Abstract

The ability of a mesoscale atmospheric model to reproduce the spatial distribution of the precipitation of the Caribbean island of Puerto Rico during an early rainfall season month (April) is evaluated in this paper, taking the month of April 1998 as the primary test case, and analyzed in detail with subsequent simulations for April 1993. The monthly accumulated rainfall was simulated using the Regional Atmospheric Modeling System (RAMS), and the results were validated with precipitation data from 15 cooperative stations located throughout the island. The monthlong numerical simulation for April 1998 replicated the observed precipitation pattern, including the general spatial distribution, and daily and monthly totals, to varying degrees of accuracy. At specific locations, errors ranged from 2% in the rainy mountains to 82% in the San Juan metropolitan area, with a general tendency of the model to produce lower precipitation values throughout the simulation domain. An error analysis proved that the accuracy of the simulation is independent of elevation. The station data showed two dominant precipitation events during the month of April 1998: one on 2 April and the other on 16 April. The model was able to replicate the precipitation observed during the first precipitation event with less precision than for the second event. This might be attributed to the model’s inability to capture the large-scale forcing that produced the recorded amounts of rainfall observed during the second precipitation event. The results for total accumulated precipitation for April 1993 were very similar to the results for the April 1998 simulation, for both the spatial distribution and total values of rainfall.

1. Introduction

Climate patterns over the tropical islands of the Caribbean are mostly associated with global-scale oscillations, synoptic-scale easterly trade winds, orographic effects, ocean-induced circulations, and convective-scale instabilities related to intense surface heating (Malmgren et al. 1998; Taylor et al. 2002). There is a need to improve our understanding of local- and synoptic-scale contributions to precipitation on tropical islands for practical and scientific purposes. A better understanding and improved prediction capability of the daily and monthly variability of precipitation may also help to assess the possible impacts of climate and environmental changes in the Caribbean basin. These changes could occur at global scales, like greenhouse gas emissions and global warming, and at regional and local scales, such as land cover and land use changes for agriculture and urbanization, deforestation, etc. Recent research has revealed that global climate changes due to increases in greenhouse gases and local land use changes may be influencing the Caribbean climate as evidenced by, for example, higher temperature, precipitation anomalies, wind pattern disruptions, and hydrological cycle and water budget imbalances (e.g., Velazquez-Lozada et al. 2006).

Recent efforts have been made to understand the regional climate of the Caribbean region and its two-way coupling to the global climate system (Enfield and Alfaro 1999; Gianinni et al. 2000; Taylor et al. 2002). Most of the reported studies relied upon coarse-resolution datasets and output from general circulation models (GCMs) of the atmosphere and, in some cases, of the oceans (Polcher and Laval 1994; Blake et al. 1998; Chen and Taylor 2002). Because of their coarse resolution, GCM output and standard archived global observational atmospheric and ocean data need to be downscaled to the desired finer horizontal resolution using semiempirical (Kidson and Thompson 1998) or numerical approaches using mesoscale atmospheric models (e.g., Kao and Bossert 1992; Pielke et al. 1999). A first step toward our ultimate goal of studying and understanding the regional–local impact of climate and environmental change over the Caribbean is to investigate the ability of a mesoscale atmospheric model to reproduce the spatial and temporal pattern of different atmospheric variables at the appropriate resolution in the region of interest. The work presented in this paper focuses on the objective of assessing the ability of a regional atmospheric model to reproduce the observed monthly accumulated precipitation in the Caribbean island of Puerto Rico during the month of April, which falls in the middle of the Caribbean early rainfall season. As a result of this first analysis, the ability of the mesoscale atmospheric model to replicate precipitation events on the island, induced by both local convective- and synoptic-scale forcing, is also investigated.

The most important features of the Caribbean topography are presented in Fig. 1. The highest peaks are shown in the island of La Hispaniola, in the center of the domain, where the mountains are as high as 3125 m in the Central Mountain Range, with lowlands on the coastlines and on the east of the island. Cuba, the largest of the Greater Antilles and located to the northwest of the basin, is mostly flat with limited elevations on the south side of the island. The main features of the island of Puerto Rico, the focus of this work, are represented in Fig. 2, and include the Central Mountain Range and the Luquillo Mountains, the easternmost topography element located at approximately 18°15′N and 65°45′W and the only tropical rain forest administered by the U.S. Forest Service. This area is also referred to as El Yunque, as it will be called hereafter. These natural features, along with the main urban centers of Puerto Rico, are indicated in Fig. 3 for reference.

Fig. 1.

Model domain used at a horizontal resolution of 25 km showing the topography ingested as part of the atmospheric model bottom boundary for grid 1 [contour interval (CI) = 150 m].

Fig. 1.

Model domain used at a horizontal resolution of 25 km showing the topography ingested as part of the atmospheric model bottom boundary for grid 1 [contour interval (CI) = 150 m].

Fig. 2.

Model domain used at a horizontal resolution of 5 km showing the topography ingested for grid 2 (CI = 75 m). The location of the stations selected for this study are presented in their geographical location and identified by the number assigned by the authors (see Table 1).

Fig. 2.

Model domain used at a horizontal resolution of 5 km showing the topography ingested for grid 2 (CI = 75 m). The location of the stations selected for this study are presented in their geographical location and identified by the number assigned by the authors (see Table 1).

Fig. 3.

Mean 1963–95 precipitation in April, from Daly et al. (2003).

Fig. 3.

Mean 1963–95 precipitation in April, from Daly et al. (2003).

An analysis of the climatology of Puerto Rico indicates that two distinct seasons characterize the yearly precipitation pattern in the island, the early and late rainfall seasons (Malmgren and Winter 1999; Daly et al. 2003), separated by what is referred to as the “midsummer drought.” The rest of the Caribbean also follows this bimodal trend (Chen and Taylor 2002). Figure 4 shows the bimodal nature of the Puerto Rico rainfall by presenting the 30-yr precipitation climatology and the 1993 and 1998 monthly precipitation totals recorded by 15 cooperative stations selected for this study. The early season covers the first 6 months of the year (from January through June), with its precipitation peak in May, and it is characterized by convective rainfall, differential heating and local transport of moisture. The late season begins after the summer months and peaks in October. Included in the late season is the hurricane season, occurring between June and November. The precipitation maximum in October occurs mostly due to enhanced low-level convergence to the east of the Lesser Antilles islands, low vertical wind shear, high sea surface temperatures, and greater amounts of deep layer moisture across the tropical North Atlantic, which is advected to the Caribbean area by the easterly trade winds (Taylor et al. 2002).

Fig. 4.

The 1993 and 1998 monthly precipitation totals and 30-yr precipitation averages for the 15 cooperative stations distributed throughout Puerto Rico.

Fig. 4.

The 1993 and 1998 monthly precipitation totals and 30-yr precipitation averages for the 15 cooperative stations distributed throughout Puerto Rico.

Malmgren and Winter (1999) identified climate zones in Puerto Rico using a principal component analysis and an artificial neural network. These authors analyzed climate data, seasonal averages of precipitation, and maximum, mean, and minimum temperatures for the period 1960–90, using data from 18 stations located over the island to determine the existence of climate zones in Puerto Rico. Four principal climate zones were identified as the interior Central Mountains, the northern and southern Coastal Plains, and the western coastline. The area of El Yunque located in the easternmost region of the island falls into the same climate zone as the Central Mountains. The 18 stations selected for the study presented in Malmgren and Winter (1999) are part of a network of 78 cooperative stations located through the island. Daly et al. (2003) mapped the mean 1963–95 precipitation in April for Puerto Rico using cooperative station data; there the authors found a strong correlation between the climatological precipitation of Puerto Rico with the island’s topography (see Fig. 3). The 15 stations used in the study presented in this paper were selected from the same network (please refer to Table 1 for information on the stations). The stations were selected on the basis of containing consistent information for the period of interest and of representing the different climatic zones suggested for Puerto Rico (Malmgren and Winter 1999).

Table 1.

Station identification and location.

Station identification and location.
Station identification and location.

The results presented here were obtained by carrying out monthlong simulations for the month of April with the purpose of analyzing the regional model performance on the production of precipitation over Puerto Rico during the early rainfall season. April 1998 was selected as the basic case to test the regional model precipitation production because 1998 was the year when record high temperatures were recorded on the island, and recorded precipitation totals were similar to the 30-yr climatology for April, even though it was an El Niño + 1 yr. Analysis of the Caribbean early rainfall season data reveals a teleconnection accounting for almost half of the season’s precipitation variability, resulting in a wetter period one to two seasons after the equatorial Pacific anomalies referred to as El Niño, which are strongly linked to positive spring north tropical Atlantic sea surface temperature anomalies (Chen and Taylor 2002). Additional simulations were performed for the month of April 1993, which recorded higher amounts of precipitation compared to the climatology and 1998.

This paper has been divided into five sections. Section 2 includes a description of the mesoscale model and configuration. The different observations used for the study, namely, the environmental setting of the two largest precipitation events and the station data, are presented in section 3. Discussion and validation of the simulation results are given in section 4, and section 5 offers a summary of the work conducted.

2. Model description and configuration

The Regional Atmospheric Modeling System (RAMS) is a highly versatile numerical code developed for simulating and forecasting meteorological phenomena; RAMS 4.3 was used in this research (Pielke et al. 1992; Cotton et al. 2003; Walko et al. 1995b; Walko and Tremback 1995). The atmospheric model is built around the full set of nonhydrostatic, dynamical equations that governs atmospheric dynamics and thermodynamics, plus conservation equations for scalar quantities such as mass, water vapor, liquid, and ice hydrometeor mixing ratios. These equations are complemented by a large selection of parameterizations available in the model.

The research was conducted with two grids making use of the two-way interactive grid nesting capability of the mesoscale model (Walko et al. 1995b; Clark and Hall 1991). Grid 1 covers the Caribbean basin (∼25°–12°N latitude and 83°–59°W longitude) at a horizontal resolution of 25 km. Grid 2, which is nested within grid 1, covers the island of Puerto Rico at 5-km horizontal resolution. The model grids with the model topography are shown in Figs. 1 and 2. For the vertical coordinate, both grids have the same specification. A grid spacing of 100 m was used near the surface, stretched at a constant ratio of 1.1 until ΔZ reached 1000 m. The model depth is 22.83 km with 40 vertical layers. Another grid configuration was set to test the sensitivity of the model results to the horizontal resolution ratio between consecutive grids. This test configuration used three grids with the coarser grid having a horizontal resolution of 50 km, with the intention of keeping a 5:1 ratio. Results from the two simulations did not show substantial differences (less than 1% for a 10-day simulation period). Because these two configurations gave similar, reasonable output, we chose the computationally cheaper two-grid configuration over the three-grid configuration.

Time-dependant boundary conditions were updated every 12 h. A 12-h frequency should be suitable for an ocean-dominated region like the Caribbean basin, for which the diurnal cycle will be relatively small, versus a land-dominated region. Strong nudging (900s user-specified damping time scale) was performed at the lateral boundaries, decreasing parabolically inward with the parabola vertex located five grid points into the domain. A very weak nudging (21 600-s relaxation time scale) was performed in the interior of the entire domain with no damping specified for the top of the domain. The manipulation of the interior nudging time scale offers the potential for synoptic conditions to be simulated closely to the observed large-scale atmospheric dataset used to drive the model. For weather forecasting, the data processing in RAMS for boundary and initial conditions can include assimilation of surface and rawinsonde observations, but since the focus of our work is to develop climate prediction capabilities these were not included at the data assimilation stage of the simulations.

Cumulus parameterization is turned on for grid 1 only; a simplified Kuo convective parameterization is the standard scheme used in RAMS (Kuo 1974). On grid 2 an explicit cloud microphysics scheme is used. The microphysics moisture complexity was set to the highest level. This level incorporates all categories of water in the atmosphere (cloud water, rainwater, pristine ice crystals, snow, aggregates, graupel, and hail); the module includes the precipitation process (Walko et al. 1995a). The cloud water concentration was set to 40 drops cm−3 and the mean diameter of the raindrop spectra was set to 1000 μm following Rogers and Yau (1989), where it is stated that maritime clouds have small concentrations of large drops and a broad size spectrum. Initial and time-dependant lateral boundary conditions were given by the National Centers for Environmental Prediction (NCEP) reanalysis fields (Kalnay et al. 1996). Initial volumetric soil moisture content was specified constant at 35% for 10 soil layers 0.5 m deep, and it spins up to a run specific profile during the simulation.

3. Observations

The two main observational resources used for this study consist of the Cooperative Station Network (COOP) and the environmental setting of the two largest precipitation events identified by the station data as depicted by synoptic-scale fields derived from NCEP data.

a. Cooperative station data

The daily and monthly precipitation totals from a network of stations located throughout Puerto Rico were obtained from the Southeast Regional Climate Center and the Puerto Rico and U.S. Virgin Islands Climate Office. From the network of over 60 stations that collected data for 1998, 15 stations were selected for the analysis on the basis of the completeness of the data recorded and station location. These 15 stations are located along the coastline and in the interior of the island and represent all the climatic zones suggested by Malmgren and Winter (1999). Figure 2 shows the distribution of the 15 stations selected and Table 1 gives their identification, latitude–longitude location, and elevation.

Figure 5 shows the daily precipitation totals for April 1998 and 1993 of the 15 COOP stations, calculated as the sum over all stations (for the daily average over the 15 stations, the reader is referred to Fig. 12). From Fig. 5, several important precipitation events could be identified during the month based on the daily precipitation total. The two main precipitation events, however, occurred during the 2nd (167.56 mm) and between the 16th (278.38 mm) and 17th (183.38 mm) of April 1998. The later one is considered the most dominant event. Station numbers 2, 6, 11, and 15, located along the north coast, and stations 13 and 14, located in El Yunque, showed significant precipitation for 16 April 1998. The westernmost stations recorded an average of 30 mm for that day, while the east coast and Central Mountain stations had an average of 48 mm. For 17 April 1998, stations 9 and 14 showed precipitation on the order of 36 mm, and station 3 recorded 30 mm of rainfall, showing that the event was islandwide.

Fig. 5.

Daily precipitation totals recorded by the 15 stations selected for the study for April 1998 and 1993.

Fig. 5.

Daily precipitation totals recorded by the 15 stations selected for the study for April 1998 and 1993.

Fig. 12.

(left) Station and model daily precipitation, averaged over the 15 locations shown in Fig. 2 and the closest grid points to each station on the 5-km grid. (right) The difference between the model and the stations. (top) April 1998 and (bottom) April 1993.

Fig. 12.

(left) Station and model daily precipitation, averaged over the 15 locations shown in Fig. 2 and the closest grid points to each station on the 5-km grid. (right) The difference between the model and the stations. (top) April 1998 and (bottom) April 1993.

The second dominant precipitation event occurred earlier in the month and at fewer stations. Seven stations showed precipitation for 2 April 1998. Stations 5, 7, and 8 (central-south) showed an average of 28 mm, and station 13 (El Yunque vicinity) received 17 mm of rain. The stations that showed the highest difference between their relative precipitation maxima and minima were stations 12, 13, and 14, which are located in the El Yunque area.

An interesting case occurs at the end of the month of April 1993, namely during 28–30 April 1993 when the highest precipitation totals for both cases presented in this study were recorded. However, individual precipitation events were not examined for April 1993 because they were not island-wide events for that month, a criterion we used to define individual events. The main focus of the work presented here was to simulate the monthly total and average rainfall in Puerto Rico; future analysis of individual convective precipitation events could be useful.

The monthly precipitation totals recorded by the stations were interpolated to a regular 5-km grid, plotted on a map of Puerto Rico and presented in Fig. 6 for clarification and for easy viewing of the monthly observed precipitation spatial distribution for April 1998. A bilinear interpolation was used; this method fits a bilinear surface through existing data points where the value of an interpolated point is a combination of the values of the four closest points. Here it is clear that the two precipitation maxima fall on the two most prominent topographic features, namely, the Central Mountains and El Yunque, and have a southern shift when compared to the precipitation climatology shown in Fig. 3. On an individual station basis the highest recorded accumulated precipitation for April 1998 was at Toro Negro Forest in the Central Mountains (station 8), which recorded 319.79 mm of rain. Other prominent observations were at station 5 in Adjuntas (195.58 mm), station 13 in Pico del Este (260.35 mm), and station 14 in Paraiso (El Yunque, 180.85 mm). The lowest accumulated precipitation was recorded at station 4 (45.47 mm) and the next driest was station 10 (47.24 mm), located in the southern localities of Ensenada and Guayama, respectively.

Fig. 6.

Monthly precipitation totals for April 1998 as recorded by the 15 stations selected for the study; the values were interpolated to a regular 5-km grid and plotted on a map of Puerto Rico.

Fig. 6.

Monthly precipitation totals for April 1998 as recorded by the 15 stations selected for the study; the values were interpolated to a regular 5-km grid and plotted on a map of Puerto Rico.

b. Synoptic setting of main precipitation events for April 1998

It is necessary to analyze the synoptic events that contributed to the rainfalls events of the test cases considered. The 1200 UTC 2 April 1998 synoptic pattern shows a surface low east of the Caribbean islands, with a weak temperature gradient prevailing over the region at this level (Fig. 7a). The 850-mb map (Fig. 7b) shows a cyclonic vortex to the northeast of Puerto Rico, with cold-air advection approaching the island of Puerto Rico from the north combined with a southerly warm and moist airflow coming from the southern region of the Caribbean. A very strong 300-mb low pressure sits to the northeast of the Caribbean, almost in the same position as the low-level vortices, with a well-defined trough (Fig. 7c). This indicated the likely presence of an enhanced convergence at the low levels over, or slightly south of, Puerto Rico, which is further enhanced by the large-scale negatively tilted trough to the northeast of the Caribbean.

Fig. 7.

Synoptic surface and upper-air analyses for 1200 UTC 2 Apr 1998: (a) sea level pressure (solid lines) and temperature (dashed lines), CI = 1 mb and 2°C, respectively; (b) 850-mb analysis of geopotential height (solid) CI = 20 m, temperature (dashed) CI = 2°C, and relative humidity (solid-thick) CI = 15%; and (c) 300-mb analysis with geopotential height CI = 40 m; the dashed line indicates the high-level trough axis.

Fig. 7.

Synoptic surface and upper-air analyses for 1200 UTC 2 Apr 1998: (a) sea level pressure (solid lines) and temperature (dashed lines), CI = 1 mb and 2°C, respectively; (b) 850-mb analysis of geopotential height (solid) CI = 20 m, temperature (dashed) CI = 2°C, and relative humidity (solid-thick) CI = 15%; and (c) 300-mb analysis with geopotential height CI = 40 m; the dashed line indicates the high-level trough axis.

The 16 April 1998 situation is quite different. Sea level pressure plots at 0600, 1200, and 1800 UTC (Fig. 8) show a weak tropical wave passing right over the island of Puerto Rico, moving slightly northward. This setting can be noticed when looking at the undulation of the 1540-m contour line in the 850-mb analysis (Fig. 9a). The tropical wave was interacting with a major upper-level trough, enhancing air parcel instability over the area (Fig. 9b). This is the day of the month when the most precipitation was observed over the island of Puerto Rico (∼278 mm of rain; see Fig. 5).

Fig. 8.

Synoptic surface analyses for 16 Apr 1998 at (a) 0600, (b) 1200, and (c) 1800 UTC. Sea level pressure (solid lines) and temperature (dashed lines) CI = 1 mb and 2°C, respectively.

Fig. 8.

Synoptic surface analyses for 16 Apr 1998 at (a) 0600, (b) 1200, and (c) 1800 UTC. Sea level pressure (solid lines) and temperature (dashed lines) CI = 1 mb and 2°C, respectively.

Fig. 9.

Synoptic upper-air analyses for 1200 UTC 16 Apr 1998: (a) 850-mb analysis of geopotential height (solid) CI = 20 m, temperature (dashed) CI = 2°C, and relative humidity (solid-thick) CI = 15%; (b) 300-mb analysis with geopotential height CI = 40 m; the dashed line indicates the upper-level trough axis.

Fig. 9.

Synoptic upper-air analyses for 1200 UTC 16 Apr 1998: (a) 850-mb analysis of geopotential height (solid) CI = 20 m, temperature (dashed) CI = 2°C, and relative humidity (solid-thick) CI = 15%; (b) 300-mb analysis with geopotential height CI = 40 m; the dashed line indicates the upper-level trough axis.

4. Model results

a. April 1998

The model was initialized at 0000 UTC 27 March and integrated for 35 days until 0000 UTC 1 May for the April simulations. All the results presented and the analysis refers to the 5-km grid covering Puerto Rico, unless otherwise noted. The spatial distribution of the accumulated total precipitation for April 1998 is shown in Fig. 10, an indication of a rainfall pattern dominated by orographic lifting. The model was able to capture the general regional character of the precipitation identified by the 15 stations, with an overall tendency of the model to simulate lower precipitation in the area of interest when compared to observations. The general pattern is for higher precipitation over the Central Mountains and El Yunque, followed by the northern and western coastlines, and finally the south, which is the driest part of the island climatologically. This pattern is similar to the precipitation mapping performed by Daly et al. (2003), where a rate of increase of precipitation with elevation of approximately 140% of the 30-yr mean per kilometer of elevation was reported (see Fig. 3). It is noted that the model precipitation maxima falls on the Central Mountains ridge, and the climatology in Fig. 3 has a bias for maximum precipitation in the northern flanks of the mountain range. The model produced an axis of flow convergence that follows the ridge of the Central Mountain Range linking precipitation to moisture flow over the topography (see Fig. 11a), and therefore was more inclined to produce precipitation in that region. Since precipitation was produced by individual events, the flow pattern at the lowest model atmospheric level for 16 April 1998 is shown in Fig. 11b. Here we can see a more northerly pattern of the wind pattern, transporting the available moisture up the mountain slopes, further describing the southward shift of the simulation precipitation maxima.

Fig. 10.

Model accumulated total precipitation (mm) for April 1998 (CI = 20 mm).

Fig. 10.

Model accumulated total precipitation (mm) for April 1998 (CI = 20 mm).

Fig. 11.

Model wind vector field at the lowest atmospheric σ level, time averaged for the (top) complete April 1998 simulation and (bottom) the 16 Apr 1998 day, showing the flow of moisture (thick lines represent model relative humidity contours, CI = 3%, averaged for the time period of each panel) over topography (shaded contours as in Fig. 2).

Fig. 11.

Model wind vector field at the lowest atmospheric σ level, time averaged for the (top) complete April 1998 simulation and (bottom) the 16 Apr 1998 day, showing the flow of moisture (thick lines represent model relative humidity contours, CI = 3%, averaged for the time period of each panel) over topography (shaded contours as in Fig. 2).

Table 2 shows a comparison between the monthly accumulated precipitation observed by the 15 stations selected for the study and the model results for the grid cell closest to each station. Model simulations agree best with observations over the Central Mountains (points near stations 8 and 12) and El Yunque vicinity (points near stations 13, 14, and 15). The highest simulated precipitation total was at station 13, near Pico del Este, with 268 mm of rainfall. It was also the simulated point closest to the observation, departing from the station record by 2.9%. The points near the Roosevelt Roads and Paraiso stations, 14 and 15 respectively, simulated 147 and 110 mm each, and show absolute errors of 16.4% and 16.7%, respectively. The second closest point to the observed record was located near the San Lorenzo station (number 12) with an absolute error of 7.3% and 108 mm of simulated rainfall. The model was able to perform satisfactorily in general in the four climate zones existing within the domain as identified by Malmgren and Winter (1999).

Table 2.

Comparison of station data vs model results for April 1998 and April 1993.

Comparison of station data vs model results for April 1998 and April 1993.
Comparison of station data vs model results for April 1998 and April 1993.

The point near the San Juan station, located in the northeast coast, deviated the most from its corresponding station (close to 82% below the corresponding station). The reasons that may have contributed to this low precipitation versus observation are that at the current configuration and horizontal resolution, the model might not be capturing the complexity of the San Juan Metropolitan Area, the largest urban center of the island that generates a strong urban heat island with its subsequent mesoscale and local circulations (González et al. 2005; Velazquez-Lozada et al. 2006). However, a closer look at the difference between the COOP stations and the simulation results (Fig. 12) shows that most of the error for the majority of the stations across the island occurs as the result of lower simulated precipitation values during the 16 April 1998 rainfall event, the strongest of the month. We assume that a better replication of the large-scale fields across the Caribbean area during that day should have yielded better overall monthly results.

Figure 12 shows the daily observed and model precipitation averages, where the variable was averaged over the 15 locations seen in Fig. 2. The difference between the model and observations (model − observed) is also presented. The model reproduced the general daily accumulated precipitation pattern across the island of Puerto Rico for April 1998 (Fig. 12), with the only major discrepancy being 16 April, as stated previously. Over the month, the model averages 1.06 mm of rain per day below the observed precipitation, with a deviation from observed values of 5 mm day−1. The precipitation simulation for most of the days appears in temporal agreement with the station data. This is the case for the crests that occurred during 2, 11, and 20 April 1998 (see Fig. 12) whereas other days lagged for one day, possibly related to the weak nudging time scale used in the interior of the domain. To give a better idea of the behavior of the regional atmospheric model in regard to accuracy and the topographic influence, a scatterplot of errors calculated for both April runs (1998 and 1993) against elevation is presented in Fig. 13. Here we can see that for both simulations there was a wide range of errors for low elevation, and the errors seems to be decreasing with elevation, although the number of stations shown in Fig. 13 are not enough to state this in a conclusive manner. Given the small amount of reliable and complete stations and observational sources, a horizontal map of absolute errors for the case of April 1998 is presented in Fig. 14, which shows some interesting features.

Fig. 13.

Comparison of the error between station observations and model-simulated precipitation plotted against elevation for the runs April 1998 and April 1993.

Fig. 13.

Comparison of the error between station observations and model-simulated precipitation plotted against elevation for the runs April 1998 and April 1993.

Fig. 14.

Horizontal distribution of absolute errors calculated for the April 1998 accumulated precipitation totals. The error for each of the 15 stations was interpolated to a 5-km grid.

Fig. 14.

Horizontal distribution of absolute errors calculated for the April 1998 accumulated precipitation totals. The error for each of the 15 stations was interpolated to a 5-km grid.

As mentioned before, the model was observed to perform better along regions of elevated terrain and high amounts of recorded precipitation. It was also noted that the point near San Juan was the least accurate, as shown in Fig. 14. But other spots of poor performance also arise from Fig. 14, like the one seen in the western coast around the station of Mayagüez (station number 1). The western coast of Puerto Rico is subject to a classical sea–land-breeze circulation that produces heavy and short-lived rainfall in the early afternoon local time. At 5-km horizontal resolution, the model is unable to produce this sea breeze pattern and its associated convection and cloud formation, and is relying more on the upslope flow in the northern flank of the Central Mountains to produce precipitation by generating the orographic lifting and moisture transport in that region.

Regarding the precipitation events identified by the stations, the model was able to replicate the first major precipitation event as it reproduced the same synoptic features seen in the observations (Figs. 6 and 15). For the case of 16 April 1998, the model did not reproduce the situation described in section 3b. The tropical wave crossing over Puerto Rico is much weaker in the model results for the same time periods (Fig. 16), although the upper-air pattern (i.e., the 300-mb trough; not shown) is shown clearly by the model. This is likely the reason for the lower values of the modeled precipitation when compared with the amount recorded by the COOP stations for that particular day. Some of the factors that may have contributed to this model behavior are the use of strong nudging near the boundaries but weak nudging in the interior of the domain only every 12 h (believed to be causing the apparent 1-day lag noted above). The analysis presented in this section should not be viewed as an error analysis because one should not expect the model to reproduce all the details of the observed flow for long runs with the relaxation time scale used in the interior of the domain. For the nudging specification used it would be difficult for the model to exactly replicate the large-scale conditions observed. Also the use of the Kuo convective scheme in a two-way interactive grid configuration might be undersimulating precipitation in the parent grid and contaminating precipitation in the finer grid.

Fig. 15.

Model surface and upper-air analyses for 1200 UTC 2 Apr 1998: (a) sea level pressure (solid lines) and temperature (dashed lines), CI = 1 mb and 2°C, respectively; (b) 850-mb analysis of geopotential height (solid) CI = 20 m, temperature (dashed) CI = 2°C, and relative humidity (solid-thick) CI = 15% intervals; and (c) 300-mb analysis with geopotential height CI = 40 m.

Fig. 15.

Model surface and upper-air analyses for 1200 UTC 2 Apr 1998: (a) sea level pressure (solid lines) and temperature (dashed lines), CI = 1 mb and 2°C, respectively; (b) 850-mb analysis of geopotential height (solid) CI = 20 m, temperature (dashed) CI = 2°C, and relative humidity (solid-thick) CI = 15% intervals; and (c) 300-mb analysis with geopotential height CI = 40 m.

Fig. 16.

Model-simulated surface analyses for 16 Apr 1998 at (a) 0600, (b) 1200, and (c) 1800 UTC. Sea level pressure (solid lines) and temperature (dashed lines), CI = 1 mb and 2°C, respectively.

Fig. 16.

Model-simulated surface analyses for 16 Apr 1998 at (a) 0600, (b) 1200, and (c) 1800 UTC. Sea level pressure (solid lines) and temperature (dashed lines), CI = 1 mb and 2°C, respectively.

b. April 1993

To further investigate the ability of the mesoscale model used to simulate precipitation in Caribbean islands during the early rainfall season, one additional run was configured and conducted for the month of April 1993. A comparison of the results for April 1993 and its corresponding station values is presented in Table 2 and Fig. 12.

Here it is clearly seen that for April 1993 the model produced the same spatial pattern of high precipitation totals in areas of elevated terrain, and relatively low precipitation in the southern coastal region. The results agreed fairly well with the observed accumulated totals as recorded by the station network but with a wide range of errors when the comparison was performed. The error pattern, however, was strikingly similar in its spatial distribution with the errors calculated for the April 1998 case. It can be noticed that the 1993 case did not present a major precipitation event as the 1998 case, and that the daily pattern was less variable for April 1993 than for April 1998. This produced daily precipitation simulation for 1993 that agreed better with observations. This reinforces the hypothesis of the mesoscale model performing better for locally produced precipitation, induced primarily by orographic lifting and differential heating, than for large-scale synoptic induced precipitation events.

5. Summary and conclusions

A regional atmospheric model was used as the main research tool to simulate the monthly precipitation pattern of the island of Puerto Rico during the Caribbean early rainfall season. Results from the RAMS model, the numerical code selected for this study, were validated with the recorded precipitation values from a network of cooperative stations located throughout the island. The station data identified two dominant precipitation events during April 1998, namely, on 2 and 16 April 1998. The same stations were used to collect the precipitation information for April 1993. The station data also show that the local topography has a strong influence in the observed monthly precipitation pattern across the island.

The regional atmospheric model reproduced, to varying degrees of success, the total amounts of rainfall observed for most of the COOP stations for April 1998 and April 1993. The model-resolved precipitation pattern over Puerto Rico appears to be influenced by the island topography and presence of urban areas. This tendency is due primarily to orographic lifting of the easterly flow. As the topography becomes steeper the updraft becomes stronger and helps induce strong convection. The end result is more frequent rainfall occurrence on the ridge and crest of topographic features, as in the case over the Central Mountains and El Yunque. In the case of the urban areas, the model might not be producing the enhanced convection due to an increase in sensible heating in these areas; it may also be due to the model’s inability to capture complex urban landscapes and coastal moisture effects. The model also showed satisfactory results in capturing the precipitation event that occurred on 2 April 1998. However, it could not replicate the second and strongest event in the same month (16 April 1998). This likely caused the majority of simulated precipitation errors during the month for some stations. This work offered a look at the difficulties of predicting precipitation, particularly over a tropical island with complex terrain.

Acknowledgments

This research was conducted with the support of the NASA-EPSCoR program of the University of Puerto Rico. The atmospheric model simulations were performed at the High Performance Computing Facilities in Rio Piedras. The COOP station precipitation data were obtained from the Puerto Rico and U.S. Virgin Islands Climate Office. The authors thank J. Stalker, A. Velázquez, and P. Mulero for their assistance and discussion on the modeling effort and data analysis.

REFERENCES

REFERENCES
Blake
,
R.
,
R.
Khanbilvardi
,
L.
Errand
,
D.
Rind
, and
C.
Rosenzweig
,
1998
:
Assessing climate change impacts in the Rio Cobre basin, Jamaica.
Hydrology in the Human Tropic Environment, A. I. Johnson and C. A. Fernández-Jáurgui, Eds., IAHS Publication 253, 185–193
.
Chen
,
A. A.
, and
M. A.
Taylor
,
2002
:
Investigating the link between early Caribbean rainfall and the El Niño+1 year.
Int. J. Climatol.
,
22
,
87
106
.
Clark
,
T. L.
, and
W. D.
Hall
,
1991
:
Multi-domain simulations of the time dependant Navier-Stokes equations: Benchmark error analysis of some nesting procedures.
J. Comput. Phys.
,
92
,
456
481
.
Cotton
,
W. R.
, and
Coauthors
,
2003
:
RAMS 2001: Current status and future directions.
Meteor. Atmos. Phys.
,
82
,
5
29
.
Daly
,
C.
,
E. H.
Helmer
, and
M.
Quiñones
,
2003
:
Mapping the climate of Puerto Rico, Vieques, and Culebra.
Int. J. Climatol.
,
23
,
1359
1381
.
Enfield
,
D. B.
, and
E. J.
Alfaro
,
1999
:
The dependence of Caribbean rainfall on the interaction of the tropical Atlantic and Pacific Oceans.
J. Climate
,
12
,
2093
2103
.
Gianinni
,
A.
,
Y.
Kushnir
, and
M. A.
Cane
,
2000
:
Interannual variability of Caribbean rainfall, ENSO, and the Atlantic Ocean.
J. Climate
,
13
,
297
311
.
González
,
J. E.
, and
Coauthors
,
2005
:
Urban heat islands developing in coastal tropical cities.
Eos, Trans. Amer. Geophys. Union
,
86
,
397
403
.
Kalnay
,
E.
, and
Coauthors
,
1996
:
The NCEP/NCAR 40-Year Reanalysis Project.
Bull. Amer. Meteor. Soc.
,
77
,
437
471
.
Kao
,
C. Y.
, and
J. E.
Bossert
,
1992
:
Regional climate simulations: A GCM/mesoscale model interface.
Trends Atmos. Sci.
,
1
,
309
323
.
Kidson
,
J. W.
, and
C. S.
Thompson
,
1998
:
A comparison of statistical and model-based downscaling techniques for estimating local climate variations.
J. Climate
,
11
,
735
753
.
Kuo
,
H. L.
,
1974
:
Further studies of the parameterization of the influence of cumulus convection on large-scale flow.
J. Atmos. Sci.
,
31
,
1232
1240
.
Malmgren
,
B. A.
, and
A.
Winter
,
1999
:
Climate zonation in Puerto Rico based on principal components analysis and an artificial neural network.
J. Climate
,
12
,
977
985
.
Malmgren
,
B. A.
,
A.
Winter
, and
D.
Chen
,
1998
:
El Niño–Southern Oscillation and North Atlantic Oscillation control of climate in Puerto Rico.
J. Climate
,
11
,
2713
2717
.
Pielke
,
R. A.
, and
Coauthors
,
1992
:
A comprehensive meteorological modeling system—RAMS.
Meteor. Atmos. Phys.
,
49
,
69
91
.
Pielke
,
R. A.
,
R. L.
Walko
,
L. T.
Steyaert
,
P. L.
Vidale
,
G. E.
Liston
,
W. A.
Lyons
, and
T. N.
Chase
,
1999
:
The influence of anthropogenic landscape changes on weather in south Florida.
Mon. Wea. Rev.
,
127
,
1663
1673
.
Polcher
,
J.
, and
K.
Laval
,
1994
:
The impact of African and Amazonian deforestation on tropical climate.
J. Hydrol.
,
155
,
389
405
.
Rogers
,
R. R.
, and
M. K.
Yau
,
1989
:
A Short Course in Cloud Physics.
3rd ed. Butterworth-Heinemann, 290 pp
.
Taylor
,
M. A.
,
D. B.
Enfield
, and
A. A.
Chen
,
2002
:
Influence of the tropical Atlantic versus the tropical Pacific on Caribbean rainfall.
J. Geophys. Res.
,
107
.
3127, doi:10.1029/2001JC001097
.
Velazquez-Lozada
,
A.
,
J. E.
Gonzalez
, and
A.
Winter
,
2006
:
Urban heat island effect analysis in San Juan, Puerto Rico.
Atmos. Environ.
,
40
,
1731
1741
.
Walko
,
R. L.
, and
C. J.
Tremback
,
1995
:
RAMS, The Regional Atmospheric Modeling System 4.3/4.4: Model input namelist parameters. Atmospheric, Meteorological, and Environmental Technologies, 50 pp. [Available online at http://www.atmet.com/html/docs/rams/ug44-mod-namelists.pdf.]
.
Walko
,
R. L.
,
W. R.
Cotton
,
J. L.
Harrington
, and
M. P.
Meyers
,
1995a
:
New RAMS cloud microphysics parameterization. Part I: The single-moment scheme.
Atmos. Res.
,
38
,
29
62
.
Walko
,
R. L.
,
C. J.
Tremback
, and
F. A.
Herstenstein
,
1995b
:
RAMS: The regional atmospheric modeling system: Technical description. Atmospheric, Meteorological, and Environmental Technologies, 50 pp. [Available online at http://www.atmet.com/html/docs/rams/rams_techman.pdf.]
.

Footnotes

Corresponding author address: D. Comarazamy, Mechanical Engineering Department, University of Puerto Rico, Mayagüez, Puerto Rico. Email: comarazamy@me.uprm.edu