a. Precipitation evolution

Figure 2 compares the observed, modeled, and reanalysis monthly rainfall fields in the NAM region. In the observations, precipitation over land is from the NCEP gauge data, and precipitation over the oceans is from the PERSIANN estimates. The NCEP–National Center for Atmospheric Research (NCAR) Regional Reanalysis (NARR) data are produced with reinitializations every 2.5 days (with an additional 12-h run for spinning up) and also leveraged by frequent three-dimensional observation data assimilation (3DDA) including precipitation and many other physical variables (Mesinger et al. 2005). For the use of 3DDA, the daily NCEP gauge data over land and the (satellite based) Climate Prediction Center (CPC) Merged Analysis of Precipitation pentad data over oceans are disaggregated into hourly values. Figure 2 shows that the reanalysis and the observation monthly rainfall are highly correlated. By ingesting observation data into the modeling, the reanalysis data aim at capturing the regional meteorological and hydrological features of the weather and climate systems. As mentioned above, the MM5 modeling is operated under the conditions of every fifth-day reinitiations and without using observation data assimilation. In general, the model reproduces the monsoon precipitation characteristics in the three summer months over land. However, over the eastern Pacific Ocean, the model generates excessive precipitation, mainly because the use of the Eta PBL scheme in domain 1 that is known to produce unrealistic rainfall predictions in the subtropical oceanic region (see above) and to affect domain 2 through the west boundary. Over land, the model illustrates the evolution characteristics of the monsoon rainfall, matching the summary based on observations (Douglas et al. 1993; Higgins and Shi 2001; Cavazos et al. 2002). In June, rainfall is concentrated on a band from southern Mexico along the coastal line of the Gulf of Mexico to the Great Plains. Meanwhile, the model shows that monsoon rainfall starts growing northward along the western slope of the Sierra Madre Occidental (SMO). Arizona is dry in June. In comparison with the observations, the model underestimates not only the major rainband in northeastern Mexico, Texas, and the Great Plains, but also NAM rainfall over the western coast of Mexico. In July, although the model still underestimates rainfall, it shows that the NAM precipitation regime reaches northwestern Mexico and continues extending northeastward, passing Arizona and New Mexico into the high plains of Colorado and Kansas. In contrast, the rainfall band along the coastal line of the Gulf of Mexico to the Great Plains declines. In August, monsoon precipitation becomes weak along the western coast of Mexico, but continues growing in Arizona. Meanwhile, precipitation over northeastern Mexico, Texas, and the Great Plains regains strength compared with the decline in July.

These monthly precipitation variations are associated with the synoptic progression of circulations over the conterminous United States and Mexico, as described by the diagnostic studies cited above. Figure 3 presents the differences among 200-mb streamlines, 850-mb wind vectors, and precipitation over land in two consecutive months. From June to July (Fig. 3a), the changes in the upper-tropospheric winds (200-mb streamline) show a pattern similar to that derived from the long-term composition of NCEP–NCAR reanalyses (Higgins et al. 1999, their Fig. 8c): a broad cyclone over the north-central United States and east-to-west streamlines over the tropical region, which indicates the monsoon anticyclone’s northward migration from southern Mexico in June to the vicinity of northwestern Mexico by July. At the 850-mb level, the model shows typical NAM low-tropospheric wind variations, including an increase in the tropical westerly, a decrease in the southerly over the Gulf of Mexico and the Great Plains, and a decrease in the northwesterly west of Baja California. Correspondingly, the precipitation pattern illustrates “tripole” interactive relationships: an increase (exceeding 1 mm day⁻¹) over the NAM region, an out-of-phase decrease over the Great Plains/northern tier, and an in-phase increase over the eastern United States.

The circulation and precipitation changes from July to August (Fig. 3b) reveal pronounced reverse patterns in comparison with the June–July variations, which are also similar to the NCEP–NCAR reanalysis compositions (Higgins et al. 1999) and indicate that the NAM grows from maturity to decline. At 200 mb, an anticyclone replaces the cyclone in the subtropical U.S. region, with tropical streamlines running in the opposite direction, indicating “a tendency of the monsoon anticyclone to begin its southward track” (Higgins et al. 1999). The 850-mb winds show the strengthening of the tropical easterly and the Pacific northwesterly (west of Baja California). The Great Plains and the eastern United States experience complicated local wind variations. Surface precipitation variations are correlated with these circulation changes. The monsoon rainfall regime decays except for northwestern Mexico, Arizona, and southern Nevada, where NAM precipitation continues growing. Meanwhile, over northeastern Mexico, Texas, and Great Plains, precipitation strengthens to the west and weakens to the east. Over the eastern United States, precipitation decreases in the northern region but increases in the southern and coastal regions. Over southern Mexico, the RCM predicts increased rainfall in August, in contrast with the decrease observed from the NCEP gauge data (not shown). This positive bias in model precipitation is associated with the synoptic circulation and with the model’s deficiencies in overestimating tropical rainfall. During the NAM’s mature period, the subtropical high (near the plateau) dominates southwestern North America and results in the prevalence of warm southerly winds from the tropical Pacific over southern Mexico. This synoptic pattern predisposes the model to produce rainfall over the region (Li et al. 2004).

In diagnostic studies, the date of monsoon onset is found to be a useful indicator to estimate many important features (strong–weak, wet–dry) of the coming monsoon and winter (Higgins et al. 1997; Higgins and Shi 2000). Based on historic observational data, Higgins et al. (1997, 2000) have defined the monsoon onset date for the southwestern United States as the first day after June 1 when the daily mean precipitation over the Arizona–New Mexico (AZNM) region (32°–36°N and 112.5°–107.5°W; see Fig. 1) exceeds 0.5 mm day⁻¹ for three consecutive days. The time series of daily precipitation over the AZNM region in Fig. 4a shows a dry June, and a wet July and August. The model predicted two rain events during 23–30 June that exceeded 0.5 mm day⁻¹ and lasted for 3 days. However, the observational data show that both rainfall events have much smaller amounts and shorter durations, thereby refuting the premise about the 2004 NAM onset. If we exclude these two rainfall events in June, the model and observational data agree upon 11 July as the onset day. After the monsoon onset, the observational data show that the AZNM region experienced a continuous “wet episode” that included four heavy rain (>1 mm day⁻¹) intervals: 11–19 and 23–29 July, and 2–7 and 13–20 August. The model time series shows the same dry-to-wet transition on 11 June, as well as the four heavy rain intervals in 11–19 and 23–28 July, 31 July–7 August, and 13–17 August. Clearly, although the model predictions show errors in rainfall amounts and timing, they nonetheless capture the critical diagnostic intraseasonal features of precipitation over the AZNM region.

Figure 4 plots the time series of modeled and observed daily rainfall over five 2° × 2° boxes along the southwestern United States and northwestern Mexico (see Fig. 1 for locations). The five boxes are located at

• A (33°–35°N, 108°–110°W)—Mogollon Rim on the Arizona–New Mexico border;

• B (32°–34°N, 111°–113°W)—Sonora Desert in southwestern Arizona;

• C (28°–30°N, 108°–110°W)—Sonora, northwestern Mexico;

• D (24°–26°N, 106°–108°W)—Sinaloa, western Mexico; and

• E (20°–22°N, 102°–104°W)—west-central Mexico.

The observational data on rainfall over boxes A and B (inside the AZNM region) shows that the model generated three unrealistic heavy rainfall events in June in box A, the Mogollon Rim (not in box B, Sonora Desert). The last two events raise the mean rainfall intensities over the AZNM region to greater than 0.5 mm day⁻¹ for 3 days, which could lead to misidentification of the monsoon onset date, as discussed above. This problem is due to the K–F convective parameterization scheme, which tends to predict excessive precipitation over mountains (Wang and Seaman 1997; Liang et al. 2004). In July and August, the model showed four major rainy periods in box A and three in lower-elevation box B that generally matched the observations.

Boxes C and D are located in the NAME intensive field observation area. Box C covers 23 NERN rain gauges from the T5 and T6 transects, and box D includes 12 NERN gauges from the T2 and T3 transects. The precipitation observations from both the NCEP gauge data and the NERN gauge network are plotted in Fig. 4b (NERN, lines with triangles; NCEP, lines with diamonds). These observations are highly correlated with each other, but the NERN daily rainfall often seems higher and earlier (∼1 day) than that observed from the NCEP gauge data. The PERSIANN satellite rainfall estimates over the boxes more closely match the timing of heavy rainfall observed from the NERN network than do the NCEP gauge data (not shown). This is because the NERN network has installed many rain gauges in high-elevation areas in regions indicated by the boxes, while the NCEP gauges are concentrated largely in accessible low-elevation areas. The mountain areas receive more rainfall at an earlier time than the low-elevation areas (see the discussion in section 4c). Monsoon rainfall over boxes C and D starts earlier and is much heavier, with more variability than the rainfall over boxes A and B. The model seems to lack agility in simulating such quick variations, resulting in precipitation underestimation.

Box E represents the area in the south of the Gulf of California and a portion of inland west-central Mexico. The simulation traced the observed rainfall perturbations in general, but missed many severe rainfall events.

b. Moisture surge over the Gulf of California

The moisture surge over the Gulf of California is an important intraseasonal feature of the NAM system. Previous studies indicate that most moisture for monsoon rainfall in the southwestern United States is transported from the tropical Pacific through surges (Adams and Comrie 1997; Stensrud et al. 1997; Anderson and Roads 2002; Douglas and Leal 2003; Higgins et al. 2004). Douglas and Leal (2003) identify the conditions to define a surge based on radiosonde data at Empalme, Mexico. Higgins et al. (2004) introduce diagnostic criteria to identify surges occurring over Arizona based on surface meteorological observations at Yuma, Arizona:

• (a) dewpoint temperature from below the climatological mean [15.7°C (60.1°F)] to higher than the mean and sustained for at least three consecutive days,

• (b) wind (at 700 and 200 hPa) speed exceeds the climatological mean (>4 m s⁻¹) and remains elevated for a few days (a week or more for strong surges), and

• (c) wind direction from northerly to southerly.

In Fig. 5, the modeled daily dewpoint and meridional wind at 10-m height in the grid box that covers Yuma are plotted with the corresponding surface observation data at the Yuma Valley station (information online at http://ag.arizona.edu/azmet/). The observational data from the other two stations, Gila and Mesa, were checked, and they are almost identical to the data at Yuma Valley. The ground observation data enable the identification of four surges during July and August: surge 1, 11–15 July; surge 2, 17–26 July; surge 3, 8–16 August; and surge 4, 28–30 August. Correspondingly, the surges determined from the model data are as follow: surge 1, 13–14 July; surge 2, 23–26 July; surge 3, 9–16 August; and surge 4, 28–29 August. Both surges 1 and 2 are identified with different onset dates and durations when using modeled and observed data. The cause of the differences was investigated using the diary of the “NAME Previous Day Summary,” written by the NAME committee members (available online at http://www.joss.ucar.edu/cgi-bin/catalog/name), during the NAME intensive observation period. The diary’s summary reported that on 11 July, “Dewpoint at 12Z [0000 UTC] remained in the low to mid 50’s across southern AZ” and on 13 July, “Flow in the Gulf of CA had increased from the south setting up what will most likely become Gulf Surge #1. At 12Z, the dewpoint at Yuma had increased to 70.” Apparently surge 1 started on 13 July, which agrees with the model prediction. After 14 July, the summary recorded no surge until 23 July, when it reported that “Tier 1 is under a strong gulf surge.” Again, the model onset date for surge 2 is correct. The model seems to work well in predicting 2004 NAM surge events. We suspect that the ground measurements of daily dewpoint and winds in the dataset may not be reliable because a large amount of missing data was found in the files for these variables possessing high diurnal variations. Therefore, one should be cautious about using ground observation data to identify surge onset and ending dates for the NAM.

Anderson et al. (2000) have identified two types of surges: 1) tropical cyclone surges, which are associated with the passage of tropical disturbances (such as hurricanes and tropical storms) to the south of Baja California, and 2) easterly trough surges, which are associated with the westward propagation of a low-level trough from the SMO to the eastern Pacific. Surge 1 was clearly a tropical cyclone surge. On 12 July, Tropical Storm Blas was moving northward near the mouth of the Gulf of California, and surge 1 started on 13 July. Figure 6 displays the progression of surge 1 in 700-mb mixing ratio, geopotential height deviation, wind, and geopotential height on 12–14 July, accompanied by the PERSIANN rainfall, which shows Blas’s migration. The model’s description of the surge event agrees with previous studies. In Fig. 6, the isobars of 700-mb height (middle column) delineate the movements of the cyclone Blas and the monsoon anticyclone. When the cyclone moved northward to the gulf and dispersed west of southern Baja California, the anticyclone moved westward from Louisiana to central Texas. During the surge period, over the Gulf of California, southeasterly winds strengthened (right column) and moisture became concentrated (middle column). The initiation of the surge event started on 13 July (Fig. 6f) when Blas brought strong pressure-gradient forcing near the mouth of the gulf, resulting in a surge front that propagated through the gulf over a span of 18–30 h (Stensrud et al. 1997). Once the surge front passed, the southeasterly winds over the gulf continued for another day. The winds were maintained (Fig. 6i) by a geostrophic balance between a low pressure center located to the southwest of the gulf and a high pressure area situated over the SMO (Anderson et al. 2000). Figures 6b, 6e and 6h (middle column) show that in the surge event, a large amount of moisture accumulated over the gulf and northern Mexico, then spread to Arizona, New Mexico, and southern Colorado. When this paper was prepared, we were unable to find the NAME field measurement data (as well as the NARR data) to evaluate the modeled surge processes (therefore, we will not discuss the model results for other surge events). When the observation field data and reanalysis data are available they should provide the basis for an interesting future study.

c. Diurnal variability of precipitation and low-level flow

Evidence from satellite and ground observations as well as model simulations demonstrates that NAM rainfall is strongly modulated by diurnal variations (e.g., Negri et al. 1993; Dai et al. 1999; Sorooshian et al. 2002; Berbery 2001; Mo and Juang 2003; Li et al. 2004). Based on hourly data from the NERN gauge network, Gochis et al. (2004) found that diurnal cycles of precipitation frequency and intensity have distinct relationships to terrain elevation. Figure 7 compares the 3-hourly rainfalls of July at six elevation bands (bins) along the western slope of the SMO with the NERN observations. The model results demonstrate pronounced diurnal cycles, but exhibit two major errors in comparison with the NERN results: 1) the model diurnal cycles at all elevations except the lowest (elevation <500 m) share a similar diurnal phase variability, with peak rain occurring at around 1700 mountain standard time (MST). In contrast, the observed diurnal cycles across the SMO slope show a shift in diurnal peak from local afternoon (∼1700 MST) over the highest elevation to late evening (∼2300 MST) over the lowest, and 2) the model underestimates the diurnal peak intensity over all elevation bands. In particular, the model rainfall over the lowest-elevation band (<500 m) shows a low and flat distribution throughout the period of a day without visible diurnal variability, which differs from the observed diurnal variability with its late evening peak.

To understand the cause of the diurnal precipitation feature over the western slope of the SMO, Fig. 8 presents 3-hourly diurnal patterns of the streamline in July at σ = 0.9535 (∼500 m above the terrain surface). From local late morning (1100 MST or 1800 UTC) to later night (2300 MST or 0600 UTC), sustaining sea breezes from the Gulf of California and the eastern Pacific invade western Mexico and bring rich moisture to the western slope of the SMO. During this period, the convergence line (the dotted lines) of the lateral flows shifts from near the SMO peak (the Continental Divide) to near the coastline. The convergence line indicates the location where vertical motion or convection occurs. Therefore, the model does predict a migration of convection activity from local afternoon over the highest-elevation band to late night over the lowest-elevation band, which matches the observed shift of the rain peaks. The well-simulated convection migration (the convergence line) over the SMO’s west slope suggests that the errors in diurnal rainfall cycles result from a deficiency in the convective parameterization scheme.

Figure 8 shows that in the morning [0500–1100 MST (1200–1800 UTC)], instead of producing westward land breezes across the coast to the Gulf of California, the low-level flows turn to southeasterly winds. A low-level jet (LLJ) from northern Mexico across the Mexico–U.S. border in Arizona and New Mexico reaches the southwest United States [Fig. 8 at 0800 MST (1500 UTC)], bringing moisture to the region. In the afternoon and night [1400–2300 MST (2100–0600 UTC)], another LLJ from the northern end of the gulf flows into Arizona. This LLJ is considered the major moisture source of monsoon rainfall in the southwestern United States (Stensrud et al. 1995; Douglas and Li 1996; Anderson et al. 2000). However, based on observations at Yuma, Arizona, Douglas and Li (1996) have determined that the timing of the maximum LLJ from the northern Gulf of California occurs in the early morning hours (1200 UTC), not in the afternoon, as the model shows. This has been one of the key research issues in the NAM modeling study. The results of Saleeby and Cotton (2004) are comparable to ours. The modeling of Anderson et al. (2000) shows that the LLJ occurs over the western portion of the northern gulf and the foothills of the SMO at 1200 UTC, then extends across the gulf at 0000 UTC. Stensrud et al. (1997) show that LLJ development is related to the moisture surge over the Gulf of California. Figure 8 demonstrates that the core NAM region of northern Mexico and the southwestern United States is between two synoptic flows with opposite directions: the northwesterly, from the eastern Pacific (on the west side), and the southeasterly, from the Gulf of Mexico (on the east side). The diurnal variation of the low-level flow can be affected sensitively by simulating the diurnal variations of the synoptic flows (Fig. 8 shows that the southeasterly from the Gulf of the Mexico varies more than the northwesterly from the eastern Pacific).

Our MM5 modeling fails to simulate late evening peak rainfall over the lowest-elevation band. Figure 9 shows the modeled diurnal variability of the mean soundings in July over a coastal site in northern Mexico (28.8°N, 110.6°W; elevation, 389 m). In the morning, there is insufficient convective available potential energy (CAPE) in the atmospheric column to support convective rainfall. From afternoon to the later night, the CAPE above the level of free convection (LFC) reaches values greater than 1000 J kg⁻¹, indicating high buoyant energy forms in the atmosphere column. However, a negative energy layer, the convective inhibition (CIN), exists below the LFC. Under such conditions, the hostile dry layer and CIN make it difficult for convective rainfall to develop. A recent study (Li et al. 2006) using cloud microphysics at a high grid resolution (3 km) shows that this CIN may be diminished by downward outflow from the preceding convection, which occurs over the upper (mountain) slope.