You are on page 1of 19

INTERNATIONAL JOURNAL OF CLIMATOLOGY

Int. J. Climatol. 26: 2161–2179 (2006)


Published online 6 July 2006 in Wiley InterScience
(www.interscience.wiley.com) DOI: 10.1002/joc.1361

A REGIONAL CLIMATE STUDY OF CENTRAL AMERICA USING THE MM5


MODELING SYSTEM: RESULTS AND COMPARISON TO OBSERVATIONS
JOSE L. HERNANDEZ,a JAYANTHI SRIKISHEN,b DAVID J. ERICKSON III,a, * ROBERT OGLESBYb and DANIEL IRWINb
aClimate and Carbon Research Institute, National Center for Computational Sciences, Oak Ridge National Laboratory, Oak Ridge,
Tennessee 37830, USA
b Marshall Space Flight Center, National Aeronautics and Space Administration, Huntsville, Alabama, USA

Received 26 September 2005


Revised 16 February 2006
Accepted 10 April 2006

ABSTRACT
The Mesoscale Modeling system, version 3.6 (MM5) regional modeling system has been applied to Central America
and has been evaluated against National Oceanic and Atmospheric Administration/National Climatic Data Center
(NOAA/NCDC) daily observations and the Global Precipitation Climatology Project (GPCP) precipitation data. We
compare model results and observations for 1997 and evaluate various climate parameters (temperature, wind speed,
precipitation and water vapor mixing ratio), emphasizing the differences within the context of the station dependent
geographical features and the land use (LU) categories. At 9 of the 16 analyzed stations the modeled temperature, wind
speed and vapor mixing ratio are in agreement with observations with average model-observation differences consistently
lower than 25%. MM5 has better performance at stations strongly impacted by monsoon systems, regions typified by
low topography in coastal areas and areas characterized by evergreen, broad-leaf and shrub land vegetation types. At
four stations the model precipitation is about a factor of 3–5 higher than the observations, while the simulated wind
is roughly twice what is observed. These stations include two inland stations characterized by croplands close to water
bodies; one coastal station in El Salvador adjacent to a mountain-based cropland area and one station at sea-level. This
suggests that the model does not adequately represent the influence of topographic features and water bodies close to
these stations. In general, the model agrees reasonably well with measurements and therefore provides an acceptable
description of regional climate. The simulations in this study use only two seasonal maps of land cover. The main model
discrepancies are likely attributable to the actual annual cycle of land–atmosphere vapor and energy exchange that has a
temporal scale of days to weeks. These fluxes are impacted by surface moisture availability, albedo and thermal inertia
parameters. Copyright  2006 Royal Meteorological Society.

KEY WORDS: climate; regional modeling; Central America; diagnostic analysis; land use

1. INTRODUCTION

The evaluation of models against observations is fundamental in climate studies since it reveals various
misrepresentations in parameterization schemes leading to bias in estimating environmental properties and
fluxes at a variety of time and spatial scales. Regional models used in climate and forecasting allow the
study of the evolution of such geophysical properties and air–land interaction processes because of coupling
an atmospheric model to a land surface model (LSM). Research on lower atmosphere and land surface
property exchange have revealed the importance of considering several parameters, such as surface albedo,
evapotranspiration, roughness length, soil properties and vegetation type. These variables are required to
have an adequate description of energy and water vapor exchange that ultimately influences the atmospheric
boundary layer dynamics (Charney et al., 1977; Dickinson, 1983; Avissar and Verstraete, 1990; Henderson-
Sellers et al., 1993; Yongjiu et al., 2003). Current high-resolution regional models coupled with improved

* Correspondence to: David J. Erickson III, Climate and Carbon Research Institute, National Center for Computational Sciences, Oak
Ridge National Laboratory, Oak Ridge, TN 37831-6016, 865-574-3136, USA; e-mail: ericksondj@ornl.com

Copyright  2006 Royal Meteorological Society


2162 J. L. HERNANDEZ ET AL.

LSMs have shown a reasonable representation of land–atmosphere interactions by incorporating a range of


different land surface properties. These improvements have allowed a realistic depiction of diurnal, short-
term and seasonal cycles of heat, momentum and water (Betts et al., 1997; Ek et al., 2003; Chen and Dudhia,
2001b). Investigations evaluating the interactions among microphysics, cumulus, radiation, planetary boundary
layer (PBL) and surface processes have contributed to the creation of state-of-the-art models like MM5.
The fifth-generation Mesoscale Model (Anthes and Warner, 1978; Grell et al., 1994) is used worldwide
in forecasting and climate studies. The MM5 offers several options related to the coupling to LSMs and
schemes of atmospheric physics, giving reasonable estimates for land surface heat fluxes and atmospheric
solar radiative transfer (Chen and Dudhia, 2001a). For instance, the MM5 land surface schemes range from
simple five-layer vertical diffusion descriptions to sophisticated multilayered models allowing the prediction
of soil moisture, temperature profiles and interactions with the PBL. Even with these advances in modeling,
simulations of small-scale (≤20 km) processes and properties under different model options and environmental
settings present a challenge. We carried out simulations with MM5 to explore the performance of this model
in an annual simulation for Central America using observations from stations deployed under distinct land-
cover categories and elevations. The dry and wet seasons are also investigated with the Global Precipitation
Climatology Project (GPCP) data (Huffman et al., 2001) and model-observation comparisons were completed
for the rate of precipitation.
Seasonal controls on Central American regional climate are known, e.g. the Pacific and Atlantic monsoon
systems, trade winds, topographic effects, seasonal displacement of the Intertropical Convergence Zone
(ITCZ) and land surface type. However, many uncertainties remain on subseasonal to interannual timescales.
For example, on intraseasonal timescales, the midsummer drought (MSD) on the Pacific side of Central
America is well documented, but little is known about why it occurs (Magaña et al., 1999). In Central
America a bimodal annual cycle in precipitation with intensification around the May–June–July (MJJ)
and the August–September–October (ASO) time periods are observed. The Atlantic and Pacific Ocean
hydrologic cycles obviously are important in regional rainfall activity in Central America. Hastenrath (1976)
studied the driest and wettest summers in the Caribbean using rainfall records from 1911 to 1972 to find
a correlation between elevated sea-surface temperatures (SSTs) in the equatorial Pacific with a drier season
in the Caribbean. Rogers (1988) documented a drying influence of El Niño during the late rainfall season
over Central America. Positive anomalies in the sea-surface temperature of the tropical North Atlantic are
associated with an enhancement of precipitation in the MJJ time period. Taylor et al. (2002) suggest that a
decrease in rainfall during the ASO time period is strongly influenced by El Niño/La Niña events and that
there is a robust relationship between an east–west gradient of SST with rainfall rates during the ASO time
period.
A reliable evaluation of regional climate over the last several decades in Central America requires maps of
land-cover categories that reflect changes in land surface features corresponding to the study period. Global
forest assessments from the Food and Agriculture Organization (FAO) indicate that the Central American
region has undergone dramatic changes in land cover due to deforestation over the last several decades. These
human induced alterations to vegetation coverage cause changes in land surface geophysics and the surface
energy balance, resulting in significant climate change in tropical regions (Snyder et al., 2004). The Forest
Resources Assessment (FRA, 2000a,b) show that annual rates of deforestation in Central America during
the 1980s and 1990s were among the highest in the world, particularly in small countries like El Salvador
(4.6%), Belize (2.3%) and Guatemala (1.7%). Archard et al. (2002) estimated changes in humid tropical forest
distributions using satellite imagery to refine the calculation of carbon fluxes in the global carbon budget.
Archard et al. (2002) found several hot spots of deforestation in Central America, particularly in Honduras,
Nicaragua, Belize and Guatemala with almost the same loss rates as estimated for East Asia, a region that
has experienced the highest rates of deforestation on Earth. A dramatic example of a land use (LU)–climate
relationship is illustrated by the tropical montane cloud forests (TMCF), which are typically located high
in mountains where orographic clouds form due to the forced rising of trade winds. Lawton et al. (2001)
investigated TMCF and showed that deforested lowlands in Costa Rica remain relatively cloud free, while
the forested counterparts develop a dry season of cumulus clouds. They evaluated simulations from regional
models with realistic moisture advection across the model boundaries to compare results using forested and
Copyright  2006 Royal Meteorological Society Int. J. Climatol. 26: 2161–2179 (2006)
DOI: 10.1002/joc
A REGIONAL CLIMATE STUDY USING THE MM5 MODELING SYSTEM 2163

pasture surface cover. In the present study, MM5 is coupled to a five-layer soil vertical diffusive model and
a simple cumulus parameterization (Grell and Devenyi, 2002). The model uses information on land surface
characteristics from two seasonal maps of 25 vegetation and land-cover categories from the United States
Geological Survey (USGS). MM5 is currently using the most recent version of land-cover maps for regional
climate models from USGS, which has been generated using the 1993 normalized difference vegetation index
(NDVI) measured by the advanced very high-resolution radiometer (AVHRR) on board of National Oceanic
and Atmospheric Administration (NOAA) satellites (Loveland et al., 2000). In our work, we want to study the
performance of MM5 to evaluate the daily to annual variability of the regional climate in Central America. We
base this evaluation on model-observation comparisons of four parameters: temperature, wind speed, water
vapor mixing ratio and precipitation.
Climate studies from high-resolution regional models are very scarce in the Central American region. Here,
we present a diagnostic study of the MM5 modeling system adapted to describe the regional climate in Central
America (92.5° –77 ° W; 7° –18.5 ° N) for a one-year period (1997) at a resolution of 20 km. The National Center
for Environmental Protection (NCEP) reanalysis data (Kalnay et al., 1996) were used as boundary conditions
in the annual integration. MM5 allows nested simulations, which in our study were set to 60 and 20 km.
Since we focus on local to small region climate variability and diagnostic analysis, we prefer to use the
20 km resolution results in our investigation. We compare MM5 results with NOAA/NCDC observations
and GPCP monthly data in the seven countries in Central America and southern Mexico. Our analysis
includes an examination of climate statistics at 16 observational stations. We explore model-measurement
discrepancies within the context of land-cover features (including vegetation type) of the regions surrounding
the meteorological stations. We examine temperature, wind, water vapor mixing ratio and precipitation rates
and relate these to atmospheric transport and land surface climate parameters.

2. MODEL DESCRIPTION AND OBSERVATIONAL DATA

The MM5 modeling system is designed to simulate mesoscale and regional-scale atmospheric circulations. It
basically consists of a series of components that horizontally interpolate terrestrial data; interpolates global
or meteorological records on pressure levels to a predefined horizontal grid; establishes lateral and initial
conditions; and finally carries out regional simulations under various land and atmospheric physics options.
Lateral and lower boundary conditions are necessary in weather forecasting and regional climate simulations
and the NCEP reanalysis provides those conditions for the annual integration carried out in this work.
Soil and vegetation features are incorporated in the description of the atmosphere–land interactions through
parameterizations of plant mediated effects in the land surface scheme used. We use the MM5 version 3.6
in this study to conduct an annual simulation using settings described and referenced in the MM5 user’s
guide (Dudhia et al., 2005). We configured our model experiments using the cloud radiation scheme that
accounts for longwave and shortwave interactions with explicit cloud and clear air (Dudhia, 1989), a simple
single-cloud scheme useful for small grid sizes (Grell et al., 1994), a simple ice scheme where the clouds, rain
water and nonconvective precipitation are resolved through microphysics adding ice phase processes (Dudhia,
1989). We used the boundary layer parameterization of Hong and Pan (1996). A five-layer soil model with
levels at 1, 2, 4, 8 and 16 cm that employs a one dimensional heat diffusion equation is applied to predict
the vertical soil temperature profile (Dudhia, 1996). This model estimates the temperature profile considering
the thermal inertia and moisture availability provided as initial conditions for each category in the USGS LU
maps. According to Chen and Dudhia (2001b), although there is no explicit representation of vegetation in
this model, MM5 behaves well with regard to many atmospheric variables. The land surface scheme in MM5
results in simulations similar to advanced LSMs in forecasting simulations. We recognize the importance of
having improved representations of land surface processes through an advanced LSM; however, they require
additional fields and they are computationally expensive particularly for long climate simulation as proposed
here.
We used meteorological station measurements and GPCP precipitation data to evaluate the MM5 model
results. The meteorological measurements are taken from the Global Surface Summary of Day, which is
Copyright  2006 Royal Meteorological Society Int. J. Climatol. 26: 2161–2179 (2006)
DOI: 10.1002/joc
2164 J. L. HERNANDEZ ET AL.

supported by the National Climatic Data Center (NCDC) and is available at http://www.ncdc.noaa.gov/. The
data are from 1994 to the present, with daily average records updated weekly from a network of about 8000
stations around the world. To carry out the model evaluation we consider 16 stations in Central America that
are characterized by different geographical conditions: plains, mountains, coastal regions on the Pacific and
Atlantic oceans and the various land-cover categories. There are more than 16 stations for Central America
in the NCDC data set; however, several of these stations suffer from incomplete records with gaps of several
months. For this reason we select a subset with the longest records during the study period, year 1997.
Because of the limited number of stations in the region, this study of MM5 model evaluation is meant to
present preliminary conclusions and guidelines for future research. The GPCP monthly precipitation data
begin in 1979 and go through 2005. These precipitation estimates have been constructed by merging infrared
and microwave satellite estimates of precipitation with rain gauge analysis data from more than 6000 stations.
Adler et al., (2003) presents an exhaustive analysis of the GPCP data and estimates a random and sampling
error of 10–30% over regions of significant rainfall (>100 mm/month), particularly in the tropics. We compare
GPCP precipitation rates with those derived from MM5 convective and nonconvective precipitation.

3. STUDY AREA

The Mesoamerican region (Sader et al., 2001), which is part of the corridor between North and South America,
consists of seven countries (Panama, Costa Rica, Nicaragua, Honduras, El Salvador, Guatemala and Belize)
and southern Mexico. Mesoamerica is bordered by the Atlantic Ocean and Pacific Ocean, which influence
the climate variability in both coastal regions as some observational and reanalysis studies have shown. For
instance, Enfield and Alfaro (1999) used coarse (2.5° × 2.5° latitude–longitude) satellite-raingauge model
analysis data to find that the climatological behavior of the rainy season in Central America is more strongly
related to the tropical Atlantic sea-surface temperature anomaly than its counterpart in the tropical Eastern
Pacific. In a detailed description of the interannual variability, Taylor et al. (2002), found that the influence of
the tropical Atlantic wanes in the late season, when the equatorial Pacific and Atlantic become relatively more
important modulators. Magaña et al. (1999) considered 1° data derived from satellite and observational stations
to study the annual cycle of precipitation over the Caribbean and Central America. According to Magaña
et al. (1999) a bimodal distribution of precipitation exists with maxima in the June and September–October
time periods and a minimum in July–August not associated with the ITCZ, but controlled by trade wind
intensification and the topography of Central America. Although the regional climate is in general well-
defined, there are some particular climate features that depend on the combined effects of the proximity to
the coasts, the meridional migration of the ITCZ and topography (<4200 m in Guatemala, <3800 m in Costa
Rica, <2850 m in Honduras and <400 m mostly in the study region). For instance, in Guatemala (in the north
portion of the study region) there are two regimes of precipitation. Over the plains on the Atlantic coast of
Guatemala the rainy season lasts 7 months while on the Pacific coast of Guatemala, dominated by the Sierra
Madre, there are only 2 months of rainfall activity. In Nicaragua (in the midlatitudes of Mesoamerica), the
Pacific region experiences two well-defined dry and rainy seasons of 6 months each with annual precipitation
of 1000–2000 mm/year. In the same country, the large plains in the region bordering the Atlantic Ocean
experience longer rainy seasons (9–11 months) with annual precipitation that exceeds by a factor of 2 those
characteristics of the near Pacific coastal region. Panama, at the lowest latitudes, has precipitation patterns
more influenced by the ITCZ, with a minimum of rainfall around July and two rainy seasons in the April–June
and August–November time periods. As expected in coastal regions, the monsoon strongly influences the
strength, duration and spatial distribution of temperature, winds and precipitation through differential heating
between land and ocean. In addition, Central America has a variety of land-cover types (mainly crop lands,
mixed land types and forests) and water bodies (e.g. the Nicaragua and Managua lakes) influencing the
land-atmosphere exchange of heat and moisture.
Figure 1 presents the distribution of LU derived from the USGS, which is divided into 25 categories,
and the borders of the seven countries in Central America. Figure 2 shows the percentage of land-cover
types in the study region indicating large areas covered by tropical forest (evergreen broad-leaf, needle leaf
Copyright  2006 Royal Meteorological Society Int. J. Climatol. 26: 2161–2179 (2006)
DOI: 10.1002/joc
A REGIONAL CLIMATE STUDY USING THE MM5 MODELING SYSTEM 2165

Central America: Land Use Categories

Drylnd crop
18
Irrg crop
mixdry/irrig
crop/grass
16 crop/wood
grassland
Shrubland
Mix shrb/gr
14
Savanna
Latitude

Decis broadlf
Decis needle
12 Evergr br−lf
Evergr nd−lf
Mix forest

10 Herb wetland
Wd wetland
Sparse Veg
8 Herb tundra
Wd tundra
Mix tundra
−94 −92 −90 −88 −86 −84 −82 −80 −78
Longitude

Figure 1. The land use and land-cover categories in the seven countries in Central America used to drive the MM5 modeling system.
This USGS 25-vegetation based categories map shows that deciduous broad and needle leaf, crop grass and dry land crop are the most
frequent categories in the region

and mixed forest; with 42%, 4% and 5.8%, respectively), agriculture areas (dry land crop, crop wood with
22.4% and 13.7%, respectively), savannas (4.7%), grassland (2.8%) and shrub lands (2.2%). Figure 3 shows
the Central American topography and locations of the meteorological stations. Topography mainly consists
of large plains and low hills (<700 m) in Panama, Nicaragua, Belize, northern parts of Costa Rica and
Guatemala, Eastern Honduras and Belize. Two cordilleras, Talamanca (in Costa Rica and Panama) and Sierra
Madre (in Guatemala), dominate the mountain systems in Central America and have the highest elevations
at Volcan Tajamulco (4211 m), Volcan Baru (3475 m) and Cerro Chirripo (3810 m).

4. RESULTS AND DISCUSSION

The next eight plots present and summarize model results and attendant diagnostic analyses at 16 meteorolog-
ical stations for four variables: temperature, wind speed, water vapor mixing ratio and precipitation. Table I
contains the statistical summary for all stations and variables. The stations were chosen among a set of 30
available stations since they represent the longest continuous regional records in 1997 and are deployed in
areas with a variety of land-cover categories. Temperature, wind speed and precipitation are explicitly defined
in both measurements and model output variables; however, observed water vapor mixing ratio (Q) is derived
from the dew point and sea-level pressure measurements. Standard formulas were used for vapor pressure in
mb (Equation (1)) and mixing ratio in Kg/Kg (Equation (2)).

e = 6.1078 × 10TD×A/TD+B where TD is dew point (C) (1)


Q = .62197e/P − e where P is total pressure (2)
Copyright  2006 Royal Meteorological Society Int. J. Climatol. 26: 2161–2179 (2006)
DOI: 10.1002/joc
2166 J. L. HERNANDEZ ET AL.

USGS Land Use Categories Distribution in CA


45

40

35
Percentage of land cells

30 (*) Cells for ocean and water bodies


were masked
25

20

15

10

0
Urban
Dry land Crop. Past
Irrig. Crop. Past.
Mix. Dry/Irrg. C.P.
Crop Gras Mosc.
Crop./Wood Mosc.
Grassland
Shrubland
Mix.Shrb./Grs.
Savanna
Decids. Broadlf
Decids.Needlf
Evergrn. Broadlf
Evergrn Needlf
Mixed Forest
* Water Body
Herb. Wetland
Wood Wetland
Bar.Sparse Veg
HerbTundra
Wooden Tundra
Mixed Tundra
Bare Grnd Tundra
Snow or Ice
No Data
Figure 2. The percentages of land cover in Central America using the 25 USGS land use categories. The region is mostly covered by
tropical forest (52%; evergreen broad-leaf, needle leaf and mix forest categories), agriculture areas (36%; dry land, crop), savannas
(4.7%), grassland (2.8%) and shrub lands (2.2%)

A and B are constants, 7.5 and 237.3, respectively. Observed Q is compared against 2 m model vapor
mixing ratio. The modeled wind speed and temperature correspond to 10 m and 2 m levels, respectively.
The observational data sets are compared to the variables on the model grid map by using those grid
cells contained in a radius of 0.1° (10.84 km) around the observational stations. The model time series
was obtained by computing the average of those grid cells in the previously defined radius. We selected
0.1° because it represents a threshold radius below which the average includes no data at those station
locations close to bodies of water. The statistics analysis presented here includes only data on land close
to the stations and eliminates nearby data on water body grid cells. The model resolution is variable 0.15°
to 0.175° or (16.7–19.5 km) in the region and is nominally reported to be 20 km. Before the discussion
of results, we want to discuss the limitations and expectations in the modeling experiment presented in
this work. First, it is very difficult for any model in annual or longer simulations to follow local daily
variability since surface conditions (like land coverage and their physical characteristics) are changing
during the year with a direct impact on land–atmosphere exchange of properties. Having simulations
considering land-cover changes accordingly, the study period would have improved model results. Second,
when deciding on the spatial resolution and the possible use of any advanced LSM there is an increase
in computational cost for practical simulations. In the simulations carried out in this work, we anticipate
that the model could not follow a perfect representation of observed climate parameters, but we expect
the model reasonably simulate mesoscale circulation patterns and the correct magnitude of the annual
variability.
Copyright  2006 Royal Meteorological Society Int. J. Climatol. 26: 2161–2179 (2006)
DOI: 10.1002/joc
A REGIONAL CLIMATE STUDY USING THE MM5 MODELING SYSTEM 2167

Figure 3. The regional topography of Central America. Central America is mostly coastal plains and low lands (<700 m) and Cordillera
systems reaching heights of 4200 m in Guatemala, 3800 m in Costa Rica and 2850 m in Honduras

Figure 4 presents the MM5 simulation results during the dry (16–31 January) and two typically observed
rainy (16–30 June and 16–30 October) seasons using the average of the 3-h model records in 1997. Further
discussion on dry and rainy seasons will be accompanied with precipitation plots. The patterns of temperature
at 2 m clearly show differences in the dry and rainy season and depict temperatures commensurate with the
expected cold environments surrounding the two cordillera systems. The seasonal differences reach up to
9 ° C in mountainous highlands (>1000 m) and are less than 4° in coastal plain areas. The seasonal difference
of temperatures between dry and rainy season is normally expected to be lower at low latitudes and the
model results are in accordance with this view. In Guatemala (higher latitude) the differences are evident,
reaching about 10 ° C at low elevations; however, at similar elevations in Panama (lower latitude) such seasonal
differences oscillate about 2.5 ° C. In fact, the daily variation of temperature in the isthmus (6–10 ° C) is
typically four times higher than the seasonal variation. Regions adjacent to the Atlantic and Pacific Oceans
such as the low lands in Nicaragua (midlatitudes) also show small variation of temperature over the seasonal
cycle with an average around 23.5 ± 3 ° C. Average regional water vapor mixing ratios are smaller in January
(about 13 g/kg) and reach the lowest values (8–10 g/kg) at high elevations in all seasons. It is clear that
the coastal plains (in Nicaragua and Belize), and low lands (in northern Guatemala and Panama), experience
higher water vapor mixing ratios (>14 g/kg). There is a pronounced difference of water vapor mixing ratio
between dry (January) and rainy seasons (June and October). This is because of the occurrence of large spatial
patches of high water vapor mixing ratios in October as expected because this is when the impact of the ITCZ
is greatest. The average winds fluctuate between 1 and 5 m/s over the year over Central American lands with
intensification over the cordillera system in Guatemala, Honduras and central Nicaragua. Over the two ocean
basins, winds show areas of intensified winds (>6 m/s) in October, which is a normal characteristic of the
observed annual cycle.
Copyright  2006 Royal Meteorological Society Int. J. Climatol. 26: 2161–2179 (2006)
DOI: 10.1002/joc
2168

Table I. Summary statistic, 16 meteorological stations, physiographic features and annual average

Stat Z(m) Lat(° ) Lon(° ) Lnd use Temp (° C) Wind (m/s) MixRat (g/kg) Prec (mm/day)
Obs MM5 Obs MM5 Obs MM5 Obs MM5

Be1: Drylnd crop

Copyright  2006 Royal Meteorological Society


5 17.53 −88.3 27.5 ± 2.1 26.3 ± 2.2 3.3 ± 1.4 3.8 ± 1.2 17.9 ± 2.2 17.4 ± 1.7 9.9 ± 14.6 17.9 ± 23.8
Be2: 1 16.1 −88 .25 Water 25.5 ± 1.7 25.2 ± 1.5 1.9 ± 1.3 4.3 ± 1.8 15.3 ± 3.7 16.9 ± 1.4 5.7 ± 10.6 19.7 ± 34.7
Gu1: 1 15.71 −88.6 everg br-lf 27.0 ± 2.4 24.9 ± 2.3 2.6 ± 1.3 2.6 ± 0.9 17.9 ± 2.3 16.6 ± 1.7 14.5 ± 28.0 11.6 ± 21.8
Gu2: 2 13.91 −90.81 Drylnd crop 28.1 ± 1.7 26.5 ± 1.5 2.8 ± 1.0 2.8 ± 0.8 18.3 ± 1.8 16.9 ± 1.2 19.7 ± 39.0 36.1 ± 65.6
Gu3: 1489 14.58 −90.51 Urban 19.6 ± 1.5 17.2 ± 2.1 5.1 ± 1.8 4.4 ± 1.8 10.3 ± 1.4 12.4 ± 1.4 6.9 ± 8.5 12.4 ± 22.3
Ni1: 50 12.15 −86 .16 Crop/wood 27.7 ± 1.5 26.0 ± 1.5 1.4 ± 0.8 3.5 ± 1.2 16.1 ± 2.0 17.0 ± 1.0 9.7 ± 11.1 48.7 ± 108.0
Ni2: 90 12.1 −85 .36 Drylnd crop 29.2 ± 1.8 24.7 ± 1.5 2.4 ± 0.8 4.7 ± 1.7 16.7 ± 1.6 16.7 ± 1.0 2.6 ± 5.4 10.8 ± 22.8
CR1: −4 10 −83.05 Everg br-lf 27.2 ± 1.6 25.0 ± 1.2 2.7 ± 1.0 2.4 ± 0.6 19.8 ± 2.0 16.6 ± 1.1 19.1 ± 25.7 6.0 ± 10.1
Ho1: 3 15.73 −86.86 Drylnd crop 27.5 ± 2.1 23.7 ± 2.2 2.8 ± 1.5 3.8 ± 1.1 17.4 ± 1.6 16.2 ± 1.3 15.4 ± 17.9 17.9 ± 40.5
Ho2: 13 15.21 −83.8 Shrubland 27.8 ± 1.5 27.0 ± 1.6 3.1 ± 1.1 3.7 ± 1.1 18.5 ± 1.4 17.8 ± 1.2 11.4 ± 21.3 13.2 ± 24.7
Everg nd-lf mix
J. L. HERNANDEZ ET AL.

Ho3: 1079 14.78 −88.78 21.0 ± 2.4 20.3 ± 2.5 2.0 ± 1.0 3.0 ± 1.4 12.7 ± 1.5 13.9 ± 1.3 12.6 ± 18.4 6.9 ± 12.9
forest
Pn1: 11 9.05 −79.36 Shrubland Drylnd 27.9 ± 1.2 26.8 ± 1.4 2.1 ± 0.8 3.3 ± 1.4 17.7 ± 1.3 17.9 ± 1.0 8.9 ± 10.1 10.7 ± 19.8
crop
Pm2: 26 8.4 −82.41 Shrubland 28.8 ± 1.5 28.2 ± 1.6 2.9 ± 1.1 3.5 ± 2.0 17.5 ± 2.0 16.7 ± 1.1 21.7 ± 20.5 6.9 ± 26.1
Pm3: 13 8.98 −79.55 Drylnd crop 27.6 ± 1.0 26.8 ± 1.4 3.6 ± 1.0 3.0 ± 1.2 18.1 ± 1.4 17.9 ± 1.0 11.5 ± 15.3 7.7 ± 11.9
ES1: 621 13.7 −89.11 Drylnd crop 24.9 ± 1.7 23.1 ± 1.5 3.6 ± 1.5 3.2 ± 1.5 15.0 ± 1.9 15.5 ± 1.4 6.4 ± 12.7 19.5 ± 54.1
ES2: 15 13.56 −89 .83 Water Drylnd crop 29.3 ± 1.5 25.1 ± 1.1 1.5 ± 0.6 3.6 ± 1.9 18.6 ± 2.2 16.0 ± 1.4 17.1 ± 32.4 15.4 ± 30.9

DOI: 10.1002/joc
Int. J. Climatol. 26: 2161–2179 (2006)
A REGIONAL CLIMATE STUDY USING THE MM5 MODELING SYSTEM 2169

1997 MM5 Average


Temp(C) Mix Rat(g/kg) Wind Spd(m/s)
18 18 18
16 16 16
Jan, 16−31

14 14 14
12 12 12
10 10 10
8 8 8
−90 −85 −80 −90 −85 −80 −90 −85 −80

18 18 18
16 16 16
Jun, 16−30

14 14 14
12 12 12
10 10 10
8 8 8
−90 −85 −80 −90 −85 −80 −90 −85 −80

18 18 18
16 16 16
Oct, 16−30

14 14 14
12 12 12
10 10 10
8 8 8
−90 −85 −80 −90 −85 −80 −90 −85 −80

6 15 18 21 24 27 30 8 11 14 17 20 1 3 6 8 10

Figure 4. MM5 results: 2-week averages of 3-h data for 2 m-temperature, 2 m-water vapor mixing ratio, wind speed. Each row represents
periods during the typical dry (January) and rainy seasons (June: early wet season and October: wettest season) in Central America.
Significant changes are evident in each season and at different climate environments and elevations

The 3-h time step simulation of temperature, water vapor mixing ratio, wind speed and accumulated
precipitation are compared to their corresponding daily observations in Figures 5–8 throughout the 360 day
integration. These figures present the time series for each station and their corresponding names, for example,
ES2 is El Salvador second station. Since the model output is formatted as the accumulated precipitation until
any integration time, we compute daily precipitation by calculating the difference between the first and last
time of any particular day. The total daily accumulation of precipitation is the addition of both the convective
and nonconvective precipitation. A summary of the comparative performance showing the percentage
difference (100 × [model-observation]/observation) is presented in Figure 9. Simultaneous observation and
model records (as presented in Figures 5–8) are used to compute correlation coefficients for each variable
and station. These correlation coefficients are presented in Figure 10.
Figure 5 shows the 2 m air temperature from model and observations. In general, the daily model average
temperatures follow the same pattern of daily-observed temperatures in the year cycle, regardless if they are
inland or coastal (on the Atlantic or Pacific Ocean border). The model estimates of temperature tend to be
roughly 5% lower than observations. Where the model results are appreciably different (7–10% difference)
than observations, in Ni2, CR1, Ho1 and ES2, the model produces consistently lower daily estimates reflecting
a possible systematic bias. It is not clear what drives this bias considering the large variety of environmental
conditions where these stations are deployed. For instance, three of these stations (Ni2, Ho1 and ES2) are
Copyright  2006 Royal Meteorological Society Int. J. Climatol. 26: 2161–2179 (2006)
DOI: 10.1002/joc
2170 J. L. HERNANDEZ ET AL.

1997 Daily averages: Temperature

Temp (°C) 30 30 30 30

20 20 20 20

10 Be1 10 Be2 10 Gu1 10 Gu2


100 200 300 100 200 300 100 200 300 100 200 300

30 30 30 30
Temp (°C)

20 20 20 20

10 Gu3 10 Ni1 10 Ni2 10 CR1


100 200 300 100 200 300 100 200 300 100 200 300

30 30 30 30
Temp (°C)

20 20 20 20

10 Ho1 10 Ho2 10 Ho3 10 Pm1


100 200 300 100 200 300 100 200 300 100 200 300

30 30 30 30
Temp (°C)

20 20 20 20

10 Pm2 10 Pm3 10 ES1 10 ES2


100 200 300 100 200 300 100 200 300 100 200 300
day day day day
Red: MM5 2m Temp. Blue: NCDC Surface Temp.

Figure 5. Diagnosis of 2 m height MM5 temperature and NOAA/NCDC compiled observations at 16 meteorological stations in Central
America during 1997. There is good performance of the model to represent 2 m-temperature in the region. The modeled 3-h temperature
oscillates around observed daily averages and they behave in the same fashion over the annual cycle

in inland or coastal areas surrounded by agriculture regions, while station CR1 is bordered by tropical
forest (evergreen broad-leaf). The model agrees very well (to about 2%) in humid and coastal regions
such as Panama, and also in less humid higher latitudes, as can be seen in stations Be1, Be2, Gu1, Gu2,
Ho2 and Ho3. Since all stations show lower than 10% deviations from the observations, it is reasonable
to assert that the reanalysis boundary conditions and physical options used in this year long simulation
make a reasonable representation of 2 m air temperature for Central America. In addition, under different
environmental conditions (land surface categories and topography) the model performs well in describing this
variable.
The 10 m height modeled winds and observations are presented in Figure 6. In general, the model daily
average wind speed tracks the observed annual behavior; however, in some instances the model over-predicts
the observed magnitudes. The agreement is good at Be1, Gu1, Gu2, Gu3, CR1, Ho2, Pm2 and Pm3. These
stations are located in coastal regions near the Atlantic or Pacific Ocean basins and they are mostly at
low elevation (Z < 20 m), with the exception of Gu3, at 1489 m. Their land-cover categories span several
vegetated and agriculture ecosystems (evergreen broad-leaf, shrub land and dry land crops) and include the
urban category (Gu3). The MM5 has the highest overestimates (factor of 2) in the four inland or coastal
stations: Be2, Ni1, Ni2 and ES2 located in low lands surrounded by water or agricultural environments. In
summary, the wind speed from model and observations are in general agreement for the daily averages and
Copyright  2006 Royal Meteorological Society Int. J. Climatol. 26: 2161–2179 (2006)
DOI: 10.1002/joc
A REGIONAL CLIMATE STUDY USING THE MM5 MODELING SYSTEM 2171

1997 Daily averages: Wind Speed


15 15 15 15
Be1 Be2 Gu1 Gu2
WndSp(m/s) 10 10 10 10

5 5 5 5

0 0 0 0
100 200 300 100 200 300 100 200 300 100 200 300

15 15 15 15
Gu3 Ni1 Ni2 CR1
WndSp(m/s)

10 10 10 10

5 5 5 5

0 0 0 0
100 200 300 100 200 300 100 200 300 100 200 300

15 15 15 15
Ho1 Ho2 Ho3 Pn1
WndSp(m/s)

10 10 10 10

5 5 5 5

0 0 0 0
100 200 300 100 200 300 100 200 300 100 200 300

15 15 15 15
Pm2 Pm3 ES1 ES2
WndSp(m/s)

10 10 10 10

5 5 5 5

0 0 0 0
100 200 300 100 200 300 100 200 300 100 200 300
day day day day
Red: MM5 10m Wind Speed Blue: NCDC Surface Wind

Figure 6. Diagnosis of 10 m height MM5 wind speed and NOAA/NCDC compiled observations at 16 meteorological stations in
Central America during 1997. In general, the MM5 describes well the surface wind following the observed daily averages. The model
overestimates wind speed in Be2, Ni1, Ni2 and ES2 stations, which have in common some characteristics; low elevations and surrounded
by water bodies and crops

yearly timescale variability and as a consequence, the model shows good performance describing this variable
under the different environments in the region.
Figure 7 displays the model-observation evaluation for water vapor mixing ratio. The model provides water
vapor mixing ratio as one of the output variables; however, the observational data was calculated as a function
of observed dew point and pressure as per the following Equations (1) and (2). Water vapor mixing ratio
is the meteorological parameter that was best simulated in our experiment, with bias ≤10% among 14 of
the 16 stations. The stations GU3 (at 1489 m, the highest station) and CR1 (at −4 m, the lowest station)
that experience very different environmental conditions, displayed slightly higher bias ranging 10–20%. We
consider Be2 an isolated case where model-observation discrepancies in the second half of the year could
be likely attributable to a malfunction of the measurement devices in the station. Since lower than 10 g/kg
mixing ratio are unlikely to occur in the region, at Be2 we reject intervals of low mixing ratio values. This
quality control procedure will influence the comparative performance plots and statistical analysis presented
later and is used only at this station. Overall, the model water vapor mixing ratio agrees with the observed
annual variation and daily averages in all the stations. Figures 4 and 7 jointly confirm that MM5 simulates the
mesoscale patterns expected in the annual cycle with lower mixing ratios (10–13 g/kg) at higher elevation
(Z > 600 m) and higher values (15–19 g/kg) in coastal low elevation and plain areas.
Copyright  2006 Royal Meteorological Society Int. J. Climatol. 26: 2161–2179 (2006)
DOI: 10.1002/joc
2172 J. L. HERNANDEZ ET AL.

1997 Daily averages: Mixing Ratio


25 25 25 25
MxRt(g/kg) Be1 Be2 Gu1 Gu2
20 20 20 20
15 15 15 15
10 10 10 10
5 5 5 5
100 200 300 100 200 300 100 200 300 100 200 300

25 25 25 25
Gu3 Ni1 Ni2 CR1
MxRt(g/kg)

20 20 20 20
15 15 15 15
10 10 10 10
5 5 5 5
100 200 300 100 200 300 100 200 300 100 200 300

25 25 25 25
Ho1 Ho2 Ho3 Pn1
MxRt(g/kg)

20 20 20 20
15 15 15 15
10 10 10 10
5 5 5 5
100 200 300 100 200 300 100 200 300 100 200 300

25 25 25 25
Pm2 Pm3 ES1 ES2
MxRt(g/kg)

20 20 20 20
15 15 15 15
10 10 10 10
5 5 5 5
100 200 300 100 200 300 100 200 300 100 200 300
day day day day
Red: MM5 Mixing Ratio Blue: Mix.Rat. estimated from NCDC obs.

Figure 7. Diagnosis of 2 m height MM5 mixing ratio and NOAA/NCDC compiled observations at 16 meteorological stations in Central
America during 1997. In general the MM5 performs very well in describing the observed annual pattern of this parameter

Figure 8(a) shows the model-observation evaluation for daily accumulated precipitation. MM5 computes
separately convective and nonconvective precipitation, and we add these two components to obtain total
model precipitation. The model precipitation is the parameter that has the highest frequency of significant
discrepancies with measurements. The time series presented here only shows those days when model and
observation are both indicating significant precipitation is occurring. Overall, MM5 over-predicts precipitation,
however, at some stations the model performs reasonably well. There is a group of stations with extreme
differences: Ni1 (402%), Ni2 (309%), Be2 (246%) and ES1 (203%); another group with large differences
around 75% that includes Be1, Gu2, Gu3, CR1 and Pm2 and finally, stations with error lower than 25%
(Gu1, Ho1, Ho2, Pm1 and ES2). A common feature of those stations with extreme values is that water bodies
surround them. Apparently, MM5 tends to overestimate precipitation under the chosen physical options at
such stations. The stations in Nicaragua (Ni1 and Ni2) are surrounded by Nicaragua and Managua lakes,
while Be2 (Caye Hunting) in Belize is located on the Atlantic Ocean coast at sea-level. Figure 8(b) presents
a more detailed inspection of the model convective and nonconvective components at those stations with
elevated precipitation. Both convective and nonconvective model components follow the features displayed
by observations, particularly the enhanced precipitation in September–October during the rainy season at
Ni1. In Be2 the model also matched the observed precipitation over the first two months of the year.
However, in Panama, a region more influenced by the ITCZ, the model predicts much lower precipitation
than expected. Validation studies of MM5 in the United States, such as Chen and Dudhia (2001b) have
Copyright  2006 Royal Meteorological Society Int. J. Climatol. 26: 2161–2179 (2006)
DOI: 10.1002/joc
A REGIONAL CLIMATE STUDY USING THE MM5 MODELING SYSTEM 2173

(a) 1997 Daily Accumulated Precipitation

Prec (mm/day)
150 Be1 150 Be2 150 Gu1 150 Gu2
100 100 100 100
50 50 50 50
0 0 0 0
100 200 300 100 200 300 100 200 300 100 200 300
Prec (mm/day)

150 Gu3 150 Ni1 150 Ni2 150 CR1


100 100 100 100
50 50 50 50
0 0 0 0
100 200 300 100 200 300 100 200 300 100 200 300
Prec (mm/day)

150 Ho1 150 Ho2 150 Ho3 150 Pn1


100 100 100 100
50 50 50 50
0 0 0 0
100 200 300 100 200 300 100 200 300 100 200 300
Prec (mm/day)

150 Pm2 150 Pm3 150 ES1 150 ES2


100 100 100 100
50 50 50 50
0 0 0 0
100 200 300 100 200 300 100 200 300 100 200 300
day day day
Red: MM5 Convective + NonConvective acum. prec. Blue: NCDC daily acum. prec.

Figure 8. (a) Diagnosis of precipitation: The total precipitation from the MM5 model is derived from the addition of nonconvective
and convective precipitation. Extreme differences correspond to Be2, Ni1, Ni2 and Es2. Several previous MM5 based investigations
documented a domain resolution dependency on precipitation biases. (b) Diagnosis of precipitation separately for convective and
nonconvective precipitation in the lowest performance stations. The components of model precipitation in general contribute in the
same proportion to total precipitation, but in the wettest season at Ni1 the nonconvective precipitation is about twice than the convective
precipitation

reported too high simulated total rainfall and over-prediction of aerial extent for light precipitation. Previous
investigations have shown that under different environmental conditions MM5 reproduces basic features
in time and location of precipitation, however; amounts are higher than measurements and the model
performance is affected by effects of horizontal resolution on orographic rainfall (Chiao et al., 2004).
The performance of MM5 using different resolutions has been an issue of investigation over the last
several years. Related papers and presentations at MM5 workshops are available at the model web page:
http://www.mmm.ucar.edu/mm5/workshop/workshop-papers ws03.html. For instance, Zhong et al. (2004),
used MM5 on 12 km and 4 km resolutions and found over-prediction of precipitation at both resolutions;
however, the 12 km results were greater by a factor of 2–4 than those obtained with the 4 km domain,
which shows a factor of 1.3 to two times higher than observations. In some stations in Central America
MM5 has reasonable estimates with differences compared to observations of less than 25%. These stations
include Gu1, Ho1, Ho2, Pm1 and ES2. Most of these stations are bordered by tropical forest (evergreen
broad-leaf and mixed forest), shrub lands at low elevation or are close to ocean basins (Atlantic and
Pacific).
Figure 9 shows the performance of MM5 compared to the station-by-station observational data set. This
plot summarizes the discussion of Figures 5–8. The model precipitation has the highest deviation from
observations among the four analyzed variables, followed by wind speed. The extreme values model-data
mismatch for precipitation are noted in the upper-left corner caption. The vertical limits are adjusted to
Copyright  2006 Royal Meteorological Society Int. J. Climatol. 26: 2161–2179 (2006)
DOI: 10.1002/joc
2174 J. L. HERNANDEZ ET AL.

(b) NCDC Obs. and MM5 Conv. Precipitation


100 100
Prec(mm/day) Ni1: 0.0037962 Ni1: −0.0029986
50 50

0 0
100 200 300 100 200 300

100 100
Prec(mm/day)

Be2: 0.13945
Be2: 0.35532
50 50

0 0
100 200 300 100 200 300

100 100
Prec(mm/day)

ES1: −0.084416 ES1: −0.025712


50 50

0 0
100 200 300 100 200 300

100 100
Prec(mm/day)

Pm2: 0.20035 Pm2: 0.40262


50 50

0 0
100 200 300 100 200 300
day day
Red: MM5 Model Blue: NCDC Obs.

Figure 8. (Continued)

see the distribution of deviations for most of variables and stations excluding extreme values. According
to the analysis in Figure 9, the model describes the four analyzed fields with error lower than 25% in ten
of the stations. At five stations (Gu1, Ho1, Ho2, Pm1 and ES2) MM5 achieves reasonable agreement with
observed precipitation with an error less than ±25%. MM5 does quite well simulating water vapor mixing
ratio and temperature with an average error of less than 10% for all stations. Our analysis, time series
plots and statistics (averages, standard deviation and correlation coefficients) is based on concurrent records
for all variables. As we show, MM5 used with the current configuration is tracking the trends in annual
variability for three variables (temperature, wind and water vapor mixing ratio). The correlations between
observations and the model are expected to improve by using advanced LSMs and land use maps that are
exactly representative of the studied period. At this stage a moderate positive correlation higher than 0.5, would
represent a reasonable result. Figure 10 shows the correlation coefficients for all stations and variables and it
is intended to quantitatively summarize the performance of the model. The correlation coefficients presented
in Figure 10 confirm our discussion of Figures 5–8(a) in terms of a better agreement for water vapor mixing
ratio and temperature but lower realism for wind speed and precipitation rate. Further investigation is needed
to explore new physical options. We particularly believe that considering an advanced LSM for soil description
and moisture schemes can improve the performance of the model.
Table I presents the statistics summary (annual average and standard deviation) of the model output and
observations. This table together with Figures 9 and 10 are intended to capture the common features for
Copyright  2006 Royal Meteorological Society Int. J. Climatol. 26: 2161–2179 (2006)
DOI: 10.1002/joc
A REGIONAL CLIMATE STUDY USING THE MM5 MODELING SYSTEM 2175

Comparative Performance: all stations and Variables


200
+Precipitation
+Temperature +
Extreme percentage of difference in precipitation at: Wind speed
Be2(246%); Ni1(402%); Ni2(309%); ES1(203%)
+Mixing rate
Percentage of Difference 150

100

75

50

25

−25

−50

−75
Be1 Be2 Gu1 Gu2 Gu3 Ni1 Ni2 CR1 Ho1 Ho2 Ho3 Pm1 Pm2 Pm3 ES1 ES2

Figure 9. Comparative performance of the model at all stations for temperature, wind speed, mixing ratio and total precipitation. In
9 of the 16 stations the percent difference for temperature wind and mixing ratio is lower than 25%. It is difficult to define the drivers
of precipitation model bias since this parameter is also well described in five stations with error <25%

regions of best and worst model performance. Station location, altitude and surrounding LU cover are pre-
sented to facilitate the analysis of possible influences from natural controls on the estimation of the surface
climate parameters. To emphasize our findings displayed in Figures 5–8, Table I shows several highlighted
lines. Rows in italics represent the worst performance of the model (Be2, Ni1, Ni2 and ES2), where precip-
itation and wind speed are overestimated by the model by more than a factor of 2. Rows in bold black show
the best model descriptions (Gu1 and Ho2) where model-observation differences are less than 25% for all
four variables.
To fully explore the precipitation rates produced by MM5 for Central America, Figure 11 compares 1997
monthly precipitation rate distributions from MM5 and GPCP satellite and direct measurements blended
data. As seen in Figure 8(b), nonconvective precipitation tends to be higher than convective precipitation in
some instances making total model precipitation higher than expected. As seen in Figure 8(a) the model
gets reasonable results (Gu1, Ho1, Ho2, Pm1 and ES2) when convective + nonconvective (CP + NCP)
is taken. We now show the separate distribution of convective precipitation only and the distribution
from convective + nonconvective precipitation during dry (January) and wet (June and September–October)
seasons. These seasons are clearly seen in the GPCP data. The ITCZ can explain most of the annual variation
of precipitation in the study area. Minimal precipitation rates (monthly average <3 mm/day) all over the
Central American region are observed in January when the ITCZ is located at lower latitudes barely influencing
southern Central America. CP and CP + NCP display similar spatial patterns as the GPCP data, but with higher
magnitude than the observations. June shows the increased precipitation over the Pacific basin for CP + NCP
and CP + NCP resembles more accurately the GPCP distributions on land over Nicaragua, Honduras and
Guatemala. The September–October wet season shows larger oceanic areas of high precipitation in the model
results. However, CP is essentially half of GPCP and the CP + NCP distributions over land are in better
agreement.
One important point that influences model precipitation is that our 1-year simulation uses only two
seasonal maps to describe land surface features. This is the standard methodology for using MM5 (Chen
and Dudhia, 2001a). Every land-cover category has properties such as albedo, soil moisture availability,
Copyright  2006 Royal Meteorological Society Int. J. Climatol. 26: 2161–2179 (2006)
DOI: 10.1002/joc
2176 J. L. HERNANDEZ ET AL.

Correlation Coeficients: All stations and variables


1

+Mixing rate +Wind speed + Temperature + Precipitation


0.8

0.6
Correlat. Coef.

0.4

0.2

−0.2

−0.4
Be1 Be2 Gu1 Gu2 Gu3 Ni1 Ni2 CR1 Ho1 Ho2 Ho3 Pm1 Pm2 Pm3 ES1 ES2

Figure 10. Correlation coefficients for all stations and variables. Model-observation relationship for water vapor mixing ratio has the
highest correlations (0.5–0.75). Lowest correlations are observed for precipitation

emissivity, roughness length and thermal inertia that the model uses during integration to calculate the
surface energy and mass fluxes that have an impact on precipitation. Clearly, the lack of information on
a finer time resolution for land cover is likely a cause of bias in the model precipitation. Monthly land-cover
maps would be appropriate for additional annual simulations. As continuation of the experiment presented
here we are exploring new experiments and carrying out a sensitivity analysis to study the influence of
LU category distributions and higher spatial resolution around stations in Central America. In summary, we
found noticeable changes in LU distribution around the stations under higher resolutions 3 and 9 km. Setting
the initial conditions for MM5 implies interpolation from very high (30 s) spatial resolution LU maps. In
these new experiments, model precipitation show an improvement following the spatial patterns of satellite
precipitation data and the magnitudes of precipitation are closer to observations as well. Another issue is that
our model experiment is long compared to short forecast simulations. That can explain why mesoscale features
may be smoothed out during data preprocessing. In addition, convective and nonconvective parameterization
schemes are approximations of real atmospheric processes. Although these schemes can produce generally
correct trends, they are unlikely to capture short-term events (<3 h) at local stations unless data assimilation
is done. We did no data assimilation in our experiment.

5. CONCLUSIONS

This study has presented an evaluation of a regional climate model (MM5) in a region exposed to a
complex set of physical, topographic and environmental controls. The MM5 model system was tested
against NOAA/NCDC daily observations in Central America for the year 1997 and compared to GPCP
precipitation rate data. Temperature, wind speed, mixing ratio and precipitation were studied to explore
the model performance and the ability to describe the regional climate variability. We have investigated
the observed differences within the context of the station dependent geographical features and the USGS
Copyright  2006 Royal Meteorological Society Int. J. Climatol. 26: 2161–2179 (2006)
DOI: 10.1002/joc
A REGIONAL CLIMATE STUDY USING THE MM5 MODELING SYSTEM 2177

1997 Monthly Precipitation Rate (mm/day)


GPCP MM5: Conv. Prec. MM5: Con+NonCor
18 18 18
15 15 15
Jan

12 12 12
9 9 9
6 6 6
−96 −92 −88 −84 −80 −76 −72 −96 −92 −88 −84 −80 −76 −72 −96 −92 −88 −84 −80 −76 −72

18 18 18
15 15 15
Jun

12 12 12
9 9 9
6 6 6
−96 −92 −88 −84 −80 −76 −72 −96 −92 −88 −84 −80 −76 −72 −96 −92 −88 −84 −80 −76 −72

18 18 18
15 15 15
Sep

12 12 12
9 9 9
6 6 6
−96 −92 −88 −84 −80 −76 −72 −96 −92 −88 −84 −80 −76 −72 −96 −92 −88 −84 −80 −76 −72

18 18 18
15 15 15
Oct

12 12 12
9 9 9
6 6 6
−96 −92 −88 −84 −80 −76 −72 −96 −92 −88 −84 −80 −76 −72 −96 −92 −88 −84 −80 −76 −72

0.25 0.5 1 3 6 9 12

Figure 11. Comparison of monthly precipitation rates from GPCP data and the MM5 output in four periods characterized as dry season
(January) and the rainy seasons (June, September and October). MM5 convective precipitation (second column) presents lower than
observed GPCP precipitation but follow the trend in the annual cycle. Convective + nonconvective precipitation describes in a better
way the annual cycle on land and show intensive precipitation on the Pacific basin during September–October wettest season

LU categories. On the basis of this evaluation, we studied the influence of physiogeography and other
ecosystem drivers in the distribution and variability of surface climate parameters. In general, the model
has a good performance in describing mesoscale and local climate variability for temperature, wind speed
and water vapor mixing ratio, responding to topographic features, coastal processes and external controls
like the ITCZ. At the meteorological stations, the modeled temperature, wind speed and water vapor
mixing ratio are in good agreement with corresponding observations, showing an average percentage of
difference lower than a 25% in 9 of the 16 surveyed stations. MM5 has the best performance (all variable
errors <25%) at stations Gu1 and Ho2 and they are located within monsoon systems surrounded by
low topography in Atlantic coastal areas with land-cover categories of evergreen broad-leaf and shrub
land.
The stations that show high model-observation differences mostly correspond to those stations where
overestimates in precipitation and wind were found (Be2, Ni1, Ni2 and Es2). These extreme case stations
Copyright  2006 Royal Meteorological Society Int. J. Climatol. 26: 2161–2179 (2006)
DOI: 10.1002/joc
2178 J. L. HERNANDEZ ET AL.

are located in low elevation lands surrounded by water or agricultural environments, inland or coastal areas
near the Atlantic or Pacific Ocean. Model accumulated precipitation is the parameter that shows the highest
discrepancies, reaching a factor of 3–5 at four stations (Be2, Ni1, Ni2 and ES1). There were five stations
(Gu1, Ho1, Ho2, Pm1 and ES2) where total model precipitation agrees with the observations with an error
of <25%. Since the model experiments carried out in this study use only two seasonal maps of land cover
and the higher discrepancies occur in precipitation, the error is likely attributable to the annual variation of
land cover which controls land–atmosphere vapor and energy exchange through surface moisture availability,
albedo and thermal inertia parameters.

ACKNOWLEDGEMENTS

The submitted manuscript has been authored by four contractors of the U.S. Government under Contract
No. DE-AC05-000R22725. Accordingly, the U.S. Government retains a non-exclusive, royalty-free license to
publish or reproduce the published form of this contribution, or allow others to do so, for U.S. Government
purposes. The authors wish to thank the support for this investigation from the USAID PASA Contract No.
596-P-00-03-00144-00. This research used computational resources from Oak Ridge National Laboratory,
Computer Science and Mathematics division.

REFERENCES
Adler RF, Huffman GJ, Chang A, Ferraro R, Xie P, Janowiak J, Rudolf B, Schneider U, Curtis S, Bolvin D, Gruber A, Susskind J,
Arkin P, Nelkin E. 2003. The version 2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979-
present). Journal of Hydrometeorology 4: 1147–1167.
Anthes RA, Warner TT. 1978. Development of hydrodynamic models suitable for air pollution and other mesometeorological studies.
Monthly Weather Review 106: 1045–1078.
Archard FF, Hugh E, Stibig H-J, Mayaux P, Gallego J, Richards T, Malingreau J-P. 2002. Rates of the world’s humid tropical forests.
Science 297: 999–1002.
Avissar R, Verstraete MM. 1990. The representation of continental surface process in atmospheric models. Reviews of Geophysics 28:
35–52.
Betts A, Chen F, Mitchel K, Janjic Z. 1997. Assessment of the land surface and boundary layer model in two operational versions of
the NCEP ETA model using FIFE data. Monthly Weather Review 125: 2896–2916.
Charney JG, Quirk WK, Chow SH, Kornfield J. 1977. A Comparative study of the effects of albedo change on drought in semiarid
regions. Journal of Atmospheric Sciences 34: 1366–1385.
Chen F, Dudhia J. 2001a. Coupling and advanced land-surface/hydrology model with the Penn State-NCAR MM5 modeling system.
Part I: model implementation and sensitivity. Monthly Weather Review 129: 569–585.
Chen F, Dudhia J. 2001b. Coupling an advanced land surface-hydrology model with the Penn State/NCAR MM5 modeling system.
Part II: preliminary model validation. Monthly Weather Review 129: 587–604.
Chiao S, Lin Y-L, Kaplan M. 2004. Numerical study of the orographic forcing of heavy precipitation during MAP IOP-2B. Monthly
Weather Review 132: 2184–2203.
Dickinson RE. 1983. Land surface processes and climate-surface albedos and energy balance. Advances in Geophysics 25: 305–353.
Dudhia J. 1989. Numerical study of convection observed during the winter monsoon experiment using a mesoscale two-dimensional
model. Journal of the Atmospheric Sciences 46: 3077–3107.
Dudhia J. 1996. A multi-layer soil temperature model for MM5. Preprints, The Sixth PSU/NCAR Mesoscale Model Users Workshop,
22–24 July 1996, Boulder, CO, 49–50. Available from: http://www.mmm.ucar.edu/mm5/mm5v2/whatisnewinv2.html. [Date of access
February 16, 2006].
Dudhia J, Gill D, Manning K, Wang W, Bruyere C. 2005. PSU/NCAR mesoscale modeling system tutorial class notes and users’ guide.
(MM5 Modeling System Version 3). http://www.mmm.ucar.edu/mm5/mm5v3/v3model.html. [Date of access February 16, 2006].
Ek MB, Mitchell KE, Lin Y, Rogers E, Grunmann P, Koren V, Gayno G, Tarpley JD. 2003. Implementation of Noah land surface
model advances in the National Centers for Environmental Prediction operational mesoscale Eta model. Journal of Geophysical
Research 108(D22): 8851, DOI: 10.1029/2002JD003296.
Enfield DB, Alfaro EJ. 1999. The dependence of Caribbean rainfall on the interaction of the tropical atlantic and pacific oceans. Journal
of Climate 12: 2093.
Forest Resources Assessment. 2000a. Global Forest Resources Assessment 2000. http://www.fao.org/docrep/004/y1997e/y1997e00.htm#
Contents. [Date of access December 6, 2005].
Forest Resources Assessment. 2000b. Comparison of forest area and forest area change estimates derived from FRA 1990 and FRA
2000. ftp://ftp.fao.org/docrep/fao/006/ad068e/AD068E.pdf. [Date of access December 6, 2005].
Grell G, Devenyi Y. 2002. A generalized approach to parameterizing convection combining ensemble and data assimilation techniques.
Geophysical Research Letters 29(14): 1963.
Grell G, Dudhia J, Stauffer D. 1994. A description of the fifth-generation Penn State/NCAR Mesoscale Model (MM5). NCAR Tech
note NCAR/TN-398 + STR, 117.
Hastenrath S. 1976. Variations in low-latitude circulation and extreme climatic events in the tropical Americas. Journal of Atmospheric
Sciences 33: 202–215.

Copyright  2006 Royal Meteorological Society Int. J. Climatol. 26: 2161–2179 (2006)
DOI: 10.1002/joc
A REGIONAL CLIMATE STUDY USING THE MM5 MODELING SYSTEM 2179

Henderson-Sellers A, Yang ZL, Dickinson RE. 1993. The project for the intercomparison of land-surface parameterization schemes.
Bulletin of the American Meteorological Society 74: 1335–1349.
Hong SY, Pan L. 1996. Nonlocal boundary layer vertical diffusion in a medium-range forecast model. Monthly Weather Review 124:
2322–2339.
Huffman GJ, Adler RF, Morrissey MM, Curtis S, Joyce R, McGavock B, Susskind J. 2001. Global precipitation at one-degree daily
resolution from multi-satellite observations. Journal of Hydrometeorology 2: 36–50.
Kalnay E, Kanamitsu M, Kistler R, Collins W, Deaven D, Gandin L, Iredell M, Saha S, White G, Woollen J, Zhu Y, Leetmaa A,
Reynolds B, Chelliah M, Ebisuzaki W, Higgins W, Janowiak J, Mo KC, Ropelewski C, Wang J, Jenne R, Joseph D. 1996. The
NCEP/NCAR 40-Year reanalysis project. Bulletin of the American Meteorological Society 77(3): 437–472.
Lawton RO, Nair US, Pielke RA Sr., Welch RM. 2001. Lowland deforestation on nearby montane cloud forests. Science 294: 584–587.
Loveland TR, Reed BC, Brown JF, Ohlen DO, Zhu J, Yang L, Merchant JW. 2000. Development of a global land cover characteristics
database and IGBP discover from 1-km AVHRR data. International Journal of Remote Sensing 21(6/7): 1-303–1-330.
Magaña V, Amador JA, Medina S. 1999. The midsummer drought over Mexico and Central America. Journal of Climate 12: 1577–1588.
Rogers JC. 1988. Precipitation variability over the Caribbean and Tropical Americas associated with the Southern oscillation. Journal
of Climate 1: 172–182.
Sader SA, Hayes DJ, Irwin DE, Saatchi SS. 2001. Preliminary forest cover estimates for Central America (1990’s) with reference to the
proposed Mesoamerican Biological Corridor. In American Society for Photogrammetry and Remote Sensing (ASPRS), 2001 Annual
Meeting, St. Louis, Missouri, April 23–26. ASPRS , Bethesda, MD, CD-ROM.
Snyder P, Foley JA, Hitchman MH, Delire C. 2004. Analyzing the effects of complete tropical forest removal on the regional
climate using a detailed three-dimensional energy budget: An application to Africa. Journal of Geophysical Research 109: D21102,
DOI:10.1029/2003JD004462.
Taylor MA, Enfield DB, Chen AA. 2002. Influence of Tropical Atlantic versus the tropical Pacific on Caribbean rainfall. Journal of
Geophysical Research 107: C9–3127, DOI: 10.1029/2001JC001097.
Yongjiu D, Zeng X, Dickinson RE, Baker I, Bonan GB, Bosilovich MC, Denning AS, Dirmeyer PA, Houser PR, Niu G, Oleson KW,
Schlosser CA, Yang Z-L. 2003. The common land model. American Meteorological Society 84: 101 301 023, DOI: 10.1175/BAMS-
84-8-1013.
Zhong HJ, Bian X, Charney J, Heilman W, Potter B. 2004. Evaluation of Real-Time High Resolution MM5 Predictions Over the Great
Lakes Region. Presentation at MM5 Modeling Workshop: Boulder, CO, February 2004.

Copyright  2006 Royal Meteorological Society Int. J. Climatol. 26: 2161–2179 (2006)
DOI: 10.1002/joc

You might also like