. For that purpose, the
Amazon is necessary as a first step to model used here is set up to represent the
achieve better long-term regional landscape heterogeneity and convection at a
simulations (Takle et al. 1999). very high resolution, it covers a large area,
Several studies have been performed with and it is compared with state-of-art
mesoscale models to understand the effects observations collected during the LBA
of land-cover change on the dynamics of WetAMC.
local circulations and convection in the
Amazon (Silva Dias and Regnier 1996; 2. Numerical Experiments
Souza et al. 2000; Wang et al. 2000; Baidya The Regional Atmospheric Modeling
Roy and Avissar 2002; Silva Dias et al. System (RAMS) (Pielke et al. 1992; Cotton
2002). These numerical studies provided et al. 2003) is used here together with
important preliminary understanding of the observations collected during the LBA
convection dynamics in that region. WetAMC in an area where deforestation has
However, the majority of these studies generated a patchy landscape of forest and
focused on the dry season, and/or were pasture (see Fig. 1). RAMS is a three-
limited to only a few cases. Furthermore, the dimensional model consisting of a set of
models used have not been carefully prognostic equations, including dynamics,
evaluated, partly due to the scarcity of data. thermodynamics, and hydrometeor
But, to use confidently a model to study microphysics. These equations are solved
precipitation in the Amazon basin a detailed numerically using finite difference schemes
evaluation of its capability to simulate the applied to a staggered Arakawa C-grid. The
model contains several interacting sub-
 4

models that simulate soil-vegetation-

atmosphere exchange of heat and water
(Avissar and Pielke 1989; Walko et al.
2000a); surface-layer turbulent processes
(Louis 1979); boundary-layer turbulent
processes (Mellor and Yamada 1974;
Deardorff 1980); solar and thermal radiation
transfer and its interaction with
hydrometeors (Harrington 1997); and cloud
microphysics and precipitation (Walko et al.
1995; Walko et al. 2000b). RAMS scored
quite well in various evaluation studies, such
as large-eddy simulations (Avissar et al.
1998), mesoscale systems (Silva Dias and
Regnier 1996; Weaver and Avissar 2001; FIG. 2. Measured relative humidity profiles
Baidya Roy and Avissar 2002; Nair et al. in Rondônia at the Jaru Forest (green), Ouro
2003), and regional climate (Takle et al. Preto D Oeste (red) and Rolim de Moura
1999; Liston and Pielke 2001; Gandu et al. (blue) (See Fig. 1 for locations) at 0800 LT
2004; Hasler et al. 2005). (1200 UTC) on (a) 04 February, (b) 06
February, (c) 14 February, and (d) 23
February. A polynomial function was used
to extrapolate between observation points.

The cloud microphysics package in

RAMS is based on a set of prognostic
equations for the prevailing hydrometeors
(Walko et al. 1995; Meyers et al. 1997;
Walko et al. 2000b). Water is partitioned in
up to eight forms: vapor, cloud droplets,
liquid rain, pristine ice, snow, aggregates,
graupel and hail. Cloud droplets and rain are
liquid water that may be super-cooled.
Pristine ice, snow and aggregates are
assumed to be completely frozen, while
graupel and hail are mixed-phase categories,
capable of comprising ice only or a mixture
FIG. 1. Physiographical characteristics of of ice and liquid. The model calculates the
the studied area in Rondônia (Brazil). Forest concentration of vapor in an atmospheric
is green and deforested areas are white. This column and prognoses the formation of
domain is centered at [11o22 S, 62o00 W] different hydrometeors. Their size
and covers the region in which most of the distribution is determined using a two-
meteorological instruments were deployed moment statistical scheme (Meyers et al.
during the LBA WetAMC. The circle 1997). The model also follows the exchange
represents the S-POL radar scanning area. of water and energy as hydrometeors
interact with vapor and between themselves
(Walko et al. 2000b).
 5

The domain simulated here consists of conditions are calculated following the
250 by 250 horizontal grid elements, each approach proposed by Klemp and
representing a 1 km by 1 km area, centered Wilhelmson (1978). With this approach, a
at 11o22 S, 62o00 W in Rondônia, Brazil normal velocity component is specified at
(Fig. 1). This domain was chosen for two the boundary (20 m/s in our case), which
reasons. First, it possesses the major features allows most disturbances to propagate
believed to have a significant impact on outside the simulated domain without
precipitation variability in Rondônia (i.e., reflecting strongly towards the internal
topography and land-cover variability). domain. We also explored using the global
Second, it covers the region in which most NCEP reanalysis data (Kalnay et al. 1996) to
of the meteorological instruments were force the model at its lateral boundaries.
deployed during the LBA WetAMC However, probably due to its too coarse
including the S-band polarization (S-POL) resolution (which is 2.5 degree) for the small
radar that is located at its center. At a domain simulated here, results were
horizontal grid size of 1 km and a vertical generally less good. Note that the profiles
grid size of about 100 meters in the measured at Ouro Preto D Oeste have a
boundary layer, RAMS resolves cumulus, larger humidity in the boundary layer. Betts
thus avoiding the complications encountered et al. (2002) identified and discussed this
in subgrid-scale parameterizations used in wet bias at this measurement site. RH
coarse-resolution models. Subgrid profiles at Ouro Preto D Oeste were
turbulence is parameterized with the Mellor- corrected with the measurements collected
Yamada (1974) scheme. in the atmospheric boundary layer (ABL)
In the vertical, the grid structure has with a tethered balloon (Betts et al. 2002).
higher resolution near the ground surface At this location, the atmosphere is often
and is stretched with height, starting 30 m saturated. Tests using saturated profiles as
above ground and progressively increasing initial conditions in RAMS produce
the grid elements up to a size of 1.5 km near unrealistic immediate cloudiness over the
the top of the domain, which is set at a entire domain, which reduce the amount of
height of 22.5 km. solar radiation reaching the surface and
Homogeneous initial conditions are induce a negative land-atmosphere feedback
assumed in all numerical experiments. Silva that affects the development of the ABL.
Dias et al. (2002) found that this The average relative humidity in the first
initialization method is best suited for short 1000 m of the atmosphere for February 4, 6,
simulations, and it is extensively used in 14, and 23 (hereinafter F04, F06, F14 and
convection-resolving models (Emanuel F23) is 86%, 84%, 78%, and 90%,
1994, pg 307). respectively, indicating that F04 and F23
Figure 2 shows observed profiles of were wetter than the other two days.
relative humidity (RH) at three locations in The integration time step of the model is
Rondônia at 0800 LT (i.e., 1200 UTC) for 5 seconds and the simulated period is 12
different days of the WetAMC. The model is hours for each case study, starting at 0800
initialized with homogeneous conditions LT. Thus, only the day-time evolution of the
using the atmospheric profiles retrieved convective boundary layer is simulated.
from the radiosondes launched at Jaru Forest The soil numerical grid is also stretched
because it is available for all the studied and consists of 12 layers, starting from the
cases and compares well with the sounding surface down to a depth of 4 meters, with a
from Rolim de Moura. Lateral boundary higher resolution near the surface. Dawson
 6

(1993) and Nepstad et al. (1994) found that rainfall obtained by integrating 10-minute
vegetation in the Amazon basin can uptake rainfall rate estimates is presented in Figure
water from deep soil layers, thus requiring 3. Light rainfall accumulated all over the
the grid selected here. Soil moisture is domain and high rainfall accumulated near
initialized homogeneously over the entire the western hills on F04, F06 and F14. On
simulated domain based on a profile F23, rainfall accumulated more in the
observed at Ouro Preto D Oeste (Rondônia) southeast region of the domain. These
during the WetAMC (Alvala et al. 2002). accumulations seem to be associated with
Accordingly, at the first level below the the large-scale winds in the free-atmosphere
surface, the initial soil moisture was set to that are responsible for the storms
39% of saturation, increasing linearly to propagation.
45% of saturation at the deepest layer. A
sandy-clay-loam soil texture was assumed in TABLE 1. Vegetation parameters used in
accordance with the predominant soil RAMS to characterize forest and
observed in this region (Alvala et al. 2002). pasture in the Amazon basin.
To realistically represent the landscape of Parameter Forest Pasture
the region, vegetation characteristics Albedo .13 .18
estimated from Landsat satellite images by Emissivity .95 .96
Calvet et al. (1997) were adopted for our Leaf Area Index 5.0 2.0
simulations (Fig. 1). Two types of land Vegetation Fraction .98 .80
covers were defined: evergreen forest Roughness Length (m) 2.0 .02
(tropical forest) and short grass (pasture). To Displacement Height (m) 20.0 .2
model these vegetation types, several Root Depth (m) 4.0 1.0
parameters estimated from in situ Maximum Stomatal .0035 .01
measurements (Gash and Nobre 1997) and -1
Conductance (ms )
tested in recent numerical experiments
(Gandu et al. 2004) were adopted. They are The model set-up described in this
summarized in Table 1. section is assumed to be the control run for
The four days selected to evaluate the each case study and is used to evaluate the
model (F04, F06, F14, and F23) were model performance against the observations.
characterized by local formation of Modeling convection is highly sensitive to
convection under different synoptic-scale initial conditions (Lorenz 1963). Therefore,
meteorological conditions. High-quality in addition to the control simulations, a few
observations were recorded for these four sensitivity tests are performed using various
days. Cases having propagation of large- initial conditions, especially for those
scale convective systems into the domain of estimated with a lack of a high level of
study were avoided because they would confidence (e.g., soil moisture and relative
require an even larger simulation domain humidity).
with too many grid points to be realistically Coarse-grid simulations have the
simulated with readily available computing advantage of low computational cost. While
resources. Analysis of the S-POL radar a few numerical experiments have been
reflectivity (NCAR 1999) shows that storms performed to simulate the rainy season in
were propagating from the east on F04 and Rondônia at different grid resolutions
F14, from the west on F23 and were quasi- (Wang et al. 2000; Silva Dias et al. 2002), a
stationary (slight westward propagation) on careful evaluation of model performance as
F06. Corresponding total accumulated a function of grid resolution has not yet
 7

conducted. This issue is explored here using estimates but their comparison shows similar
grid sizes of 2, 4 and 20 km in addition to structure during the onset stage of precipitation
the control simulations that use a 1-km grid caused by topographic forcing. These
mesh. differences can be partly explained by the
inaccuracy of initial and boundary
conditions in the simulations. But it is very
important to keep in mind that convection is
chaotic in nature and, therefore, the timing
and location of a precipitation cell cannot be
expected to be exactly similar in the model
and the observations. Only their statistical
properties are comparable. Furthermore, in
the tropics, mesoscale systems are mostly
triggered by local convection, which has a
characteristic size of the order of 2-20 km
(i.e., meso- scale) and a very large Rossby
radius of deformation. Therefore, their
predictability is considerably lower than at
mid-latitudes, where meso- systems that
are more baroclinic and are shaped by
geostrophic winds, dominate.

FIG. 3. Accumulated precipitation (mm)

derived from the S-POL radar between 0800
LT and 2000 LT on (a) 04 February, (b) 06
February, (c) 14 February and (d) 23
February. The blue arrows represent the
horizontal wind used to initialize the model
above the atmospheric boundary layer (at a
height of 1642 m). Contour lines represent
topography higher than 300 m.

3. Results
a. Control simulation
Figure 4 presents the spatial distribution
of accumulated precipitation for the control
simulation of each one of the four case
studies. Only the inner 200 km domain is
displayed for a clearer comparison with the FIG. 4. Same as Fig. 3 but for the simulated
radar estimates shown in Fig. 3. The wind precipitation with RAMS. The additional
directions at two heights emphasize the orange arrows are for the initial wind at 314
various meteorological conditions prevailing m.
during the rainy season in Rondônia. There
is no good spatial correlation between the Figure 5 shows the domain-average,
simulated and S-POL radar-derived rainfall accumulated rainfall simulated with RAMS
 8

as compared to corresponding estimates accumulation of rainfall, the simulations and

from the S-POL radar. Averages are the radar provide similar results over the
computed for the region encircling the radar domain, when precipitation is within the 3 -
location, with a radius of 100 km (see Fig. 30 mm range. At several locations, the radar
1). The control simulations produce does not detect any precipitation, while the
accumulated rainfall very similar to the model produces light accumulation.
radar estimates. Correlation coefficients
between model results and observations are
.98, .98, .95 and .98 for F04, F06, F14, and
F23, respectively. In general, the onset of
rainfall in the model is delayed as compared
to that observed with the radar. Later in the
afternoon, the simulated rainfall is stronger
than that seen by the radar, and this
compensates for the late onset of the
simulated precipitation.

FIG. 6. Frequency distribution of

accumulated rainfall between 0800 and 2000
LT on (a) 04 February, (b) 06 February, (c)
14 February and (d) 23 February.

The S-POL radar is based on the

polarimetric method that is sensitive to
raindrop deformation to estimate the rainfall
rate. This method is not sensitive to small
droplets and, therefore, this radar does not
provide a reliable estimate of light rainfall.
Also, RAMS produces heavy accumulation
at several locations that is not captured by
FIG. 5. Domain-average accumulated the radar. Comparing radar data with
precipitation simulated with RAMS and raingauge observations during the
estimated from the S-POL radar on (a) 04 WetAMC, Carey et al. (2000) showed that
February, (b) 06 February, (c) 14 February the radar under-estimates events of high
and (d) 23 February. rainfall rate. Furthermore, RAMS, like any
other model, is based on various
Figure 6 shows the frequency of grid approximations and it is unlikely to simulate
elements of accumulated rainfall obtained perfectly the real precipitation. Table 2
from the simulations and from the radar. For indicates that, in general, RAMS simulates
a more appropriate comparison between well not only the mean accumulation but
them, the model results that were obtained also its median, tri-mean, median absolute
on a 1 x 1 km grid were regrided into a 2-km deviation (MAD) and inter-quartile range
data set. Although there are discrepancies in (IQR). But on F14, the discrepancies in light
regions receiving light and heavy rainfall are also reflected in these other
 9

statistical properties. In this case the model wind is from the southeast. On F14 (Fig.
produces high rainfall accumulations over 7c), in the layer between 2 and 4 km,
small areas while the radar shows weak observed winds are from the east, while the
rainfall distributed over large areas. simulated predominant wind is from the
Nevertheless, similar means are obtained northeast. These discrepancies may be due to
with both methods (Tables 2 and 3). the use of homogeneous initial conditions and
the lack of large-scale atmospheric gradients that
TABLE 2. Simulated (with RAMS) and possibly exist over the simulated domain but are
S-POL-radar-derived statistical moments of not accounted for properly with the
homogeneous forcing used in our numerical
rainfall total accumulation at 2000 LT.
experiments. Furthermore, the observations
S-POL RAMS are instantaneous values at the given sites,
F04 Mean 3.9 3.7 while the modeled winds are spatially
Median 1.2 1.4 averaged over the entire domain and time
Tri-mean 2.0 2.2 step. Thus, they are not strictly similar.
MAD 1.1 1.0 Using radar and satellite data, Laurent et
IQR 3.8 3.4 al. (2002) showed that storms propagate
F06 Mean 3.1 3.3 along with mid-level winds. Weisman and
Median 0.5 0.5 Klemp (1986) concluded that, in general,
Tri-mean 1.3 1.0 weak wind shear is associated with
MAD 0.5 0.3 individual convective cells, while strong
IQR 2.7 1.4 shear induces multi-cell convective systems.
F14 Mean 4.8 4.7 Thus, a misrepresentation of the wind shear
Median 1.4 0.9 may affect the storm evolution and the
Tri-mean 2.9 1.9 spatial distribution of rainfall, as discussed
MAD 1.4 0.7 in detail in Section 3.e.
IQR 6.0 3.0
F23 Mean 5.9 6.4
Median 2.4 1.9
Tri-mean 3.4 3.0
MAD 2.0 1.1
IQR 5.8 4.1

Figures 7 and 8 show the simulated

vertical profiles of predominant wind
direction and domain-average wind speed,
as compared to observations at 1700 LT.
The predominant wind is assumed to be that
wind whose direction is the most frequent at
a given atmospheric level. RAMS simulates
well this variable in most cases, but there are
FIG. 7. Simulated predominant wind direction as
a few discrepancies that may affect the compared to observations at 1700 LT on (a) 04
spatial distribution of rainfall. For example, February, (b) 06 February, (c) 14 February, and (d)
on F04 (Fig. 7a), the observed winds are 23 February. The predominant wind at each
from the northeast in the layer between 1 atmospheric level is defined as the most frequent
direction at that level.
and 3 km, while the model predominant
 TABLE 3. Simulated rainfall area (km2), domain-average rainfall (mm) for locations with
precipitation greater than 1mm, and maximum accumulated precipitation (mm) obtained at 2000
LT.
S-POL Cntrl RH+10% RH-10% WetSoil DrySoil Fog
F04 Area 16592 18980 24776 9276 15984 12384 22764
Rainfall 7.07 5.68 5.75 3.89 8.79 3.54 5.98
Max 67.38 67.55 60.10 38.57 100.2 36.56 54.47
F06 Area 12760 9840 15740 5684 8684 6856 10632
Rainfall 6.91 9.74 7.03 8.53 14.59 4.84 8.50
Max 52.89 78.23 112.7 53.36 95.33 46.39 95.28
F14 Area 17300 14320 24112 5592 13732 16712 18164
Rainfall 9.05 9.76 9.26 8.95 11.18 3.20 10.32
Max 73.94 149.9 160.7 61.97 128.0 22.50 202.09
F23 Area 21708 23496 25296 9608 25204 17568 22240
Rainfall 7.23 8.18 9.13 5.80 4.01 9.10 4.72
Max 72.99 79.98 96.77 65.90 65.80 85.29 57.29

tropics. Therefore, evaluating the potential

impact of biased initialization of
atmospheric moisture on the simulated
rainfall is important. Here, this was achieved
by raising the model s initial relative
humidity profile by 10% in one case and
decreasing it by the same amount in another
case. As it can be seen in Fig. 5, higher
initial atmospheric moisture results in earlier
and larger accumulation of rainfall in all
cases. It is interesting to note that this higher
atmospheric moisture produces about 50%
more rainfall on F14, and rainfall starts
about 1-2 hours earlier than in the control
FIG. 8. Simulated domain-average wind case. Overall, higher initial atmospheric
speed profiles (black) as compared to moisture results in stronger storms, with
observations at 1700 LT on (a) 04 February, more rainfall that covers larger areas (Table
(b) 06 February, (c) 14 February, and (d) 23 3). By contrast, decreasing the model s
February. Observations are from Jaru Forest initial relative humidity by 10% results in
(green), Rolim de Moura (blue), and Ouro reduced and delayed rainfall. Thus,
Preto D Oeste (red). A polynomial function misrepresenting the initial moisture profile,
was used to extrapolate between observation such as the possible wet bias observed in
points. Ouro Preto D Oeste, leads to large errors in
modeling the amount and timing of rainfall.
Early-morning, low-level clouds (i.e.,
b. Sensitivity to Initial Atmospheric Moisture
fog) are commonly observed in the region.
Betts et al. (2002) emphasized that
Their formation is due to the cool, moist
atmospheric moisture is highly susceptible
conditions near the surface at this time of the
to measurement errors, especially in the
day. Imposing an initial profile of relative
 11

humidity at saturation in the boundary layer precipitation (Chang and Wetzel 1991;
only (namely the lowest 500 meters of the Beljaars et al. 1993; Pielke et al. 1997). Due
atmosphere) generates low-level clouds to its significant impact on
almost immediately in RAMS. Such fogs evapotranspiration and soil thermal
also affect the timing of precipitation and its properties, one can speculate that it has a
accumulation (Fig. 5). Interestingly, in some significant impact on hydrometeorological
cases, morning fogs increase rainfall conditions in the Amazon basin as well.
accumulation (F4 and F14) while in others, Here, we check its impact by increasing it
they decrease it (F06 and F23), revealing by 15% over the entire domain in one case
complex interactions between the surface, and decreasing it by the same amount in
clouds, and rainfall in this wet environment. another.
Similar to the higher relative humidity
case discussed previously, a wetter soil
produces more rainfall, except on F23 (Fig.
5d and Table 3). In that case, the
combination of high moisture in the soil and
the atmosphere creates stratiform clouds and
rain, in contrast to the convective type
obtained with the dry soil. This is illustrated
in Fig. 9, which depicts a vertical cross
section of cloud mixing ratio in the
atmosphere above the hilly terrain of the
northeastern part of the simulated domain.
The narrow and deep convective cell has a
rain mixing ratio greater than 0.6 g kg-1,
while the shallow, widely spread stratiform
cloud only reaches a value of 0.2 g kg-1.
Figure 9 also shows that, at 1200 LT, the
stratiform cloud reduces more significantly
the solar radiation reaching the ground
surface. The downward long-wave radiation
is about 460 Wm-2 for convective clouds and
about 440 Wm-2 for stratiform clouds,
emphasizing a stronger effect of cloud type
on solar radiation than on long-wave
FIG. 9. Simulated vertical, west-east cross radiation. Reduced energy at the surface
section of (a) rainfall (g kg-1), and (b) inhibits the formation of strong convection
downward solar radiation at the surface, Rs, and leads to a stratiform type of cloud and
at latitude 10.8oS at 1200 LT on 23 rainfall. Later in the day (around 1400 LT),
February, resulting from the initialization of the convective rain is associated with
a dry soil (continuous line) and wet soil vertical motions greater than 3 ms-1, while
(dashed line). those obtained in the stratiform clouds are
an order of magnitude smaller. At higher
altitude (6 km) the vertical motion is larger
c. Sensitivity to Initial Soil Moisture Content than 8ms-1, indicating very strong
Initial soil moisture is known to have a convective activity there.
strong effect on the modeling of
 12

Formation of moist convection is highly

dependent on the partition of energy at the
ground surface (including vegetation) into
sensible and latent heat fluxes. While the
transfer of sensible heat between the surface
and the atmosphere promotes the
development of convective thermals, latent
heat flux supplies the atmosphere with
moisture. Figures 10 and 11 present the
domain-average sensible and latent heat
fluxes simulated with RAMS for all the
control and sensitivity experiments
performed in our study, together with
observations.
In the first two hours of the simulations
of the high- and low-atmospheric moisture
experiments, heat fluxes are very similar to
the control case. But after 1000 LT, a
FIG. 10. Simulated domain-average sensible distinct pattern is noticed between the high-
heat flux for the control and sensitivity and low-initial atmospheric moisture
experiments on (a) 04 February, (b) 06 simulations. In the high relative humidity
February, (c) 14 February, and (d) 23 case, clouds and rain cover larger areas as
February. Black crosses represent available compared to the control and the dry
observations at (a) Jaru Forest, and (b)-(d) atmospheric simulations. As a result, less
Ouro Preto D Oeste. solar radiation reaches the surface and both
the sensible and latent heat fluxes are
As compared to dry soils, wet soils delay reduced (Figs. 10 and 11). Although less
the onset of rainfall because the available energy is available at the surface, large
energy at the ground surface is used release of latent heat of condensation is
primarily for evaporation, thus inhibiting the produced during precipitation, thus
sensible heat that drives convection, as was maintaining convection. In the tropics, this
already explained by Sud et al. (1993) and release of heat is a major source of energy
Betts et al. (1996). Rickenbach (2004) for the atmosphere. In the fog case, the
analyzed the diurnal formation of rain in the sensible and latent heat fluxes are reduced as
Amazon and found that occurrence of compared to the control simulations,
nocturnal rainfall delayed the next day reflecting the impact of cloudiness on the
convection. Our results are also supported by available radiative energy at the surface. It is
the work of Qian et al. (2004) who
interesting to note that a drier atmospheric
profile, while producing less rainfall, allows the
demonstrated that water recycling has an
surface to release more latent heat in the
important effect on rainfall distribution. These atmosphere during the afternoon hours, thus
results emphasize the complexity of the increasing the atmospheric relative humidity that
interactions taking place in the land- tends to increase rainfall. Thus, it has a
atmosphere system of this region, where negative feedback on the water recycling.
water recycling is an important factor of its Soil moisture has a strong impact on the
hydrometeorology (Eltahir 1996). partition of energy absorbed at the surface
d. Sensible and latent heat fluxes into sensible and latent heat fluxes. Dry soils
 13

cause a strong release of sensible heat and a Figures 12 and 13 depict the spatial
weak release of latent heat (Figs. 10 and 11). distribution of accumulated precipitation for
This is due to the effects of the moisture the cases of high- and low-moisture content
stress on the vegetation, which close its in the atmosphere and the soil.
stomata thus reducing (or even inhibiting in
extreme cases) transpiration. The sensible
heat peaks around 1200 LT, following the The low relative humidity experiments
solar cycle. In this case, the low cloud-cover produce less rainfall accumulation and
fraction allows more solar radiation to reach precipitation tends to occur mostly on the
the surface. Wet soils result in low surface lee side of the mountains (Fig. 12a-d). This
sensible heat and high latent heat fluxes. The is mainly visible over the hills located in the
lack of strong convection in that case is northeastern part of the domain. High
probably the main reason for the delayed atmospheric humidity results not only in
rainfall, as illustrated in Figure 5. early and enhanced rainfall as illustrated in
Fig. 5, but it also affects the location of its
maximum accumulation (Fig. 12e-f). As
compared to the rainfall obtained for the
simulations initialized and forced with low
relative humidity, precipitation is obtained
earlier and at upwind locations. In general,
the storms are advected by the predominant
wind in the free atmosphere (illustrated by
arrows in Fig. 12). Accordingly, rain
accumulates in regions where the wind is
coming from. Indeed, it rains more in the
eastern part of the domain in those cases
when easterly winds are dominant (F04,
F06, and F14), and in the northwestern part
of the domain on F23. In addition,
precipitation spreads over larger areas as
compared to the control run and the dry
FIG. 11. Same as Fig. 10, but for the latent atmosphere case (Table 3).
heat flux. Dry soils produce early convection and
induce rainfall in the hills where thermals
Heat and moisture fluxes were difficult to are better organized by the dynamical effects
measure during the LBA/WetAMC of topography (Gopalakrishnan et al. 2000)
campaign due to frequent rainfall. During (Fig.13a-d). On F14, the lack of moisture in
rainy periods the eddy-correlation system the soil and the low atmospheric relative
often fails to provide reliable data. However, humidity produce light rainfall. But on F23
limited available observations indicate that the combination of high sensible heat flux
simulations are comparable with and atmospheric moisture produce strong
observations at the towers. This is convective cells. High soil moisture delays
particularly well illustrated on the clearest and shifts the rainfall accumulation
day (F06) at Ouro Preto D Oeste (Figs. 10b downwind (Fig. 13e-h). High amounts of
and 11b). rainfall accumulate in the northwestern part
e. Spatial variability of the domain on F04 and F14, and on the
 14

lee side of the northeast mountains on F06

and F23. This also suggests that a better
knowledge of initial soil-moisture conditions
are likely to improve the simulations, as was
shown in previous studies (Cheng and
Cotton 2004; Weaver 2004).

FIG. 13. Same as Fig. 12, but for low (left)

and high (right) initial soil moisture.

f. Sensitivity to grid size

Previous numerical experiments
FIG. 12. Same as Fig. 4, but for low (left) conducted to simulate the hydrometeorology
and high (right) initial relative humidity on of the Amazon basin have used grid
(a, e) 04 February, (b, f) 06 February, (c, g) elements larger than 2 km (Silva Dias and
14 February, and (d, h) 23 February. Regnier 1996; Wang et al. 2000; Silva Dias
et al. 2002). Here, we evaluate the impact of
grid size on the model performance.
Figure 14 presents the domain-average,
accumulated rainfall simulated with RAMS
 15

set-up with different grid-element sizes as resolution simulations result in slower

compared to the corresponding estimates evolution of convective storms. They
derived from the S-POL radar. In general, concluded that grid elements on the order of
coarser grids (i.e., 2 and 4 km) tend to 2-4 km would be sufficient to resolve mid-
produce less rainfall as compared to latitude meteorological events. The
observations. However, on F14, high rainfall sensitivity analysis performed here indicates
is produced in the center of the simulated that much better results are obtained with a
domain (Fig. 15c), increasing the mean grid whose elements are not larger than
accumulation. In that case two factors affect 1km.
precipitation there, the low initial relative Figure 15 displays the accumulated
humidity (see Fig. 2c) and the easterly winds rainfall for the simulations using grid
in the entire atmosphere (see Fig. 7c). Early elements of 2 and 4 km. These coarse-grid
in the morning, the dry atmosphere inhibits simulations do not resolve some of the
cloud formation allowing solar radiation to meso- scale (i.e., 2-20 km) features
reach the surface and to heat it. Also, the captured with the S-POL radar and the high-
easterly winds transport the convective cells resolution (control) simulations. Also, the
away from the mountains where they are slower evolution of convection produces
generated towards the center of the domain. rainfall accumulation more downstream as
compared to the control simulation. As the
grid-element size increases, less rainfall
accumulates near the hills.

4. Summary and conclusions

A series of numerical simulations were
performed to evaluate the capability of the
Regional Atmospheric Modeling System
(RAMS) to simulate the evolution of
convection in the Amazon basin during the
rainy season and to elucidate some of the
complex land-atmosphere interactions
taking place in that region. Overall, we
demonstrated that RAMS can simulate
FIG. 14. Simulated domain-average properly the domain-average accumulated
accumulated rainfall with different rainfall in Rondônia, when reliable initial
numerical grid sizes on (a) 04 February, (b) profiles of relative humidity and soil
06 February, (c) 14 February, and (d) 23 moisture are provided. An increase
February. (decrease) of the initial relative humidity by
only 10% generates significantly more (less)
The same case simulated with a 4-km and earlier (delayed) rainfall. Other models,
grid results in weaker and downstream such as the operational ECMWF Integrated
rainfall accumulation. Coarser grids delay Forecast System (IFS), simulated too early
rainfall onset by about 3-4 hours. Additional rainfall in the wet season in Rondônia (Betts
simulations using grid elements of 20 km do and Jakob 2002). In another study, Silva
not produce any rainfall accumulation (not Dias et al. (2002) used RAMS to simulate a
shown). Weisman et al. (1997) and Salvador squall line that formed in February 1999 for
et al. (1999) also indicated that coarse- the same region that was simulated here and
noticed a delayed precipitation as compared
 16

to observations. The tests performed here observations and the model. Rather, the
elucidate some of the factors that can affect statistical properties of the rainfall pattern need
quite significantly timing, rate and location to be duplicated by the model. It should also be
of rainfall. mentioned that landscape features tend to
anchor convective rain and the model does
respond to this mechanism. Our sensitivity
tests show that initial relative humidity and
soil moisture affect the timing and spatial
rainfall accumulation. In general, more
water in the soil and/or the atmosphere
produces more rainfall. But these conditions
affect the onset of rainfall in opposite ways;
while higher relative humidity leads to early
rainfall, higher soil moisture delays its
formation.
The impact of soil moisture content on
the timing and rainfall location creates a
negative feedback that works to homogenize
the spatial distribution of rainfall and land
water content. Indeed, a wet soil delays
convection and produces more precipitation
downstream. These results agree with
previous simulations of theoretical
landscapes, which show that convective rain
developed over the dryer part of the
landscape, creating a negative feedback
(Chen and Avissar 1994; Avissar and Liu
1996; Emori 1998). This probably has
significant ecological implications that it
would be interesting to investigate with an
appropriate vegetation dynamics model
coupled to RAMS. We are currently in the
process of developing such a coupled model.
Early morning atmospheric humidity
appears to be quite important for the
simulation of rainfall in this region. A
nocturnal rainfall event raises the soil
FIG. 15. Same as Fig. 12, but for grid sizes moisture and delays the onset of the next
of 2 km (left) and 4 km (right). day convective rain. Otherwise, the moisture
remains in the atmosphere and leads to
The simulations described here earlier rainfall the next morning. Thus,
emphasize that RAMS can provide a good models that are unable to represent nocturnal
estimate of the total area experiencing rainfall in the Amazon will likely fail to
rainfall. But it is important to keep in mind that simulate rainfall due to misrepresentation of
convection is chaotic in nature and, therefore, early morning atmospheric and soil
one cannot expect to obtain the exact same moisture. The implications of these daily
timing and location of a precipitation cell in the changes in timing and location for long-term
 17

hydroclimate simulations of the Amazon and will report our results in subsequent
remain to be studied. publications.
The land-surface energy-budget partition
into sensible and latent heat fluxes affects Acknowledgments. This research was
the amount of heat and moisture released supported by NASA under Grants NAG 5-
into the atmosphere and, as a result, the 8213 and NAG 5-9359, by the NSF under
vertical gradients and the atmospheric Grant ATM-0346554, and by the Gordon
stability. This also affects whether clouds and Betty Moore Foundation. The views
and precipitation will be convective or expressed herein are those of the authors and
stratiform in nature. Convective cells grow do not necessarily reflect the views of
deep, have strong upward motion, are more NASA, NSF or the Moore Foundation.
effective in creating rain, and allow more
solar radiation to pass through their cloud- REFERENCES
free downward branches, thus enhancing the Alvala, R. C. S., and Coauthors, 2002:
radiative energy absorbed at the ground Intradiurnal and seasonal variability of
surface. By contrast, stratiform clouds are soil temperature, heat flux, soil moisture
shallow but they cover larger areas and content, and thermal properties under
reduce the amount of solar radiation that forest and pasture in Rondonia. J.
reaches the surface, leading to light vertical Geophys. Res., 107,
motion and rain. The demonstrated doi:10.129/2001JD000599.
capability of RAMS to simulate these two Avissar, R., and R. A. Pielke, 1989: A
types of raining clouds is quite important parameterization of heterogeneous land
because the land-cover change in this region surfaces for atmospheric numerical-
may affect their relative distribution and, as models and its impact on regional
a result, the local hydroclimate. meteorology. Mon. Wea. Rev., 117, 2113-
Sensitivity tests to model resolution show 2136.
that a grid size of not more than 1 km is Avissar, R., and Y. Q. Liu, 1996: Three-
needed to simulate properly the averaged dimensional numerical study of shallow
rainfall accumulation in this region. This is convective clouds and precipitation
explained by the fact that large amounts of induced by land surface forcing. J.
precipitation are produced by relatively Geophys. Res., 101, 7499-7518.
small convective cells, which are neither Avissar, R., E. W. Eloranta, K. Gurer, and
parameterized nor explicitly resolved at G. J. Tripoli, 1998: An evaluation of the
grid-element sizes larger than 2 km. large-eddy simulation option of the
Finally, the numerical experiments regional atmospheric modeling system in
performed in this study indicate the simulating a convective boundary layer:
existence of complex hydrometeorological A FIFE case study. J. Atmos. Sci., 55,
feedback in the land-atmosphere system, 1109-1130.
involving atmospheric and soil moisture. To Avissar, R., and C. A. Nobre, 2002: The
understand the hydroclimatological Large-Scale Biosphere-Atmosphere
implications of these processes, it is Experiment in Amazonia (LBA). J.
necessary to perform long-term, high- Geophys. Res., 107,
resolution simulations, which require large doi:10.1029/2002JD002507.
computing resources. We are in the Avissar, R., P. L. Silva Dias, M. A. Silva
processes of performing such simulations Dias, and C. A. Nobre, 2002: The Large
Scale Biosphere Atmosphere Experiment
in Amazonia (LBA). Insights and future
 18

research needs. J. Geophys. Res., 107, lba/spol_rain_info/PrelimRptLBASPOLr

doi:10.129/2002JD002704. ain.htm.]
Avissar, R., and D. Werth, 2005: Global Chang, J. T., and P. J. Wetzel, 1991: Effects
hydroclimatological teleconnections of spatial variations of soil-moisture and
resulting from tropical deforestation. J. vegetation on the evolution of a prestorm
Hydrometeor., 6, 134-145. environment - a numerical case-study.
Baidya Roy, S., and R. Avissar, 2002: Mon. Wea. Rev., 119, 1368-1390.
Impact of land use/land cover change on Chen, F., and R. Avissar, 1994: The impact
regional hydrometeorology in Amazonia. of land-surface wetness heterogeneity on
J. Geophys. Res., 107, mesoscale heat fluxes. J. Appl. Meteor.,
doi:10.129/2000JD000266. 33, 1323-1340.
Beljaars, A. C. M., P. Viterbo, M. J. Miller, Cheng, W. Y. Y., and W. R. Cotton, 2004:
and A. K. Betts, 1993: The anomalous Sensitivity of a cloud-resolving
rainfall over the United States during July simulation of the genesis of a mesoscale
1993: sensitivity to land surface convective system to horizontal
parameterization and soil moisture heterogeneities in soil moisture
anomalies. Mon. Wea. Rev., 124, 362-383. initialization. J. Hydrometeor., 5, 934-
Betts, A., J. H. Ball, and J. Fuentes, 2002: 958.
Calibration and Correction of Chou, S. C., A. M. B. Nunes, and I. F. A.
LBA/TRMM ABRACOS pasture site Cavalcanti, 2000: Extended range
merged dataset. [Available online from forecasts over South America using the
http://lba.cptec.inpe.br.] regional ETA model. J. Geophys. Res.,
Betts, A. K., J. H. Ball, A. C. M. Beljaars, 105, 10147-10160.
M. J. Miller, and P. A. Viterbo, 1996: Chou, S. C., C. A. S. Tanajura, Y. K. Xue,
The land surface-atmosphere interaction: and C. A. Nobre, 2002: Validation of the
A review based on observational and coupled ETA/SSiB model over South
global modeling perspectives. J. America. J. Geophys. Res., 107,
Geophys. Res., 101, 7209-7225. doi:10.129/2000JD000270.
Betts, A. K., and C. Jakob, 2002: Evaluation Cotton, W. R., and Coauthors, 2003: RAMS
of the diurnal cycle of precipitation, 2001: Current status and future
surface thermodynamics, and surface directions. Meteor. Atmos. Phys., 82, 5-
fluxes in the ECMWF model using LBA 29.
data. J. Geophys. Res., 107, Cutrim, E., D. W. Martin, and R. Rabin,
doi:10.129/2001JD000427. 1995: Enhancement of cumulus clouds
Calvet, J. C., R. C. S. Alvala, G. Jaubert, C. over deforested lands in Amazonia. Bull.
Delire, C. Nobre, I. Wright, and J. Amer. Meteor. Soc., 76, 1801-1805.
Noilhan, 1997: Mapping surface Dawson, T. E., 1993: Hydraulic lift and
parameters for mesoscale modeling in water use by plants: Implications for
forested and deforested Southwestern water balance, performance, and plant-
Amazonia. Bull. Amer. Meteor. Soc., 78, plant interactions. Oecologia. 95, 565-
413-423. 574.
Carey, L. D., R. Cifelli, W. Petersen, and S. Deardorff, J. W., 1980: Stratocumulus-
Rutledge, 2000: TRMM - LBA rainfall capped mixed layers derived from a 3-
estimation using the S-POL radar. dimensional model. Bound.-Layer
[Available online from Meteor., 18, 495-527.
http:/radarmet.atmos.colostate.edu/trmm_
 19

Druyan, L. M., M. Fulakeza, and P. simulated warm and transition season

Lonergan, 2002: Dynamic downscaling arctic stratus., Atmospheric Department,
of seasonal climate predictions over Colorado State University, 288 pp.
Brazil. J. Climate. 15, 3411-3426. Hasler, N., R. Avissar, and G. E. Liston,
Durieux, L., L. A. T. Machado, and H. 2005: Issues in simulating the annual
Laurent, 2003: The impact of precipitation of a semi-arid region in
deforestation on cloud cover over the Central Spain. J. Hydrometeor., 6, 409-
Amazon arc of deforestation. Remote 422.
Sens. Environ., 86, 132-140. Henderson-Sellers, A., K. McGuffie, and H.
Eltahir, E. A. B., 1996: Role of vegetation in Zhang, 2002: Stable isotopes as
sustaining large-scale atmospheric validation tools for global climate model
circulations in the tropics. J. Geophys. predictions of the impact of Amazonian
Res., 101, 4255-4268. deforestation. J. Climate. 15, 2664-2677.
Emanuel, K. A., 1994: Atmospheric Henderson-Sellers, A., and A. J. Pitman,
Convection. Oxford University Press Inc., 2002: Comments on "Suppressing
580 pp. impacts of the Amazonian deforestation
Emori, S., 1998: The interaction of cumulus by the global circulation change". Bull.
convection with soil moisture Amer. Meteor. Soc., 83, 1657-1661.
distribution: An idealized simulation. J. Horel, J. D., J. B. Pechmann, A. N.
Geophys. Res., 103, 8873-8884. Hahmann, and J. E. Geisler, 1994:
Gandu, A. W., J. C. P. Cohen, and J. R. S. Simulations of the Amazon Basin
Souza, 2004: Simulation of deforestation circulation with a regional model. J.
in eastern Amazonia using a high- Climate. 7, 56-71.
resolution model. Theor. Appl. Climatol., INPE, 2002: Monitoring the Brazilian
78, 123-135. Amazon gross deforestation. [Available
Gash, J. H. C., and C. A. Nobre, 1997: online from
Climatic effects of Amazonian http://www.grid.inpe.br/amz/prodes2000/
deforestation: Some results from html/pag_2.htm.]
ABRACOS. Bull. Am. Meteor. Soc., 78, Kalnay, E., and Coauthors, 1996: The
823-830. NCEP/NCAR 40-year reanalysis project.
Gopalakrishnan, S. G., S. B. Roy, and R. Bull. Amer. Meteor. Soc., 77, 437-471.
Avissar, 2000: An evaluation of the scale Klemp, J. B., and R. B. Wilhelmson, 1978:
at which topographical features affect the The simulation of three-dimensional
convective boundary layer using large convective storm dynamics. J. Atmos.
eddy simulations. J. Atmos. Sci., 57, 334- Sci., 35, 1070-1096.
351. Laurent, H., L. A. T. Machado, C. A.
Goulding, M., R. Barthem, and E. J. G. Morales, and L. Durieux, 2002:
Ferreira, 2003: The Smithsonian Atlas of Characteristics of the Amazonian
the Amazon. Smithsonian Books, 253 pp. mesoscale convective systems observed
Halverson, J. B., T. Rickenbach, B. Roy, H. from satellite and radar during the
Pierce, and E. Williams, 2002: WETAMC/LBA experiment. J. Geophys.
Environmental characteristics of Res., 107, doi:10.1029/2001JD000338.
convective systems during TRMM-LBA. Liston, G. E., and R. A. Pielke, 2001: A
Mon. Wea. Rev., 130, 1493-1509. climate version of the regional
Harrington, J. Y., 1997: The effects of atmospheric modeling system. Theor.
radiative and microphysical processes on Appl. Climatol., 68, 155-173.
 20

Lorenz, E. N., 1963: Deterministic non- regional climate change. J. Climate. 4,

periodic flow. J. Atmos. Sci., 20, 131- 957-988.
140. Pielke, R. A., and Coauthors, 1992: A
Louis, J. F., 1979: Parametric model of comprehensive meteorological modeling
vertical eddy fluxes in the atmosphere. system - RAMS. Meteor. Atmos. Phys.,
Bound.-Layer Meteor., 17, 187-202. 49, 69-91.
Mellor, G. L., and T. Yamada, 1974: A Pielke, R. A., T. J. Lee, J. H. Copeland, J. L.
hierarchy of turbulence closure models Eastman, C. L. Ziegler, and C. A. Finley,
for planetary boundary layers. J. Atmos. 1997: Use of USGS-provided data to
Sci., 31, 1791-1806. improve weather and climate simulations.
Meyers, M. P., R. L. Walko, J. Y. Ecol. Appl., 7, 3-21.
Harrington, and W. R. Cotton, 1997: New Qian, J. H., W. K. Tao, and K. M. Lau,
RAMS cloud microphysics 2004: Mechanisms for torrential rain
parameterization .2. The two-moment associated with the Mei-Yu Development
scheme. Atmos. Res., 45, 3-39. during SCSMEX 1998. Mon. Wea. Rev.,
Misra, V., P. A. Dirmeyer, B. P. Kirtman, H. 132, 3-27.
M. H. Juang, and M. Kanamitsu, 2002: Rao, V. B., I. F. A. Cavalcanti, and K. Hada,
Regional simulation of interannual 1996: Annual variation of rainfall over
variability over South America. J. Brazil and water vapor characteristics
Geophys. Res., 107, over South America. J. Geophys. Res.,
doi:10.129/2001JD900216. 101, 26539-26551.
Molinari, J., and M. Dudek, 1992: Richey, J. E., R. L. Victoria, E. Salati, and
Parameterization of convective B. R. Forsberg, 1991: The
precipitation in mesoscale numerical biogeochemistry of a major river system:
models: A critical review. Mon. Wea. The Amazon case study. Biogeochemistry
Rev., 120, 326-344. of Major World Rivers, J. E. Richey, Ed.,
Nair, U. S., R. O. Lawton, R. M. Welch, and John Wiley & Sons, 57-74.
R. A. Pielke, 2003: Impact of land use on Rickenbach, T. M., 2004: Nocturnal cloud
Costa Rican tropical montane cloud systems and the diurnal variation of
forests: Sensitivity of cumulus cloud field clouds and rainfall in southwestern
characteristics to lowland deforestation. Amazonia. Mon. Wea. Rev., 132, 1201-
J. Geophys. Res., 108, 1219.
doi:10129/2001JD001135. Roads, J., and Coauthors, 2003:
NCAR, 1999: SPOL TRMM-LBA Brazil International Research Institute/Applied
1999. [Available online from Research Centers (IRI/ARCs) regional
http://www.atd.ucar.edu.] model intercomparison over South
Negri, A. J., R. F. Adler, L. M. Xu, and J. America. J. Geophys. Res., 108,
Surratt, 2004: The impact of Amazonian doi:10.1029/2002JD003201.
deforestation on dry season rainfall. J. Salvador, R., J. Calbo, and M. M. Millan,
Climate. 17, 1306-1319. 1999: Horizontal grid size selection and
Nepstad, D. C., and Coauthors, 1994: The its influence on mesoscale model
role of deep roots in the hydrological and simulations. J. Appl. Meteor., 38, 1311-
carbon cycles of Amazonian forests and 1329.
pastures. Nature. 372, 666-669. Silva Dias, M. A., and P. Regnier, 1996:
Nobre, C. A., P. J. Sellers, and J. Shukla, Simulation of mesoscale circulations in a
1991: Amazonian deforestation and deforested area of Rondonia in the dry
 21

season. Amazonian Deforestation and hydrology models for environmental

Climate, R. L. Victoria, Ed., J. Wiley & modeling. J. Appl. Meteor., 39, 931-944.
Sons, 531-547. Walko, R. L., W. R. Cotton, G. Feingold,
Silva Dias, M. A., and Coauthors, 2002: A and B. Stevens, 2000b: Efficient
case study of convective organization computation of vapor and heat diffusion
into precipitating lines in the Southwest between hydrometeors in a numerical
Amazon during the WETAMC and model. Atmos. Res., 53, 171-183.
TRMM-LBA. J. Geophys. Res., 107, Wang, J. F., R. L. Bras, and E. A. B. Eltahir,
doi:10.129/2001JD000375. 2000: The impact of observed
Souza, E. P., N. O. Renno, and M. A. Silva deforestation on the mesoscale
Dias, 2000: Convective circulations distribution of rainfall and clouds in
induced by surface heterogeneities. J. Amazonia. J. Hydrometeor., 1, 267-286.
Atmos. Sci., 57, 2915-2922. Weaver, C. P., and R. Avissar, 2001:
Sud, Y. C., W. C. Chao, and G. K. Walker, Atmospheric disturbances caused by
1993: Dependence of rainfall on human modification of the landscape.
vegetation - Theoretical considerations, Bull. Amer. Meteol. Soc., 82, 269-281.
simulation experiments, observations, Weaver, C. P., 2004: Coupling between
and inferences from simulated large-scale atmospheric processes and
atmospheric soundings. J. Arid. Environ., mesoscale land-atmosphere interactions
25, 5-18. in the U.S. Southern Great Plains during
Takle, E. S., and Coauthors, 1999: Project to summer. Part I: case studies. J.
Intercompare Regional Climate Hydrometeor., 5, 1223-1246.
Simulations (PIRCS): Description and Weisman, M. L., and J. B. Klemp, 1986:
initial results. J. Geophys. Res., 104, Characteristics of isolated convective
19443-19461. storms. Mesoscale Meteorology and
Tanajura, C. A. S., 1996: Modeling and Forecasting, P. S. Ray, Ed., American
analysis of the South American summer Meteorological Society, 793.
climate, Department of Meteorology, Weisman, M. L., W. C. Skamarock, and J.
University of Maryland, 164 pp. B. Klemp, 1997: The resolution
Vital, H., and K. Stattegger, 2000: Sediment dependence of explicitly modeled
dynamics in the lowermost Amazon. J. convective systems. Mon. Wea. Rev.,
Coastal Res., 16, 316-328. 125, 527-548.
Walko, R. L., W. R. Cotton, M. P. Meyers, Werth, D., and R. Avissar, 2002: The local
and J. Y. Harrington, 1995: New Rams and global effects of Amazon
cloud microphysics parameterization .1. deforestation. J. Geophys. Res., 107,
The single-moment scheme. Atmos. Res., doi:10.129/2001JD000717.
38, 29-62.
Walko, R. L., and Coauthors, 2000a:
Coupled atmosphere-biophysics-
This document was created with Win2PDF available at http://www.win2pdf.com.
The unregistered version of Win2PDF is for evaluation or non-commercial use only.