Data from NASA’s TRMM satellite and NOAA’s GOES satellites were used to survey the orographic organization of cloud precipitation in central and southern Mexico during the monsoon with two main objectives: 1) to investigate large-scale forcing versus local landform controls, and 2) to compare the results with previous work in the Himalayas. At large scales, the modes of spatial variability of cloudiness were estimated using the empirical orthogonal function (EOF) analysis of GOES brightness temperatures. Terrain modulation of synoptic-scale high-frequency variability (3–5- and 6–9-day cycles normally associated with the propagation of easterly waves) was found to cause higher dispersion in the EOF spectrum, with the first mode explaining less than 30% of the spatial variability in central and southern Mexico as opposed to 50% and higher in the Himalayas. A detailed analysis of the first three EOFs for 1999, an average La Niña year with above average rainfall, and for 2001, a weak La Niña year with below average rainfall, shows that landform (mountain peaks and land–ocean contrast) and large-scale circulation (moisture convergence) alternate as the key controls of regional hydrometeorology in dry and wet years, or as active and break (midsummer drought) phases of the monsoon, respectively. The diurnal cycle is the dominant time scale of variability in 2001, as it is during the midsummer drought in all years. Strong variability at time scales beyond two weeks is only present during the active phases of the monsoon. At the river basin scale, the data show increased cloudiness over the mountain ranges during the afternoon, which moves over the low-lying regions at the foot of the major orographic barriers [the Sierra Madre Occidental (SMO)/Sierra Madre del Sur (SMS) and Trans-Mexican Volcanic Belt (TMVB)], specifically the Balsas and the Rio de Santiago basins at nighttime and in the early morning. At the ridge–valley scale (∼100–200 km), robust day–night (ridge–valley) asymmetries suggest strong local controls on cloud and precipitation, with convective activity along the coastal region of the SMO and topographically forced convection at the foothills of headwater ridges in the Altiplano and the SMS. These day–night spatial shifts in cloudiness and precipitation are similar to those found in the Himalayas at the same spatial scales.
In mountainous regions, the relation between landform and the space–time characteristics of precipitation over a wide range of scales (i.e., precipitation organization) plays an important role in regional water resources, hydropower harvesting, landscape hydro-ecology, and agricultural activities in general (Bandyopadhyay 2001). The fundamental hypothesis that landform provides the stage and the organizing controls of clouds and precipitation processes has been investigated by many (Bhushan and Barros 2007; Xie et al. 2006; Fuhrer and Schär 2005; Kirshbaum and Durran 2005a, b; Sato and Kimura 2005; Barros et al. 2004; Lang and Barros 2002; Berbery 2001; Miniscloux et al. 2001; Barros et al. 2000; Tucker 1993; Landin and Bosart 1989; Riley et al. 1987, among others). The working premise is that a characterization of orographic precipitation regimes can be progressively achieved through the recent progress in remote sensing observations [e.g., from satellites such as the Tropical Rainfall Measuring Mission (TRMM), and the National Oceanic and Atmospheric Administration (NOAA) Geostationary Operational Environmental Satellite (GOES) series, and others] that make it possible to monitor precipitation processes in regions of complex terrain, such as the Western Cordillera of North America, where ground-based data are scarce because of the remoteness and limited access to these locations.
Although the measurement and retrieval uncertainty of remote sensing estimates of clouds and precipitation in regions of complex terrain can be large (e.g., Barros et al. 2006), satellite-based observations are currently the most uniform and accessible data available and, more importantly, they provide a unique macroscale view of precipitation patterns that cannot be achieved with point-based observations.
The specific objective of this work is twofold. First, we examine the synoptic and local processes responsible for precipitation over Mexico during the northern summer. The motivation is to track the evolution of the organization of convective activity over parts of western, central, and southern Mexico and to differentiate between those modes of variability controlled by large-scale circulations and those controlled locally and regionally by landform. For this purpose, cloudiness data from the GOES satellite series were used to estimate the space–time modes of the variability of precipitation. Following the work of Barros et al. (2004), the working assumption is that cloudiness fields can be used as a proxy of potential rain-producing weather systems (e.g., cloudiness is viewed as a dynamic tracer of the space–time variability of rainfall). Second, we contrast the results of this study with previous work in the Himalayas (Barros et al. 2004) for an intercomparison of the space–time fingerprints of terrain controls on the space–time organization of tropical monsoon rainfall in mountainous regions.
The Himalayas can be described as a “continental” orographic monsoon regime because of the affect of the large landmass that separates the mountains from the Bay of Bengal to the east and the Arabian Sea to the west (Lang and Barros 2002). In Mexico, however, the mountains are aligned along the coast, with minimal land–atmosphere interactions with approaching air masses prior to reaching the mountains, and thus it can be described as a “maritime” orographic regime, in the sense that rainfall events are directly dependent on the adjacent ocean. Another difference between the two regions is that the synoptic flow from the Indian monsoon impacts the Himalayas from a predominantly southerly-southeasterly direction perpendicular to the mountain range and unhindered by other significant topographic features. Interactions between the topography and the large-scale flow are more complicated over Mexico as a result of the geographic layout of the mountains, which are impacted by the flow from the Gulf of Mexico and the Caribbean to the east and the ITCZ and the eastern Pacific warm pool to the south and west, respectively.
The large-scale context of regional climate in central Mexico is presented in section 2. The remote sensing data used in this study is described in section 3. Section 4 reports the analysis of large-scale space–time variability of cloud (and precipitation) fields from IR measurements of cloud-top temperatures obtained from GOES. Section 5 focuses on the affect of landform controls on the diurnal cycle of clouds and precipitation. Final conclusions are summarized in section 6.
2. Large-scale context
The utility of studying tropical orographic precipitation regimes over Mexico stems from the landform presenting an intricate arrangement of mountain ranges orientated latitudinally [the Sierra Madre del Sur (SMS) and the Trans-Mexican Volcanic Belt (TMVB)] and longitudinally [the Sierra Madre Occidental (SMO) and the Sierra Madre Oriental (SMOr)] in a monsoon region.
The interactions between the various mountain ranges and large-scale flow produce strong regional divisions in Mexico (Fig. 1). Figure 1a shows the North American Regional Reanalysis (NARR; Mesinger et al. 2006) average horizontal 850-hPa wind speed and direction over Mexico on 31 June 1999. The wind pattern is characteristic of the active phase of the monsoon, and it is similar to the 1968–88 composite generated from the National Centers for Environmental Prediction (NCEP) reanalysis reported by Higgins et al. (1999). The formation of the low-pressure trough associated with the North American monsoon (NAM) is seen along the SMO of western Mexico as well as the divergence of easterly flow from the Gulf of Mexico near the eastern Mexican coast (Berbery 2001).
Strong convergence over the SMO including moisture flux from the Gulf of California and easterly flow from the Gulf of Mexico are typical circulation features of NAM, and the effects on local weather have been the focus of several studies in northwestern Mexico, more recently during the North American Monsoon Experiment (NAME; Reyes and Cadet 1988; Douglas et al. 1993; Negri et al. 1993, 1994; Farfán and Zehnder 1994; Douglas 1995; Adams and Comrie 1997; Douglas and Englehart 1998; Higgins et al. 1999; Berbery 2001; Gutzler 2004; Gochis et al. 2006; Lang et al. 2007). NAME was an intensive field campaign conducted using an extensive network of rain gauges, weather stations, enhanced upper air soundings, ground-based radars, and aircraft measurements to improve precipitation forecasts (Gochis et al. 2004; Higgins et al. 2006).
During the NAM, precipitation occurs along the upper western slopes of the SMO in the late afternoon (Reyes and Cadet 1988; Douglas et al. 1993; Negri et al. 1993, 1994; Farfán and Zehnder 1994; Garreaud and Wallace 1997; Berbery 2001; Gochis et al. 2004), but there is a lack of convective activity over eastern and central Mexico (Douglas et al. 1993; Berbery 2001). Because of the dry conditions over central and eastern Mexico and the significant moisture convergence over the SMO (Berbery 2001), the Gulf of California and the eastern Pacific Ocean to the west and south of Mexico, respectively, are key sources for moisture during the monsoon, whereas the contribution of the Gulf of Mexico to NAM is less clear (Douglas et al. 1993; Stensrud et al. 1995; Adams and Comrie 1997). There is a divergence pattern over the eastern coast of Mexico associated with the easterly flow from the Gulf of Mexico (Fig. 1a). The upper branch is related to the Great Plains’ low-level jet in North America, whereas the southern branch affects central and southern Mexico near the Isthmus of Tehuantepec (IOT) and along the SMS. Bhushan and Barros (2007) describe a numerical study of a strong monsoon storm originating in the Gulf of Mexico that moves westward and interacts with terrain features in central and southern Mexico, leading to significant cloudiness and precipitation along its northwest–southeast path and eventually reaching the SMO.
The ITCZ also plays an important role in the climate of Mexico during summer. Greater amounts of rainfall occur in southern Mexico as the ITCZ assumes a more northerly position during the Northern Hemisphere summer that is consistent with different monsoon regimes above and below 26°N (Hu and Feng 2002). This northerly position combined with modulation by easterly waves from the Caribbean Sea and the lower branch of easterly flow from over the Gulf of Mexico have a direct influence on the SMS, initiating precipitation over the southern slopes and northward over the TMVB and parts of central Mexico. Precipitation peaks in June and September when convective activity and easterlies are strongest, with a break period (i.e., the midsummer drought) in late July and early August (Magaña et al. 1999; Curtis 2004) as a result of the presence of an anticyclone southwest of Mexico over the ITCZ. Figure 1b shows the NARR (Mesinger et al. 2006) average horizontal 850-hPa wind speed and direction over Mexico on 31 July 1999 during the break phase of the monsoon. Note the northwesterly and westerly flow over the SMO and SMOr, respectively, whereas the strong easterly flow extends well into the Pacific in the 10°–20°N band. Also seen is a drastic weakening in the low-level trough over the SMO, especially when compared to late August (Fig. 1c), when the large-scale circulation recovers the broad features characteristic of the active phases of the monsoon.
The propagation of easterly waves (periods of 3–5 days and longer) from across the tropical Atlantic to the eastern Pacific has also been found to affect precipitation patterns over parts of western Mexico. Shapiro (1986) suggested that the fast westward propagation of these synoptic-scale disturbances could be associated with strong vertical coupling from diabatic heating. Using a simple model and comparing results against climatology to support this hypothesis, Shapiro et al. (1988) demonstrated that the westward tilt combined with the height of the easterly waves extends down from 200 to 700 hPa at higher latitudes (e.g., 19°N in their modeling experiments), heights that are consistent with local terrain elevations. Although Thorncroft and Hodges (2001) identified a limited number of African easterly waves that cross over to the Pacific above 15°N, they also found a high incidence of cyclonic activity associated with easterly wave activity at the southern and western edges of the SMO and TMVB, respectively. Their results support the idea that interactions between terrain heating and synoptic-scale disturbances not necessarily associated with NAM could play an important role in explaining the temporal variability of precipitation in central and southern Mexico. In support of the observations above, Lang et al. (2007) also found that precipitation over northwest Mexico during NAME is correlated with the passage of easterly waves.
Observations from the TRMM precipitation radar (PR) are used to estimate the intensity and distribution of rainfall in three-dimensional space inside a storm cloud. The horizontal resolution of the PR is 5 km, with a total swath width of 247-km postboost from the TRMM satellite in August 2001 corresponding to 4 and 220 km, respectively, preboost (available online at http://trmm.gsfc.nasa.gov). The TRMM Microwave Imager (TMI) is a passive microwave sensor that provides observations to derive quantitative estimates of water vapor, cloud water, and rain rate in the atmosphere. The TMI operates at five distinct frequencies, the highest being 85.5 GHz (5 and 5.1-km ground resolution pre- and postboost, respectively). At this frequency, the radiation emission from rainfall and cloud drops is minimal compared to the ice scattering effects from precipitation-sized ice, which results in a depression of brightness temperatures proportional to factors including ice particle number concentration, density, and size distribution. The presence of ice detected in this way is viewed as in indicator of deep clouds, and thus a proxy of the presence of convective activity. To remove surface polarization effects, polar-corrected brightness temperatures are used (Spencer et al. 1989) in an algorithm developed by Nesbitt et al. (2000) that classifies precipitation features into relatively shallow rainfall events [precipitation feature 1 (PF1)] or strong convective systems (PF2) based on these data. The spatial location of these features is represented by the location of their geometric centroid. This algorithm was used to classify precipitation events over western and central Mexico during seven years of the monsoon (1998–2004).
Another dataset used was NOAA’s GOES series of satellites, which allows for the constant monitoring of convective development and movement. Channel 4 data (10.2–11.2 μm) from the GOES imager (an imaging radiometer) provides a measurement of cloud-top brightness temperatures with an absolute accuracy up to 1 K, and horizontal and temporal resolutions of 4 km and under 26 min, respectively. The temporal resolution of 3 h used in this study is insufficient to monitor the evolution of specific convective events, but it does allow for tracking the propagation of areas of convective activity as the diurnal cycle progresses. For years prior to 2003, GOES-8 data were used, which was positioned over the equator at 75°W, whereas GOES-12 data, at the same location, were used for 2003 and 2004. GOES data used in the current study were obtained from NOAA’s Comprehensive Large Array-data Stewardship System Web site (available online at www.lass.noaa.gov).
The study region used for TRMM and GOES data analysis is shown in Fig. 1. Data for this and all other topographic maps were obtained as digital elevation maps (DEMs) from the Global 30 Arc-Second (1 km2) Elevation Dataset (GTOPO30) of the U.S. Geological Survey (USGS).
4. Large-scale space–time variability
a. Infrared brightness temperatures
With the purpose of displaying the entire set of GOES IR data over one rainy season, two Hovmöller diagrams of cloud brightness temperatures taken at three-hour intervals and averaged latitudinally from 16° to 22°N and from 22° to 28°N, respectively, over the longitude range of the study region were constructed for June, July, and August (JJA) of 1999 (Fig. 2). Note the westward propagation of convective activity as well as the diurnal cycle. During the first week of June, the initiation of convection over the TMVB and the SMS (16°–22°N) occurs east of 104°W longitude, whereas convective activity over the SMO (22°–28°N) does not occur until mid-June. This evolution pattern is in agreement with the results of Higgins et al. (1999). Enhanced convective activity persists over all mountain ranges through late July.
Cloudiness decreases dramatically over the TMVB and the SMS from late July to mid-August but remains active over the SMO. This period is synonymous with Mexico’s midsummer drought (MSD), effectively a break phase of the NAM. Convective activity resumes over the entire region after mid-August. In contrast with the dynamic succession of 10–20-day periods of active and break phases of the monsoon in the northern India convergence zone (NICZ; see Webster et al. 1998; Barros et al. 2004; Chiao and Barros 2007), the NAM exhibits two long active periods that are interrupted in early July by the MSD that vary in duration from a few days in northern and central Mexico to several weeks in southern Mexico and Central America (Magaña et al. 1999). The MSD is linked to the weakening of land–sea temperature (and moist static energy) contrasts and subseasonal changes in the spatial distribution of oceanic convective activity, which explains the northward decrease of the duration of the MSD (Curtis 2004).
An assessment of interannual variability among specific rainy seasons that exhibited contrasting hydrometeorological behaviors was also pursued. Whereas a detailed study on interannual variability would require data over several decades—not possible with the record length of using only the remote sensing data available today—the comparison between contrasting hydrometeorological years provides some new insights into our understanding of wet and dry hydrological regimes in the region. Previously, El Niño and La Niña events have been associated with dry and wet summer monsoons, respectively, in Mexico (Douglas and Englehart 1998; Higgins et al. 1999; Englehart and Douglas 2002; Brito-Castillo et al. 2003), especially during the warm phase of the Pacific decadal oscillation (PDO). To illustrate this relationship and to compare wet versus dry years using remote sensing data, the years 1998 (strong El Niño), 1999 (average La Niña), and 2001 (weak La Niña) were selected for detailed analysis. During this period, however, the influence of the PDO from 1998 to 2001 is small because of the lack of a strong warm phase (available online at http://jisao.washington.edu/pdo/). Higgins et al. (1999) suggest that low SST anomalies (SSTAs) during a La Niña event produce an enhanced land–sea contrast along the coast of Mexico favorable to increased cloudiness and rainfall. Figure 3 displays the average June SSTA fields for 1998, 1999, and 2001. Indeed, there are significant differences in the eastern Pacific SST above 20°N with strong negative anomalies (strong land–sea contrast) in 1999. Below 20°N, both the eastern Pacific and the Caribbean exhibit lower SSTs in 1998 and 2001, though the differences are much smaller than for 1999 above 20°N. Whereas the 1998/99 contrast between an El Niño and a La Niña year is expected, the large variability between 1999 and 2001, two years in the same ENSO phase, may also reflect from interannual variability in easterly waves and low-frequency modulation of convective activity and precipitation processes by the North Pacific oscillation (NPO), or more generally by tropical modes of decadal variability seen elsewhere in North America and globally (Gershunov and Barnett 1998; Hu and Feng 2002; Pierce 2005; Douville et al. 2006). Understanding this interplay of interannual and decadal variability on monsoon precipitation over Mexico requires further research, and thus the remainder of this section will only focus on 1999 and 2001.
In Fig. 4, GOES IR brightness temperatures provide a clear comparison of late afternoon cloud formation [1745 local standard time (LST)] averaged over June of 1999 and 2001. The location of the study region used for this and all other GOES analysis is illustrated by the large square in Fig. 1. Note the localized cold cloud tops almost perfectly aligned with the topographic divides along the SMO, SMS, and TMVB during 2001, which are consistent with convective activity modulated by the terrain in the absence of strong synoptic forcing (and moisture convergence), whereas there is more widespread cloud cover seen during 1999. A survey of seasonal rainfall estimates from the TRMM 3B43 merged products indicates much lower rainfall in 2001 than in 1999 (available online at http://trmm.gsfc.nasa.gov/3b43.html) as shown for the month of July in Figs. 5a and 5b. The 1999–2001 contrast can be explained by the regime proposed by Hu and Feng (2002), which is that cooler SSTs coexisted with a northward shift of the ITCZ that resulted in more rainfall in south-central Mexico, a scenario that matches the rainfall and SST anomaly fields in 1999.
b. EOF analysis
EOF (empirical orthogonal function) analysis is widely used in climate research to identify stationary physical modes of variability (North et al. 1982; Behera et al. 2003, among many others). Here, we rely on EOF analysis to investigate the spatial variability of cloudiness anomalies using the covariance method (Von Storch and Zwiers 1999). Following closely the work of Barros et al. (2004), anomaly time series of the three-hourly IR fields S′t were calculated first for each pixel (S′t = St − μ̂, where μ̂ is the mean field over the period of analysis). The anomalies can be expressed as a summation of a finite series of independent patterns:
where ei is the estimated EOFs, and αi,t are the EOF coefficients (ECs) known as principal components. The EOFs are orthogonal vectors and the lag-0 sample cross correlations of the ECs are all zero. The ECs can thus be obtained by projecting the anomalies onto the EOFs and minimizing the squared difference between the right- and left-hand sides of Eq. (1). This is equivalent to calculating the eigenvalues (ECs) and unit eigenvectors (EOFs) of the covariance matrix of the anomalies. Each eigenvalue (EC) corresponds to the fraction of total variance of the anomaly fields that is described by the corresponding eigenvector (EOF). By convention, the EOFs are ranked according to the magnitude of the corresponding EC values, that is, the first EOF is the eigenvector corresponding to the largest eigenvalue (first EC); the second EOF provides the mode of variability with greatest variance that is not correlated to the first EOF, and so on. Results are shown for the anomaly fields of GOES cloud brightness temperatures for the months of June and July of 1999 and 2001.
The distribution of the relative contribution of the first 10 EOF modes to the overall spatial variance (i.e., the covariance matrix spectrum) is shown in Figs. 6a and 6b. Results of Barros et al. (2004) for June 2000 in the Himalayas are also shown, so a simple comparison can be made between the two regions. The percentage of spatial variance over Mexico, explained by each mode during June and July, is similar for both years. Note the standard error scales with (2/N)1/2 (North et al. 1982) and with 240 samples per month (e.g., 240 = 30 days × 8 images per day) correspond to 6.5%. It is not possible, therefore, to establish unambiguously the independence of the fourth and higher modes.
Previously, Englehart and Douglas (2002) used 50 yr of monthly rainfall data for June–September, collected at several stations throughout Mexico. They found that the first five modes explained about 38.1% of the variance, which is comparable to the combined variance explained by modes 1–3 in our analysis. The first mode for each month, on the other hand, explains less variance than what was calculated for the Himalayas, with July showing a relative difference of nearly 50%. The mesoscale topographic features in Mexico are substantially more complex than in the Himalayas, especially the particular geometric layout of the three mountain ranges. Direct oceanic influence from both eastern and western seaboards results in more complex interactions among low-frequency synoptic-scale disturbances and topography, which are reflected in more complex patterns of convective activity.
June was chosen to display the spatial patterns for the most significant EOFs. EOF1–EOF3 for June of 1999 and 2001 are shown in Fig. 7. The EOF1 for both years (Figs. 7a and 7b) shows sharply contrasting patterns: the EOF for 1999 consists of one large feature that spans the entire region from the ocean to the eastern continental divide in central Mexico; the EOF for 2001 shows strong spatial correlation with the layout of the terrain, and in particular the alignment of the peak amplitudes with the mountains ridges, thus exhibiting characteristics associated with daytime heating and cooling patterns (e.g., land–ocean contrast and solar forcing and land–atmosphere interactions). We propose that the EOF1 pattern in 1999 is indicative of strong large-scale moisture convergence to the region from the eastern Pacific and the ITCZ, whereas the EOF1 in 2001 is indicative of landform controls on the organization of cloudiness and precipitation in the presence of weak large-scale forcing.
The EOF2s (Figs. 7c and 7d) show a northwest–southeast dipole separating the NAM signature from the signature of land modulation given off by convection over the SMS and TMVB. Similar patterns of convective organization and landform dependency are observed in 1999 and 2001, a result that is similar to other monsoon regions (e.g., the Himalayas). However, again there is lack of a distinct land–ocean boundary in the 1999 features as opposed to 2001. Dramatic differences can be seen in the third EOF. In 1999, EOF3 (Fig. 7e) shows a strong north–south dipole with centers over the eastern Pacific Ocean near the ITCZ and over northern and northwest Mexico, whereas the pattern is consistent with topographic controls in 2001 and the dipole is oriented in the east–west direction (Fig. 7f). Overall, these results suggest that during La Niña events, there is a strong influence of the eastern tropical Pacific Ocean, and the entire region tends to have a more active monsoon season, whereas during a weak La Niña landform controls dominate and strong topographically forced convective activity is highly constrained to the mountains of south central Mexico. Further analysis (not shown) of the EOF1s for dry hydrometeorological regimes (e.g., 1998 and during the midsummer drought) shows that the spatial patterns are consistent with those in June 2001.
The wavelet power spectrum of the seasonal time series of the EOF1 coefficients shows that variability at time scales longer than the diurnal cycle is confined to the later part of second active phase of the monsoon in 2001, whereas it is distributed throughout the season in 1999 (Figs. 8a and 8b). There is a peak of variability at 3–5 days in 2001, which is much stronger and appears at the 6–9 day time scale in 1999. By contrast, the diurnal cycle is much stronger in 2001 than in 1999 compared to other time scales. Again, this behavior is consistent with the diurnal cycle of cloudiness in regions of complex terrain in the absence of strong large-scale moisture convergence. Finally, the amplitude of the power spectrum at longer time scales (30–60 days) in 1999, which is absent in 2001, further indicates the dominant role of large-scale moisture convergence from the eastern Pacific and the ITCZ. The longer time scales of variability in 2001 occur in the latter portion of the monsoon as it retreats, and they do not exceed the 16-day scale associated with the intramonthly Rossby mode.
The subseasonal variability at longer time scales may reflect different modes of interaction between large-scale moisture convergence patterns, easterly waves, and the regional orography, with the latter dominating in the weak La Niña year of 2001. Note the diurnal cycle in the wavelet spectra is also stronger in 1999 during the July–August period, coinciding with the midsummer drought. Overall, during dry hydrometeorological regimes, the diurnal cycle is stronger and the intramonthly variability associated with longer time scales (16 days or longer) is low.
In synthesis, two hydrometeorological regimes are observed: northwest Mexico controlled by NAM dynamics and south and central Mexico affected by the ITCZ and the Gulf of Mexico. This analysis concurs with Hu and Feng (2002) and Gochis et al. (2006), except that Gochis et al. (2006) used streamflow data to determine there were three precipitation regimes, including one in eastern Mexico. The latter findings may have included more uncertainty with regard to the spatial patterns as a result of the lack of data over the SMO and the interpolation of sparse data. As the length of the remote sensing data record increases to support true climatological studies over the study region, a more extensive analysis on the important mechanisms responsible for the spatial patterns of convection during wet and dry seasons should be performed that can elucidate interannual and decadal variability. Nevertheless, this study clearly indicates that during a wet hydrometeorological year, the spatial variability of cloudiness and precipitation during the monsoon are controlled by large-scale circulation patterns interacting with the terrain, whereas during a dry hydrometeorological year landform is the key agent modulating cloudiness and rainfall.
5. Landform controls on the diurnal cycle of precipitation
a. Precipitation feature analysis
Barros et al. (2004) compared the algorithms of Nesbitt et al. (2000) using TRMM data against an IR algorithm to detect and classify convective systems using infrared satellite data (e.g., Evans and Shemo 1996) and established that the TRMM-based approach is unique in its ability to detect small-scale convective activity (∼100 km2) in mountainous regions in contrast with the infrared-based approach (∼10 000–100 000 km2). Similar to Barros et al. (2004), TRMM data, specifically PR near-surface precipitation reflectivities and TMI 85.5-GHz polar-corrected temperatures (PCTs), were used to detect precipitation in areas of shallow versus deep convection. The location of the study region used for TRMM PF analysis is represented by the small square in Fig. 1 and was chosen because of the dramatic contrasts in landform, including an enclosed deep-valley feature (the Balsas basin) and a convergent network of smaller tributary valleys to the Rio Grande de Santiago (RGS). The region is also close to Mexico City (19°26′N, 99°08′W). A disadvantage of using TRMM data in mountainous regions is that the low sampling rate (i.e., long satellite revisit times) is compounded because precipitation at high elevations, especially light rainfall, tends to be confounded with other effects (e.g., ground clutter, partial beam filling at the edges of clouds, blowing snow, snow on the ground, and others) as shown in Barros et al. (2006). This problem was addressed by creating monthly composites of all the PFs detected over several years (1998–2004), which allow inquiry of subseasonal time scales but interannual variability cannot be assessed.
The observed values of PR reflectivities and PCTs were used to identify PFs according to the following criteria: for PF1, PR ≥ 20 dBZ in at least four contiguous data bins (area ≥ 75 km2); and for PF2, PR ≥ 20 dBZ with PCTs ≤ 250 K in at least one bin, or PCTs ≤ 250 K in four contiguous data bins (Nesbitt et al. 2000). In essence, as discussed in section 3, classification into PF1 or PF2 reflects the differences in the magnitude of ice scattering. PF1s apply to precipitation in areas of shallow or no convection, while PF2s apply to areas of enhanced convection and intense precipitation. Cecil et al. (2005) and Lang et al. (2007), using TRMM data globally and ground-based radar observations, respectively, during NAME, used this algorithm as a basis for classifying precipitation features under different conditions. They reported that the areal extent of the overwhelming majority of features (>99%) is less than 1000 km2 with maximum characteristic lengths below 30 km. To capture both the location and characteristic length scale of precipitation features while capturing a number large enough to produce robust statistics, we cluster the features using a 0.5° × 0.5° grid resolution, roughly about 50 km to organize and classify the precipitation features on the terrain.
Figures 9 and 10 show the centroid positions of all PF1s and PF2s detected between June and August from 1998 to 2004 at time intervals (a) 0000–0600 UTC and (b) 1200–1759 UTC, respectively. Note the relationship between the elevation of the terrain and the location of the PFs as well as the contrast between daytime and nighttime. During the midday hours, the PFs cluster along the valley ridges and over steep mountain slopes and are absent in the deep valleys. During the early morning hours, there is a significant decrease in precipitation features’ density, although relatively more PFs occur in the valleys and the foothills of coastal slopes and fewer along the ridges. These phenomena are most prominent in the valley of the RGS, the Balsas basin, and the surrounding mountains and agree with the preliminary results of NAME elsewhere as described by Gochis et al. (2004). Barros et al. (2004) noted the PF2s of central Nepal tend to specifically cluster along the ridgelines during the afternoon, whereas the PF1s are more uniformly distributed on valley walls and mountain slopes. These results are likely because of the combined effect of large-scale moisture pathways, which was described earlier, and the diurnal cycle of local circulations: mountain (valley) winds and sea (land) breezes near the SMO and SMS as the land temperature increases (decreases) during the daytime (nighttime), respectively, compared to nearby valleys and water bodies (illustrated in Fig. 11).
b. PF concentration versus landform
An expanded region of study that includes all mountain ranges was selected so the combined effect of large- and small-scale forcing can be examined over a wider range of elevations and locations. Based on elevation and location (landform), the study region was divided into 14 classes that describe the major physiographic regions, which were divided into the prevalent altitudinal gradients as shown in Fig. 12. This number of classes distinguishes among the relative position of the ridges in eastern (Sierra Madre Oriental), western (Sierra Madre Occidental), and southern Mexico (Sierra Madre del Sur) and the central highlands (Mexican Altiplano) as well as the Balsas and Rio Grande de Santiago basins.
Figures 13a and 13b display the number of PF1s observed in each class during the time interval of June–August from 1998 to 2004 in the late afternoon (1500–1600 LST) and early morning (0000–0300 LST). During the afternoon, the maximum number of PF1s can be seen in class 9–class 11 (1500–3000 m) on the Mexican Altiplano. A high number of occurrences is also detected along the southern slopes of the SMS (class 3–class 5), at midelevations on the western slopes of the SMO (class 7–class 8), and in the valley of the RGS (class 1). A general increase in PFs with elevation can be seen at all locations in the afternoon, which is consistent with the topographically modulated radiative forcing of convection, whereas during nighttime the features tend to cluster along midelevation protruding ridges (class 5 and class 13) in the SMO and SMOr, respectively, similar to the Himalayas (Barros et al. 2004).
Although the Balsas basin (class 2) experiences precipitation in the afternoon, the TRMM data shows that the maximum number of PF1s is found here in the early morning hours, whereas precipitation elsewhere is largely suppressed. This phenomenon is similar to the foci of heavy precipitation detected by Barros et al. (2004) against the foothills of the Himalayas and along enclosed deep valleys in the Tibetan Plateau. The distribution of PF2s (Figs. 13c and 13d) follows the same pattern, the major difference being a decrease in the number of PF2s with elevation in the SMS and for elevations below 1500 m. There is also a greater percentage of PF2s occurring in the RGS (class 2) against the surrounding foothills during nighttime. The valley of the Rio Grande de Santiago is open to moisture from the Pacific Ocean during the daytime and is susceptible to nighttime moisture convergence because of drainage valley flows. The diurnal cycle of RGS precipitation closely trails the diurnal cycle of land–sea breezes along the coast and mountain–valley circulations near the headwaters (e.g., Bhushan and Barros 2007). In contrast, the Balsas basin is enclosed by mountain ranges to its north and south and must depend on katabatic valley flow during the nighttime for precipitation to occur (class 1).
Because TRMM data were used to illustrate the effects of large- and small-scale forcing mechanisms on Mexico precipitation, data from the GOES satellites were used to monitor the formation and propagation of convective activity and to calculate spatial modes of cloudiness variability (see section 4). Following the work of Barros et al. (2004), the working assumption is that cloudiness works as a spatial proxy of precipitation processes. Figure 14 shows the brightness temperature fields at 1745 UTC and 0245 UTC averaged for July during the peak of NAM from 1998 to 2004. Bhushan and Barros (2007) used a coupled land–cloud model at high spatial resolution and showed that simulated moisture convergence patterns are consistent with the space–time distribution of satellite observations of clouds in the region overall and with precipitation features. The GOES data in Fig. 14a show three lines of enhanced convective activity that are linked to the precipitation features observed during the daytime in regions over the SMO, SMS, and the TMVB, which is consistent with the left-hand side panels of Figs. 13a and 13b. Figure 14b reveals the cloudiness associated with topographically forced convective activity responsible for the PFs in the early morning hours in the Balsas basin and in the Rio Grande de Santiago as well as offshore along coastal concavities favoring increased local convergence, whereas precipitation is suppressed elsewhere, which is consistent with the right-hand side panels of Figs. 13a and 13b.
6. Discussion and conclusions
Remote sensing data from the National Aeronautics and Space Administration’s (NASA) TRMM satellite and NOAA’s GOES satellite network were used to examine the influences of landform and large-scale circulation features on the spatial and temporal variability of cloudiness and precipitation in central and southern Mexico. Areas experiencing either shallow rainfall (PF1) or enhanced convective activity (PF2) were identified in central Mexico for the rainy seasons (June–August) of 1998–2004. All precipitation features (PFs) were predominantly found along the valley ridges during the afternoon hours, with PF2s being clustered closer to the peaks of the ridges than the PF1s. During the night, PFs are limited to the deepest valley features of the region, the Balsas basin and the Rio Grande de Santiago (RGS). The nighttime PFs are caused by moisture convergence as a result of sinking valley winds (Balsas basin) and land breezes (RGS). The sea breeze transports moisture to the RGS basin during the daytime, leading to an enhancement of convective activity and a peak in the number of precipitation features detected in the late afternoon similar to those at high elevations.
Remote sensing data from the GOES were used to track the initiation and propagation of areas of cloudiness and to determine whether they are associated with local landform features or large-scale circulation (global and synoptic). The data show increased cloudiness over the Sierra Madre Occidental (SMO), the Sierra Madre del Sur (SMS), and the Trans-Mexican Volcanic Belt (TMVB) during the afternoon and over the Balsas Basin and the RGS during the early morning. Organizing the results into a Hovmöeller diagram permitted the tracking of the evolution of the diurnal cycle and the westward propagation of convective activity over Mexico, including a 1–2-week lag between the NAM onset in northwest and southern Mexico. In addition, a period of high brightness temperatures in the Hovmöller diagram revealed the midsummer drought.
An assessment of interannual variability was also performed through the intercomparison of years of contrasting rainfall over Mexico and SSTAs off the west coast. Cooler SSTs corresponded with higher convective activity as a result of the greater land–sea contrast during the daytime, but the geographical patterns are substantially different between 1999 and 2001, reflecting the relative influence of large-scale forcing versus regional landform controls. Albeit over a limited number of years (1998–2004), our analysis shows that whereas temperature contrasts between land and adjacent ocean waters are directly related to cloudiness and precipitation, the interannual variability in the magnitude and spatial arrangement of these contrasts (and therefore precipitation) cannot be explained simply by ENSO activity. Instead, interannual variability in the interaction of synoptic-scale disturbances (e.g., easterly waves) with orography and low-frequency variability (e.g., tropical decadal variability; North Pacific oscillation) may play an important part in modulating decadal precipitation variability (20–30 yr time scales; see also Hu and Feng 2002). This nonstationarity begs the question of what is the role of low-frequency variability (e.g., tropical decadal variability) in explaining long-term precipitation variability that is relevant for water resources management purposes (20–50 yr). A detailed study of interannual variability using only TRMM and GOES data is not currently possible because of the short history of the datasets; therefore, other sources of data are required for these studies.
The ultimate goal is to develop a remote sensing basis for freshwater prospecting from space. Our focus in this manuscript is on elucidating the role of topography in organizing cloudiness and precipitation over Mexico, a maritime monsoon region, and how this compares to a previous empirical study in the Himalayas, a continental monsoon region (Barros et al. 2004). At large scales, the spatial variability of cloudiness was estimated through the empirical orthogonal function (EOF) analysis of the GOES brightness temperature profiles. In contrast with the succession of 10–20-day active and break phases of the monsoon on the Indian subcontinent, subseasonal variability of the monsoon in central Mexico is characterized by two long active phases with high-frequency variability interrupted by one break period, the midsummer drought in July and August. The predominance of high-frequency variability (a succession of 3–5- and 6–9-day cycles) has an impact on the spatial patterns of cloud and precipitation processes during the monsoon, which is well illustrated by the differences in the EOF spectra, with a significantly higher dispersion of the modes of variability in Central Mexico than in the Himalayas (e.g., on a monthly basis, the first EOF mode typically accounts for ∼30% vis-à-vis ∼50% of overall variance, respectively). This increase in complexity and high-frequency variability reflects the multiple pathways by which convective activity in the Caribbean Sea and the Gulf of Mexico in the east, the ITCZ and the Pacific Ocean off the western and southern coasts, respectively, and complex orographic land–atmosphere interactions affect precipitation patterns in Mexico.
The first EOF pattern in 1999, a La Niña year with above average rainfall, shows one continuous regional-scale feature that extends from central Mexico toward the ITCZ, spanning both ocean and land areas. Wavelet analysis of the first principal component in 1999 shows strong variability at the 30–60-day time scales, thus consistent with large-scale intramonthly moisture convergence events. The first EOF pattern in 2001, a weak La Niña year with below average rainfall, also displays one singular regional-feature, but it is aligned with the SMS and the TMVB. This morphology is aligned with mountain geometry and exhibits strong land–ocean contrast, which suggests dominant landform controls on the formation and organization of cloudiness. In 2001, the diurnal cycle is the dominant time scale of variability, and longer time scales do not go beyond two weeks and only so late in the monsoon season. The morphology of the second EOF patterns does not change significantly from one year to the next; thus it can be linked to landform controls on the diurnal cycle of solar radiation and topographically forced convective activity, although the land–ocean contrasts are more diffuse in 1999 likely as a result of a large-scale influence. The third EOF patterns are dipoles with strong year-to-year variations. When the phase of the La Niña is strong (i.e., 1999), the entire region tends to have a more convectively active summer season, and the dipole is oriented in the north–south direction. Whereas during a weak La Niña (i.e., 2001), strong convective activity is limited to south central Mexico, and the dipole is oriented in the west–east direction. Overall, the results support the different contributions of landform and large-scale moisture convergence controls to regional hydrometeorology in dry and wet hydrometeorological years, respectively.
An advantage of using remote sensing data over point-based measurements from field experiments is the temporal and spatial consistency over large regions and intercomparison among different hydrometeorological regimes. In addition to new findings, one contribution of this work is to show that using these data, we were able to independently confirm and use different methods results previously reported in the literature, which further indicates the great potential of remote sensing for integrative science as more data become available. Furthermore, when used with ground-based measurements, satellite data can indicate where errors and uncertainties in the field data may exist. Typical problems with precipitation and streamflow data are well identified in the literature for the region including diversity in the temporal and spatial scales of the data (Gutzler 2004), interpolation errors where data are missing (Gochis et al. 2006), and the brevity of data collection, or existing historical records (Gochis et al. 2004). Diagnostic analyses using remote sensing data as in the current study can provide valuable guidance by providing physically based spatial constraints (spatial modes of variability consistent at the regional scale) in the process of assembling these datasets.
This work was funded in part by NASA Grant NNGO04GP02G through the work of the second author and by the Pratt School of Engineering at Duke University. We are grateful to Stephen Nesbitt and Axel Graumann for their help with data manipulation and retrieval. Constructive comments and suggestions by three anonymous reviewers were very helpful in improving the manuscript.
Corresponding author address: Dr. Jason Giovannettone, Institute of Water Resources, Hydrologic Engineering Center, 609 2nd Street, Davis, CA 95616. Email: Jason.P.Giovannettone@usace.army.mil