1. Introduction

The North American monsoon (NAM) is a remarkable phenomenon that results in a pronounced increase of precipitation from extremely dry spring conditions to much wetter summers over much of the semiarid and arid areas of the southwestern United States (SW) and northwestern Mexico (NW Mexico). The North American monsoon system (NAMS), which develops over low-latitude continental regions in response to seasonal changes of the thermal contrast between the continent and adjacent oceanic regions, is a major component of North American continental warm season precipitation regimes (Webster 1987). Monsoon rainfall usually starts in June or July, and lasts several months until mid-September. Some 40%–80% of the annual precipitation in the SW and NW Mexico falls during the monsoon period (Douglas et al. 1993; Stensrud et al. 1997).

Compared to its Asian monsoon sister, NAMS is less impressive partly because of the much smaller area that is affected (Tang and Reiter 1984). Nonetheless, it has tremendous local socioeconomic impact especially considering the rapid population increase in the last few decades in the affected region (Castro et al. 2001). Therefore, improved seasonal predictability of the NAM precipitation would have not only important scientific implications, but important practical ones as well. Improved predictability of NAMS precipitation is also motivated by the fact that NAMS has not only local affects, but is also linked to warm season precipitation over much of the United Sates and Mexico (Higgins et al. 1997; Mo et al. 1997).

Douglas et al. (1993) document that the precipitation over SW is but the northwestern extremity of a much more pronounced phenomenon centering over NW Mexico. The onset of NAMS is characterized by heavy rainfall over southern Mexico, which quickly spreads northward along the western slopes of the Sierra Madre Occidental (SMO) (Douglas et al. 1993) into Arizona and New Mexico in early July (Douglas et al. 1993; Stensrud et al. 1995). This northward extension can be represented as the decreasing ratio of July–September (JAS) rainfall to annual precipitation from the region surrounding the southern Gulf of California, along the axis of SMO into Arizona and New Mexico (Douglas et al. 1993). Gutzler (2004) also identifies a region of NW Mexico (24°–30°N) as the “core” of the continental interannual variability of precipitation in the NAM domain, which captures the maximum in continental interannual variability across the monsoon domain and exhibits significant persistence of precipitation anomalies from the early summer season to the late season. Gutzler’s core monsoon region is very similar to the core of the NAMS region in northwestern Mexico (which we term MSa here; shown later in Fig. 2). Even though the NAM is most spatially consistent over NW Mexico with greater variability to the north, research to date has focused largely on the northern part of the NAM domain, which includes Arizona and New Mexico.

As an ocean–atmosphere–land coupled system, NAMS exhibits apparent dependence on (land and ocean) surface conditions (Adams and Comrie 1997). For this reason, premonsoon land surface and oceanic conditions are promising predictors for NAM precipitation at seasonal lead times, especially where these predictors [like soil moisture (S_m) and SST] are temporally persistent. Therefore, before building a useful seasonal monsoon climate prediction capability, we first need to explore and understand the possible relationships between the NAM precipitation over NW Mexico and antecedent surface conditions. To date, the role of land surface mechanisms in NW Mexico as they relate to NAMS predictability has received relatively little attention, quite likely due to the absence of suitable data (especially S_m).

A few studies have documented the role of remote SST anomalies (SSTA) in the equatorial Pacific as they affect NAMS precipitation anomalies. Higgins et al. (1999) examined the relationship between anomalous monsoon behavior over Arizona–New Mexico, NW Mexico and southwest Mexico, and ENSO signals. They found that wet (dry) monsoons in southwest Mexico tend to occur during La Niña (El Niño), which they attributed partly to the impact of local SSTA on the land–sea thermal contrast, hence the strength of the monsoon. They also found a weak (nonsignificant) association between dry monsoons in NW Mexico and El Niño. Hu and Song (2002) showed that south-central Mexico monsoon rainfall is highly affected by interannual variations in the SST and in the location of the intertropical convergence zone (ITCZ) in the eastern tropical Pacific. Cooler (warmer) than normal SST coexisted with the more northern (southern) position of ITCZ and more (less) monsoon rainfall in central-south Mexico.

There are even fewer studies of the land surface feedback mechanisms over NW Mexico. Matsui et al. (2003) investigated the influence of land–atmosphere interactions on the variability of the NAMS by testing a hypothesis regarding the connection between observed land surface variables [April snow water equivalent (SWE), surface (skin) temperature (T_s)] and precipitation in the NAM region, including NW Mexico, for the period 1979–2000. Their results showed that there is a weak negative relationship between 1 April SWE in the southern Rocky Mountain region and subsequent spring temperatures that persist into June in the NAM region. They concluded that this inverse relationship could not directly influence monsoon rainfall in July and August because it disappears during the monsoon season. The Matsui et al. (2003) study appears to be the only attempt to evaluate the connection between monsoon onset and/or intensity and antecedent land surface conditions using observations. A major reason for the absence of other studies is the scarcity of long-term observations of the relevant land surface variables (S_m, T_s, SWE) over the NAM region.

Small (2001) performed experiments using the fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5), coupled land–atmosphere model with the Oregon State University (OSU) land surface scheme, to examine the influence of S_m anomalies on the variability of the NAM. His modeling results show an inverse relationship between southern Rocky Mountains antecedent season snowpack and monsoon rainfall. However, the S_m prescribed in Small’s simulations seems excessive (exceeds field capacity).

Past studies have been constrained by the lack of long-term, consistent observations of land surface variables that might be predictors of the land surface state, and hence the onset and strength of NAMS. Zhu and Lettenmaier (2007) describe a long-term gridded set of observed surface climate variables, and model-derived [using the Variable Infiltration Capacity (VIC) macroscale hydrological model] land surface states and fluxes for a domain consisting of all of Mexico for the period 1925–2004. The VIC model is forced with observed precipitation, and the model-simulated runoff plausibly matches observations for 14 test river basins distributed across Mexico. Therefore, evapotranspiration, averaged appropriately, is arguably realistically reproduced as well. On this basis, and given the physically based model parameterizations in VIC for S_m and energy fluxes, we argue that the other surface flux and state variables (such as S_m) in the dataset should be reasonably well represented as well. In this study, we supplement the Zhu and Lettenmaier (2007) data with the retrospective North American Land Data Assimilation System (N-LDAS) dataset of Maurer et al. (2002) over the continental United States for 1950–2000.

In a previous similar study over SW (Zhu et al. 2005), we proposed a land surface feedback hypothesis (Fig. 1): anomalous winter precipitation (P) leads to more winter and early spring SWE, hence more spring and early summer S_m, and lower spring and early summer T_s. These conditions induce a feedback to the atmosphere that results in a weaker NAM onset (and less rainfall) and vice versa. In this study, we start with the same hypotheses to investigate the monsoon onset variability in NW Mexico. Sections 2 and 3 describe the data and methodology utilized. In section 4 we classify and define the extreme wet and dry, and early and late, monsoon years. In section 5 we identify the possible winter or spring precipitation and spring snowpack related regions linked to monsoon onset date and we determine their dynamical links. Based on the land surface feedback mechanism hypothesized in our previous study (Zhu et al. 2005), in section 6 we begin to examine three links in this feedback chain. Specifically, we test a possible linkage between winter P and spring SWE and spring S_m. In section 7, we evaluate the second and third links in our hypothesis and demonstrate that the spring soil moisture–spring surface temperature–monsoon onset hypothesis is viable, in contrast to the results of Zhu et al. (2005) for SW. In section 8 we confirm the existence of the land–sea thermal contrast concept for the initiation of monsoon onset by composite analysis of sea surface temperature and land surface temperature in May. In section 9 we conduct a preliminary analysis of the possible role of the atmospheric circulation anomalies in regulating the premonsoon land surface temperature conditions, which in turn modulate monsoon onset. Conclusions are presented in section 10.

2. Data

The land domain in this study extends from 14° to 50°N and includes the conterminous United States, parts of Canada, and all of Mexico (Fig. 2). The primary source of land surface data for the continental United Sates and parts of Canada is Maurer et al. (2002). Monthly P in this dataset is gridded from climatological (daily) observations of precipitation. Surface air temperature is gridded from daily observations, while downward solar and longwave radiation data and humidity are derived using algorithms described in Maurer et al. (2002) based on surface air temperature. Surface wind is taken from the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis (Kalnay et al. 1996). Surface skin temperature T_s, S_m in three vertical layers, and SWE are derived from the VIC land surface model, which is designed both for off-line, or stand-alone use to simulate the water and energy budgets of large continental areas. Detailed information about the dataset for the U.S. portion of the domain can be obtained from Maurer et al. (2002), while information about the VIC model is given in Liang et al. (1994), Abdulla et al. (1996), Nijssen et al. (2001), and Pan et al. (2003).

For Mexico, we utilize a dataset described by Zhu and Lettenmaier (2007), which was developed using methods similar to those employed by Maurer et al. (2002) and which covers the period 1925–2004. In the Zhu and Lettenmaier (2007) dataset, P was extracted from three different daily station data sources: recently released (in 2005) Servicio Meterorológico Nacional of Mexico (SMN) long-term improved surface station data (pre-1940–2003), SMN daily precipitation data (1995 to near–real time), and NW Mexico North American Monsoon Experiment (NAME) Event Rain Gauge Network (NERN) daily precipitation data (2002 to the present) with 86 stations across the Sierra Madre Occidental (Gochis et al. 2003, 2004). The latter two data sources compensate somewhat for the scarcity of precipitation stations in the first source in northern Mexico from 1995 on.

Because the Maurer et al. (2002) data span a shorter time period than the Zhu and Lettenmaier (2007) data, for this analysis we use 1950–99 as the study period. Prior to conducting the analysis, we aggregated the ⅛° data to 1° spatial resolution for ease of computation.

To explore the relationship between midtropospheric conditions and surface land conditions, we use monthly mean 500-mb geopotential heights (Z500) from the NCEP–NCAR reanalysis data (Kalnay et al. 1996) provided by the National Oceanic and Atmospheric Administration–Cooperative Institute for Research in Environmental Sciences (NOAA–CIRES) Climate Diagnostics Center, Boulder, Colorado, via their Web site (http://www.cdc.noaa.gov/). The global Z500 dataset covers the period 1948–2003 and has a horizontal resolution of 2.5° latitude × 2.5° longitude. In this paper, we defined a study domain for Z500 analysis as 10° –50°N, 67.5°–135°W covering the land domain (Fig. 2) and the adjacent ocean region. We downloaded the Niño-3 index from the National Weather Service’s Climate Prediction Center (CPC; information online at http://www.cpc.ncep.noaa.gov/data/indices/sstoi.indices). The SST dataset used is the extended reconstructed sea surface temperature dataset (ERSST) produced by the National Climate Data Center (NCDC) for the period of January 1854–June 2005 with a resolution of 2° (Smith and Reynolds 2004). Our SST domain was −10°–50°N, 60°W– 180°, which covers the equatorial, tropical, and the North Pacific Ocean and the Caribbean Sea.

3. Methodology

The NAM domain was originally classified into four monsoon subregions (Fig. 2) per Comrie and Glenn’s (1998), who based definition of the subregions on the seasonality and variability of JJAS monsoon precipitation from 1961 to 1990 through the use of principal components analysis. Using their notation, there are four major monsoon subregions: monsoon south (MSa and MSb), monsoon west (MW), monsoon north (MN), and monsoon east (ME) (Fig. 2). The regions MW, MN, and ME lie in SW, Msa, and MSb in NW Mexico. MSa and MSb receive 80% of their annual precipitation in summer (Fig. 3) and constitute the core region. In this paper, we divided the original Comrie and Glenn MS region into two subregions—MSa and MSb—with 25°N as the division line. Below 25°N the seasonal distribution of precipitation is much more affected by tropical storms than is the region above 25°N (Englehart and Douglas 2001). Hu and Song (2002) also found that south-central Mexico (approximately that part of Mexico below 25°N) is modulated by different processes than is northwestern Mexico. They argued that while the south-central Mexico monsoon rainfall is affected significantly by JAS ITCZ location and SSTA in the eastern equatorial Pacific, NW Mexico together with SW are modulated, at least in part, by the land memory effect. Cavazos and Hastenrath (1990) and Higgins and Mo (1998) also demonstrated that during the summer southwestern Mexico (below 25°N) exhibits a stronger ocean connection with ENSO than did northwestern Mexico. Therefore, we concentrate here on the role of antecedent land surface conditions on the monsoon rainfall variability of MSa.

We created seasonal indices of area-averaged total JJAS P and monsoon onset date for MSa as the starting point of our analysis (Fig. 4). The correlation between total JJAS P and onset date is −0.46, suggesting that early monsoon years tend to be wetter than late monsoon years, as has been also documented for Arizona and New Mexico by Higgins and Shi (2000). Winter is defined as January–March (JFM) and spring as April–May (AM) antecedent to the JJAS monsoon season. We also utilized mean monthly values of T_s, S_m, SWE, and Z500 for all the grid points of the domain. We use simple lag correlations and composite analyses to examine the relationship between JJAS rainfall and antecedent winter and spring P, T_s, S_m, SWE, and Z500. A Student’s t test is used to test the significance of spatial correlations, assuming one degree of freedom per year.

4. Classification of extreme years

To remove the seasonal mean and local rainfall effects, we first calculated standardized rainfall anomalies for each grid point of the domain and then calculated an average over the MSa region (Fig. 5). Wet (dry) monsoon years in MSa are defined as those years characterized by rainfall anomalies greater than 0.75 or less than −0.75 standard deviations. The wet years in Fig. 5 are 1958, 1966, 1984, 1986, and 1990, and the dry years are 1951, 1969, 1973, 1979, 1987, and 1995.

Monsoon onset (Fig. 4) was determined using a threshold-crossing procedure for the daily regional precipitation time series. As defined in Higgins et al. (1999), the magnitude and duration criteria for the onset are 1.0 mm day⁻¹ and 5 days. The start of the monsoon occurs when rainfall exceeds 1.0 mm day⁻¹ for five consecutive days after 1 May. The variability of the monsoon onset date in MSa is high, ranging from 1 June to 12 July with the mean onset date on 26 June. If the extreme five early and five late monsoon years (Fig. 4) are defined as the top and bottom 10% of the 50-yr period, then the early monsoon years are 1972, 1976, 1977, 1984, and 1996, and the late years are 1955, 1983, 1987, 1992, and 1995.

Even though the correlation between monsoon magnitude and monsoon onset (Fig. 4) is statistically significant at the 95% level, very few years are common between the early and wet years, and the late and dry years, suggesting that antecedent land surface conditions contribute only partially to the total monsoon precipitation variability. It is possible that once the monsoon starts, other synoptic and intraseasonal mechanisms become important such as the effect of evaporation in the dynamics of the regional flow and pressure gradients (e.g., Kanamitsu and Mo 2003), the passage of moisture surges, and the number of hurricanes. Thus, the onset date seems to be more heavily influenced by premonsoon seasonal conditions than the magnitude of the monsoon precipitation. In later sections, we will evaluate the feedback mechanism for both monsoon magnitude and onset date.

5. Winter precipitation: Snow–monsoon connection

In our previous study (Zhu et al. 2005), we defined a winter precipitation predictor region over the southwest United States that was negatively correlated with MW summer monsoon precipitation. We also showed [as was found by other authors, e.g., Higgins and Mo (1998) and Hu and Song (2002)] that this inverse relationship is not robust over time; it is strong during 1965–90 and weak otherwise. Similarly for MSa, we also try to construct a winter precipitation index as a possible predictor for monsoon onset date and summer P. The correlation map between JFM winter P and MSa monsoon onset date (Fig. 6) shows that there exists during winter a positively correlated region in SW and NW Mexico, and parts of Texas, Oklahoma, Kansas, and Nebraska. A wet winter tends to be followed by late onset of the monsoon, and a dry winter by an earlier start of the monsoon. Based on the correlations in Fig. 6 we defined a winter index region (Fig. 7a) and construct a corresponding winter P index time series (Fig. 7b). In the mid-1970s there was a dramatic shift to much larger winter P totals, possibly associated with North Pacific decadal fluctuations (e.g., Dettinger et al. 1998; Gershunov and Barnett 1998; Gershunov and Cayan 2003). To determine the temporal persistence of the correlation between winter precipitation and MSa onset date, the 15-yr moving correlation is shown in Fig. 7c. The strongest correlation is observed for about 15 yr during 1965–79, a much shorter time period than we found for MW (Zhu et al. 2005). With respect to monsoon magnitude, no consistent significant correlation between antecedent winter or spring P and summer monsoon P is found for a long enough time and spanning over a large enough area to form a useful predictive relationship.

Similarly to our MW study (Zhu et al. 2005), we hypothesize that there is a link between winter SWE (related to winter precipitation) and monsoon onset in MSa because the winter P-related region in Fig. 7a covers high elevations of the Rocky Mountains. Through sliding correlation analysis between SWE in this region and monsoon onset in MSa (not shown here), we found the same winter snow-related region (Fig. 8a) in the southern Rocky Mountain as for MW (Zhu et al. 2005). This snow–monsoon link could be nonlocal because the winter snow accumulation area is not necessarily located in the same region as the monsoon rainfall response area (Barnett et al. 1989). The 15-yr moving-average correlation (Fig. 8b) indicates that the most significant positive relationship between JFM SWE in the southern Rocky Mountain and monsoon onset date is during the period from 1960 to 1980. Compared to MW, the MSa winter precipitation and snow–summer monsoon onset link is weaker and shorter. According to hypotheses advanced by Higgins and Mo (1998) and Hu and Song (2002) for SW, the relationship between winter precipitation and summer monsoon is partly due to a land memory effect. Hu and Song (2004) argue that land process effects may have existed even during those periods when there was no correlation between winter P and monsoon onset date. Similarly, we argue that for MSa the land surface feedback may exist for summer monsoon initiation based on this positive winter precipitation–summer monsoon relationship, even though its strength apparently varies over time. The variation of this relationship is partly caused by the interference of SST persistence in the Niño-3 region, which we discuss in section 8. In the next section, we begin to investigate this land memory effect for MSa.

6. Spring land memory

Similar to our MW study (Zhu et al. 2005), we propose a winter P, SWE–S_m–T_s–NAM feedback mechanism (Fig. 1) for the positive relationship between winter P and SWE and monsoon onset date. Heavy winter snowfall and precipitation lead to wetter spring soil, and cooler surface temperatures, which may induce a weak land–sea temperature contrast and, hence, a late monsoon onset. This hypothesis includes three possible mechanisms: 1) land surface memory through persistence, 2) land–atmosphere interaction, and 3) thermal contrast before the onset of the monsoon. We begin our hypothesis testing for extreme late and early monsoon years. Figure 9 shows the JFM winter P relative anomaly composite map in extreme late and early monsoon years for MSa. Late monsoon years are preceded by wet winters over large areas of SW, the midwestern United States, and northern Mexico; the reverse is true for early years. Late monsoon years show a stronger spatial signal than early monsoon years, covering large areas extending into Nebraska, South and North Dakota, and eastern Mexico. The late monsoon composite anomaly map is consistent with the correlation map shown in Fig. 6, except that the positive signal in most of Mexico and the midwestern United States (Fig. 9) disappears in normal years (Fig. 6).

Figure 10 shows May total-column S_m anomalies for extreme monsoon onset years. We choose May as the key premonsoon season because the earliest onset date for MSa is 1 June (see section 4). The May S_m anomaly pattern (Fig. 10) is similar to the winter P pattern (Fig. 9) in SW and NW Mexico, with late monsoon years showing anomalously wet spring S_m and vice versa for early monsoon years, implying that May S_m has a memory of the winter–spring P anomaly. Figure 3a indicates that, on average, there is little precipitation during winter and spring in NW Mexico; however, Figs. 9 and 10 suggest that during extreme monsoon years the impact of anomalous winter and spring P may persist into the premonsoon season in May. The gridpoint local correlation map between JFM P and May total-column S_m (Fig. 11a) and first-layer S_m (Fig. 11b) further confirm the existence of such a land surface effect. JFM P is highly correlated with May total column S_m over most of the United States and Mexico especially for SW and NW Mexico. Texas and the U.S. Great Plains show no significant winter precipitation signal in the first Sm layer, possibly because the major precipitation season in that region is during spring (Mock 1996). It is impressive that even the May first-layer S_m has a strong memory effect from winter P in SW and northern Mexico, which may exert an energy link to the atmosphere during the monsoon season.

7. Spring land surface thermal condition: Monsoon onset

Figure 12 shows the May T_s anomaly composite map during late and early monsoon years. As expected from our hypothesis, May T_s exhibits a strong inverse relationship with May Sm (Fig. 10); in late monsoon years, May T_s is colder than normal in large areas of the southwest United States and northwestern Mexico, with some grid cells showing less than −1°C anomalies for wet soil conditions; the reverse pattern in SW is true for early monsoon years. This T_s anomaly pattern is consistent with the thermal contrast concept for driving the onset of the monsoon with colder T_s delaying monsoon onset and warmer T_s initiating an earlier monsoon onset. This is further confirmed by the correlation pattern between May T_s and the monsoon onset date shown in Fig. 13, which shows a negative signal from the Sierra Madre Occidental into the Rocky Mountains with strong values in some parts of the Sierra Madre and especially over New Mexico and Colorado.

It is surprising that compared to most of SW, the desert regions of southern Arizona, southeastern California, and NW (Sonora) Mexico, which are well known for the development of a surface thermal low (Rowson and Colucci 1992), show a strong warm signal only in early monsoon years, and a weak signal in late years. This finding does not disprove the importance of desert regions for the formation of the thermal low, but rather indicates its low covariability with monsoon onset. Stated another way, even though the desert region is very important to the formation of the thermal low over SW (Rumney 1968) because of its extremely high temperatures during the premonsoon season, it seems to have little value for the prediction of monsoon onset.

To better understand the land–atmosphere link in this logic chain, it is necessary to test whether the negative relationship between May first-layer S_m and May T_s exists during the entire study period (not only during extreme monsoon years). Figure 14 shows that this negative relationship exists over most of the continental United States and Mexico, except in the Sonora–Arizona desert and parts of the core monsoon region. The weak signal over the desert regions may be partially a result of the low evaporation in that area in all years, as discussed in our previous paper (Zhu et al. 2005, their Fig. 16). Because the premonsoon thermal condition in this area is not strongly related to soil wetness in most years, we believe that T_s in the desert regions is more constrained by energy available from radiant fluxes, which are related largely to cloud cover. This suggests the influence of large-scale circulation (in contrast to land surface feedbacks) as discussed in section 9.

The lack of significance over most parts of the core monsoon region in Figs. 13 and 14 suggests that, with the exception of extreme monsoon years, the thermal contrast signal may be nonlocal (note, e.g., the significant correlation in Fig. 13 for the larger area including northern Mexico, New Mexico, and Colorado). This is understandable because the summer thermal low pressure system dominates a much larger area than just the desert region of SW and NW Mexico (Barry et al. 1981), and this larger area may well contribute to the formation of monsoon circulation.

Overall, the correlation signal in Fig. 14 and the composite maps in Figs. 10 and 12 indicate that the spring soil wetness condition affects the premonsoon surface thermal condition over the key area in New Mexico and Colorado, thus influencing the monsoon onset. Therefore, the hypothesized land surface feedback mechanism (Fig. 1) is confirmed for MSa monsoon onset. On the other hand, for monsoon magnitude (JJAS P), we did not find a strong signal when using JFM P and spring S_m as the predictors. However, Fig. 15 shows a strong positive relationship between monsoon strength and May T_s over Arizona, New Mexico, and part of NW Mexico, which partially overlaps with the significantly correlated region of May S_m and T_s in Fig. 14. This suggests that antecedent land surface conditions also influence the monsoon magnitude, even though this feedback is not as strong as for monsoon onset.

8. Land–sea thermal contrast

The combination of seasonally warm land surfaces in lowlands and elevated areas together with atmospheric moisture supplied by nearby maritime sources is conducive to the formation of a monsoonlike system (e.g., Li and Yanai 1996; Adams and Comrie 1997). For the Asian monsoon, Flohn (1957) suggested that the seasonal warming of the Tibetan Plateau and the consequent inverse thermal gradient south of 35°N were responsible for its onset. Much later, Fu and Fletcher (1985) showed that the interannual variability of the monsoon rainfall in India was highly correlated to the thermal contrast between the Tibetan Plateau and the equatorial Pacific Ocean. Similarly, seasonal changes in the thermal contrast between land and adjacent oceanic regions are a cornerstone of the conceptual basis for understanding the onset of NAMS (Adams and Comrie 1997; Higgins and Mo 1998; Gutzler 2000).

Figures 16a and 16b show SST and Ts anomaly composite maps for early and late monsoon years in MSa. The early monsoon years are characterized by warmer land over SW and NW Mexico, and a weak positive SST signal over the subtropical eastern Pacific, indicating a thermal gradient from the ocean to SW. On the Pacific side, the thermal difference between adjacent oceanic regions and SW is about 1.2°C larger in early monsoon years than in normal years. In contrast, late monsoon years show a reverse differential thermal contrast (of about 2°C), with a stronger warming over the eastern Pacific and cooling over northern Mexico and SW. Late monsoon years also exhibit a strong positive sea surface temperature signal over the eastern equatorial Pacific, clearly reflecting an El Niño–type signal. These signatures are further enhanced in the difference map (Fig. 16c) between late and early monsoon years, which shows an inverse thermal contrast during late years with a thermal difference of 2.5°C. The mean anomaly pattern over the Pacific Ocean (Figs. 16b and 16c) that characterizes late monsoon years resembles the constructive phase of the El Niño/+PDO (e.g., Gershunov and Barnett 1998).

In contrast, early monsoon years are characterized by more neutral ENSO conditions (Fig. 16a). Table 1 shows the correlation between JFM, AMJ Niño-3, and several monsoon indices. In general, the winter P index tends to be positively and significantly correlated with winter Niño-3, except for the 1965–79 period, when winter P and the monsoon onset date are significantly correlated (Fig. 7c). Both JFM Niño-3 and AMJ Niño-3 are significantly correlated with MSa monsoon onset during the period 1980–99. Hu and Song (2004) postulate that an active land surface memory from winter to summer (e.g., 1965–79) may be due to a weak SST persistence, which is partially consistent with the weak winter to spring Niño-3 persistence shown Fig. 17. In contrast, after the 1977 decadal shift (+PDO) there is a strong SST anomaly persistence from winter to spring in the Niño-3 region. According to Hu and Song (2004), the influence of this persistence (through atmospheric teleconnections) would override the land surface memory, which would explain (a) the insignificant correlation between winter P and monsoon onset after 1980 (Fig. 7c) and (b) the strong correlation between Niño-3 and both winter P and monsoon onset. However, Hu and Song’s hypothesis does not explain the inactive land memory observed before 1965. Thus, the interseasonal persistence of the land memory on the summer monsoon needs further investigation.

If we focus only on premonsoon conditions, it seems that the most consistent SST pattern associated with monsoon onset is the differential heating between SW and northern Mexico, and the adjacent eastern Pacific Ocean, which would create the low-level pressure gradients to initiate the monsoon (Fig. 16). In contrast, the Ts anomaly maps for wet and dry years (not shown) in May do not show a clear thermal contrast between land and sea, possibly because the total monsoon precipitation is not only modulated by premonsoon seasonal conditions, but also by other land surface conditions during the evolution of the monsoon season, as was suggested in section 4.

9. Role of 500-hPa geopotential heights

In our previous study for MW (Zhu et al. 2005), we found a strong positive relationship between June surface air temperature and the large-scale midtropospheric circulation, thus concluding that the main controlling factor for the intensity of premonsoon surface air temperature anomalies could be the large-scale atmospheric circulation condition, at least in extreme years. For MSa, in addition to analyzing the land surface conditions, we also explored the possible influence of the upper-tropospheric circulation on premonsoon seasonal T_s. Figures 18a and 18b display the 500-mb geopotential height (Z500) anomaly composite maps in extreme late and early monsoon years for JFM and May. During late (early) monsoons the winter circulation is characterized by negative (positive) geopotential height anomalies over the western United States and NW Mexico (Fig. 18a), similar to the tropical–Northern Hemisphere teleconnection pattern (Mo and Livezey 1986). These patterns suggest a southward- (northward-) displaced winter subtropical westerly jet preceding late (early) monsoons, which are consistent with the winter precipitation anomalies documented in Fig. 9. A southward-displaced winter jet stream is typical of, but not limited to, strong and moderate El Niño events (e.g., Barnston et al. 1991; Cavazos and Rivas 2004). By May, the Z500 anomaly maps (Fig. 18b) indicate that the westerlies are still much farther south (negative anomalies) during late monsoons than during early monsoons (positive anomalies), suggesting a possible large-scale atmospheric persistence from winter to late spring. The May Z500 anomaly maps exhibit a positive correlation with May T_s anomalies (Fig. 12) with low (high) Z500 anomalies over SW associated with negative (positive) T_s anomalies during late (early) monsoons. The subsidence associated with higher Z500 in early monsoon years may cause high T_s due to enhanced adiabatic compression and surface shortwave absorption associated with subsidence-related clear skies, and the reverse is true for late years. The higher Z500 anomalies over SW during early monsoon years are also indicative of a farther north monsoon anticyclone, clear skies (positive outgoing longwave radiation anomalies, not shown), and a warmer troposphere (larger tropospheric thickness). Thus, the patterns in Fig. 18b suggest a stronger land–sea thermal contrast during early monsoons than in late monsoons consistent with the results presented before. The correlation map between Z500 and Ts (not shown here) shows that they are highly positively correlated with each other (as would be expected from a hydrostatic relationship) over almost all of the continental United States including the desert SW, not only in extreme years (Fig. 18) but also in normal years. These correlation coefficients across the whole domain (i.e., >0.6 over SW and >0.4 over NW Mexico, not shown here) are stronger than the linkage between S_m and T_s shown in Fig. 14, suggesting that although the land surface memory may exert some influence on T_s, the large-scale circulation may play a more important role in the premonsoon land–sea thermal contrast, which in turn affects the monsoon onset.

10. Conclusions

As a follow-on study to our earlier research (Zhu et al. 2005) of MW, we have explored here the role of antecedent land surface conditions on the onset and intensity of the monsoon in the core of the North American monsoon region in NW Mexico. We focus in particular on a domain termed MSa, which is expected not to be strongly affected by tropical storms that might complicate interpretation of teleconnections.

Based on our previous study of MW (Zhu et al. 2005), we proposed a similar winter P–spring S_m–early summer T_s–monsoon onset land surface feedback hypothesis for MSa. In our MW study, we confirmed the existence of a land memory associated with winter precipitation and snow over MW. However, the land surface feedback hypothesis broke down due to the small contribution of land surface memory to the surface thermal condition at the onset of the monsoon and, hence, to monsoon strength. In contrast to the MW results, we find for MSa evidence of a land feedback mechanism wherein anomalous winter precipitation in SW and NW Mexico affect spring Sm there and, hence, May T_s especially over New Mexico and Colorado, at the onset of the monsoon. We have shown here that the signal is especially evident with respect to the (date of) onset of the monsoon, and less so for monsoon magnitude. It is interesting that unlike the widely accepted monsoon generation concept that focuses on the desert region of SW and NW Mexico, we find that a larger region over SW (especially New Mexico and Colorado) significantly affects the variability of monsoon onset and strength.

We also confirmed that the monsoon driving force concept based on land–sea temperature contrasts is linked to the monsoon onset in MSa. We found that the most important factor driving the monsoon onset is the thermal contrast between SW and NW Mexico and the adjacent ocean conditions with early (late) monsoons initiated by a stronger (weaker) land–sea thermal contrast. Our analysis suggests that the SST ENSO-related links to the monsoon onset vary with time. When the Niño-3 SST anomalies significantly persist from winter through spring, the land surface memory may be overridden by the influence of the atmospheric teleconnections (e.g., Hu and Song 2004). We found that the large-scale circulation, arguably related in part to cloud cover and its effect on net radiation, and hence on surface temperature, influences the monsoon onset and strength in MSa. In fact, the role of the large-scale circulation may well be larger than the apparent land surface feedback effect.