1. Introduction

The onset of the North American monsoon system (NAMS) abruptly changes the hydroclimatology of a large region extending from northwestern Mexico to the southwestern United States (see Fig. 1). The summer season over this region is characterized by a reversal in circulation, as the midtropospheric winds shift from westerly to easterly due to the northward displacement of the subtropical ridge (Douglas et al. 1993). The circulation reversal is accompanied by an abrupt increase in precipitation. As early as May, intense rainfall begins over southern Mexico and, by July, the NAMS has migrated north into Arizona and New Mexico (Douglas et al. 1993). Most of the moisture at upper levels originates in the Gulf of Mexico and enters the region from the east, while lower-level moisture of oceanic sources originates predominantly from the tropical Pacific Ocean and the Gulf of California (Schmitz and Mullen 1996; Adams and Comrie 1997; Higgins et al. 2003). Much like the larger South Asian monsoon, the North American system is characterized by seasonal heating of the land surface lowlands and elevated areas (Tang and Reiter 1984; Adams and Comrie 1997). But, while the land surface energy budget is known to play a crucial role in the development of the monsoon, many lingering questions remain regarding the effect of land surface water and energy fluxes in monsoon initiation and sustenance (Higgins et al. 2003; Higgins and Gochis 2007). Land surface memory provides some predictive potential in the monsoon regions. In much the same way as increased Eurasian snow cover leads to a decrease in monsoon rainfall over Southeast Asia (Barnett et al. 1989), positive (negative) soil moisture anomalies in the southwestern United States, generally related to winter snowfall anomalies, suppress (enhance) summer monsoon rainfall (Gutzler and Preston 1997; Small 2001). On the other hand, soil moisture anomalies located within the NAMS domain (Fig. 1) generate a positive feedback between soil moisture and summer precipitation (Small 2001), as positive anomalies potentially enhance rainfall by decreasing boundary layer height, increasing moist static energy, and increasing instability (Eltahir 1998; Betts et al. 1996). Vegetation clearly plays an important role in controlling surface moisture and energy fluxes to the atmosphere, as areas with abundant transpiration within the NAMS domain tend to have higher net radiation (Watts et al. 2007). The relationship is further complicated by topography, as the strength of land–atmosphere coupling is modulated by the elevation of the terrain, which also controls the ecosystem type (Vivoni et al. 2007).

While most studies have focused on the effect that soil moisture has on the overlying atmospheric energy budget, surface moisture in the NAMS region also directly contributes to precipitation by providing evaporative sources for precipitable water. Bosilovich et al. (2003) used a general circulation model (GCM) with water vapor tracers to identify the source regions for the NAMS precipitation. They found that during May, June, and July continental moisture sources from Mexico are the dominant source of water vapor for the western Mexico region (roughly corresponding to the NAMS region), contributing to approximately 30% of the total precipitation within the region. Their study also concludes that the wettest monsoons have the largest continental sources, while drier monsoons have less local sources of precipitation. In this work, we will quantify precipitation recycling using the Dynamic Recycling Model (DRM) developed by Dominguez et al. (2006) [see Dominguez and Kumar (2008, hereafter Part I) for a detailed explanation]. While the water vapor tracer analysis of Bosilovich et al. (2003) includes more detailed physics than the analytical model, our work uses the higher-resolution NARR assimilated data, which provides improved estimates of land–atmosphere variables compared to the GCM and is able to resolve the Gulf of California. Our work focuses primarily on the daily-to-intraseasonal time scale in order to better understand the physical mechanisms that drive recycling spatiotemporal variability at these time scales. Furthermore, we will perform a detailed analysis of the source and sink regions of recycled precipitation within the NAMS domain.

The goal of this work is to better understand the role that precipitation recycling plays in the NAMS hydrometeorologic system. Some of the questions we would like to address are: How much of the total NAMS precipitation comes from evapotranspiration within the NAMS domain? Furthermore, as a link between the long-memory soil moisture storage and the atmosphere, what is the role of precipitation recycling in NAMS sustenance? Which are the primary source and sink regions for recycled precipitation? What is the spatial distribution of precipitation recycling? How are vegetated environments affected by recycling of moisture?

This is the second of a two-part study on precipitation recycling over North America. While the first part analyzed the central U.S. plains, a region of abundant soil moisture, here we focus on the arid and semiarid regions of southwestern United States and northwestern Mexico. As we have shown in Part I, in the central U.S. plains the dominant mechanism is a negative feedback that ensures, even when advected precipitation is low, the land continues to feed moisture into the overlying atmosphere and contribute to rainfall. The NAMS, on the contrary, shows a predominantly positive feedback where precipitation enhances evapotranspiration and subsequent precipitation of recycled origin. However, as we will show in this work, during long monsoons the NAMS region exhibits a similar negative feedback mechanism as for the central U.S. plains. During long monsoons, precipitation of recycled origin is significantly enhanced during periods of low advected moisture, thus bringing additional rainfall that creates favorable hydroclimatological conditions to the vegetated regions.

Section 2 gives a brief overview of data and methodology used in this work. The reader is referred to Part I for a more detailed description of the recycling model, the multivariate singular spectrum analysis (M-SSA) statistical technique, and the data. Section 3 focuses on the average hydroclimatology of the NAMS region and uses M-SSA to characterize the seasonal variability of different land–atmosphere hydrologic variables, including recycling. Section 4 further explores the variability within the monsoon system and the role that precipitation recycling plays in the sustenance of the NAMS. Our goal in this section is to unravel the physical mechanisms that drive precipitation recycling in the region through a detailed spatial and temporal variability analysis of the land–atmosphere variables. Summary and conclusions of the work are discussed in section 5.

2. Data and methodology

As in Part I, we use data from the North American Regional Reanalysis (NARR) (Mesinger et al. 2006) which was developed as an improvement upon the earlier National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) Global Reanalysis, and has focused significant effort on improved hydrologic modeling. Twenty-two years (1985–2006) of warm season (1 April–30 September) 3-hourly data was obtained from NARR, and daily values were later derived from this data. The NARR grid points corresponding to the NAMS region are depicted in Fig. 1. Because the assimilation process depends on the quality and quantity of observations, there is a sharp difference in the quality of the NARR data between Mexico and the United States. This can clearly be seen in the precipitation and evapotranspiration fields throughout this work. Furthermore, it is important to reiterate that NARR is an assimilated dataset and, without any climatological observations of evapotranspiration (ET), the estimates of ET in the dataset are subject to error (Nigam and Ruiz-Barradas 2006). Because we lack the data to calculate the error in NARR ET, we have estimated the uncertainty in the recycling ratio (RR) as a function of the ratio of “actual” ET over NARR-calculated ET (ETnarr) for average conditions over the NAMS region; the analysis is shown in Fig. 2. Using average NARR ET (ET/ETnarr = 1), precipitable water, and winds over the region we calculate a regional recycling ratio of around 15.6%. However, if the actual ET were 50% larger, RR would be 22.6%, while a 50% decrease in ET would give us 8.2% RR. Consequently, for the range of values in the NAMS, we have an approximate linear relationship between ET and NARR. It is important to point out that the primary focus of this analysis is to understand the mechanisms that drive variability in precipitation recycling: if NARR ET has a consistent bias over the region, this will in no way affect the variability analysis we have performed. Unfortunately, at this point there is no way to analyze the effect that nonlinear ET errors will have on our RR calculations. For this reason we underscore that the precipitation recycling analysis reflects the hydrologic cycle as obtained with the dynamically consistent NARR data.

The geographical classification system used in this research is the Level II ecological regions of North America (CEC 1997). The NAMS region, as delineated by Gutzler (2004), is composed of six very distinct Level II ecoregions (see Fig. 1) described in CEC (1997). The driest ecoregions are the Sonoran and Chihuahuan Deserts where the predominant arid-region vegetation includes shrubs, grasses, and cacti. Originally, ecoregion 10.2 included both the Mojave and Sonoran Deserts but, since the Mojave Desert receives most of its rainfall during the winter months (Warner 2004), we excluded this area. At higher elevations (1100–2500 m) we find the semiarid western Sierra Madre piedmont, where blue grama short grass is dominant except for scrublands and forests in the transition zones leading to the western Sierra Madre and upper Gila Mountains. The high mountain ranges, with elevations up to 3000 m, have temperate climates ideal for evergreen or deciduous forests, with conifers and oaks. Cloud forests are common in the upper elevations. Descending from the mountains to the lower altitudes, we find the Pacific coastal plains hills and canyons. Seasonally dry tropical forest vegetation is abundant in the region where low deciduous and subdeciduous forests dominate. This seasonally dry tropical region receives intense rainfall during the summer months and experiences dramatic green-up (Higgins et al. 2003).

3. North American monsoon

4. Recycling and the monsoon

Precipitation recycling variability in the monsoon region is inextricably linked to precipitation and evapotranspiration variability. While rainfall provides moisture for precipitation that can later be evaporated, it also increases atmospheric humidity. As low-level humidity increases, the vertical gradient of specific humidity decreases, bringing a proportional drop in evapotranspiration. Consequently, precipitation brings competing mechanisms that affect the variability of precipitation recycling: therefore, the important question is what mechanisms dominate? Figure 8 shows the average daily summer season (1 June–30 September) recycling ratio, precipitation, and evapotranspiration from 1985 to 2006. There is a statistically significant positive correlation between the recycling ratio with both precipitation (0.56 correlation coefficient) and evapotranspiration (0.63 correlation coefficient): both correlations are above the 95% threshold.

Looking back at Fig. 7, we see that the duration of the monsoon varies considerably from year to year. The summers with longest monsoons (above one standard deviation) are 1986, 1991, and 2000, while 1989, 1992, and 2005 present the shortest monsoons. Figure 9 presents the daily recycling ratio and precipitation of recycled origin for these six selected years and for the average of the 22-yr period. Precipitation of recycled origin is defined as the recycling ratio multiplied by the daily precipitation. As shown in the previous analysis, the recycling ratio and the precipitation of recycled origin are significantly enhanced as the monsoon enters the region. In agreement with the analysis of Bosilovich et al. (2003), the climatological recycling ratio shows a sharp increase in continental sources of moisture as the warm season progresses and then a slow decrease toward the end of the season. Our results are consistently about 10% smaller than in Bosilovich et al. (2003, see their Fig. 8a). This is mainly due to the larger region that they are considering; the lower resolution of the GCM analysis, which does not resolve the Gulf of California; and the assumption of a well-mixed atmosphere in the dynamical recycling model. The main inconsistencies between both studies are the months of May and June when we predict much smaller recycling ratios than in Bosilovich et al. (2003); the reason for this difference might be the larger continental region that they considered. The bottom panel of Fig. 9 also presents the average monthly recycling ratio for the region. Interestingly, the calculation of the recycling ratio using monthly average values of winds, humidity, and precipitation yields higher recycling ratios than the daily analysis. As shown in Part I, the recycling ratio is inversely proportional to wind velocity: because the monthly average dampens the wind variability in this region, the recycling ratios will be over estimated. Owing to the recycling ratio dependence on the highly variable wind velocities, a more realistic monthly recycling ratio will be obtained by averaging the daily recycling ratio values, as opposed to calculating the value of r using average monthly fields.

The summers of 1986, 1989, 1991, 2000, and 2005 show a pronounced increase in recycling in early July. For these years, the recycling ratios during the monsoon are consistently above 15% and can be as high as 25%. On the other hand, 1992 presents significant recycling ratios in the spring (late April, early May). In fact, it has the highest spring recycling of the 22-yr period, while recycling during the monsoon season is not as high as for other years.

As in the previous climatological analysis, we now perform M-SSA on each of the three years with longest monsoons and the three years with shortest monsoons. An interesting pattern is revealed by the analysis (Fig. 10). The summers of 1986, 1991, and 2000 present a characteristic pattern of rainfall and recycling comprising a precipitation peak at the beginning of the season and then a second precipitation peak toward the end of the season. Evapotranspiration continues to be high during the midsummer dry period, and we see a peak in the recycling ratio during mid-August. The selected years of short monsoon duration do not present this precipitation–evapotranspiration–recycling pattern. The bimodal distribution of summer rainfall is characteristic of the central and southern part of Mexico and most of Central America (Magaña et al. 1999), where the local name for the midsummer drought period is “canicula” or “veranillo” (which means the small dry season in Spanish). Douglas and Englehart (1996) also identified the characteristic double peak in summer precipitation using observed daily rainfall data in the Mexican state of Oaxaca. Our analysis shows that during monsoons of long duration this pattern migrates northward and can be seen throughout the NAMS domain. Furthermore, our results show that during long monsoons recycling ratios are enhanced during the drier midsummer period between precipitation peaks.

Because the actual amount of recycled precipitation is obtained by multiplying the recycling ratio by the total precipitation, the actual amount of recycled rainfall during the dry intermediate period is low. The importance of this intermediate period is that evapotranspiration from the soil and vegetation continues to provide moisture for precipitation even though the large-scale advective sources have dramatically decreased. While in the previous section we have shown that, on average, there is a positive feedback between soil moisture/evapotranspiration and precipitation (see Fig. 4 and discussion in section 3), the years of monsoon with longest duration show a different picture. They have an intermediate period of negative feedbacks; that is, reduced total precipitation promotes precipitation of recycled origin. To further explore this point, we examine other land–atmosphere variables and how they progress in relation to precipitation and recycling ratio during the 1986 long monsoon (Fig. 11). The first precipitation peak in 1986 is preceded by a temperature peak. In the midsummer drought period, we see a decrease in soil moisture with an increase in sensible heat and increased cloud base height. However, evapotranspiration slowly keeps increasing and continuously feeding into the overlying atmosphere: by mid-August the recycling ratio reaches its highest value. We can see that the peak in recycling coincides with the highest negative zonal flux anomaly, indicating that the air is mostly coming from the east. The second precipitation peak, of predominantly advective origin, comes at the end of the season during late August.

b. Moisture paths

When performing the recycling ratio calculation using the formulation of Dominguez et al. (2006), we are effectively following the atmospheric moisture along the paths defined by the u and υ velocities (where u and υ are, respectively, the zonal and meridional moisture flux divided by the total precipitable water, as shown in Part I). Throughout the trajectory, we calculate the ratio of moisture from evaporative origin (ε) to precipitable water in the column (ω), and integrate it from the time the moist air enters the region until the water precipitates (time τ). The local recycling ratio (R) at each grid point is a function of the integrated ratio:

View Expanded

The paths defined by the u and υ velocities are particularly interesting as they represent the moisture-weighted trajectory of the air. The value of ε/ω along the path will indicate the regions along this trajectory that significantly contribute to the local recycling ratio; these will be the dominant source regions of evaporative moisture.

Figure 15 presents a detailed analysis for 10 August 1986. This day was selected because it presents one of the highest recycling ratios of 1986, and the paths are less dense than for 15 August (the day with highest recycling ratios), making it easier to visualize the paths. Although we cannot make generalizations from only one day of the analysis period, Fig. 15 clearly shows that the spatial variability of evapotranspiration translates into different ε/ω ratios throughout the paths. In Fig. 15c we have selected three paths and denoted the ratio ε/ω for each grid point in the path as a shaded bar. The darkest shading denotes the regions that make the largest contribution of evapotranspired moisture. For these three paths, the air enters the region through the east, where there is little contribution of ET as the air crosses the Chihuahuan Desert: however, as the air traverses the Sierra Madre Occidental and the western Pacific coastal plains we see how evapotranspiration from these areas significantly contributes to the moisture content in the atmospheric column.

5. Conclusions

The North American monsoon system abruptly transforms the hydrology of the North American southwest. In this work, we take a set of 16 land–atmosphere variables to represent the hydroclimatology of the monsoon and use M-SSA to extract nonlinear trends in the data while at the same time selecting only the variability common to all of the variables. Our study uses the North American Regional Reanalysis (NARR) data throughout the analysis and, while this is currently the best available dataset for extended spatiotemporal analysis over North America, it is important to keep in mind that this is an assimilated dataset and hence is only as good as the underlying model and observations. The most uncertain variable in our analysis is evapotranspiration; consequently, any consistent bias in ET will translate into changes in the recycling calculations.

Our main focus is to understand precipitation recycling as part of the dynamic NAMS system. The first five M-SSA modes of average 1985–2006 daily data capture many of the characteristics of the monsoon that have been widely discussed in the literature, such as the abrupt increase in precipitation and precipitable water, the temperature peak prior to the rains, the shift from westerly to easterly winds and consequent shift in zonal moisture flux, and the increase in northerly moisture flux confined to the Gulf of California coastal region. Furthermore, we see an increase in evapotranspiration and recycling ratio and a decrease is sensible heat and the cloud base height.

The climatological analysis of NAMS precipitation recycling reveals a positive feedback mechanism between monsoon precipitation and subsequent increase in precipitation of recycled origin. The study by Koster et al. (2004) has shown that there is a large region extending from the U.S. Great Plains down to northern Mexico where there is a strong impact of soil moisture on precipitation. Our analysis over the NAMS region, which is embedded within the region delineated by Koster et al. (2004), shows that precipitation recycling is one of the mechanisms responsible for this positive feedback. In agreement with previous work, our study finds that evapotranspiration within the NAMS region significantly contributes to monsoon rainfall after NAMS onset. Recycling ratios during the monsoon are consistently above 15% and can be as high as 25%. While monsoon rainfall and evapotranspiration are predominantly located in the southwestern part of the domain, recycling is enhanced northeast of this region, indicating a relocation of soil moisture farther inland to drier regions in the north.

Focusing our analysis on the three years with longest monsoons in the 1985–2006 period, an interesting asynchronous pattern between precipitation and recycling ratio is revealed. The longest monsoons present a characteristic double peak in precipitation. Intense rainfall during late June and early July is followed by a period of dry conditions and a subsequent peak in late August precipitation. Contrary to what one might expect, recycling peaks during the intermediate dry period. Further inspection of other land–atmosphere variables during a long monsoon reveals that the period of decreased precipitation is accompanied by changes in many other land–atmospheric conditions that lead to the recycling peak. Our results present one of the feedback mechanisms modulating precipitation during monsoons of long duration. After the first precipitation pulse, atmospheric conditions are drier and evapotranspiration is enhanced during mid-August. The moisture of evaporative origin, originating predominantly in tropical forests in the southwest part of the domain, is then transported north and east where it later falls as precipitation, contributing to as much as 25% of the total precipitation in the NAMS domain. Using NDVI data, we argue that precipitation of recycled origin brings moisture to regions that would otherwise be dry and, hence, helps maintain vegetation greenness during the midsummer dry period. In this sense, precipitation recycling is a mechanism for ecoclimatological stability in the NAMS region.

As the seasonally dry tropical forests of Mexico continue to be severely degraded by human activities—particularly agriculture, cattle ranching, and timber extraction—the region faces imminent changes to existing vegetation cover and evapotranspiration regimes. According to Masera et al. (1997), the mid-1980s was a period of intense vegetation perturbation with deforestation of around 668 000 ha, 75% of which was concentrated in tropical forests. This led to a deforestation rate of 1.9% yr⁻¹ for deciduous tropical forests and 2.0% yr⁻¹ in evergreen tropical forests. Trejo and Dirzo (2000) found an annual deforestation rate of seasonally dry tropical forest in the state of Morelos, Mexico, of 1.4% yr⁻¹. What effect does land cover change in Mexico have on evapotranspiration and subsequent precipitation of recycled origin? We believe this is an intriguing question for future studies aimed at understanding the effect of land cover change on NAMS hydroclimatology.