A Regional Climatography of West Nile, Uganda, to Support Human

VOLUME 51
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
JULY 2012
A Regional Climatography of West Nile, Uganda, to Support Human Plague Modeling
ANDREW J. MONAGHAN,* KATHERINE MACMILLAN,1 SEAN M. MOORE,*,1 PAUL S. MEAD,1
MARY H. HAYDEN,* AND REBECCA J. EISEN1
1
* National Center for Atmospheric Research,# Boulder, Colorado
Centers for Disease Control and Prevention, Division of Vector Borne Infectious Diseases, Fort Collins, Colorado
(Manuscript received 13 September 2011, in final form 22 February 2012)
ABSTRACT
The West Nile region in northwestern Uganda is a focal point for human plague, which peaks in boreal
autumn and is spread by fleas that travel on rodent hosts. The U.S. Centers for Disease Control and Prevention is collaborating with the National Center for Atmospheric Research to quantitatively address the
linkages between climate and human plague in this region. The aim of this paper is to advance knowledge of
the climatic conditions required to maintain enzootic cycles and to trigger epizootic cycles and ultimately to
target limited surveillance, prevention, and control resources. A hybrid dynamical–statistical downscaling
technique was applied to simulations from the Weather Research and Forecasting Model (WRF) to
generate a multiyear 2-km climate dataset for modeling plague in the West Nile region. The resulting dataset
resolves the spatial variability and annual cycle of temperature, humidity, and rainfall in West Nile relative to
satellite-based and in situ records. Topography exerts a first-order influence on the climatic gradients in West
Nile, which lies in a transition zone between the drier East African Plateau and the wetter Congo Basin, and
between the unimodal rainfall regimes of the Sahel and the bimodal rainfall regimes characteristic of
equatorial East Africa. The results of a companion paper in which the WRF-based climate fields were applied
to develop an improved logistic regression model of human plague occurrence in West Nile are summarized,
revealing robust positive associations with rainfall at the tails of the rainy season and negative associations
with rainfall during a dry spell each summer.
1. Introduction
Human plague is an often-fatal, fleaborne disease
caused by the bacterium Yersinia pestis. The rural West
Nile region of northwestern Uganda represents an epidemiological focus for plague, where humans are most
often exposed to Y. pestis during plague outbreaks in local
rat populations in which large numbers of rats die. The
infectious fleas on the dead rats are forced to seek alternative hosts, including humans. Previous work by the U.S.
Centers for Disease Control and Prevention (CDC) has
suggested that human plague cases in West Nile—which
typically occur at elevations above 1300 m and exhibit
a distinct seasonal cycle that corresponds to the annual
#
The National Center for Atmospheric Research is sponsored
by the National Science Foundation.
Corresponding author address: Andrew J. Monaghan, National
Center for Atmospheric Research, P.O. Box 3000, Boulder, CO
80307-3000.
E-mail: [email protected]
rainy season from August through October—are linked to
spatial and temporal climatic variability (Winters et al.
2009; Eisen et al. 2010; MacMillan et al. 2011). The CDC is
therefore developing models to examine quantitatively
the linkages between human plague case occurrence and
climatic variability in the West Nile region. The motivation for the work is to identify climatic variables that are
predictive of the spatial distribution of human plague
cases. Doing so advances our knowledge of the conditions
required to maintain enzootic cycles and may ultimately
be used to target limited surveillance, prevention, and
control resources to areas most at risk and to predict the
future range of human plague.
The fine-spatial-resolution climate data that are required for the CDC modeling effort were, until recently,
not available over the West Nile region because of the
sparse and unreliable observational network and because
existing atmospheric model datasets (such as atmospheric
reanalyses) are too coarse to resolve the complex orography and climatic gradients in the region. An elevation
gradient of ;1000 m exists across the ;75 km 3 100 km
study region, which is situated on the East African Plateau
DOI: 10.1175/JAMC-D-11-0195.1
Ó 2012 American Meteorological Society
1201
1202
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
and is bounded by the Albert Nile River to the east and
the Democratic Republic of Congo (DRC) to the west
(Fig. 1). Lake Albert, the northernmost in the chain of
the African Great Lakes that dot the Western Rift
Valley, lies to the south. To address the need for reliable
climate data to support human plague modeling efforts,
the Research Applications Laboratory of the National
Center for Atmospheric Research has simulated a high
quality, fine-resolution climate dataset for the greater
West Nile region. The dataset, which contains monthly
fields with 2-km spatial resolution for the period 1999–
2009, was generated with a hybrid dynamical–statistical
downscaling technique that is based on the Weather
Research and Forecasting Model (WRF; Skamarock
and Klemp 2008). In the following sections we 1) provide a literature-based overview of the climate of the
broader East African region and its primary large-scale
drivers, 2) describe the WRF-based dataset and assess
its veracity in comparison with satellite-derived and in
situ records, 3) employ the WRF-based dataset to examine the general climatic features of the greater West
Nile study region, and 4) summarize the successful application of the dataset toward modeling the spatial
distribution of plague cases.
2. Overview of equatorial East African climate
The climate of equatorial East Africa (EEA: Uganda,
Kenya, and Tanzania) is broadly influenced by three
different air masses at low levels of the atmosphere: the
northeast monsoon, the southeast monsoon, and westerly to southwesterly flow often referred to as the Congo
air mass (e.g., Basalirwa 1995; Nicholson 1996). Both
monsoons transport relatively stable, dry air because
of their divergent nature (Griffiths 1972). Therefore,
rainfall associated with the monsoonal flow generally
requires mechanical uplift, either at the boundary between the monsoons—the intertropical convergence
zone (ITCZ)—or from localized topographic lifting
(Nicholson 1996). Rainfall seasonality is thus linked to
the transition seasons during which the ITCZ moves
northward or southward through EEA and flow associated with the monsoons becomes more zonally (onshore) oriented (e.g., Hills 1979). Superimposed on the
monsoonal flow, episodic westerly incursions from the
humid and unstable Congo air mass can bring rainfall in
all but the driest months during boreal winter, during
which the ITCZ is far south of EEA (Thompson 1957;
Griffiths 1959; Okoola 1999). Interannual variations in
the relative influences of all three air masses in EEA are
governed by sea surface temperatures (SSTs) in the
tropical Atlantic, Pacific, and Indian Ocean basins and
their subsequent effect on the positions of the subtropical
VOLUME 51
anticyclones (e.g., Mutai and Ward 2000; Camberlin
et al. 2001).
Rainfall, because of its paramount influence on the
regional economies and livelihoods (e.g., Conway et al.
2005) and its comparatively large spatial and temporal
variability, is the focus of the majority of meteorological
literature that covers East Africa. In contrast, nearsurface temperature receives less attention because its
seasonal and interannual fluctuations are typically small
and its spatial variability is mainly a function of topography (Griffiths 1972).1 Although the influence of the
synoptic-scale atmospheric circulation is largely consistent throughout EEA, the spatial variability of rainfall is
inhomogeneous because of local-scale features such as
topography, large lakes, land cover, and proximity to
the coast (Plisnier et al. 2000; Anyah and Semazzi 2004;
Oettli and Camberlin 2005; Moron et al. 2007; Camberlin
et al. 2009; Mölg et al. 2009; Hession and Moore 2011;
Moore et al. 2010). The seasonal cycle of rainfall varies
substantially by region as well. There are between 26
and 52 distinct rainfall zones in EEA (10–14 for Uganda
alone), with each one being distinguished by the timing
and frequency (uni-, bi-, or trimodal) of their annual
peaks (Griffiths 1972; Ogallo 1989; Basalirwa 1995). For
simplicity, a bimodal annual rainfall cycle is often employed to broadly categorize EEA climate into four
seasons, per the East African Meteorological Department (1963): 1) a dry period from December to February when the ITCZ is far south of EEA; 2) the ‘‘long
rains’’ season from March to May that coincides with the
northward progression of the ITCZ through EEA during boreal spring; 3) the second dry period that occurs
during June–August, when the ITCZ generally lies
northward of EEA; and 4) the ‘‘short rains’’ season from
September to November, associated with the southward
passage of the ITCZ back through EEA during boreal
autumn.
Large-scale influences on EEA climate variability—in
particular, rainfall—have primarily been explained in
terms of teleconnections arising from SST variability
in the tropical Atlantic, Pacific, and Indian Oceans (e.g.,
Mutai and Ward 2000). The influence of tropical Atlantic Ocean SSTs on EEA rainfall is manifested primarily through modulating the north–south position
of the ITCZ, which in turn has an impact on low-level
wind fields (Camberlin et al. 2001; Balas et al. 2007). In
1
Despite the comparatively modest amount of literature on
African temperature variability, it is noteworthy that recent studies
suggest that African agricultural yields may diminish substantially
under modest climatic-warming scenarios, even if rainfall remains
optimal for crops (e.g., Lobell et al. 2011).
JULY 2012
MONAGHAN ET AL.
1203
FIG. 1. (a) The WRF configuration with gradual nesting of spatial resolution from 54 (outer map) to 18 (orange
box), 6 (yellow box), and 2 (red box) km. (b) Terrain height (m MSL) on the 18-km domain. (c) Terrain height and
geographic features on the 2-km domain (the study region is outlined with the black dashed box). The color scale is
cropped below 700 m and above 2100 m to emphasize regional features.
general, warm SST anomalies off the coast of equatorial
West Africa, combined with neutral to negative SST
anomalies in the subtropical Atlantic, are associated
with enhanced rainfall in EEA during both the long- and
short-rains seasons, with nearly zero lag (Nicholson and
Entekhabi 1987; Nicholson 1996). The Atlantic–EEA
teleconnection appears to be less robust, however, when
compared with the impacts of the Indian and Pacific
Oceans on EEA climate (Camberlin et al. 2001; Mutai
et al. 1998), which are described next.
The influence of the tropical Pacific Ocean on EEA
climate is through the El Niño–Southern Oscillation
(ENSO), the quasiperiodic (every 4–7 years) ocean–
atmosphere oscillation manifested in shifting sea surface
temperature anomalies in the tropical Pacific Ocean
(Walker 1924; Bjerknes 1969). In particular, ENSO
1204
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
warm events (above-average SSTs in the eastern tropical
Pacific) are positively correlated (zero-to-several-months
lag) with rainfall anomalies during the latter part of the
short-rains season, from October through December,
throughout EEA (Nicholson and Entekhabi 1987; Ogallo
1988; Nicholson and Kim 1997; Phillips and McIntyre
2000; Mutai and Ward 2000). This seasonal ENSO teleconnection has been linked to a wave train of horseshoe-shaped convective anomalies emanating from the
central Pacific Ocean (Mutai and Ward 2000). In contrast,
during the interim rainy season and the early portion of
the short-rains season, from July through September,
ENSO warm events tend to be negatively correlated
with rainfall anomalies, especially in the westernmost
regions of EEA, such as Uganda (Ogallo 1988; Phillips
and McIntyre 2000). In the dry and long-rains seasons of
the year from January to May, ENSO correlations with
EEA rainfall in general are comparatively weaker and
more spatially variable (Indeje et al. 2000; Camberlin
and Philippon 2002).
Tropical Indian Ocean variability is often manifested
as an ENSO-like coupled ocean–atmosphere oscillation
known as the Indian Ocean dipole (IOD), which is characterized by warm (cool) SST anomalies in the northwestern (southeastern) equatorial Indian Ocean during
positive phases (Saji et al. 1999; Webster et al. 1999).
Greater rainfall occurs in EEA during positive IOD
phases, especially for the short-rains season in boreal
autumn, because of enhanced convergence and convection over the warm SST pool in the northwestern
equatorial Indian Ocean and adjacent EEA (Birkett
et al. 1999; Hastenrath 2001, 2007; Ummenhofer et al.
2009; Hastenrath et al. 2010). The opposite is also true
for negative IOD phases, for which enhanced divergence
and subsidence over the same region lead to drought
conditions in EEA (Hastenrath et al. 2007, 2010). Black
et al. (2003) suggest that only ‘‘extreme’’ positive IOD
events, which persist for several months, lead to enhanced short rains over EEA, because only in these
events are the associated equatorial easterly anomalies
strong enough to enhance convection over EEA. Most
studies argue that the comparative influence of the IOD
on EEA rainfall is stronger than that of ENSO (e.g.,
Latif et al. 1999; Goddard and Graham 1999; Philippon
et al. 2002). However, although earlier studies suggest
that IOD variability is largely due to internal Indian
Ocean dynamics and thus is independent of ENSO
(Webster et al. 1999; Saji et al. 1999), more recent studies
suggest that the IOD and ENSO signals are closely coupled (Black et al. 2003; Black 2005; Ummenhofer et al.
2009). In section 4 the linkages between rainfall and the
IOD, ENSO, and tropical Atlantic SST variability are
examined for the West Nile study region.
VOLUME 51
3. Methods
a. Atmospheric model configuration
The Advanced Research WRF version is a fully compressible conservative-form nonhydrostatic atmospheric
model suitable for both research and weather-prediction
applications that has demonstrated ability for resolving
small-scale phenomena and clouds (Klemp et al. 2007;
Skamarock and Klemp 2008). WRF has multiple options
and choices for physical parameterizations (i.e., radiation,
cloud physics, cumulus clouds, and planetary boundary
layer turbulence) and is optimized for each application
on the basis of the prevailing weather conditions, location, and goals of the project. To simulate realistically
the two-way land surface interactions with the atmosphere, WRF is coupled to the ‘‘Noah’’ land surface
model (LSM) (e.g., Chen and Dudhia 2001). The Noah
LSM provides WRF with fluxes of energy and water
from the land surface, while also maintaining stores of
water and energy in four soil layers to a depth of 2 m.
A four-domain nested WRF configuration was employed,
with 56-, 18-, 6-, and 2-km spatial resolutions, respectively,
from the outermost to innermost domains (Fig. 1). Oneway nesting was employed. There were 35 vertical levels
from the surface to 50 hPa, with 7 levels in the lowest
1000 m. The initial and boundary conditions were
provided by the National Centers for Environmental
Prediction–U.S. Department of Energy Atmospheric
Model Intercomparison Project Reanalysis II (NCEP–
DOE II; Kanamitsu et al. 2002). NCEP–DOE II was
chosen because of the veracity of the results it produced
in the tropics for a recent downscaling experiment (Rife
et al. 2010; Monaghan et al. 2010). Daily-updated 0.258
SSTs for the oceans were specified with version 2 of the
National Oceanic and Atmospheric Administration
(NOAA) optimum-interpolation SST dataset (Reynolds
et al. 2002). Inland lake surface temperatures were updated monthly with 0.058 monthly mean skin temperature
data (product ‘‘MOD11C3’’) derived from the Moderate
Resolution Imaging Spectroradiometer (MODIS) instrument flown aboard the National Aeronautics and
Space Administration (NASA) Terra satellite (Wan
2009). We employed the MOD11C3 fields because preliminary simulations indicated that rainfall in the West
Nile region is sensitive to lake surface temperatures in
nearby Lake Albert, one of the African Great Lakes.
The MOD11C3 skin temperatures yielded more-realistic
(warmer) lake surface temperatures than those from the
coarser-resolution NCEP–DOE II skin temperatures,
leading to improved rainfall simulations by resolving a
dry rainfall bias. Anyah and Semazzi (2004) also noted
the strong sensitivity of EEA rainfall to lake surface
temperatures.
JULY 2012
1205
MONAGHAN ET AL.
TABLE 1. Physical parameterizations used in WRF simulations.
Parameterization
Choice
Citation
Cloud microphysics
Longwave radiation
Shortwave radiation
Surface layer
Land surface
Planetary boundary layer
Cumulus clouds
Thompson
RRTM
Dudhia
Eta similarity
Noah LSM
Mellor–Yamada–Janjić
Grell–Devenyi
Thompson et al. (2004)
Mlawer et al. (1997)
Dudhia (1989)
Janjić (2002)
Chen and Dudhia (2001)
Janjić (2002)
Grell and Devenyi (2002)
Three-hourly ‘‘spunup’’ skin temperature, snow (water equivalent), soil temperature, and soil moisture fields
from the 0.0258 Global Land Data Assimilation System
(GLDAS; Rodell et al. 2004), which employs the Noah
LSM that is also used in WRF, were used to initialize the
land surface and subsurface states for the Noah LSM in
WRF at the beginning of each month-long simulation.
These fields are available beginning on 24 February
2000. Therefore, because our simulations began in January of 1999, for the monthly simulations prior to March
of 2000 we employed the GLDAS data on the equivalent day for the first available year for which data exist.
For example, we used the 30 April 2000 GLDAS fields
to initialize the land state for the month-long simulation
beginning on 30 April 1999. Rainfall totals during the
first available year of GLDAS (March 2000–February
2001) were average for 9 of the 12 months and were below
average for July, September, and October. Rainfall totals
during the months prior (January 1999–February 2000)
were nearly all average. Therefore, the impact on soil
conditions of using GLDAS fields from later months to
initialize the LSM in earlier months was likely minimal.
The chosen model physical parameterizations are
summarized in Table 1. Modifications were made to the
longwave physical parameterization, the Rapid Radiative Transfer Model (RRTM; Mlawer et al. 1997), so
that carbon dioxide (CO2) was changed from a timeinvariant value of 360 ppm to annually varying values
ranging from 368 to 387 ppm for 1999–2009 on the basis
of the Mauna Loa, Hawaii, CO2 observations (Keeling
et al. 2009). Several options were tested for simulating
cumulus-scale precipitation, including 1) the updated
Kain–Fritsch parameterization (Kain 2004) for all four
domains, 2) the Grell–Devenyi parameterization (Grell
and Devenyi 2002) for all four domains, and 3) for the
6- and 2-km domains, no cumulus parameterization (i.e.,
assuming cumulus precipitation is resolved at these fine
spatial scales; in these cases the Kain–Fritsch parameterization was turned on in the outer 18- and 56-km
domains). It was found that the Grell–Devenyi parameterization for all four domains produced monthly
rainfall totals (spatial patterns and magnitude) that
most closely matched satellite–gauge precipitation
datasets. For the simulations in which the cumulus
parameterization was turned off on the 6- and 2-km
domains, monthly rainfall totals were unrealistically
low. The other parameterizations listed in Table 1 were
chosen because they yielded realistic simulations of
near-surface temperature and precipitation fields in
our preliminary simulations when compared with validation datasets (see section 4), and these are the meteorological fields of primary interest for modeling cases
of human plague.
The final production simulations were performed for
month-long intervals; that is, 134 one-month simulations were performed for 1999–2009, each beginning
on the last day of the previous month to allow 24 h for
spinup of the modeled fields. Because of computationalresource constraints, we performed 1 yr of monthly
simulations—2006—with all four domains turned on;
the additional 10 yr of simulations were performed with
only the outer two domains (56 and 18 km) turned on,
and the comparatively expensive inner domains (6 and
2 km) turned off. A statistical technique, presented below,
was subsequently employed to downscale the 18-km
output to 2 km, for the 10-yr period in which the inner
domains were turned off, by calibrating the 18-km output with the 2-km output for the overlap year of 2006.
Henceforth, we refer to the 2-km WRF-based dataset
resulting from the dynamical–statistical technique as
WRF-DS. The analysis of the climate of the West Nile
region of Uganda presented in this paper focuses on
the WRF-DS results because they are being employed
for the plague-modeling effort and because a detailed
study of the general climate of the West Nile region
covered by the WRF-DS domain has not previously
been performed.
b. Statistical downscaling
We statistically downscaled the WRF-simulated monthly
mean temperature Tavg, maximum temperature Tmax,
minimum temperature Tmin, relative humidity RHavg,
specific humidity Qavg, and total rainfall Ptot from 18
to 2 km using the bias-correction spatial disaggregation (BCSD) method of Wood et al. (2004). BCSD is
well suited for downscaling climate model output of
1206
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 51
FIG. 2. (left) The 2-km domain WRF-simulated 2003 monthly Tavg (8C) for (top) February and (bottom) August. (middle) As in the left
panels, but for WRF-DS Tavg. (right) The difference of WRF-DS minus WRF-simulated Tavg.
comparatively coarse spatial resolution in regions where
in situ data are sparse, because finer-scale climate-model
fields can be used as the training dataset, in lieu of
gridded point-based observational data (which are not
available in our region of interest). Wood et al. (2004)
found that BCSD yielded better results for downscaled
rainfall and temperature fields when compared with
spatial linear interpolation and spatial disaggregation
without bias correction.
The BCSD method was applied to generate daily 2-km
WRF-DS fields for 1999–2009. Monthly and annual WRFDS means and totals were then calculated from the daily
fields. One important assumption of our bias correction is
that the 2-km fields with which we are calculating the
bias provide an accurate (unbiased) representation of
the rainfall, temperature, and humidity fields. Although
this clearly is not a perfect assumption, we demonstrate in
section 4 that our results compare well to satellite-based
and in situ observations of rainfall, temperature, and
humidity, especially given that the satellite-based and in
situ observations are subject to their own sources of
uncertainty. Another important assumption is that our
statistical downscaling technique, which is calibrated for
each month of 2006, is applicable to months in other
years. This assumption is implicitly tested by our validation analysis in section 4, but we test it here too. As
was done for 2006, we turned on all four domains for the
February and August WRF simulations for 2003 to
perform an independent comparison of the WRF-DS
fields with purely dynamically downscaled fields. February and August were chosen because they represent
two extremes in the annual rainfall cycle and nearly two
extremes in the annual temperature cycle. The year 2003
was randomly chosen. The WRF-simulated fields versus
WRF-DS fields for February and August of 2003 are
shown for Tavg in Fig. 2 and Ptot in Fig. 3. The WRFsimulated fields and WRF-DS fields visually appear to
be nearly identical for both Tavg and Ptot. The domainwide February (August) Spearman rank pattern correlations are 0.995 (0.997) for Tavg and 0.944 (0.857)
JULY 2012
MONAGHAN ET AL.
1207
FIG. 3. As in Fig. 2, but for Ptot (kg m22). Differences are expressed as a percentage of WRF-simulated Ptot.
for Ptot, confirming the similarity between the WRFsimulated fields and WRF-DS fields. The difference
fields for Tavg are generally less than 60.48C over the
domain and are nearly zero within our study region
north of Lake Albert. The difference fields for Ptot are
large in February because they are expressed as percentages of total WRF-simulated rainfall, which is
very low: only 56% of grid boxes in the study region
fall within 625% of the simulated values. When expressed in absolute terms (not shown), however, the
differences are clearly small: 99% of grid boxes within
the study region are within 65 mm. In August, when
rainfall is near its peak, the percentage differences are
smaller than for February: 75% of grid boxes within
the study region fall within 625% of the simulated
values. In summary, this comparison for 2003 indicates
that the BCSD statistical downscaling technique is
robust and that the resulting uncertainty that is introduced into the 2-km meteorological fields by choosing
2006 as the calibration year is reasonable in comparison
with the uncertainty among the satellite-based and in
situ estimates of Tavg and Ptot that will be presented in
section 4.
c. Comparison and validation data
Validating the WRF-DS results is challenging in
northwestern Uganda because there is a dearth of in situ
data, and therefore we employ the strategy of using
several temperature and precipitation datasets to help to
characterize uncertainty. The datasets used for comparison with and validation of the WRF-DS results are
summarized in Table 2. The first four rows summarize
the spatially explicit precipitation datasets. The primary
precipitation dataset we employ is the Climate Prediction Center Morphing Technique (CMORPH), which
uses precipitation estimates from low-orbiting-satellite
microwave observations whose features are transported
by spatial propagation information from geostationary
infrared satellite data (Joyce et al. 2004). The version
of CMORPH used in this study has ;8-km resolution,
making it the best suited for comparison with the 2-km
WRF-DS rainfall. The second precipitation dataset is
1208
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 51
TABLE 2. Datasets used for comparison with and validation of WRF-DS.
Dataset
Variables
CMORPH
Rainfall (;8 km hourly)
TRMM
ARC
WorldClim
MODIS
Arua weather
obs
Citation
Span
Download source
Joyce et al. (2004)
Dec 2002–present
Rainfall (product 3B42; 0.258 daily)
Rainfall (0.18 daily)
Huffman et al. (2007)
Love et al. (2004)
Jan 1998–present
Jan 1982–present
Rainfall and near-surface temperature
(30 arc-s monthly)
MOD11C skin temperature
(0.058 monthly)
Rainfall, near-surface air temperature,
and humidity
Hijmans et al. (2005)
—
http://www.cpc.ncep.noaa.gov/
products/janowiak/cmorph_
description.html (long-term
record by request)
http://mirador.gsfc.nasa.gov
ftp://ftp.cpc.ncep.noaa.gov/fews/
AFR_CLIM/DATA/
http://www.worldclim.org/
Wan (2009)
Mar 2000–present
Unpublished
Jan 1999–present
the 0.258 Tropical Rainfall Measuring Mission Project
(TRMM) 3-hourly ‘‘TRMM and others rainfall estimate’’
(Huffman et al. 2007). The TRMM product combines
microwave and infrared information from a variety of
satellite instruments. The third precipitation dataset is
the 0.108 NOAA African Rainfall Climatology (ARC)
product (Love et al. 2004), which is a more temporally
stable version of the NOAA/Climate Prediction Center
operational daily precipitation estimate that supports
the U.S. Agency for International Development Famine
Early Warning System Network (Xie and Arkin 1996).
ARC rainfall estimates are based on the 3-hourly Geostationary Operational Environmental Satellite precipitation index (Janowiak and Arkin 1991) and Global
Telecommunications System data. The fourth precipitation dataset is the WorldClim 30 arc-s (;1 km) product,
a long-term estimate of monthly global precipitation that
is based on station records interpolated with thin-plate
smoothing splines (Hijmans et al. 2005). Dinku et al.
(2007) evaluated the first three datasets versus a relatively dense station gauge network over the Ethiopian
highlands in East Africa for 10-day rainfall totals. They
found that CMORPH and TRMM compared reasonably to the gauge data according to a variety of statistical
measures while ARC had a relatively large mean error
and low correlation coefficient when compared with the
gauge data. The authors cautioned that the accuracy of
the precipitation datasets is probably regionally variable
because of the different methods and data sources each
employs. Asadullah et al. (2008) evaluated five rainfall
datasets, including CMORPH, TRMM, and ARC, over
Uganda and found that CMORPH and TRMM correlated most closely with station records of rainfall amount
and occurrence. Like Dinku et al., however, Asadullah
et al. (2008) suggested that more than one rainfall dataset
should be used for any application because of the relative
strengths and weaknesses of each. Therefore, we employ
all four precipitation datasets described here to partially
ftp://e4ftl01u.ecs.nasa.gov/
MOLT/MOD11C3.005
— (obtained from CDC)
account for methodological and data uncertainty in the
estimates.
We also employ two spatially explicit temperature
datasets. The first is the WorldClim 30 arc-s long-term
monthly-average near-surface temperature product
(Hijmans et al. 2005), which was created with the same
method that is described above for the WorldClim
precipitation. The second is the 0.058 monthly daytime
and nighttime clear-sky land surface skin temperature
from the MODIS instrument aboard the NASA Terra
satellite. We also use monthly temperature and precipitation observations from the only known continuously
operating in situ weather station in the West Nile region:
the Arua, Uganda, airport (3.058N, 30.98E; Fig. 1). The
reliability of the Arua data cannot be confirmed; as will
be shown subsequently, however, the Arua data compare reasonably to the other datasets in Table 2 when
those data are interpolated to the Arua coordinates. It is
noteworthy that we examined the regional representativeness of the Arua record prior to undertaking the
subsequent analysis. Monthly areal-averaged values of
temperature, humidity, and rainfall for the West Nile
region from the gridded datasets correlate strongly with
point values from Arua. Therefore, the Arua record is
considered to be representative of the entire West Nile
region in terms of its temporal variability.
4. Results and discussion
a. Climate of the West Nile region
The spatial plague modeling of MacMillan et al. (2012)
that is summarized in section 4c necessarily employed the
10-yr-average (1999–2008) 2-km WRF-DS fields for Ptot,
Tavg, Tmax, Tmin, RHavg, and Qavg. Therefore, in this
section we seek to describe the spatial variability and
annual cycle of the WRF-DS variables, specifically for
the West Nile region, from a climatological (i.e., multiyear average) perspective and to assess their veracity
JULY 2012
MONAGHAN ET AL.
relative to a variety of other datasets. Our climatography
is based on the full 11-yr WRF-DS dataset spanning 1999–
2009 (the 2009 simulations were performed in support of
related plague-modeling efforts but were not used for the
spatial plague-modeling effort described in section 4c).
Maps of 1999–2009 annual Tavg for WRF-DS and
WorldClim are presented in the top panels of Fig. 4. The
spatial patterns of monthly Tavg are similar to the annual
fields, and therefore they are not shown. In general, the
WRF-DS and WorldClim maps are remarkably similar
and show clearly the first-order effect of topography on
the spatial variability of Tavg. WRF-DS tends to have
a larger range between high and low temperatures along
the elevation gradients than do the WorldClim data.
This may be related to the differential spatial impact of
humidity and clouds on Tavg. Whereas WRF-DS resolves
these moisture variations, the interpolation method employed to generate the WorldClim data does not explicitly account for moisture variability, and therefore
the WorldClim spatial temperature patterns more closely
follow topographic variations. Within the West Nile study
region, annual Tavg in WRF-DS decreases from about
268C along the Albert Nile on the eastern boundary
(about 600 m MSL), to about 198C along the southwestern boundary with the DRC (about 1600 m MSL),
representing a substantial temperature gradient within a
relatively modest distance of about 75 km.
The annual cycle of Tmax, Tmin, and Tavg at Arua is
shown in the bottom panel of Fig. 4 for the observations,
WRF-DS, WorldClim, and MODIS. The MODIS data
are surface skin temperatures and thus are not strictly
comparable to the other near-surface temperature fields.
The MODIS data are included only to provide confirmation that the observations—which are subject to uncertainty that is due to aging instrumentation that must
be manually read and digitized—are reasonable. It is
also noteworthy that the Arua observations were likely
used in the creation of WorldClim and that therefore
WorldClim may not provide an independent measure
of temperature at Arua in that regard. Relative to the
observations and WorldClim, WRF-DS Tmax has a cold
bias ranging from about 18C during the dry season
[December–February (DJF)] to about 28C during the
short-rains season (August–October). The cold bias is
manifested more subtly in Tavg and is nearly negligible
for Tmin. The cold bias may be related to excessive
cloudiness aloft, rather than to humidity near the surface, because there is a dry bias in the near-surface humidity fields (discussed below) that would tend to
promote warmer near-surface temperatures through
greater sensible heat flux partitioning.
The annual cycle of Tmax has an amplitude of ;58–
68C, with maxima during the dry season (DJF) and
1209
minima during the period leading up to and including
the short-rains season (June–October). Temperatures
Tavg and Tmin have annual cycles with roughly the same
phase as Tmax but with smaller amplitudes of ;38–48C
and ;28C, respectively. The smaller annual amplitude
of Tmin relative to Tmax can be explained by comparing the WRF-DS daytime [1500 local standard time
(LST)] and nighttime (0300 LST) surface energy balance
for February and August of 2006 (Table 3). During
February—a typical hot, dry month—the daytime sensible heat flux consumes the majority of the net solar
radiation (QH 5 2404 W m22; negative fluxes point
away from the surface), driving Tmax higher. In contrast,
during August—a typical cool rainy month—the daytime latent heat flux consumes most of the net solar
radiation (QE 5 2351 W m22) for evaporation, leaving
little energy to partition toward raising Tmax. This reversal of partitioning among the turbulent heat fluxes
between the dry and rainy seasons is responsible for the
comparatively large annual amplitude of Tmax. The influence of the nighttime surface energy balance on Tmin
is mechanistically different due to the absence of solar
radiation. Nighttime radiative heat loss L* is greater in
February than it is in August because higher surface
temperatures occur during the daytime in February. The
radiative heat loss is largely replaced by increased heat
flux upward from the ground to the surface QG, however—a compensating effect that explains the relatively
small annual amplitude of Tmin.
Maps of 1999–2009 annual RHavg and Qavg for WRFDS are presented in the top panels of Fig. 5. As with
temperature, monthly humidity maps are not shown
because they exhibit spatial patterns that are similar to
those of the annual maps. There are no observationally
based spatial datasets of humidity for comparison to
confirm the veracity of WRF-DS. Inspection reveals
that RHavg increases and Qavg decreases along the eastto-west elevation gradient within the West Nile study
area. This inverse relationship indicates that the spatial
variability of RHavg is strongly determined by the temperature gradient (Fig. 4). In contrast, along the gently
sloping terrain of the DRC to the west of the study area,
Qavg and RHavg are positively associated (blue-shaded
areas), presumably because the moisture gradients
are largely determined by advective moisture transport through the Congo air mass rather than being
topographically driven by temperature changes. (Figure
4 confirms that there is little temperature variability
across this region of the DRC.)
The 2007–09 annual cycles for WRF-DS and observed
daytime RHavg at Arua are shown in the bottom panel of
Fig. 5. Only 3 yr of daytime values (average of RH
at 0300 and 1500 UTC) are available in the Arua
1210
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 51
FIG. 4. (top) The 1999–2009 Tavg (8C) for the (left) 2-km WRF-DS domain and (right) WorldClim. (bottom)
The 1999–2009 mean annual cycle of Tmax, Tavg, and Tmin at Arua for the observations (OBS), WRF-DS, MODIS
(2001–09 only), and WorldClim (time span varies). Pink shading denotes 61 std dev from the observed mean.
observational record. The annual cycle of RHavg follows
that for Ptot (discussed below), having the lowest observed values (;50%) during the dry season and the
highest values (;70%) during the short-rains season.
The WRF-DS annual cycle has a larger amplitude and a
clear dry bias during the dry season. The exact reasons for
the dry bias are unclear because of the numerous potential causes (parameterization choices, uncertainty
JULY 2012
MONAGHAN ET AL.
TABLE 3. WRF-simulated components of the surface energy
budget at Arua for February and August 2006 for 0300 and
1500 LST. Notations are K* 5 net shortwave radiation; L* 5 net
longwave radiation; QG 5 ground heat flux; QH 5 sensible heat
flux; QE 5 latent heat flux; and QS 5 heat storage (residual).
Positive-valued terms are directed toward the surface.
Month
Hour
K*
L*
QG
QH
QE
QS
Feb
0300
1500
0300
1500
0
700
0
609
261
2183
239
288
44
268
25
232
19
2404
13
2135
23
250
1
2351
21
24
0
4
Aug
in land surface parameters, etc.). Ruiz et al. (2010),
however, performed a series of WRF sensitivity studies
over South America from the equator southward and
found persistent near-surface dry humidity biases that
were associated with both LSM options available in
WRF (Noah and the Rapid Update Cycle). The dry
biases were not present when the authors employed the
simple five-layer slab treatment for the surface. They
suggested the biases may be related to the initialization
of soil moisture and its evolution in the LSMs, in accordance with other recent studies (e.g., Case et al.
2008). We employed spunup soil fields from GLDAS for
our initialization of the Noah LSM, as suggested by Ruiz
et al. (2010), which did not remedy the problem. Further
investigation of the near-surface dry bias in WRF-DS is
beyond the scope of this paper.
Maps of 2003–09 annual Ptot for WRF-DS, CMORPH,
TRMM, ARC, and WorldClim are presented in Fig. 6.
The shorter period was used for the spatial plots so as to
be comparable with the CMORPH record. The patterns
of rainfall across the domain are broadly similar, but
magnitudes vary substantially. The differences are partly
due to the differential spatial resolutions of the datasets
but not entirely; WorldClim has the highest spatial resolution (;1 km) but has values that are characteristic of
lower-resolution products such as TRMM (;25 km).
One common feature among the datasets is the contrast
between the rainier Congo Basin in the western domain
that is fed by the humid Congo air mass and the drier
Western Rift Valley and East African Plateau in the
eastern domain, where orographic forcing plays an important role in mechanically lifting the relatively stable
air masses from the northeast and southeast monsoons
to generate rainfall (Nicholson 1996). The West Nile
study region straddles the boundary between these two
regimes. The sharp contrast is facilitated by the moisture
barrier formed by the mountains along the DRC–Uganda
border. An additional rain-shadow effect is caused by the
bisection of the Western Rift Valley through the East
African Plateau (location indicated by the dark green
valley of the Albert Nile River in Fig. 1c). The rain
1211
shadow is evident by the line of rainfall minima that
extend from Lake Albert northward along the Albert
Nile into southern Sudan. Another common feature
among the datasets is a convective maximum to the east
of the West Nile region (;38N, 328E). Analysis of the
diurnal development of rainfall (not shown) indicates
that this maximum is linked to midday convective activity that is triggered over the mountainous regions to
the east and northeast of the model domain—for example, along the Uganda–Kenya border (Fig. 1b). The
prevailing easterly winds at mid- to low atmospheric levels
gradually propagate the convection toward the West
Nile region. The thunderstorms continue development
as they move westward and generally become fully developed in the late-evening and early-morning hours.
Maps of 2003–09 Ptot for WRF-DS and CMORPH
for four representative months are presented in Fig. 7.
Other datasets are excluded for simplicity and because
previous studies have suggested that CMORPH yields
realistic rainfall estimates over Ethiopia and Uganda in
East Africa (Dinku et al. 2007; Asadullah et al. 2008).
The patterns and magnitudes of monthly Ptot are similar
among CMORPH and WRF-DS, except in May when
CMORPH resolves a band of larger rainfall in the West
Nile region and throughout the northeastern portion of
the model domain. The seasonal migration of the ITCZ
is evident in the sequence of months shown in Fig. 7,
from January in boreal winter when it is far to the south,
to May when it is migrating northward through the study
region, to June when it is far enough northward to reduce rainfall slightly in the West Nile region (and more
so to the southeast), and finally to August when it is
again migrating southward though the region.
The 1999–2009 annual cycles of the median rainfall
totals and frequencies at Arua are shown in Fig. 8. The
subtle bimodal nature of West Nile rainfall is evident
by a short dip in June Ptot within what is otherwise
a gradual increase in rainfall beginning in March,
peaking in August–October, and sharply dropping
off in November (Fig. 8a). The pink shading, which
denotes the bounds between the 20th and 80th percentiles in the observed rainfall totals, indicates that
there is substantial year-to-year variability during the
peak of the rainy season, with the exception of October.
The comparatively small bounds about the October
median indicate that it is a consistently rainy month
from year to year. Overall, the shape of the annual
cycle and the dominance of the autumn rainy season
suggest that the West Nile region is in a transition zone
between the bimodal rainfall regimes that dominate
the equatorial regions and the unimodal regimes that
characterize Sahel rainfall (Nicholson 2000; Conway
et al. 2005).
1212
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 51
FIG. 5. (top) The 1999–2009 annual (left) RHavg (%) and (right) Qavg (g kg21) for the WRF 2-km domain. (bottom) As in
Fig. 4, but for 2007–09 daytime (0900 and 1500 LST) RHavg at Arua for the observations and the 2-km WRF-DS domain.
The WRF-DS annual cycle of Ptot nearly mirrors the
Arua observations; however, the WRF-DS rainfall is
higher by ;50% during the rainiest months. It is not
clear what proportion of the bias is due to the model
simulations versus being due to undercatch by the Arua
rain gauge. Undercatch is a common issue with older
bucket or funnel gauges like the type used at Arua (e.g.,
Ciach 2003). That the other gridded rainfall datasets
JULY 2012
MONAGHAN ET AL.
FIG. 6. The 2003–09 mean annual Ptot (kg m22) for (a) WRF-DS, (b) CMORPH, (c) TRMM, (d) ARC, and
(e) WorldClim. The datasets are summarized in Table 2.
1213
1214
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 51
FIG. 7. The 2003–09 mean monthly Ptot (kg m22) for (top) WRF-DS and (bottom) CMORPH for December, May, June, and September.
have smaller values than WRF-DS during the rainy
season suggests that model bias plays a role, although
the fact that they resolve progressively smaller rainfall
totals as their resolution decreases suggests that spatial
resolution has an important affect on rainfall estimates
in this region. The rainfall-frequency data (Figs. 8b–d)
essentially have the same annual cycle as the totals, suggesting that the type of rainfall (i.e., drizzles vs downpours) is consistent throughout the year. WRF-DS has a
clear bias toward simulating too many days with low
rainfall (Figs. 8b,c) and too few days with high rainfall,
which is a common issue for atmospheric models (e.g.,
Dai 2006). It is also possible that the near-surface dry
bias noted above (Fig. 5) affects the WRF-DS rainfall
frequency distribution.
In summary, we find that the 2-km WRF-DS climate
fields resolve the 11-yr-average spatial variability and
annual cycle of temperature, humidity, and rainfall reasonably in comparison with satellite-based and in situ
records. Given the uncertainty in all of the datasets,
especially rainfall, differences between WRF-DS and
other data do not necessarily infer that WRF-DS is less
accurate than the data. For example, it is possible that
relative to the other datasets WRF-DS may more realistically depict the nonlinear local-scale processes that
lead to the rainfall patterns shown in Fig. 6. Regardless
of which dataset is examined, it is clear that the spatial
variability and annual cycle of the temperature, humidity,
and rainfall gradients in the West Nile region are
strongly influenced by terrain. With respect to rainfall,
the West Nile region appears to be in a transition zone
between the drier East African Plateau and the wetter
Congo Basin. In addition, the annual cycle of Arua
rainfall suggests that the region is within a north-tosouth transition zone between the unimodal rainfall regimes of the Sahel and the bimodal equatorial regimes.
Such complexity can drive substantial interannual variability, as is evident by the large 20th and 80th percentile
bounds during the rainiest months (Fig. 8), and likely
confounds the influence of large-scale climate drivers such
as ENSO, which we now examine in section 4b.
b. Tropical-ocean influence on West Nile rainfall
In this paper we apply the WRF-DS fields to model
the long-term-average spatial variability of human plague
as a function of climatic fields. In future work, we aim to
model the temporal variability of plague over the West
Nile region and, if possible, to predict the annual plague
season with several months of lead time so that years
with high plague risk can be identified in advance. In
support of these future efforts and given the linkages
between plague and rainfall (described in section 4c),
here we examine the impact of teleconnections arising
from tropical-ocean variability on rainfall in the West
Nile region up to 1 yr in advance, by correlating regional
rainfall with a variety of SST and SST–atmosphere
JULY 2012
MONAGHAN ET AL.
1215
FIG. 8. The 1999–2009 median annual cycle of (a) Ptot (kg m22) and frequency of days with rainfall of greater than
(b) 2, (c) 10, and (d) 20 mm at Arua. CMORPH data span 2003–09 only. Pink shading indicates the region between
the observed 20th and 80th percentiles.
indices for the tropical Pacific, Atlantic, and Indian
Oceans (Table 4). We employ the regionally representative time series data from the Arua observational record, WRF-DS, and satellite-based time series from
ARC and TRMM. CMORPH was not used because the
record is too short to provide meaningful results. We
compared the regional WRF-DS time series with the
point-based time series from Arua and found that the
Arua record is highly representative of the interannual
rainfall variability across the plague study region. Threemonth running means of monthly rainfall totals, as well
as frequency of days per month with rainfall .2, .10,
and .20.0 mm, were detrended and normalized and
then correlated with detrended and normalized 3-month
running means of each tropical-ocean index for 1999–
2009. There were only modest differences among the
correlation patterns for rainfall totals when compared
with rainfall frequency (the frequency data provide insight into associations that are related to rainfall intensity, such as extremes). Therefore, the results discussed
here focus on rainfall totals (i.e., Ptot).
Figure 9 shows the results for the two indices that
we found to have the most robust correlation with
West Nile rainfall: the Niño-3.4 and the tropical North
Atlantic (TNA) indices. Although most of the ENSObased indices yielded similar correlation patterns, the
correlations with West Nile rainfall were strongest for
the Niño-3.4 index, a broad measure of SSTs in the
equatorial central Pacific Ocean. Of interest is that most
of the literature suggests that the strongest ENSO teleconnection with EEA rainfall is manifested as a positive
correlation for lags from zero to several months during
1216
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 51
TABLE 4. List of climate indices that are based on tropical-ocean and ocean–atmosphere variability. All datasets were obtained from the
NOAA/Environmental Systems Research Laboratory (http://www.esrl.noaa.gov/psd/data/climateindices/list/), with the exception of the
DMI, which was obtained from the Japan Agency for Marine Earth Science and Technology (http://www.jamstec.go.jp/frcgc/research/d1/
iod/HTML/Dipole%20Mode%20Index.html).
Dataset
Acronym
Tropical northern
Atlantic index
TNA
Tropical southern
Atlantic index
Niño-1.2 index
TSA
Niño-1.2
Niño-3 index
Niño-3
Niño-3.4 index
Niño-3.4
Niño-4 index
Niño-4
Trans-Niño index
TNI
Southern Oscillation
index
Multivariate ENSO
index
SOI
Bivariate ENSO time
series
Dipole mode index
BEST
MEI
DMI
Description
Anomaly of monthly mean tropical Atlantic
SSTs from 5.58 to 23.58N and from
158 to 57.58W
Anomaly of monthly mean SST from
08 to 208S and from 108E to 308W
Monthly mean tropical Pacific SST from
08 to 108S and from 908 to 808W
Monthly mean tropical Pacific SST from
58N to 58S and from 1508 to 908W
Monthly mean tropical Pacific SST from
58N to 58S and from 1708 to 1208W
Monthly mean tropical Pacific SST from
58N to 58S and from 1608E to 1508W
Standardized Niño-1.2 minus Niño-4 with
5-month running mean, then restandardized
Difference of standardized Tahiti and Darwin
monthly mean sea level pressure
A combination of sea level pressure, near-surface
winds and air temperature, SSTs, and cloud
fraction in the tropical Pacific
A combination of the Niño-3.4 index and SOI
Monthly mean difference of tropical SSTs in the
western (108S–108N, 508–708E) and eastern
(08–108S, 908–1108E) Indian Ocean
the latter part of the short-rains season from October to
December (e.g., Ogallo 1988). In the West Nile region,
there are positive correlations during October–December,
but they are statistically insignificant among all four rainfall datasets for the 1999–2009 period. There is a clear
positive relationship between ENSO and West Nile rainfall that appears to be especially robust among the four
datasets during March–April, with rainfall lagging
ENSO by ;0–11 months. Other studies have also suggested that the EEA long rains are correlated with
ENSO (e.g., Camberlin and Philippon 2002); the results
were highly variable in space, however, and no conclusions were drawn specifically for the West Nile region.
The correlations between the TNA and West Nile
rainfall yield a pattern that is similar to those for ENSO
during approximately January–September, which is
consistent with literature that notes the correlation between tropical northern Atlantic SSTs and ENSO
(Enfield and Mayer 1997). The periods having the most
robust TNA–rainfall correlations overlap with those for
Niño-3.4 but remain statistically significant for a shorter
period. The strongest correlations occur during March–
April when rainfall lags the TNA by about 2–4 months.
Of interest is that no consistent correlations between
Indian Ocean SST variability and West Nile rainfall
Citation
Enfield et al. (1999)
Enfield et al. (1999)
NOAA Climate Diagnostics
Bulletin (various)
NOAA Climate Diagnostics
Bulletin (various)
Barnston et al. (1997)
NOAA Climate Diagnostics
Bulletin (various)
Trenberth and Stepaniak (2001)
Ropelewski and Jones (1987)
Wolter and Timlin (1998)
Smith and Sardeshmukh (2000)
Saji et al. (1999)
were found (not shown), despite strong relationships
that have been found for other regions of EEA previously (e.g., Birkett et al. 1999). It has been suggested
that only very strong Indian Ocean dipole events have
significant impacts on EEA rainfall (Black et al. 2003),
and therefore it is possible that the weak correlations
with the dipole mode index (DMI) partly result from
the relatively short 11-yr correlation window, for which
only a handful of strong IOD events occurred.
In summary, our results suggest that the signs of the
correlations between ENSO and West Nile rainfall are
similar to those found in other regions of EEA but that
the most significant relationships are for the long-rains
season in March–April rather than the end of the shortrains season as found elsewhere in EEA. The differential influence of teleconnections on West Nile rainfall
relative to other regions of EEA may be partly due to
the location of the West Nile region along the westernmost reaches of EEA and the Western Rift Valley,
where the influence of the moist, unstable Congo air
mass from central equatorial Africa likely has a larger
influence on weather. SST variability in the tropical
northern Atlantic likewise is significantly correlated
with March–April rainfall in the West Nile region 2–4
months in advance but is also correlated with ENSO
JULY 2012
MONAGHAN ET AL.
FIG. 9. Pearson’s correlation coefficients for 1999–2009 monthly Ptot at Arua (for the observations, WRF-DS, TRMM, and ARC) vs the 0–12-month lagged monthly (left) Niño-3.4
and (right) TNA indices. Three-month running means are applied to both the rainfall data and
the indices. Means are centered about the month shown on the abscissa; for example, values
centered on February (F) correspond to the JFM running-mean total rainfall. Lags are also with
respect to the center month. Hatching (in red and blue contours) indicates statistical significance (correlation r . 60.6) per the Student’s t test.
1217
1218
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
variability. Given these relationships, it may be possible to
predict the long rains in the West Nile region several
months in advance, although the utility of such predictions
in support of projecting human plague risk is uncertain.
c. Modeling human plague occurrence with
WRF-derived climate data
In this section we briefly summarize the application of
the 2-km WRF-DS climate fields to model the spatial
distribution of human plague cases in the West Nile
region. The reader is referred to the companion study by
MacMillan et al. (2012) for a detailed description of the
plague model and method.
Approximately 2000 suspected or probable plague
cases occurred in the West Nile region from 1999 to 2007
(Winters et al. 2009). To ensure an accurate plague-case
dataset, during the 2008–09 plague season, which spanned
approximately from August through March, laboratory
confirmation was conducted for clinically suspected plague cases. Likewise, control (nonoccurrence) locations
were determined by mapping locations with similar access to health care but for which no cases of clinically
diagnosed plague were reported from 1999 to 2008. A
total of 36 case and 72 control points were used in our
model development (Eisen et al. 2010).
Logistic regression modeling was applied to a variety of landscape variables (e.g., elevation) and the
1999–2008 average monthly WRF-DS Tmax, Tmin, Tavg,
RHavg, Qavg, and Ptot fields at each case and control
location. Prescreening was performed to reduce the
number of climatic variables by discarding strongly
correlated fields. Several logistic regression models
were then developed and tested for goodness of fit, a
subgroup of leading models was retained, and finally
the most parsimonious model was identified with Akaike’s
information criterion (Akaike 1974).
The leading logistic regression model reveals that
within high-elevation sites (above 1300 m) plague risk is
positively associated with rainfall during February, October, and November and is negatively associated with
rainfall during June. The model explanatory variables
are consistent with prior studies that associate plague
with higher-elevation, rainy areas in Africa (e.g., Davis
1953). The overall model accuracy is 94% (Table 5),
which is a substantial improvement relative to the 81% in
a previous model that did not employ climate data (Eisen
et al. 2010). The revised model retains sensitivity from the
previous model in which 89% of case locations were
correctly classified by the model as elevated risk. Using
WRF-DS predictors, specificity is improved by nearly
20%, with 90% of control locations correctly classified
as low risk. The negative predictive value (NPV), which
measures the percentage of time that a model correctly
VOLUME 51
TABLE 5. Comparison of performance statistics between the
MacMillan et al. (2012) and Eisen et al. (2010) plague risk models.
The MacMillan et al. (2012) model employed the WRF-DS climate
fields that are described in the study presented here. Metrics are
described in the text. All units are in percent.
Model
MacMillan
et al. (2012)
Eisen et al. (2010)
Accuracy Sensitivity Specificity PPV NPV
94
89
90
82
94
81
89
71
60
93
classifies control locations as low risk, was similar to the
previous model. The positive predictive value (PPV),
which assesses the accuracy of the model for predicting
case occurrence, is improved by 22% using WRF-derived predictors. As expected when modeling the occurrence of a rare disease, PPV is low relative to other
model diagnostics in this and the previous model.
Taken together, our model predictors suggest that
areas that receive increased but not continuous rainfall
provide ecologically conducive conditions for Y. pestis
transmission in the West Nile region. We seek to develop a hypothesis to explain why these specific rainfall
associations in February, October, November, and June
are associated with the spatial distribution of plague
cases in the West Nile region. February is the driest
month, and subsequently rainfall begins gradually increasing in March until it peaks during August–October,
and it begins decreasing again in November. Therefore,
above-average rainfall during February, and again in
October–November, may extend the growing season at
both of its tails (in turn providing more agricultural
forage near huts for plague-infested rats). June represents a comparatively dry spell between the two rainier
periods in March–May and July–October. Drier June
conditions could provide the essential increase in sunlight for photosynthetic cycles during a critical growth
stage for crops or other vegetation that in turn attract
plague-infected rats or could provide adequate conditions for drying crops for long-term storage in or near
living huts (which also attracts rats). Evidence for the
latter is suggested by Roberts (1936), who noted that
excessive rainfall in Kenya in the 1920s led to abundant
harvests but did not allow adequate time to dry crops
and prevent spoilage. An ongoing field program in
Uganda and concurrent efforts to model the temporal
variability of human plague will allow us to test and
refine our hypothesis.
5. Conclusions
In this paper we describe a 2-km WRF-based dataset generated with a hybrid statistical–dynamical
JULY 2012
1219
MONAGHAN ET AL.
downscaling technique and assess its veracity in comparison with satellite-derived and in situ records. We
find that WRF-DS resolves the spatial variability and
annual cycle of temperature, humidity, and rainfall
reasonably when compared with satellite-based and in
situ records, albeit with some biases—most notably a
near-surface humidity dry bias that may be linked to the
Noah LSM. WRF-DS, along with the observationally
based datasets, is used to examine the general climatic
features of the greater West Nile region. The climatic
gradients in the West Nile region are strongly influenced
by terrain. The West Nile region lies in a transition
zone between the drier East African Plateau and the
wetter Congo Basin. The annual rainfall cycle is largely
unimodal—similar to the Sahel to the north—with a dry
season from December to February followed by gradually increasing rainfall from March through the boreal
autumn short-rains peak. A modest dry spell in June
imparts a slight bimodal aspect to the annual rainfall
cycle that is characteristic of rainfall in many other
regions of EEA. As shown above, gaining an understanding of these climatic features, especially the rainfall
variability, has been important for interpreting our
plague-modeling results and for developing a hypothesis
to explain plague seasonality.
We examine the influence of tropical-ocean variability from the Atlantic, Pacific, and Indian Oceans
on West Nile rainfall and find a positive correlation
between March–April rains and ENSO (through the
Niño-3.4 index) that extends from zero lag to at least
several months of lag, suggesting the possibility for
seasonal predictions, although it is not yet clear whether
such predictions might be beneficial for modeling human plague. Of interest is that other studies find the
strongest associations between ENSO and EEA in boreal autumn. We propose that the differential influence
of the ENSO teleconnection relative to other regions of
EEA may be partly due to the location of the West Nile
region within the complex climatic transition zone discussed above.
We summarize the application of the 2-km WRF-DS
climate fields by MacMillan et al. (2012) to model the
spatial distribution of human plague cases in the West
Nile region. It is not clear whether biases noted in some
of the WRF-DS fields may have an impact on the plaguemodel results. Employing WRF-DS fields improves the
overall plague-model accuracy by 13% relative to a
previous model that did not employ climate data (Eisen
et al. 2010), however. Human plague cases are positively
associated with rainfall during February, October, and
November and are negatively associated with rainfall
during June (during the brief dry spell between the long
rains and short rains). Our results suggest that areas that
receive increased but not continuous rainfall provide
ecologically conducive conditions for Y. pestis transmission in the West Nile region. Continued efforts to
refine our hypothesis and improve our understanding of
the conditions required to maintain enzootic cycles may
ultimately be used to target limited surveillance, prevention, and control resources to areas most at risk or to
predict the future range of human plague.
Acknowledgments. This work was funded by the CDC
and the U.S. Agency for International Development
through an interagency agreement with the National
Science Foundation. Doctor Daniel Steinhoff provided
useful comments. Sources and citations for datasets
employed in this paper are described within. Special
thanks are given to Russell Enscore and Jeff Borchert
of CDC, and to the Ugandan Virus Research Institute,
for providing in situ weather data from the airfield at
Arua. Doctor Kevin Griffith (CDC) provided invaluable insight and assistance in coordinating field work.
REFERENCES
Akaike, H., 1974: A new look at the statistical model identification.
IEEE Trans. Autom. Control, 19, 716–723.
Anyah, R. O., and F. H. M. Semazzi, 2004: Simulation of the sensitivity of Lake Victoria basin climate to lake surface temperatures. Theor. Appl. Climatol., 79, 55–69.
Asadullah, A., N. McIntyre, and M. Kigobe, 2008: Evaluation of
five satellite products for estimation of rainfall over Uganda.
Hydrol. Sci. J., 53, 1137–1150.
Balas, N., S. E. Nicholson, and D. Klotter, 2007: The relationship of
rainfall variability in west central Africa to sea-surface temperature fluctuations. Int. J. Climatol., 27, 1335–1349.
Barnston, A. G., M. Chelliah, and S. B. Goldenberg, 1997: Documentation of a highly ENSO-related SST region in the equatorial Pacific. Atmos.–Ocean, 35, 367–383.
Basalirwa, C. P. K., 1995: Delineation of Uganda into climatological rainfall zones using the method of principal component
analysis. Int. J. Climatol., 15, 1161–1177.
Birkett, C., R. Murtugudde, and T. Allan, 1999: Indian Ocean climate event brings floods to East Africa’s lakes and the Sudd
Marsh. Geophys. Res. Lett., 26, 1031–1034.
Bjerknes, J., 1969: Atmospheric teleconnections from the equatorial Pacific. Mon. Wea. Rev., 97, 163–172.
Black, E., 2005: The relationship between Indian Ocean sea-surface
temperature and East African rainfall. Philos. Trans. Roy.
Soc., 363A, 43–47.
——, J. Slingo, and K. R. Sperber, 2003: An observational study of
the relationship between excessively strong short rains in
coastal East Africa and Indian Ocean SST. Mon. Wea. Rev.,
131, 74–94.
Camberlin, P., and N. Philippon, 2002: The East African March–May
rainy season: Associated atmospheric dynamics and predictability over the 1968–97 period. J. Climate, 15, 1002–1019.
——, S. Janicot, and I. Poccard, 2001: Seasonality and atmospheric
dynamics of the teleconnections between African rainfall and
tropical ocean surface temperature: Atlantic vs. ENSO. Int.
J. Climatol., 21, 973–1005.
1220
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
——, V. Moron, R. Okoola, N. Philippon, and W. Gitau, 2009:
Components of rainy seasons’ variability in equatorial East
Africa: Onset, cessation, rainfall frequency and intensity.
Theor. Appl. Climatol., 98, 237–249.
Case, J. L., W. L. Crosson, S. V. Kumar, W. M. Lapenta, and C. D.
Peters-Lidard, 2008: Impacts of high-resolution land surface
initialization on regional sensible weather forecasts from the
WRF model. J. Hydrometeor., 9, 1249–1266.
Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface–
hydrology model with the Penn State–NCAR MM5 modeling
system. Part I: Model implementation and sensitivity. Mon.
Wea. Rev., 129, 569–586.
Ciach, G. J., 2003: Local random errors in tipping-bucket rain gauge
measurements. J. Atmos. Oceanic Technol., 20, 752–759.
Conway, D., E. Allison, R. Felstead, and M. Goulden, 2005:
Rainfall variability in East Africa: Implications for natural
resources management and livelihoods. Philos. Trans. Roy.
Soc., 363A, 49–54.
Dai, A., 2006: Precipitation characteristics in eighteen coupled
climate models. J. Climate, 19, 4605–4630.
Davis, D. H., 1953: Plague in Africa from 1935 to 1949: A survey of
wild rodents in African territories. Bull. World Health Org., 9,
665–700.
Dinku, T., P. Ceccato, E. Grover-Kopec, M. Lemma, S. J. Connor,
and C. F. Ropelewski, 2007: Validation of satellite rainfall
products over East Africa’s complex topography. Int. J. Remote Sens., 28, 1503–1526.
Dudhia, J., 1989: Numerical study of convection observed during
the Winter Monsoon Experiment using a mesoscale twodimensional model. J. Atmos. Sci., 46, 3077–3107.
East African Meteorological Department, 1963: Climatic seasons
of East Africa. East Africa Meteorological Dept. Rep. 8, 4 pp.
Eisen, R. J., and Coauthors, 2010: Assessing human risk of exposure to plague bacteria in northwestern Uganda based on
remotely sensed predictors. Amer. J. Trop. Med. Hyg., 82,
904–911.
Enfield, D. B., and D. A. Mayer, 1997: Tropical Atlantic sea surface
temperature variability and its relation to El Niño–Southern
Oscillation. J. Geophys. Res., 102, 929–945.
——, A. M. Mestas, D. A. Mayer, and L. Cid-Serrano, 1999: How
ubiquitous is the dipole relationship in tropical Atlantic sea
surface temperatures? J. Geophys. Res., 104, 7841–7848.
Goddard, L., and N. E. Graham, 1999: The importance of the Indian Ocean for simulating precipitation anomalies over eastern and southern Africa. J. Geophys. Res., 104, 19 099–19 116.
Grell, G. A., and D. Devenyi, 2002: A generalized approach to
parameterizing convection combining ensemble and data
assimilation techniques. Geophys. Res. Lett., 29, 1693,
doi:10.1029/2002GL015311.
Griffiths, J. F., 1959: The variability of annual rainfall in East Africa. Bull. Amer. Meteor. Soc., 40, 361–362.
——, 1972: Eastern Africa. Climates of Africa, J. F. Griffiths, Ed.,
Vol. 10, World Survey of Climtology, Elsevier, 333–347.
Hastenrath, S., 2001: Variations of East African climate during the
past two centuries. Climatic Change, 50, 209–217.
——, 2007: Circulation mechanisms of climate anomalies in East
Africa and the equatorial Indian Ocean. Dyn. Atmos. Oceans,
43, 25–35.
——, D. Polzin, and C. Mutai, 2007: Diagnosing the 2005 drought in
equatorial East Africa. J. Climate, 20, 4628–4637.
——, ——, and ——, 2010: Diagnosing the droughts and floods in
equatorial East Africa during boreal autumn 2005–08. J. Climate, 23, 813–817.
VOLUME 51
Hession, S. L., and N. Moore, 2011: A spatial regression analysis of
the influence of topography on monthly rainfall in East Africa.
Int. J. Climatol., 31, 1440–1456.
Hijmans, R. J., S. E. Cameron, J. L. Parra, P. G. Jones, and
A. Jarvis, 2005: Very high resolution interpolated climate
surfaces for global land areas. Int. J. Climatol., 25, 1965–
1978.
Hills, R. C., 1979: The structure of the inter-tropical convergence
zone in equatorial Africa and its relationship to East African
rainfall. Trans. Inst. Br. Geogr., 4, 329–352.
Huffman, G. J., and Coauthors, 2007: The TRMM Multisatellite
Precipitation Analysis (TMPA): Quasi-global, multiyear,
combined-sensor precipitation estimates at fine scale. J. Hydrometeor., 8, 38–55.
Indeje, M., F. H. M. Semazzi, and L. J. Ogallo, 2000: ENSO signals
in East African rainfall seasons. Int. J. Climatol., 20, 19–46.
Janjić, Z. I., 2002: Nonsingular implementation of the Mellor–
Yamada level 2.5 scheme in the NCEP Meso Model. NCEP
Office Note 437, 61 pp.
Janowiak, J. E., and P. A. Arkin, 1991: Rainfall variations in the
tropics during 1986–1989, as estimated from observations of
cloud-top temperatures. J. Geophys. Res., 96 (Suppl.), 3359–
3373.
Joyce, R. J., J. E. Janowiak, P. A. Arkin, and P. Xie, 2004:
CMORPH: A method that produces global precipitation estimates from passive microwave and infrared data at high
spatial and temporal resolution. J. Hydrometeor., 5, 487–503.
Kain, J. S., 2004: The Kain–Fritsch convective parameterization:
An update. J. Appl. Meteor., 43, 170–181.
Kanamitsu, M., W. Ebisuzaki, J. Woollen, S. K. Yang, J. J. Hnilo,
M. Fiorino, and G. L. Potter, 2002: NCEP–DOE AMIP-II
Reanalysis (R-2). Bull. Amer. Meteor. Soc., 83, 1631–1643.
Keeling, R. F., S. C. Piper, A. F. Bollenbacher, and J. S. Walker,
2009: Atmospheric CO2 records from sites in the SIO air
sampling network. Trends: A Compendium of Data on Global
Change, Carbon Dioxide Information Analysis Center, Oak
Ridge National Laboratory, Oak Ridge, TN, doi:10.3334/
CDIAC/atg.035.
Klemp, J. B., W. C. Skamarock, and J. Dudhia, 2007: Conservative
split-explicit time integration methods for the compressible
nonhydrostatic equations. Mon. Wea. Rev., 135, 2897–2913.
Latif, M., D. Dommenget, M. Dima, and A. Grotzner, 1999: The
role of Indian Ocean sea surface temperature in forcing East
African rainfall anomalies during December–January 1997/
98. J. Climate, 12, 3497–3504.
Lobell, D. B., M. Banziger, C. Magorokosho, and B. Vivek, 2011:
Nonlinear heat effects on African maize as evidenced by historical yield trials. Nat. Climate Change, 1, 42–45.
Love, T. B., V. Kumar, P. Xie, and W. Thiaw, 2004: A 20-year daily
Africa precipitation climatology using satellite and gauge
data. Preprints, 14th Conf. on Applied Climatology, Seattle,
WA, Amer. Meteor. Soc., P5.4. [Available online at http://
ams.confex.com/ams/pdfpapers/67484.pdf.]
MacMillan, K., and Coauthors, 2011: Landscape and residential
variables associated with plague-endemic villages in the West
Nile region of Uganda. Amer. J. Trop. Med. Hyg., 84, 435–442.
——, and Coauthors, 2012: Climatic predictors of the spatial distribution of human plague cases in the West Nile region of
Uganda. Amer. J. Trop. Med. Hyg., 86, 514–523.
Mlawer, E. J., S. J. Taubman, P. D. Brown, M. J. Iacono, and S. A.
Clough, 1997: Radiative transfer for inhomogeneous atmosphere: RRTM, a validated correlated-k model for the long
wave. J. Geophys. Res., 102, 16 663–16 682.
JULY 2012
MONAGHAN ET AL.
Mölg, T., J. C. H. Chiang, A. Gohm, and N. J. Cullen, 2009:
Temporal precipitation variability versus altitude on a tropical
high mountain: Observations and mesoscale atmospheric
modelling. Quart. J. Roy. Meteor. Soc., 135, 1439–1455.
Monaghan, A. J., D. L. Rife, J. O. Pinto, C. A. Davis, and J. R.
Hannan, 2010: Global precipitation extremes associated with
diurnally varying LLJs. J. Climate, 23, 5065–5084.
Moore, N., N. Torbick, B. Lofgren, J. Wang, B. Pijanowski,
J. Andresen, D.-Y. Kimf, and J. Olson, 2010: Adapting
MODIS-derived LAI and fractional cover into the RAMS in
East Africa. Int. J. Climatol., 30, 1954–1969.
Moron, V., A. W. Robertson, M. N. Ward, and P. Camberlin, 2007:
Spatial coherence of tropical rainfall at regional scale. J. Climate, 20, 5244–5263.
Mutai, C. C., and M. N. Ward, 2000: East African rainfall and the
tropical circulation/convection on intraseasonal to interannual
timescales. J. Climate, 13, 3915–3939.
——, ——, and A. W. Colman, 1998: Towards the prediction of
the East Africa short rains based on sea-surface temperature–
atmosphere coupling. Int. J. Climatol., 18, 975–997.
Nicholson, S. E., 1996: A review of climate dynamics and climate
variability in eastern Africa. The Limnology, Climatology and
Paleoclimatology of the East African Lakes, T. C. Johnson and
E. O. Odada, Eds., Gordon and Breach, 25–56.
——, 2000: The nature of rainfall variability over Africa on time scales
of decades to millennia. Global Planet. Change, 26, 137–158.
——, and D. Entekhabi, 1987: Rainfall variability in equatorial and
southern Africa: Relationships with sea surface temperatures
along the southwestern coast of Africa. J. Climate Appl. Meteor., 26, 561–578.
——, and J. Kim, 1997: The relationship of the El Niño–Southern
Oscillation to African rainfall. Int. J. Climatol., 17, 117–135.
Oettli, P., and P. Camberlin, 2005: Influence of topography on
monthly rainfall distribution over East Africa. Climate Res.,
28, 199–212.
Ogallo, L. J., 1988: Relationships between seasonal rainfall in East
Africa and the Southern Oscillation. J. Climatol., 8, 31–43.
——, 1989: The spatial and temporal patterns of the East African
seasonal rainfall derived from principal component analysis.
Int. J. Climatol., 9, 145–167.
Okoola, E. R., 1999: A diagnostic study of the eastern Africa
monsoon circulation during the Northern Hemisphere spring
season. Int. J. Climatol., 19, 143–168.
Philippon, N., P. Camberlin, and N. Faucherau, 2002: Empirical
predictability study of October–December East African
rainfall. Quart. J. Roy. Meteor. Soc., 128, 2239–2256.
Phillips, J., and B. McIntyre, 2000: ENSO and interannual variability in Uganda: Implications for agricultural management.
Int. J. Climatol., 20, 171–182.
Plisnier, P. D., S. Serneels, and E. F. Lambin, 2000: Impact of
ENSO on East African ecosystems: A multivariate analysis
based on climate and remote sensing data. Global Ecol. Biogeogr., 9, 481–497.
Reynolds, R. W., N. A. Rayner, T. M. Smith, D. C. Stokes, and
W. Q. Wang, 2002: An improved in situ and satellite SST
analysis for climate. J. Climate, 15, 1609–1625.
1221
Rife, D. L., J. O. Pinto, A. J. Monaghan, and C. A. Davis, 2010:
Global distribution and characteristics of diurnally varying
LLJs. J. Climate, 23, 5041–5064.
Roberts, J. I., 1936: Plague conditions in an urban area of Kenya
(Nairobi Township). J. Hyg., 36, 467–484.
Rodell, M., and Coauthors, 2004: The Global Land Data Assimilation System. Bull. Amer. Meteor. Soc., 85, 381–394.
Ropelewski, C. F., and P. D. Jones, 1987: An extension of the
Tahiti–Darwin Southern Oscillation index. Mon. Wea. Rev.,
115, 2161–2165.
Ruiz, J. J., C. Saulo, and J. Nogués-Paegle, 2010: WRF model
sensitivity to choice of parameterization over South America:
Validation against surface variables. Mon. Wea. Rev., 138,
3342–3355.
Saji, N. H., B. N. Goswami, P. N. Vinayachandran, and T. Yamagata, 1999: A dipole mode in the tropical Indian Ocean. Nature, 401, 360–363.
Skamarock, W. C., and J. B. Klemp, 2008: A time-split nonhydrostatic atmospheric model for research and NWP applications. J. Comput. Phys., 227, 3465–3485.
Smith, C. A., and P. Sardeshmukh, 2000: The effect of ENSO on
the intraseasonal variance of surface temperature in winter.
Int. J. Climatol., 20, 1543–1557.
Thompson, B. W., 1957: Some reflections on equatorial and tropical forecasting. East African Meteorological Dept. Tech.
Memo. 7, 14 pp.
Thompson, G., R. M. Rasmussen, and K. Manning, 2004: Explicit
forecasts of winter precipitation using an improved bulk microphysics scheme. Part I: Description and sensitivity analysis.
Mon. Wea. Rev., 132, 519–542.
Trenberth, K. E., and D. P. Stepaniak, 2001: Indices of El Niño
evolution. J. Climate, 14, 1697–1701.
Ummenhofer, C. C., A. Sen Gupta, M. E. England, and C. J. C.
Reason, 2009: Contribution of Indian Ocean sea surface temperatures to enhanced East African rainfall. J. Climate, 22, 993–1013.
Walker, G. T., 1924: Correlation in seasonal variations of weather.
IX. A further study of world weather. Mem. Ind. Meteor.
Dept., 24, 275–332.
Wan, Z., 2009: Collection-5 MODIS land surface temperature
products users’ guide. ICESS, University of California, Santa
Barbara, 30 pp.
Webster, P. J., A. M. Moore, J. P. Loschnig, and R. R. Leben, 1999:
Coupled ocean–atmosphere dynamics in the Indian Ocean
during 1997–98. Nature, 401, 356–360.
Winters, A. M., and Coauthors, 2009: Spatial risk models for human plague in the West Nile region of Uganda. Amer. J. Trop.
Med. Hyg., 80, 1014–1022.
Wolter, K., and M. S. Timlin, 1998: Measuring the strength of
ENSO—How does 1997/98 rank? Weather, 53, 315–324.
Wood, A. W., L. R. Leung, V. Sridhar, and D. P. Lettenmaier, 2004:
Hydrologic implications of dynamical and statistical approaches to downscaling climate model outputs. Climatic
Change, 62, 189–216.
Xie, P., and P. A. Arkin, 1996: Analyses of global monthly precipitation using gauge observations, satellite estimates, and
numerical model predictions. J. Climate, 9, 840–858.