1. Introduction
The North American monsoon (NAM) system is responsible for summer precipitation over much of western Mexico and the southwestern United States (Higgins et al. 1997; Barlow et al. 1998; Higgins and Shi 2001; Gochis et al. 2002; Li et al. 2004; Matsui et al. 2005; Castro et al. 2007). It evolves in response to the thermal contrast that develops in the spring and summer between the continent and the adjacent Gulf of California and east Pacific Ocean. The core NAM region is usually considered to be the area west of the Sierra Madre Occidental divide in western Mexico (see Fig. 1) (Adams and Comrie 1997; Berbery 2001; Gutzler 2004; Gochis et al. 2006), where 50%–80% of the total annual rainfall occurs during July–September (Douglas et al. 1993; Stensrud et al. 1995; Adams and Comrie 1997; Sheppard et al. 2002; Vivoni et al. 2008). The onset of the NAM precipitation, which usually progresses northward from southwestern Mexico in early June to northwestern Mexico by early July (Higgins et al. 1999), leads to dramatic ecosystem responses in terms of vegetation greenness (see Fig. 2) (Watts et al. 2007). The period of time for vegetation greenup is relatively short—typically a few weeks or less (see Salinas-Zavala et al. 2002; Méndez-Barroso et al. 2009), depending on the precipitation onset date and the amount of precipitation in the early monsoon season.
(a) Spatial distribution of seasonal change in LAI quantified as the mean LAI difference between June and September. Contours show the percentage of total annual precipitation occurring in the monsoon season in 1950–2000. The squares are the selected subdomains within the study area (labeled BXA–BXI). Flux towers at Rayon and Tesopaco are shown. Three latitudinal bands (I, II, and III) are shown for reference. (b) Topography and surface meteorological stations used to produce the VIC model forcings.
Citation: Journal of Hydrometeorology 13, 1; 10.1175/JHM-D-11-032.1
Photographs (left) before and (right) after vegetation greenup at an eddy covariance tower in a tropical deciduous forest in Tesopaco, Sonora, Mexico (see Pérez-Ruiz et al. 2010 for a site description). Photographs courtesy of Julio C. Rodríguez, Universidad de Sonora.
Citation: Journal of Hydrometeorology 13, 1; 10.1175/JHM-D-11-032.1
The influence of vegetation greening on land surface states and fluxes during the monsoon transition has been studied in several field experiments (e.g., Kurc and Small 2004; Scott et al. 2006; Watts et al. 2007; Vivoni et al. 2008; Pérez-Ruiz et al. 2010). These studies have quantified the significant seasonal variations of land surface conditions, including soil moisture (SM), soil temperature, albedo, evapotranspiration (ET), and net ecosystem exchange, among others, across a range of sites in the NAM region. To complement these point-scale field observations, remote sensing studies have helped establish the regional spatiotemporal trends in land surface conditions over long periods (e.g., Salinas-Zavala et al. 2002; Méndez-Barroso et al. 2009; Lizárraga-Celaya et al. 2010; Méndez-Barroso and Vivoni 2010; Forzieri et al. 2011). However, the links between vegetation and land surface states and fluxes, such as SM and ET, and their role in enhancing the predictability of seasonal and interannual conditions in the NAM region are poorly understood. Land surface models (LSMs) that adequately capture seasonal vegetation variations could serve to improve the intraseasonal predictability of ET and SM in data-sparse areas within the NAM region (Lawrence and Chase 2007; Gao et al. 2008). Furthermore, the derivation of spatiotemporal SM and ET estimates from an LSM has applications to precipitation recycling studies that have typically assumed climatological values for vegetation seasonality (Anderson et al. 2004; Dominguez et al. 2008).
Regional climate models have served to quantify land surface conditions and their impact on the NAM system itself (Matsui et al. 2005). In coupled models, however, the simulated SM and ET variations may be strongly controlled by the simulated precipitation, which has a large uncertainty. Offline LSM simulations using observed rainfall are an alternative that better constrains land surface conditions—a strategy used in land data assimilation systems (Mitchell et al. 2004). One complication, however, is that the LSMs typically use a climatological cycle of vegetation parameters (e.g., Maurer et al. 2002; Rodell et al. 2004; Zhu et al. 2007). For some applications, this may be a reasonable assumption as it approximates the seasonal vegetation controls on land–atmosphere interactions. However, in the NAM region, vegetation canopies increase rapidly in leaf area index (LAI) following the onset of monsoon precipitation, and the timing of this transition varies substantially from year to year (e.g., Forzieri et al. 2011). A poor representation of seasonal and interannual variations in vegetation dynamics may result in unrealistic simulations of the spatiotemporal ET and SM patterns. Vivoni et al. (2008), for example, found that the North American Regional Reanalysis (NARR; Mesinger et al. 2006) product contained significant biases in ET and SM estimation in the NAM region, in part because of the use of climatological vegetation parameters derived from the normalized difference vegetation index (NDVI) data of Duchemin et al. (2002).
In this paper, we analyze interactions between the seasonal and interannual variations of vegetation, ET, and SM in western Mexico, and evaluate the implications of using an observed [from the satellite-based Moderate Resolution Imaging Spectroradiometer (MODIS) MOD15A2 LAI product] seasonal cycle of vegetation properties on LSM simulations over the region as compared to a climatological cycle. We also assess the performance of the Variable Infiltration Capacity (VIC) model (Liang et al. 1994) with respect to its simulations of ET and SM through the comparison of model predictions with field observations at two flux tower sites operating during the NAM periods in 2004–07. We then compare model experiments that use observed and climatological seasonal cycles of LAI in the broader NAM region over the period 2001–08. A spatiotemporal analysis of simulated SM and ET patterns is then conducted to identify locations and periods of time when an observed LAI is essential for representing process dynamics in the VIC model. Our study substantially differs from sensitivity analyses of the impact of LAI in regional climate or weather models (e.g., Kurkowski et al. 2003; Knote et al. 2009) in its focus on the effect of vegetation variations on land surface states and fluxes and their intraseasonal predictability. As a result, our study is aimed at yielding insight on how vegetation dynamics influence the spatiotemporal distribution of ET and SM, which should be broadly applicable to other LSMs that represent the effects of vegetation on land–atmosphere interactions.
2. Data and experimental design
a. Observations and model forcings
The study area consists of the NAM region from 20° to 35°N and 116° to 104°W (Fig. 1). Over this region, 30%–70% of the total annual rainfall occurred during the monsoon season from June to September over the period 1950–2000 (Zhu and Lettenmaier 2007). Seasonal vegetation greening, which can be quantified as the seasonal change (September minus June) in a vegetation index (Vivoni et al. 2008) such as LAI, was averaged from the MOD15A2 data as shown in Fig. 1a. The largest seasonal vegetation greening occurs in the foothills of the Sierra Madre Occidental occupied by subtropical and tropical deciduous ecosystems that are well tuned to the NAM rainfall (Vivoni et al. 2010b). The higher elevations of the Sierra Madre Occidental are occupied by conifer forests that have more limited changes in seasonal LAI, while the coastal areas are typically desert shrublands or irrigated agriculture that experience less dramatic changes in seasonal LAI during the NAM than do the foothills regions (Forzieri et al. 2011).
We selected nine 1° × 1° square subdomains along a latitudinal transect that parallels the western slopes of the Sierra Madre Occidental (described as BXA, BXB, … to BXI) as the focus of our investigation. These nine subdomains allow a careful investigation of the latitudinal variations in precipitation and vegetation greenup across the NAM domain. Table 1 summarizes the precipitation and vegetation characteristics of the subdomains averaged during the NAM seasons in 2001–08. Gridded daily meteorological data required to drive the VIC model (including precipitation, maximum and minimum air temperatures, and surface wind speed) at 
Locations of the nine subdomains shown in Fig. 1 (arranged from north to south) and mean precipitation and mean LAI in the monsoon season (from June to September) averaged over 2001–08.
Identification of the onset of the monsoon is of considerable importance as it triggers the rapid changes in vegetation characteristics noted above (Vivoni et al. 2008). The onset of the monsoon precipitation was determined following Higgins et al. (1997, 1999). A precipitation index for each subdomain was obtained by averaging the local
We used MODIS-based dates of vegetation greening and dormancy to trace vegetation phenology in the study region. The procedure to identify the transition dates was as follows (Fig. 3). In each subdomain, the spatially averaged 8-day LAI time series was calculated from the 1-km resolution data. The 8-day domain-averaged LAI time series was then linearly interpolated to yield a daily LAI time series for each domain. The onset dates in each region were determined using a threshold-crossing procedure. The threshold criterion was set as 30% of the seasonal LAI change, which was defined as the difference between the maximum and minimum LAI in the given year. In each subdomain, the greening onset occurs when the threshold is reached after 1 June. Vegetation dormancy was defined as occurring when the threshold was first crossed in the downward direction during the period from the date with maximum LAI to 31 December.
Schematic of the definition of vegetation greening onset and dormancy. The seasonal LAI change is defined as the difference between the maximum LAI (LAImax) and minimum LAI (LAImin) in a year.
Citation: Journal of Hydrometeorology 13, 1; 10.1175/JHM-D-11-032.1
Two eddy covariance flux tower sites have been in operation within the study area since 2004 as part of NAME: Rayon (29.74°N, 110.54°W) and Tesopaco (27.85°N, 109.30°W) (Watts et al. 2007; Pérez-Ruiz et al. 2010; Vivoni et al. 2010a) (Fig. 1a). The Rayon site is a subtropical scrubland located in subdomain BXA and the Tesopaco site is a tropical deciduous forest located in subdomain BXC. The observed daily volumetric surface SM (SM in percent at 5-cm depth) and ET (in mm day−1) were obtained from Vivoni et al. (2008). The period of record used in this study for the Rayon site was day of year (DOY) 199–290 in 2004, 184–227 in 2006, and 152–240 in 2007, while, for the Tesopaco site, it was DOY 192–275 in 2004, 151–274 in 2005, and 151–275 in 2006. The observations at the flux tower sites were used to evaluate the VIC performance in terms of SM and ET simulations. Watts et al. (2007), Vivoni et al. (2008), and Pérez-Ruiz et al. (2010) provide details on the sensor measurements and data processing methods at the two tower sites.
b. Model experiments
Among the input data required by the VIC model are surface characteristics such as topography, land cover, and soil properties. Subgrid variability in topography and vegetation is represented statistically by the model (Liang et al. 1994). Digital elevation data were obtained from HYDRO1k, a geographic database developed at the U.S. Geological Survey (USGS; Verdin and Verdin 1999). Up to five vegetation classes in each model grid cell were used. The fraction of each vegetation class in each model grid cell (vegetation class fractions are not time varying in VIC) were derived from the University of Maryland’s 1-km vegetation classification (Hansen et al. 2000). There are three soil layers in the VIC model. The top layer was specified to be the 0–10-cm depth range, which we assumed to be comparable to SM observations at 5-cm depth. The other two soil layers vary spatially from 50- to 110-cm depths. The primary parameters in VIC model are the infiltration parameter (bi), the second and third soil layer thicknesses (d2, d3), and three baseflow parameters, including the maximum baseflow velocity (Dm), the fraction of Dm (Ds), and the fraction of maximum SM content of the third layer (Ws) at which nonlinear baseflow occurs (see Liang et al. 1994 for details). Zhu and Lettenmaier (2007) calibrated the VIC model using monthly streamflow data for the 1970s–1990s in 14 river basins in Mexico, among which six basins are more or less evenly distributed in our study area. Zhu and Lettenmaier (2007) then interpolated the calibrated parameters over Mexico at 
VIC model simulations were performed at a daily time step and at
Two model experiments were performed. In both experiments, the model was initialized using the forcings described above and a climatological seasonal LAI cycle for the spinup period of 1990–2000. The model experimental period was from 2001 through 2008 to match the availability of LAI data from MODIS. The first experiment was a VIC simulation in which the LAI (LAI_O) was specified for each grid cell using observed MOD15A2 data. The second experiment was a VIC simulation in which the LAI (LAI_C) was specified using a climatological seasonal cycle calculated as the average from the MOD15A2 data over the period 2001–08. Thus, the observed experiment provides the most realistic representation of vegetation dynamics and the climatological experiment is based on the average seasonal change of vegetation, thus ignoring the interannual variability of the greening timing or magnitude. The SM and ET estimates from the two experiments were evaluated at the two flux tower sites and compared with each other. In consideration of the large variability of soil texture and the scale mismatches between the tower site and model grid cell, we used the normalized SM calculated as the ratio of volumetric SM on the observation date to the maximum volumetric SM during the study period, following Vivoni et al. (2008).
3. Results and interpretation
a. Vegetation phenology
Quantifying vegetation phenology is a first step toward understanding the impact of the observed LAI cycle on the LSM response. Figure 4 shows the monsoon precipitation onset, vegetation greening onset, and vegetation dormancy date from 2001 to 2008 for the nine subdomains, arranged from north to south. For precipitation and greening onsets, the average calendar dates are later for the northern subdomains. The mean calendar date of precipitation onset in the northern four subdomains is 17 days later than in the southern-four subdomains. These differences are similar to the mean differences (about 14 days) from 1963 to 1988 reported in Higgins et al. (1999). The mean greening onset dates are 3 and 12 July in the southern- and northern-four subdomains, respectively. This indicates that the seasonal evolution of the NAM system is characterized by the northward progression of precipitation and of vegetation greening in western Mexico. Greening onsets are about 16 days later than precipitation onset averaged over the 1000-km transect, indicating the average response time for the different ecosystems to respond to the NAM rainfall. Nevertheless, the greening onset date varies substantially from year to year, with a larger range of dates in the south (about one month) as compared to the north (about half a month). A possible explanation for this may be that the north exhibits a “pure monsoon” vegetation response, while the south is affected by other precipitation sources because of its proximity to the tropics (Englehart and Douglas 2001).
Monsoon precipitation onset date, vegetation greening onset date, and vegetation dormancy date from 2001 to 2008 for the nine subdomains, arranged from (left) north to (right) south. The symbols are the average dates and bars indicate the standard deviations of the dates.
Citation: Journal of Hydrometeorology 13, 1; 10.1175/JHM-D-11-032.1
Vegetation greenness ends sooner in the north with a mean date of 16 October at the northernmost subdomain and 10 December at the southernmost subdomain. The progression time is about 56 days on average from the northernmost to the southernmost subdomain. Vegetation dormancy also exhibits large interannual variability, and the range of the dormancy date is larger in the north (about one month) than in the south (about half a month). The length of the vegetation phenological cycle (difference between the greening onset and vegetation dormancy) is longer in the south than in the north because of the northward progression of vegetation greening and southward progression of vegetation dormancy. The length of the vegetation cycle is 105 days in the northern subdomains and 158 days in the southern subdomains, on average. The characteristic lengthening of the phenological cycle is possibly due to variations in plant strategies (from intensive to extensive water use as one progresses southward) to cope with seasonal and interannual variations in water availability (see Forzieri et al. 2011).
b. Model comparisons to field observations
The impact of observed vegetation dynamics on the land surface response is first assessed with respect to the field observations at the two tower sites. Figure 5 compares the daily ET and SM from the flux tower observations and the VIC estimates obtained at
Comparisons of simulated and observed ET and SM at (top) the Rayon site during DOY 199–290 in 2004, 184–227 in 2006, and 152–240 in 2007; and (bottom) the Tesopaco site during DOY 192–275 in 2004, 151–274 in 2005, and 151–275 in 2006. SM was normalized and scaled to [0, 1]. Shown are daily ET estimates (ET_O) and SM estimates (SM_O) using LAI_O and ET estimates (ET_C) and SM estimates (SM_C) using LAI_C.
Citation: Journal of Hydrometeorology 13, 1; 10.1175/JHM-D-11-032.1
Despite these issues, the model captured the temporal variations of the observed ET and SM, closely following the seasonal evolution of land surface states and fluxes and the response to individual storm and interstorm periods. Importantly, the model is able to reproduce the low ET and SM in the premonsoon period (May–June) and the high ET and SM during the monsoon season (July–September), indicating an adequate representation of the major features of the land surface hydrology during the NAM (Vivoni et al. 2008). This characteristic seasonal evolution is retained in the spatially averaged ET and SM within the subdomains (BXA and BXC) that encompass the two sites (not shown). Furthermore, the model captures an important distinction in the seasonality of the two sites, namely that the Rayon site is characterized by infrequent and short duration pulses of SM and ET, while Tesopaco has a sustained period of high SM and ET during the entire NAM season. In addition, the results are robust for different years.
We found large discrepancies in the ET estimates between the observed (ET_O) and climatological (ET_C) simulations using their respective LAI values (LAI_O and LAI_C), in particular for periods when the differences in the LAI between the simulations were large. For example, large ET variations occur at the beginning of the NAM when LAI estimates disagree, but these variations decrease in the late monsoon season when the relative differences in LAI were reduced and other controls on ET are present. During the monsoon transition, the specification of LAI_O is critical for accurately modeling ET. For example, at the Rayon site in 2006, the early vegetation greenup (high LAI) in late June (LAI_O), relative to LAI_C, leads to higher ET until mid-July, which is much closer to the observed ET. Similarly, for the Tesopaco site in 2005, a late greening onset delays the peak ET by nearly one month. Underlying these differences is the process representation in the model, where a sharp LAI increase during leaf emergence leads to a rapid decrease in canopy resistance to ET and, thus, higher ET. During the remaining NAM season, LAI gradually decreases or remains constant, and modeled canopy resistance decreases slowly or stabilizes (Sakai et al. 1997). This response explains these ET differences in simulations using the LAI_O and LAI_C simulations. Clearly, the model ET estimates are sensitive to LAI variations primarily at the beginning of the monsoon season when greening onset varies substantially from year to year.
c. Spatiotemporal ET variations for LAI_O and LAI_C
To further compare the two simulations, an analysis of the 8-day VIC ET estimates is performed for different years over subdomains BXA and BXC that include the two tower sites in Fig. 6. Clearly, the seasonal ET variations differ from year to year and across the subdomains, with a higher interannual difference for the northernmost BXA. Maximum values of ET (1–3 mm day−1 for Rayon and 3–5 mm day−1 for Tesopaco) are consistent with the site observations (Vivoni et al. 2008). Note that the large ET differences between the LAI_C and LAI_O cycles occur even when considering the large-scale averages (1° × 1°) that aggregated behavior over many individual LAI pixels. In years with earlier (later) vegetation greening, ET estimates with LAI_O were larger (smaller) than the estimates with a LAI_C. This indicates that the VIC model with LAI_O can better capture the timing of ET changes as compared to the model with the LAI_C cycle. Since vegetation greening onset varies substantially from year to year, ET differences among the cases are substantial during the greening period. For example, LAI_O was larger than the LAI_C (by 0.7) at the beginning of July in 2008 in subdomain BXA. Consequently, VIC underestimates ET (by 0.92 mm day−1, or 28%) for the case where interannual variability in vegetation greening is not captured. Interestingly, as the NAM season nears its end in September, the ET differences between the two cases diminish considerably, despite that ET is as high as that in July and the substantial variations are still present in LAI. This suggests that other controls on ET—due to lower radiation or lower SM, for example—are more significant during the dormancy period than are differences in vegetation greenness.
Comparisons of the 8-day ET estimates (ET_O) using LAI_O and ET estimates (ET_C) using LAI_C from (left to right) June to September of 2001–08 for the subdomains (top) BXA and (bottom) BXC.
Citation: Journal of Hydrometeorology 13, 1; 10.1175/JHM-D-11-032.1
At the regional scale, Fig. 7 shows the spatial distribution of the mean ET differences (in mm day−1 at
Spatial differences between simulated time-averaged ET using LAI_C and LAI_O from June to September in (top left to bottom right) each year (2001–08). The number in each subdomain is the day of greening onset relative to the mean greening onset date for that subdomain.
Citation: Journal of Hydrometeorology 13, 1; 10.1175/JHM-D-11-032.1
d. Links between ET, SM, and LAI
To more clearly identify linkages between ET, SM, and vegetation during the NAM, we performed an analysis of the subdomain-averaged characteristics across all study years. Figures 8a,b show the relationship between the relative difference of LAI_C to LAI_O and the resulting relative differences of the VIC-estimated ET and SM. Relative differences are calculated as the difference in the time-averaged quantity (from June to September) of the climatological minus the observed seasonal cycles divided by the observed quantity. For example, the relative difference in LAI is obtained as (LAI_C − LAI_O)/LAI_O. All quantities are spatial averages at 1° × 1° resolution. ET-relative differences have a strong positive linear relation (R = 0.93) with LAI-relative differences, showing that under (over) estimation of LAI will induce a lower (higher) prediction of ET. Relative differences of LAI_C ranged from −25% to 25% and the consequent relative differences of ET ranged from −10% to 15%, indicating that fractional variations in ET are smaller than in LAI, as confirmed by the regression slope of 0.44. On the other hand, the SM relative difference has a negative linear relation (R = 0.89) with the LAI-relative difference, showing that under (over) estimation of LAI will lead to a higher (lower) SM. Nevertheless, the relative differences of SM range only from −2% to 2%, indicating that the effect of LAI on surface SM is relatively small, with a low regression slope of 0.07. These results show that LAI differences between the simulations linearly affect ET in a much stronger fashion than SM when aggregated to the subdomain scale and averaged over the summer season.
Relationship between the relative difference of LAI and the relative differences of (a) ET and (b) SM from June to September across the nine subdomains in each year. Each symbol represents a separate year and subdomain.
Citation: Journal of Hydrometeorology 13, 1; 10.1175/JHM-D-11-032.1
We also carried out an analysis of the relative differences of ET as a function of geographic position for each year in the nine subdomains and along the north (I), intermediate (II), and south (III) bands in Figs. 9a–d. Bands I, II, and III include subdomains BXA, BXD, and BXG, respectively. For each case, we show the mean ET during the season (mm day−1) computed in the subdomain or along the latitudinal band. The relative difference of ET varies substantially from year to year and closely follows the trends in LAI (not shown). The varying range of the ET-relative difference is generally small (large) where the mean ET is large (small), corresponding to the varying range of the LAI-relative difference. As shown in Fig. 9a, the LAI_C leads to higher predictions of ET by ~15% when LAI_O is smaller than average (e.g., 2002) and lower predictions of ET by ~10% when LAI_O is larger than average (e.g., 2008) in the northernmost subdomain. The ET-relative difference varies from −8% to 8% over the other subdomains (Fig. 9a). Interestingly, the latitudinal variations (from BXA to BXI) in ET-relative differences differ for each year and sometimes exhibit minimum values near the BXD subdomain. At these intermediate latitudes, the LAI_C cycle is sufficient to capture the vegetation phenology for most years, indicating limited interannual variations when viewed as the temporal average over the season.
Relative differences of VIC-estimated ET for the simulations with LAI_C and LAI_O, (ET_C − ET_O)/ET_O, and the seasonal mean ET (mm day−1) at (a) the nine subdomains, and along three longitudinal transects at (b) the latitude band I (29.25°–30.25°N), (c) band II (26.25°–27.25°N), and (d) band III (23.25°–24.25°N). Results from 110° to 115°W correspond to the Baja California peninsula and thus are separated from the mainland by a large gap corresponding to the Gulf of California (see Fig. 1a).
Citation: Journal of Hydrometeorology 13, 1; 10.1175/JHM-D-11-032.1
Figures 9b–d present the relative differences of ET for different longitudes along the latitudinal bands in which vegetation phenology varies considerably (Forzieri et al. 2011). Note that the labels (A to I) correspond to the longitudinal position of the subdomains, but do not imply that these are positioned in that band. The largest relative ET differences are observed in band I because of its higher interannual variations in LAI with respect to LAI_C and lower overall mean ET rates. The lowest ET-relative differences occur in band III near the core NAM region, where consistent greening occurs year to year, thus minimizing the LAI variations from LAI_C. It is interesting to note the variations of the relative ET differences with longitude for each band. Two longitudinal effects are observed: 1) high relative ET differences in the Baja California peninsula (115°–110°W), and 2) high relative ET differences to the east of the Sierra Madre Occidental (east of 108°W in band I and east of 106°W in band II). These regions of high relative ET differences are indicative of stronger year-to-year variations in LAI that are not adequately captured by LAI_C and correspond well to the spatial distributions shown in Fig. 7. Note also that the estimated ET differences are small where the mean ET is larger than ~2 mm day−1, and become larger where the mean ET is less than ~2 mm day−1. These results are indicative that the model ET sensitivity to LAI is highest in areas that experience stronger interannual differences, likely because of their inconsistent rainfall during the NAM due to their positioning at the outer edges of the core region—for example, in the eastern continental interior, along the northern extent, or in the Baja California peninsula.
e. Interannual variability in ET, SM, and LAI
Given the importance of interannual variations in LAI identified previously, we explored the implications of these on the simulated ET and SM for the LAI_O simulations that provide the most realistic representation of vegetation dynamics. Figures 10a–d show the mean seasonal cycles of precipitation, SM, LAI, and ET averaged over 2001–08 for the nine subdomains. In northern areas, the NAM season was shorter and the amount of precipitation was smaller than in the southern subdomains (Fig. 10a). Surface SM showed a similar spatiotemporal pattern as precipitation, but changed more slowly, resulting in the persistence of SM after the NAM ends for most subdomains (Fig. 10b). LAI is clearly linked to precipitation, but high LAI persisted after the end of the NAM in the south, while dormancy is abrupt after the end of precipitation in the north (Fig. 10c). The latitudinal changes in ecosystem type, ranging from tropical deciduous forests in the south to subtropical scrublands in the north (Lizárraga-Celaya et al. 2010), have a strong control on LAI persistence after the monsoon.
Mean (a) 8-day precipitation, (b) SM at the top soil layer, (c) LAI, and (d) ET using LAI_O from 2001 to 2008 for the nine subdomains. SM was normalized and scaled to [0, 1]. The dark dashed line is the mean precipitation onset date, the dark solid line is the mean greening onset date, and the gray solid line is the mean vegetation dormancy date.
Citation: Journal of Hydrometeorology 13, 1; 10.1175/JHM-D-11-032.1
High SM after precipitation onset typically leads to high ET in the early NAM season (Fig. 10d). Nevertheless, high SM does not always imply a high ET as vegetation greening can be delayed relative to precipitation. As the NAM progresses, ET increases when both SM and LAI are sufficiently high to allow for plant transpiration. As evidence, note that the peak ET occurs in August prior to the peak LAI in the season—an indication that ET requires both high greenness and high SM. In the later stages of the NAM, ET decreases in the southern subdomains, despite retaining high LAI and SM, as a consequence of reductions in the available radiation. In contrast, the northern subdomains have ET that is clearly limited by SM availability and decreases in LAI. This indicates that ecosystems across the region exhibit a considerable memory of monsoon precipitation, with a higher persistence for the southern subdomains. This memory is expressed as a persistence of SM, ET, and LAI beyond the NAM period, in particular for southern subdomains. This is likely due to the capacity of deep-rooted plants to extract water from soil layers (Adegoke and Carleton 2002) and their adaptations that allow a slower rate of vegetation dormancy as compared to the greening process (see also Méndez-Barroso et al. 2009).
Figures 11a–d are similar to the prior analysis but show the interannual standard deviation of the 8-day precipitation, SM, LAI, and ET. The interannual variability of precipitation is high (from 2 to 10 mm day−1) during the monsoon season, but relatively low prior to and after the NAM (Fig. 11a). Clearly, certain subdomains and time periods experience higher interannual variations—for example, in September over the BXD–BXG subdomains. The interannual variability in SM reflects these precipitation anomalies but also exhibits high values in the period after vegetation dormancy (Fig. 11b). This is likely due to the relatively long time it takes for a positive anomaly in SM to be dissipated through ET (Koster and Suarez 2001), in particular if ET is limited by LAI, radiation, or SM.
Interannual standard deviation of (a) 8-day precipitation, (b) SM at the top soil layer, (c) LAI, and (d) ET using LAI_O from 2001 to 2008 for the nine subdomains. The dark dashed line is the mean precipitation onset date, the dark solid line is the mean greening onset date, and the gray solid line is the mean vegetation dormancy in 2001–08.
Citation: Journal of Hydrometeorology 13, 1; 10.1175/JHM-D-11-032.1
As described previously, the interannual variability in LAI peaked during the greening onset and was larger in an absolute sense in the southern subdomains (Fig. 11c). However, the relative LAI variability (measured by the coefficient of variation) is greater in the northernmost subdomain where the mean LAI is smaller. It is interesting to note that high variability in LAI is present in certain locations and periods after the greening onset, which is most likely related to differences in SM availability. Nevertheless, LAI variations tend to be small near the peak greenness and dormancy period. The interannual ET changes show the LAI signal during the greening onset (Fig. 11d). During this time, the highest interannual variability in ET occurs in subdomains BXC–BXE, suggesting a higher sensitivity to LAI in this region. Another notable feature is the transition in ET variability from north to south occurring in July–October. Clearly, the northern subdomains exhibit a higher absolute and relative interannual variability in ET, likely due to a combination of year-to-year differences in precipitation, LAI, and SM.
f. Predicting ET based on greening onset dates
Since ET variability is highly correlated with LAI in the VIC simulations, it should be possible to use a measure of vegetation phenology as an indicator of the anticipated ET during the NAM. Greening onset dates were related to the mean daily ET distributions occurring during June and July of each year in Fig. 12, obtained from the LAI_O simulation. Note the large spatial variations in the mean ET patterns for each year, with certain areas and years having averaged values from 3 to 6 mm day−1. Large ET values occur when the vegetation greening comes earlier than the mean greening date and vice versa. For example, the early onset (−16 to −22 days) in the southern subdomains in 2004 leads to large ET (3–5 mm day−1) in that year. In contrast, the late onset (3–14 days) throughout the region in 2005 resulted in lower-than-expected ET rates (1–3 mm day−1). These results indicate that the greening onset date is a possible predictor variable for the ET in the early monsoon season (June and July). Since the greening onset date and its spatial distribution is easy to detect via remotely sensed products such as MODIS, this provides an operational method to forecast ET in the early NAM period with lead times ranging from a few weeks up to 2 months. A spatially variable ET forecast of June and July conditions could be useful for water managers in the region who need to forecast streamflow levels and plan reservoir operations during the NAM season.
(top left to bottom right) Estimated mean ET during June and July (mm day−1) using LAI_O for each year (2001–08). The number in each subdomain is the day of greening onset relative to the mean greening onset date for that subdomain.
Citation: Journal of Hydrometeorology 13, 1; 10.1175/JHM-D-11-032.1
To further explore the predictability of ET based on the greening onset date, Fig. 13 shows the relationship between the relative onset date and the mean ET in June, July, and August, respectively. As before, the relative onset date is the difference between the greening onset in the current year and its average over all years. This is performed for each subdomain separately using the results from all years (each symbol is a different year). The mean June ET was highly correlated to the greening onset in the southern subdomains, while the mean July ET was correlated to the greening onset over most subdomains (see Table 2 for linear regression relations). As discussed previously, the mean ET was generally greater if the greening onset came earlier than average and vice versa. Greening onset dates were not related to mean June ET in the northern subdomains, primarily because the vegetation greening in these subdomains generally occurred after June. Similarly, the relative onset date was not a good predictor variable for mean August ET throughout the study region except for subdomain BXA, indicating that the effect of the greening time is limited to the early part of the NAM season. Further, the mean ET in August likely depends on values of precipitation, SM, and LAI that have evolved during the NAM season and are less dependent on the initial greenup period.
Relationship between the mean ET from LAI_O within each month (June, July, and August) from 2001 to 2008 and the greening onset dates (expressed relative to the mean greening onset date) for the nine subdomains.
Citation: Journal of Hydrometeorology 13, 1; 10.1175/JHM-D-11-032.1
Parameters of the linear regressions between relative greening onset date and mean ET in each month (see Fig. 13).
4. Summary and conclusions
In this study, we utilized the VIC LSM and a remotely sensed vegetation phenology dataset to investigate the links between LAI, ET, and SM in western Mexico during the NAM. LAI-based vegetation phenology from the MODIS sensor over the period 2001–08 indicates that the greening onset occurs ~16 days later than precipitation onset. In general, the NAM is characterized by the northward progression of greening onset from the southern regions by late June to the northern areas by mid-July. Vegetation greening onset and dormancy were found to vary from year to year, with stronger changes in the interannual variations in the south for the onset date and in the north for the dormancy date. Overall, the length of the precipitation duration is on average about one month shorter than the length of vegetation phenological cycle. The latitudinal variations of vegetation phenology help determine how differences in seasonal and interannual dynamics relate to the changing character of the NAM system and the ecosystems that it supports.
We evaluated VIC-estimated ET and SM using LAI_O and LAI_C cycles in comparison with observations at two eddy covariance tower sites; they showed generally good agreement despite complications in the comparison of point data with modeled spatial averages. Over the broader region, we analyzed the spatial and temporal variations of the simulated ET and SM and related them to variations in LAI. Results at the point, subdomain, and regional scales indicated that the VIC model with LAI_C tends to predict lower (higher) ET when the vegetation greening comes earlier (later) than the mean greening date. The estimated ET (SM) difference has a positive (negative) linear relationship with the relative difference of LAI from June to September, with LAI differences having a much larger impact on ET as compared to SM. In addition, the VIC simulations with LAI_C could induce about 10% bias in ET estimation in the western slopes of the Sierra Madre Occidental and about 30% bias in ET estimation in the interior continental regions east of the Sierra Madre Occidental. The effect of LAI on ET estimation is largest in areas that experience stronger interannual differences along the periphery of the NAM region, such as the eastern continental interior, along the northern extent, or in the Baja California peninsula, likely because of their inconsistent rainfall during the NAM.
We used the VIC simulation using LAI_O, which provides the most realistic representation of vegetation dynamics, to examine the spatiotemporal variations in precipitation, LAI, SM, and ET. ET was generally low before the greening onset, and abruptly increased when both SM and LAI became high, reaching its peak value in August slightly before the peak LAI. Considerably large ET persists after the end of the monsoon, indicating most ecosystems exhibited the memory of monsoon precipitation. The ET in June, July, and August was separately related to the greening onset date. The greening onset date was highly related to mean ET of June and July, but not to mean August ET. The results indicate that the onset date of the greening is a potential useful metric for predicting ET variations over western Mexico in the early monsoon season, and is less useful during the full vegetation greening phase. This suggests that remotely sensed greening onset can be used to forecast ET in the early NAM period for water management applications in the region, with lead times of a few weeks up to 2 months.
Acknowledgments
The work described in this paper was supported by NASA Grant NNSO6AA78G and NOAA CPPA Grant NA060AR4310060 to the University of Washington and NOAA CPPA Grant NA09OAR4310217 to Arizona State University. We thank Christopher J. Watts and Jaime Garatuza-Payán for providing the eddy covariance tower datasets. Thanks are also due to Elizabeth Clark of the Land Surface Hydrology Group, University of Washington, for her comments and editorial advice.
REFERENCES
Abuelgasim, A. A., Fernandes R. A. , and Leblanc S. G. , 2006: Evaluation of national and global LAI products derived from optical remote sensing instruments over Canada. IEEE Trans. Geosci. Remote Sens., 44, 1872–1884.
Adams, D. K., and Comrie A. C. , 1997: The North American monsoon. Bull. Amer. Meteor. Soc., 78, 2197–2213.
Adegoke, J. O., and Carleton A. M. , 2002: Relations between soil moisture and satellite vegetation indices in the U.S. corn belt. J. Hydrometeor., 3, 395–405.
Anderson, B. T., Kanamura H. , and Roads J. O. , 2004: The summertime atmospheric hydrologic cycle over the southwestern United States. J. Hydrometeor., 5, 679–692.
Barlow, M., Nigam S. , and Berbery E. H. , 1998: Evolution of the North American monsoon system. J. Climate, 11, 2238–2257.
Berbery, E. H., 2001: Mesoscale moisture analysis of the North American monsoon. J. Climate, 14, 121–137.
Castro, C. L., Pielke R. A. , and Adegoke J. O. , 2007: Investigation of the summer climate of the contiguous United States and Mexico using the Regional Atmospheric Modeling System (RAMS). Part I: Model climatology (1950–2002). J. Climate, 20, 3844–3865.
Dominguez, F., Kumar P. , and Vivoni E. R. , 2008: Precipitation recycling variability and ecoclimatological stability—A study using NARR data. Part II: North American monsoon region. J. Climate, 21, 5187–5203.
Douglas, M. W., Maddox R. A. , Howard K. , and Reyes S. , 1993: The Mexican monsoon. J. Climate, 6, 1665–1677.
Duchemin, B., Berthelot B. , Dedieu G. , Leroy M. , and Maisongrande P. , 2002: Normalisation of directional effects in 10-day global syntheses derived from VEGETATION/SPOT: II. Validation of an operational method on actual data sets. Remote Sens. Environ., 81, 101–113.
Englehart, P. J., and Douglas A. V. , 2001: The role of eastern Pacific tropical storms in the rainfall climatology of western Mexico. Int. J. Climatol., 21, 1357–1370.
Fensholt, R., Sandholt I. , and Rasmussen M. S. , 2004: Evaluation of MODIS LAI, fAPAR and the relation between fAPAR and NDVI in a semi-arid environment using in situ measurements. Remote Sens. Environ., 91, 490–507.
Forzieri, G., Castelli F. , and Vivoni E. R. , 2011: Vegetation dynamics within the North American Monsoon Region. J. Climate, 24, 1763–1783.
Friedl, M. A., Sulla-Menashe D. , Tan B. , Schneider A. , Ramankutty N. , Sibley A. , and Huang X. , 2010: MODIS Collection 5 global land cover: Algorithm refinements and characterization of new datasets. Remote Sens. Environ., 114, 168–182.
Gao, X., Dirmeyer P. A. , Guo Z. , and Zhao M. , 2008: Sensitivity of land surface simulations to the treatment of vegetation properties and the implications for seasonal climate prediction. J. Hydrometeor., 9, 348–366.
Garrigues, S., and Coauthors, 2008: Validation and intercomparison of global Leaf Area Index products derived from remote sensing data. J. Geophys. Res., 113, G02028, doi:10.1029/2007JG000635.
Gebremichael, M., Vivoni E. R. , Watts C. J. , and Rodríguez J. C. , 2007: Submesoscale spatiotemporal variability of North American monsoon rainfall over complex terrain. J. Climate, 20, 1751–1773.
Gochis, D. J., Shuttleworth W. J. , and Yang Z. L. , 2002: Sensitivity of the modeled North American monsoon regional climate to convective parameterization. Mon. Wea. Rev., 130, 1282–1298.
Gochis, D. J., Jimenez A. , Watts C. J. , Garatuza-Payan J. , and Shuttleworth W. J. , 2004: Analysis of 2002 and 2003 warm season precipitation from the North American Monsoon Experiment (NAME) Event Rain Gauge Network (NERN). Mon. Wea. Rev., 132, 2938–2953.
Gochis, D. J., Brito-Castillo L. , and Shuttleworth W. J. , 2006: Hydroclimatology of the North American monsoon region in northwest Mexico. J. Hydrol., 316, 53–70.
Gutzler, D. S., 2004: An index of interannual precipitation variability in the core of the North American monsoon region. J. Climate, 17, 4473–4480.
Hansen, M. C., DeFries R. S. , Townshend J. R. G. , and Sohlberg R. , 2000: Global land cover classification at 1 km spatial resolution using a classification tree approach. Int. J. Remote Sens., 21, 1331–1364.
Higgins, R. W., and Shi W. , 2001: Intercomparison of the principal modes of interannual and intraseasonal variability of the North American monsoon system. J. Climate, 14, 403–417.
Higgins, R. W., Yao Y. , and Wang X. L. , 1997: Influence of the North American monsoon system on the U.S. summer precipitation regime. J. Climate, 10, 2600–2622.
Higgins, R. W., Chen Y. , and Douglas A. V. , 1999: Interannual variability of the North American warm season precipitation regime. J. Climate, 12, 653–680.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471.
Knote, C., Bonafe G. , and Giuseppe F. D. , 2009: Leaf area index specification for use in mesoscale weather prediction systems. Mon. Wea. Rev., 137, 3535–3550.
Koster, R. D., and Suarez M. J. , 2001: Soil moisture memory in climate models. J. Hydrometeor., 2, 558–570.
Kurc, S. A., and Small E. E. , 2004: Dynamics of evapotranspiration in semiarid grassland and shrubland ecosystems during the summer monsoon season, central New Mexico. Water Resour. Res., 40, W09305, doi:10.1029/2004WR003068.
Kurkowski, N. P., Stensrud D. J. , and Baldwin M. E. , 2003: Assessment of implementing satellite-derived land cover data in the Eta Model. Wea. Forecasting, 18, 404–416.
Lawrence, P. J., and Chase T. N. , 2007: Representing a new MODIS consistent land surface in the Community Land Model (CLM 3.0). J. Geophys. Res., 112, G01023, doi:10.1029/2006JG000168.
Li, J., Gao X. , Maddox R. A. , and Sorooshian S. , 2004: Model study of evolution and diurnal variations of rainfall in the North American monsoon during June and July 2002. Mon. Wea. Rev., 132, 2895–2915.
Liang, X., Lettenmaier D. P. , Wood E. , and Burges S. J. , 1994: A simple hydrologically based model of land surface and energy fluxes for general circulation models. J. Geophys. Res., 99, 14 415–14 428.
Lizárraga-Celaya, C., Watts C. J. , Rodríguez J. C. , Garatuza-Payan J. , Scott R. L. , and Saiz-Hernandez J. , 2010: Spatio-temporal variations in surface characteristics over the North American Monsoon region. J. Arid Environ., 74, 540–548.
Matsui, T., Lakshmi V. , and Small E. E. , 2005: The effects of satellite-derived vegetation cover variability on simulated land–atmosphere interactions in the NAMS. J. Climate, 18, 21–40.
Maurer, E. P., Wood A. W. , Adam J. C. , Lettenmaier D. P. , and Nijssen B. , 2002: A long-term hydrologically based dataset of land surface fluxes and states for the conterminous United States. J. Climate, 15, 3237–3251.
Méndez-Barroso, L. A., and Vivoni E. R. , 2010: Observed shifts in land surface conditions during the North American Monsoon: Implications for a vegetation–rainfall feedback mechanism. J. Arid Environ., 74, 549–555.
Méndez-Barroso, L. A., Vivoni E. R. , Watts C. J. , and Rodríguez J. C. , 2009: Seasonal and interannual relations between precipitation, surface soil moisture and vegetation dynamics in the North American monsoon region. J. Hydrol., 377, 59–70.
Mesinger, F., and Coauthors, 2006: North American Regional Reanalysis. Bull. Amer. Meteor. Soc., 87, 343–360.
Mitchell, K. E., and Coauthors, 2004: The multi-institution North American Land Data Assimilation System (NLDAS): Utilizing multiple GCIP products and partners in a continental distributed hydrological modeling system. J. Geophys. Res., 109, D07S90, doi:10.1029/2003JD003823.
Moulin, S., Kergoat L. , Viovy N. , and Dedieu G. , 1997: Global-scale assessment of vegetation phenology using NOAA/AVHRR satellite measurements. J. Climate, 10, 1154–1170.
Pérez-Ruiz, E. R., Garatuza-Payan J. , Watts C. J. , Rodríguez J. C. , Yepez E. A. , and Scott R. L. , 2010: Carbon dioxide and water vapour exchange in a tropical dry forest as influenced by the North American Monsoon System (NAMS). J. Arid Environ., 74, 556–563.
Pisek, J., and Chen J. M. , 2007: Comparison and validation of MODIS and VEGETATION global LAI products over four BigFoot sites in North America. Remote Sens. Environ., 109, 81–94.
Quintas, I., 2000: ERIC II: Documentación de la base de datos climatológica y del programa extractor (ERIC II: Documentation of the climatologic database and data extraction program). IMTA Tech. Rep., 54 pp.
Rodell, M., and Coauthors, 2004: The global land data assimilation system. Bull. Amer. Meteor. Soc., 85, 381–394.
Sakai, R. K., Fitzjarrald D. R. , and Moore K. E. , 1997: Detecting leaf area and surface resistance during transition seasons. Agric. For. Meteor., 84, 273–284.
Salinas-Zavala, C. A., Douglas A. V. , and Diaz H. F. , 2002: Interannual variability of NDVI in northwest Mexico. Associated climatic mechanisms and ecological implications. Remote Sens. Environ., 82, 417–430.
Scott, R. L., Huxman T. E. , Cable W. L. , and Emmerich W. E. , 2006: Partitioning of evapotranspiration and its relation to carbon dioxide exchange in a Chihuahuan Desert shrubland. Hydrol. Processes, 20, 3227–3243.
Shepard, D. S., 1984: Computer mapping: The SYMAP interpolation algorithm. Spatial Statistics and Models, G. L. Gaile and C. J. Willmott, Eds., D. Reidel Publishing Company, 133–145.
Sheppard, P. R., Comrie A. C. , Packin G. D. , Angersbach K. , and Hughes M. K. , 2002: The climate of the US Southwest. Climate Res., 21, 219–238.
Shuttleworth, W. J., 1993: Evaporation. Handbook of Hydrology, D. R. Maidment, Ed., McGraw-Hill, 4.1–4.53.
Stensrud, D. J., Gall R. L. , Mullen S. L. , and Howard K. W. , 1995: Model climatology of the Mexican monsoon. J. Climate, 8, 1775–1794.
Verdin, K. L., and Verdin J. P. , 1999: A topological system for delineation and codification of the Earth’s river basins. J. Hydrol., 218, 1–12.
Vivoni, E. R., Moreno H. A. , Mascaro G. , Rodríguez J. C. , Watts C. J. , Garatuza-Payan J. , and Scott R. L. , 2008: Observed relation between evapotranspiration and soil moisture in the North American monsoon region. Geophys. Res. Lett., 35, L22403, doi:10.1029/2008GL036001.
Vivoni, E. R., Rodríguez J. C. , and Watts C. J. , 2010a: On the spatiotemporal variability of soil moisture and evapotranspiration in a mountainous basin within the North American monsoon region. Water Resour. Res., 46, W02509, doi:10.1029/2009WR008240.
Vivoni, E. R., Watts C. J. , Rodríguez J. C. , Garatuza-Payan J. , Mendez-Barroso L. A. , and Saiz-Hernandez J. A. , 2010b: Improved land–atmosphere relations through distributed footprint sampling in a subtropical scrubland during the North American monsoon. J. Arid Environ., 74, 579–584.
Watts, C. J., Scott R. L. , Garatuza-Payan J. , Rodríguez J. C. , Prueger J. H. , Kustas W. P. , and Douglas M. , 2007: Changes in vegetation condition and surface fluxes during NAME 2004. J. Climate, 20, 1810–1820.
Zeng, X., 2001: Global vegetation root distribution for land modeling. J. Hydrometeor., 2, 525–530.
Zhang, X. Y., Friedl M. A. , Schaaf C. B. , Strahler A. H. , Hodges J. C. F. , Gao F. , Reed B. C. , and Huet A. , 2003: Monitoring vegetation phenology using MODIS. Remote Sens. Environ., 84, 471–475.
Zhu, C., and Lettenmaier D. P. , 2007: Long-term climate and derived surface hydrology and energy flux data for Mexico: 1925–2004. J. Climate, 20, 1936–1946.
Zhu, C., Cavazos T. , and Lettenmaier D. P. , 2007: Role of antecedent land surface conditions in warm season precipitation over northwestern Mexico. J. Climate, 20, 1774–1791.
