Relationships between soil moisture and summer precipitation over the Great Plains and the Southwest

Authors


Abstract

[1] The relationships between evaporation (E) and summer precipitation (P) over the United States are examined using the observed precipitation data set, the National Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP-NCAR) reanalysis and the NCEP regional spectral model (RSM) simulations. The composites of P anomalies based on soil moisture and E anomalies indicate that the relationships between E and P are regionally dependent. The E anomalies over the Great Plains are associated with large P anomalies located north of the E anomalies and areas downstream along the path of the Great Plains low-level jet (GPLLJ). The impact of E anomalies on P over the Southwest is small and is only recognized during the strong moisture surge events from the Gulf of California. The vertically integrated meridional moisture fluxes [qv] associated with the GPLLJ are stronger and more persistent than the fluxes associated with the low-level jet from the Gulf of California to the Southwest (GCLLJ). The E anomalies over the Great Plains are stronger and more persistent than the E anomalies over the Southwest. Therefore the E anomalies over the Great Plains have a better chance for the local feedback and for the changes in moisture flux convergence to take effect. The impact of E anomalies on P over the Southwest is small and only recognized during strong moisture surge events from the Gulf of California. Both the surge events and E anomalies over the Southwest persist for less than one week. The time window for the large E anomalies to occur during surge events is narrow. Therefore the net impact is small.

1. Introduction

[2] During summer, soil moisture is an important factor in determining precipitation (P) over the United States. Over the central United States, the sensitivity studies of Koster and Suarez [2001] and Koster et al. [2000] suggested that the influence of soil moisture on P is strong and realistic soil moisture is needed for successful precipitation forecasts. During 1993, the forecast of the July floods was successful after initializing the soil moisture at field capacity [Beljaars et al., 1996]. The skill of the summer seasonal mean precipitation and surface temperature forecasts improves when soil moisture is added to the list of predictors [Mo, 2003]. For regional model simulations, Bosilovich and Sun [1999] and Giorgi et al. [1996] showed that soil moisture plays a role in determining the location and amounts of simulated precipitation of the 1993 floods. Kanamitsu and Mo [2003] performed model experiments to show that the local impact of E on P is small over the Southwest and soil moisture anomalies are sustained largely by the moisture surge events from the Gulf of California. These studies seem to suggest that the relationships between E and P anomalies are regionally dependent. If so, it is important to identify these linkages and to examine the physical mechanisms responsible for such association.

[3] Soil moisture and E anomalies are positively correlated over the United States. Huang et al. [1996] showed that the pattern correlation between the two over the United States is as high as 0.88. The impact of soil wetness on P is largely through E. The impact can be felt directly and indirectly. The changes of solar radiation, long wave radiation and ground heat flux coincident with large changes in soil moisture and surface evaporation are usually small in comparison with the large quasi-balancing change in surface sensible heat flux. Therefore soil moisture affects the Bowen ratio at the surface defined as the ratio of sensible to latent heat [Betts et al., 1996, 1997]. The impact of the Bowen ratio on P is direct. When the Bowen ratio is large, the sensible heat is large so the surface temperature is warmer. Betts and Ball [1995] suggested that the large Bowen ratio makes the boundary layer deeper and increases entrainment of cold and dry free atmospheric air at the top of the boundary layer. This suppresses the convection particularly in the afternoon. Schar et al. [1999] examined the relationship between soil moisture and P for summer European precipitation. They found that wet soil tends to build up a comparatively shallow boundary layer and provides a source of convective instability. Together with the increase of the net radiative energy flux, these processes increase the potential for convective activity.

[4] In addition to direct feedback, the impact of E on P can be non-local and indirect. The global model experiments performed by Fennessy and Shukla [1999] suggested that soil wetness can influence seasonal precipitation prediction only if it can alter the moisture flux convergence and the mean convective stability. Paegle et al. [1996] showed that E at the entrance region of the Great Plains low-level jet (GPLLJ) can influence rainfall downstream over the central United States. Over the Southwest including Arizona and New Mexico (AZNM), the experiments performed by Kanamitsu and Mo [2003] suggested that soil moisture anomalies over the Southwest influence P through the changes in the low-level jet from the Gulf of California to the Southwest (GCLLJ). This paper presents evidence that the impact of E on P is regionally dependent. We will focus on the contrast between the Great Plains and the Southwest. The physical mechanisms responsible for the linkages between E and P anomalies for these two regions are then discussed. The data set and the methods used are presented in section 2. Based on the unified rainfall gridded data set, the National Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP-NCAR) reanalysis and simulations from the NCEP regional spectral model (RSM), we demonstrate in section 3 that soil moisture or E anomalies over the Great Plains are associated with P anomalies downstream from the E anomalies and areas roughly along the path of the GPLLJ, while the impact of E on P over the Southwest is limited to periods wherein strong moisture surge events. The physical mechanisms are discussed in section 4. Examples are presented in section 5 to examine the relationships between E and P over the Southwest during the moisture surge or non- surge periods. Conclusions are given in section 6.

2. Data Sets and Procedures

2.1. Observations

[5] The observed precipitation data set is the unified rainfall gridded data set of Higgins et al. [2000] for the United States and Mexico. The gauge data sets were merged to create the unified data set. The combined data set has one-degree resolution. Data over Mexico have a gap between 1999 and 2000. This data set covers the period from 1950–2000 and is referred to as the unified P data. The monthly mean soil moisture at each climate division was constructed from a one layer hydrological model based on the observed monthly mean temperature and P [Huang et al., 1996]. There are 344 climate divisions in the United States and the data set covers the period from 1950–2000. For both data sets, monthly mean anomalies are defined as the departures from the climatological monthly means from 1950–2000.

[6] Because E cannot be measured directly, it is difficult to quantify its uncertainty. To assure the robustness of the results, E anomalies from different data sources were used. One data set was taken from the 0–6 h accumulations during the NCEP - NCAR reanalysis [Kalnay et al., 1996] forecast cycle for the period from 1979 to 2000. The period was chosen because the satellite data were available. The second data set was obtained from the RSM simulations described in the next section.

2.2. RSM Simulations

[7] The model simulations were performed using the RSM [Juang and Kanamitsu, 1994; Juang et al., 1997]. The model has 28 levels in the vertical. It has 50-km horizontal grid space on the polar stereographic grid. The domain covers the United States and northern Mexico (16–52°N, 65–130°W). The earlier version of the RSM has been used by Anderson et al. [2000a, 2000b, 2001] to study the diurnal forcing, the influence of large-scale circulation on the GCLLJ and rainfall over the Southwest. The major recent improvement was the implementation of the terrain and vegetation data set from the U.S. Geological Service (USGS) and related changes in the physical package. The USGS data set includes fine resolution (30') surface parameters of vegetation type, vegetation cover and soil type. The vegetation cover fraction is the monthly mean climatology. For each 50-km grid box, the most dominant vegetation type and soil type in the box are chosen as the vegetation type and soil type for that box. The vegetation cover for the grid box is the average over values in the box. They are fixed during the integration.

[8] The convection scheme is the relaxed Arakawa Schubert (RAS) scheme [Moorthi and Suarez, 1992]. The soil related parameters and surface roughness are computed from vegetation types [Chen and Dudhia, 2001a, 2001b]. The corresponding albedos are calculated according to the method outlined by Briegleb et al. [1986] and Briegleb [1992]. The evaporation is a function of soil wetness, vegetation cover. The potential evaporation is calculated based on the formula given by Mahfout and Noilhan [1991] and Chen and Dudhia [2001a]. Before the implementation of the USGS surface data sets and related model changes, the earlier version of the RSM used the vegetation type, cover, soil type, albedo and surface roughness from the relatively coarse resolution simple Sib data. The formulation of direct evaporation from bare soil is a function of soil water conductivity and diffusivity at the land surface as described by Mahrt and Pan [1984]. After the implementation, the model produces more E over the Southwest and more realistic P [Kanamitsu and Mo, 2003].

[9] For each summer month (June to September) from 1991 to 2000, the model was initialized at day 1 0Z of the month and was integrated through the entire month. The initial conditions including soil wetness and soil temperature and boundary conditions were taken from the reanalysis II [Kanamitsu et al., 2002]. After initialization, soil conditions including soil wetness and soil temperature were predicted by the model. The outputs were produced every 6 hours, but the daily means were formed for all variables. The latent heat, sensible heat fluxes, vertically integrated moisture fluxes and precipitation were accumulated for the 6-hour period.

[10] Figures 1a and 1b show the 10-year summer mean P (June–September) from the GPCP rainfall data set [Xie and Arkin, 1997] and the unified P data [Higgins et al., 2000] respectively. The GPCP mean provides information over the oceans, while the unified data set only covers the land areas. The GPCP data set on a 2.5-degree grid merges the satellite-based estimates and gauge data, but it has a relatively sparse network of stations in comparison with the unified P data set. Because of missing data over Mexico after 1998, the longer time means are also given (Figures 1c and 1d). The mean precipitation is about 3 mm d−1 over the central United States. The largest precipitation is located over the Southeast. The Southwest is relatively dry with P on average about 1 mm d−1. Over Mexico, one P maximum is located along the western slopes of the Sierra Madre Occidental (SMO) and another maximum is located along the east coast of Mexico. The comparison between the GPCP and the unified P data shows that the coarse gridded GPCP data underestimate rainfall amounts over the SMO.

Figure 1.

Mean summer (July–September) P from (a) the GPCP data set from 1991–2000, (b) the unified gridded data set from 1991–2000, (c) GPCP from 1979–2000, and (d) the unified gridded data set from 1950–2000. Contour interval is 1 mm d−1. Zero contours are omitted. Areas where values are greater than 1 (4) mm d−1 are shaded light (dark).

[11] With uncertainties in mind, we examine the RSM summer P climatology (Figure 2a). The RSM captures the major features of the monsoon rainfall over northern Mexico. Heavy precipitation is located over the western slopes of the SMO with a maximum of 9 mm d−1 and another maximum located along the east coast. The rainfall maximum is larger than the maximum of 4–5 mm d−1 depicted by the unified data (Figure 1b), but is comparable with the climatology from the Global Historical Climatology Network (GHCN) and precipitation from the Eta model [Berbery, 2001]. The model captures the relative dryness over the Southwest and wetness over the Southeast. Over the central United States, rainfall is about 3 mm d−1 close to the observations (Figure 1). The model systematically produces excessive rainfall over the Rockies and areas over the Ohio Valley and the eastern United States.

Figure 2.

(a) Mean summer (June–September) P for the period 1991–2000 from the RSM simulations. Contour interval is 1 mm d−1. Zero contours are omitted. Areas where values are greater than 1 (4) mm d−1 are shaded light (dark). (b) Same as Figure 2a, but for the vertically integrated meridional moisture flux [qv]. Contour interval 20 kg (ms)−1. Areas where values are greater than 20 (60) kg (ms)−1 are shaded light (dark). (c) Same as Figure 2a, but for E.

[12] The mean E for June–September (Figure 2c) indicates that large E over the land is usually located over the areas with large P. The magnitudes of E and P are comparable over the central United States. In July and August, E is larger than P (not shown). That is consistent with Ropelewski and Yarosh [1998]. Over the Southwest, E is about 1–2 mm d−1, which is considerably less than E over the Great Plains. The RSM is able to capture both the vertically integrated meridional moisture fluxes [qv] associated with the GPLLJ and the GCLLJ (Figure 2b). There are two [qv] maxima along the path of the GPLLJ. One maximum is located over the Gulf coast between the boundary of Mexico and Texas and another one is located over Oklahoma and Kansas. The [qv] associated with the GCLLJ is much weaker with a maximum located at the boundary between California and Arizona. The moisture fluxes over the Gulf of California cannot be resolved by the NCEP reanalysis [Higgins et al., 1997; Schmidt and Mullen, 1996]. Therefore the RSM simulated moisture fluxes are used to study the physical mechanisms. For E, P and the vertically integrated moisture fluxes from the RSM, anomalies are defined as departures from the 10-yr summer (JJAS) mean (Figure 2).

[13] The simulated P (open circles) is compared with the observed P (dark circles) for selected summers from 1 June (day 1) to 30 September (day 122) (Figure 3). For 1993, the model captures large rainfall episodes in late June and July (days 20–50) during the 1993 floods and relative dry period in August (days 62–80). For 1995, the model is able to capture the dryness over the Southwest in June and July (days 1–59) and rainfall episodes in August and September (days 75–120), but it misses the first rainfall peak in early August (days 60–70). The model captures the wetness in July and August 1999 (days 32–90) over the Southwest. The RSM does not always capture rainfall amounts or every rainfall episode, but overall, it is able to capture the rainfall variability and large extreme rainfall events.

Figure 3.

(a) P averaged over the area (36–42°N, 90–100°W) from the unified gridded data set (open circles) and from the RSM simulation (dark circles) for 1993 from day 1 (1 June) to day 122 (30 September). (b) Same as Figure 3a, but for P averaged over the Southwest (32–36°N, 105–115°W) for 1995. (c) Same as Figure 3b, but for 1999. The unit is 1 mm d−1.

2.3. Procedures

[14] To test the regional dependence of the relationships between E and P, composites were formed based on the monthly mean soil moisture (SM) index or the daily E index. There are only monthly mean SM data available from 1950 to the present. There is no daily SM available for this data set. A monthly mean soil moisture index (SMI) was formed by averaging SM anomalies over a given area [Huang et al., 1996]. The standard deviation was computed for the index. Positive (negative) events are defined based on the threshold criterion. When the SMI is above one standard deviation (or below negative one standard deviation), that month is classified as the positive (negative) event. We then composited P from the unified data for positive and negative events separately. In these composites, data from the RSM are not used.

[15] Unfortunately, soil wetness was not archived from the RSM simulations. Because of uncertainty in E, we composited P based on daily E indices from both the RSM simulations for the period 1991–2000, and the NCEP-NCAR reanalysis for a longer period from 1979–2000. A daily E index was formed by averaging E anomalies over a given area. When the daily E index crosses the threshold of above 85% (or below 15%), that day is considered as the onset day (day 0) of the positive (negative) event. Composites of daily mean P, hydrological or atmospheric variables for June–September based on positive and negative events were obtained from the onset day (day 0) to day 29 based on the E index. When the composite of the 30 -day mean from day 0 to day 29 was formed, we assured that each map only entered the composite once.

[16] The statistical significance of the composite difference between positive and negative events was assessed by the Monte Carlo method. We randomly composited equal number of positive and negative events from the same data set. The process was repeated 500 times. We then counted the number of times that the composite based on any given index is larger than the composite of randomly selected maps. The composite at a given grid point is statistically significant at the 5% level if less than 5% of the randomly selected cases are greater than the composite based on the index.

3. Regional Dependence of the Relationships Between Soil Moisture or E and P

[17] We composited P based on E or SM indices using different data sets to assure that results are independent of the RSM simulations and are not influenced by the uncertainty in E. For monthly means, soil moisture anomalies are positively correlated with E anomalies over the United States [Huang et al., 1996]. Therefore we can establish the association between soil moisture and P based on monthly mean SMIs for different regions of the United States. For daily data, we composited P based on daily E indices using three different combinations of data sets: (i) composites of P from the unified data based on daily E anomaly indices from the reanalysis for the period 1979 to 2000, (ii) composites of P from the unified data based on daily E anomaly indices from the RSM simulations from 1991–2000, and (iii) composites of P from the RSM simulations based on daily E anomaly indices also from the RSM for the period from 1991–2000. The composites differ in details and magnitudes, but they all show that the relationships between P and E are regionally dependent. Because the relationships hold for composites independent of the RSM data, they are not the artifacts of the RSM simulations.

3.1. Composites Based on Monthly Mean Soil Moisture

[18] Two monthly mean soil moisture indices (SMIs) were formed for the Southern Great Plains (30–35°N, 95–105°W) including Texas and Oklahoma and for the Southwest (32–36°N, 105–115°W) including Arizona and western New Mexico respectively. Both indices were computed for June–September from 1950 to 2000 based on the climate division data from Huang et al. [1996]. For each index, composites of soil moisture, P from the unified data and E from the reanalysis were obtained for positive and negative events based on the criterion of one standard deviation. Statistical significance was assessed using the Monte Carlo method. The composites for positive and negative indices show the same pattern but opposite in sign, so the composite differences are given. There are 25 (34) positive and 30 (27) negative events for the composite based on the SMI over the Southern Great Plains (Southwest).

[19] The composite difference of soil moisture between positive and negative cases based on the SMI over the southern Great Plains shows large anomalies with a maximum of 60 mm centered over Texas (Figure 4a). In comparison, the anomalies based on the SMI over the Southwest are weaker (Figure 4d). Soil moisture anomalies over the Southern Plains are associated with large P anomalies north of the E anomalies and areas downstream along the path of the GPLLJ to Indiana and Ohio (Figure 4b). There are also positive P anomalies along the Gulf States. Figure 4b resembles the P pattern associated with SM anomalies over the Southern Plains obtained from the constructed analogue method [van den Dool et al., 2003]. The composite based on the SMI over the Southwest does not show statistically significant P anomalies (Figure 4e). The composite map (Figure 4e) does not pass the field significant test of Livezey and Chen [1983]. The composites of E from the reanalysis (Figures 4c and 4f) confirm the close correspondence between E and soil moisture. Positive E anomalies are located over the areas of large positive soil moisture anomalies.

Figure 4.

Composite difference of monthly mean soil moisture anomalies from Huang et al. [1996] between positive and negative events based on the SM index for the southern Great Plains (30–35°N, 95–105°W). Contour interval is 20 mm. Areas where positive (negative) values are statistically significant at the 5% level are shaded dark (light). (b) Same as Figure 4a, but for the composite difference of monthly mean P from the unified gridded data. Contour interval is 0.3 mm d−1. Zero contours are omitted. (c) Same as Figure 4a, but for monthly mean E anomalies from reanalysis. Contour interval is 0.2 mm d−1, (d–f) same as Figures 4a–4c, but based on the SM index over the Southwest (32–36°N, 105–115°W).

3.2. Composites Based on the Daily E Indices

[20] Three daily E indices were formed for the northern Great Plains (36–42°N, 95–105°W), the southern Great Plains (30–35°N, 95–105°W), and the Southwest (32–36°N, 105–115°W) respectively. The areas are masked on Figure 5d. Figure 5 shows the composites of the P anomalies from the unified data based on the daily mean E indices from the reanalysis for 1979–2000. The E events were selected based on the criterion of the 85th percentiles. When the E index is over the threshold, that day is considered as the onset day (day 0). The composites for the positive and negative events are similar with a phase reversal, so Figure 5 shows the P differences averaged from day 0 and day 29 between positive and negative events for June–September. These calculations were repeated for different thresholds and for July to August. The magnitudes of anomalies differ, but the major conclusion that the relationships between E and P are regionally dependent does not change. For the composites based on the E index over the northern (southern) Plains, there are 452 (445) positive events and 402 (461) negative events. For composites based on the E index over the Southwest, there are total 470 positive events and 459 negative events.

Figure 5.

Composite difference of mean P from day 0 to day 29 between positive and negative events based on the E index over (a) the northern Great Plains (36–42°N, 95–105°W), (b) the southern Great Plains (30–35°N, 95–105°W), and (c) the Southwest (32–36°N, 105–115°W). The P anomalies were taken from the unified data and the E indices were calculated from the reanalysis data for 1979 to 2000. Contour interval is 0.5 mm d−1. Zero contours are omitted. Areas where positive (negative) values are statistically significant at the 5% level are shaded dark (light). (d) E index areas: Southern Plains, northern Plains and Southwest.

[21] Composites of P from both the unified data and from the RSM (Figure 6) were obtained based on the E indices from the RSM simulations for a shorter period from 1991–2000. Since the spin up time for the RSM is roughly 3 days, the first 3 days of each month are not included in the composites. For composites based on the E index for the northern (southern) Plains, there are 170 (165) positive events and 159 (171) negative events. For composites for the Southwest, there are 145 (132) positive (negative) events.

Figure 6.

(a–c) Same as Figure 5, but P anomalies were taken from the unified gridded data and the E indices were calculated from the RSM simulations for the period 1991–2000. (d–f) Same as Figures 6a–6c, but both P and E anomalies were taken from the RSM simulations.

[22] For composites based on E index for the northern Plains, they (Figures 5a, 6a, and 6d) all show similar patterns, even though the magnitudes of anomalies and the details of the pattern differ. The common features are positive P anomalies extending from the location of the E anomalies northward to Canada and eastward to the Ohio Valley. Positive anomalies are also found along the coast of Gulf of Mexico. For composites based on E from the RSM for 1991–2000 (Figures 6a and 6d), there are also positive anomalies over the western slopes of the SMO, but this feature is absent from the composite based on the reanalysis for a longer period (1979–2000) (Figure 5a).

[23] The composites based on the E index for the southern Plains (Figures 5b, 6b and 6e) all show positive anomalies located over the central United States and over the Ohio Valley downstream along the path of the GPLLJ. The composite based on the reanalysis (Figure 5b) shows positive anomalies along the Gulf of Mexico but these anomalies are much weaker on the composite maps based on indices from the RSM. There are similarities between composites based on the E indices over the northern and southern Plains because the P anomalies over both regions are influenced by the GPLLJ. The composites based on the E indices over the Southwest do not show statistically significant anomalies.

[24] Composites of P based on the soil moisture (Figure 4) or E anomalies from different sources (Figures 5 and 6) all show that the relationships between E and P are regionally dependent. The details of composites differ for different data periods and data sources, because of the uncertainty in E and P. Therefore only the common features are discussed. When the E or soil moisture anomalies are located over the Great Plains, the positive P anomalies extend from the area north of the E anomalies to the areas downstream along the path of the GPLLJ. On contrast, the P composites (Figures 5 and 6) do not have statistical significant P anomalies associated with E anomalies over the Southwest.

4. Physical Mechanisms

[25] The moisture sources and recycling rates for the central United States and the Southwest are very different. Brubaker et al. [2001] examined the sources of E over the Mississippi River basin. They found that about 36% of the warm season P is originated from E locally and 20% comes from the Gulf of Mexico and the Caribbean. Bosilovich and Schubert [2002] showed that the recycling rate over the Southwest is smaller than the rate over the Midwest.

[26] E can influence P directly through the mean convective instability or indirectly through the changes of the low-level flow and the moisture flux convergence. If soil moisture or E anomalies are strong and persistent, then E anomalies have better chance to influence P locally or downstream. The characteristic time To [Leith, 1982; Trenberth, 1985] calculated from the auto-correlations is a good measure of persistence. To calculated from the RSM simulations (Figure 7a) should be compared with To calculated from the reanalysis for summers from 1979–2000 (Figure 7c). Details differ, but both show that E over the area extending from the entrance of the GPLLJ to the central United States is about 12–16 days. To for E is about 12–14 days over the western slopes of the SMO. E is less persistent over the Southwest. To for E is only about 2–4 days over Arizona and about 4–6 days over New Mexico. The disagreement is found over the northwestern United States, where To from the reanalysis is much larger (about 22 days) than that from the RSM (14 days).

Figure 7.

(a) Characteristic time To for E from the RSM simulations for summer (June–September) 1991–2000. The contour interval is 1 day. Values greater than 8 days are shaded. (b) Same as Figure 7a, but for [qv]. Values greater than 4 days are shaded. (c) Same as Figure 7a, but for the reanalysis from 1979–2000.

[27] The major moisture sources for P over the central United States and over the Southwest are regulated by two low-level jets: the GPLLJ and the GCLLJ respectively. The climatology indicates that the [qv] associated with the GPLLJ is overall about 2.5 times stronger than the [qv] over the Southwest (Figure 2b). The [qv] along the path of the GPLLJ is also more persistent (Figure 7b). To for [qv] over the entrance of the GPLLJ is about 9–10 days. To for [qv] is about 5–7 days over the southern Plains and about 4 days over the northern Plains. In comparison, To is smaller for [qv] along the path of the GCLLJ. To is only 3 days over the Gulf of California, and it is about 3.5 days over the maximum of [qv] over the western Arizona. The reanalysis cannot resolve the [qv] associated with the GCLLJ, so To for [qv] from the reanalysis is not shown.

[28] The [qv] anomalies associated with the GPLLJ are stronger and persist longer in comparison with the [qv] anomalies associated with GCLLJ. More moisture is transported into the Great Plains by the GPLLJ constantly. Therefore E anomalies can be maintained and be more persistent than the E anomalies over the Southwest, where moisture is largely supplied by surge events carried by the GCLLJ. With these results in mind, we examine the relationships between E and P for the central United States and the Southwest separately.

4.1. Central United States

[29] The physical processes related E and P anomalies over the southern and northern Great Plains are similar. We will discuss the southern Great Plains case in details. Because [qv] anomalies are needed, all composites are based on the RSM data. Composites were obtained based on the daily E index for the southern Plains. These are the same events selected for the P composites (Figure 6e).

[30] The 30-day mean composite E difference (Figure 8a) between positive and negative events shows positive values more than 1.5 mm d−1 extending from Texas to the northern Great Plains. Positive anomalies also extend to the Ohio Valley. There are areas with large T0. The persistence of E gives more opportunity for the direct influence to occur.

Figure 8.

Composite difference of E averaged from day 0 to day 29 between positive and negative events based on the E index over the southern Great Plains (30–35°N, 95–105°W). Contour interval is 0.5 mm d−1. Areas where positive (negative) anomalies are statistically significant at the 5% level are shaded dark (light). (b) Same as Figure 8a, but for the 850 hPa height. Contour interval is 2 m. (c) Same as Figure 8a, but for the vertically integrated meridional moisture (qv). Contour interval is 10 kg (ms)−1. (d) Same as Figure 8b, but based on the E index over the northern Great Plains (36–42°N, 95–105°W). Contour interval is 2 m.

[31] In addition to the direct impact, the changes in E can alter the magnitude and position of the GPLLJ. The latent heat is largely balanced by the sensible heat because of the changes in radiation are small. The increase of E (latent heat) is largely balanced by the decrease of the sensible heat (not shown). Less sensible heat indicates cooler surface temperature over the areas with large E. The circulation changes in 850 hPa heights associated with cooler temperature (or the increases of E) are the negative anomalies located over the western region of the United States (Figure 8b). This anomaly pattern implies stronger low-level meridional winds from the Gulf of Mexico to the central United States and weaker winds from the Gulf of California to the Southwest. That means a stronger GPLLJ and a weaker GCLLJ as indicated by the [qv] difference (Figure 8c) because [qv] is dominated by low-level flow. When the E anomalies are placed in the northern Great Plains, the changes in the 850 hPa height anomalies show negative anomalies over the western region and positive anomalies to the east (Figure 8d). Both flow patterns imply the strengthening of the GPLLJ.

[32] In Figure 9, composites for the positive and negative E events based on the daily E index over the southern Great Plains are given separately. The P composites for the positive case and the negative case are similar but opposite in phase (Figures 9a and 9c). For the positive events, the moisture flux extends northward from the Gulf of Mexico to the central United States and then turns eastward downstream along the path of the GPLLJ. For the negative events, the moisture fluxes are weaker in the areas of negative E anomalies and downstream of the GPLLJ. The P anomalies for both positive and negative events are roughly located north of the E anomalies and along the path of the GPLLJ. The negative cases also show a stronger GCLLJ. The stronger moisture transport to the Southwest from the Gulf of California is consistent with more rain there.

Figure 9.

(a) Composite of P anomalies from the RSM simulations for positive events averaged from day 0 to 29 based on the E index over the southern Great Plains from the RSM simulations. The anomalies are defined as the departures from the mean averaged over 10 summers (June–September) from 1991–2000. Contour interval 0.5 mm d−1. Areas where positive (negative) anomalies are statistically significant at the 5% level are shaded dark (light). (b) same as Figure 9a, but for the vertically integrated moisture flux (Qflux) (vector) with the unit 300 kg (ms)−1. The meridional component (qv) is contoured every 10 kg (ms)−1. Values greater than 40 kg (ms)−1 are shaded. (c and d) Same as Figures 9a and 9b, but for the negative events.

[33] From Figure 9, we conclude that the largest changes in P associated with the E anomalies over the Great Plains are located downstream from the E anomalies and are influenced by the changes in the GPLLJ. When positive soil moisture or E anomalies are located over the Great Plains, the GPLLJ strengthens. E anomalies along the path of the GPLLJ change the position and magnitude of the jet. The associated changes in moisture convergence alter P. More (less) P increases (decreases) soil moisture, and therefore there is the increase (decrease) in E. The positive feedback makes E more persistent. The strengthening of E also gives more opportunity for direct influence to occur. Therefore it enhances the net impact of E on P.

4.2. The Southwest

[34] E over the Southwest is less persistent than E over the Great Plains and the [qv] anomalies are also weaker and less persistent in comparison with [qv] associated with the GPLLJ (Figure 7). The moisture sources for the Southwest are mainly from the north Pacific and from the Gulf of California. Studies from Douglas [1995], Douglas et al. [1993], Brenner [1974], Hales [1972, 1974], and many others have demonstrated that the moisture surge from the Gulf of California is the necessary condition for rain over Arizona. This suggests that if the E anomalies occur during strong surge events, they are able to maintain because the GCLLJ transports moisture into the Southwest. In this case, E anomalies may have a chance to influence P. If E increases or decreases during the periods without surge events, then E anomalies diminish in 3–4 days and there is no impact on P. The surge events are not controlled by the E anomalies over the Southwest. They are associated with large-scale flow in midlatitudes and easterly wave disturbances in the eastern Pacific [Stensrud et al., 1995, 1997]. These surge events indicated by the [qv] anomalies only last on average about 4 days. E has the decaying time of 4–6 days. The window that E can influence P during surge periods is small. Therefore the net impact is weak Next, the RSM experiments are used to support the above theory.

5. Impact of E on P Over the Southwest

[35] The July 1995 and July 1999 cases were chosen as examples. July 1995 was a dry month (Figure 3) over the Southwest. There was one rainfall period lasting from 11 July to 21 July. July 1999 was a wet month over the Southwest. Rainfall started around 3 July and lasted to the end of the month. Associated with rainfall, there were continuous surge events from the Gulf of California. We will show that the responses to E anomalies over the Southwest were very different for these two cases.

5.1. July 1995

[36] July 1995 was a dry monsoon month. The only raining period was from 11–21 July and daily rainfall was less than 2.5 mm d−1 over the Southwest (Figure 10c). For the month of July, there was only 1–2 mm d−1 rainfall over New Mexico and Arizona was largely dry (Figure 10a). The RSM simulation, which is the control experiment for this case shows that the model captures the dryness during July. The mean P has about 1–2 d−1 rainfall over New Mexico and over the eastern boundary of Arizona (Figure 10b). There was no rain over the western part of Arizona. The simulated rainfall episodes (dark crosses) start later than the observations (Figure 10c, solid line), and the magnitudes of rainfall are weaker. The daily mean of [qv] shows only three surge periods: 7–9 July, 12–15 July and 19–21 July (Figure 10d, solid line).

Figure 10.

(a) P for July 1995 from the unified gridded data. Contour interval 1 mm d−1. Values greater than 1 (4) mm d−1 are shaded light (dark). (b) Same as Figure 10a, but for the Control RSM simulation. (c) Daily observed P (open circles), from the control experiment (crosses) and from the wet experiment (solid line) averaged over the Southwest (32–36°N, 105–115°W). (d) The [qv] averaged over (32–36°N, 107–115°W) for the control (solid line) and the wet experiment (crosses). (e) E difference between the control experiment and wet experiment averaged over the Southwest (32–36°N, 105–115°W).

[37] Recall that the RSM simulation was initialized with soil moisture taken from the reanalysis 2. In the 1995 wet experiment, we increased the model soil wetness at layer 1 (0–10 cm) and layer 2 (10–200 cm) over the Southwest (32–36°N, 105–115°W) to 125% of its original values from the reanalysis II. The boundary conditions and initial surface conditions including soil wetness elsewhere were kept the same as the control.

[38] Overall, there is no change in total P over AZNM. The control experiment has the monthly mean rainfall of 0.30 mm d−1 for July, while the wet experiment has 0.29 mm d−1. Both are less than the observed 0.68 mm d−1. The evolution of P over AZNM (Figure 10c) indicates that there is more rain for the wet experiment at the beginning while the model adjusts to the added soil moisture. E anomalies decay rapidly (Figure 10e). When the first surge event occurred on 7 July, the E anomalies already decreased to less than 1 mm d−1. Therefore E anomalies have no impact on [qv] or the next P event on 11 July 1995 (Figure 10d). In this case, E anomalies are added during the period without surge events and there is no impact on P.

5.2. July 1999

[39] The situation is different for July 1999. It was a wet monsoon month with continuous raining episodes and moisture surge events. The mean P from the unified data (Figure 11a) indicates that most areas over AZNM received rainfall over 1 mm d−1 with a maximum of 4 mm d−1. Texas was relatively dry and there was about 2 mm d−1 rainfall over the central United States. The 1999 control RSM simulation with soil wetness obtained from the reanalysis II (Figure 11b) is able to capture the wetness over AZNM with a maximum of 4 mm d−1 and the dryness over Texas. During July 1999, rainfall started on 3 July and lasted for the entire month (Figure 12a, open circles). The mean [qv] for the month from the control experiment shows that the [qv] associated with the GCLLJ is strong (Figure 11f). The maximum is located near its climatological position, but the magnitude is above 80 kg (ms)−1, which is very large in comparison with the 10-year mean (Figure 2b). The strong [qv] is largely contributed by continuous surge events through the month (Figure 12d). There were only two brief periods: 8–10 July and 22–24 July that [qv] averaged 107–113°W at 32°N was negative (northerly winds).

Figure 11.

(a) P for July 1999 from the gridded precipitation data. Contour interval 1 mm d−1. Values greater than 1 (4) mm d−1 are shaded light (dark). (b) Same as Figure 11a, but for the 1999 Control experiment. (c) Difference of P for July 1999 between the control experiment and the dry experiment. Contour interval 0.5 mm d−1. Zero contours are omitted. Positive values are shaded. (d) Same as Figure 11c, but for E difference between the control and the dry experiment at day 1. Contour interval 0.5 mm d−1. (e) Same as Figure 11d, but the E difference averaged over July 1999. (f) Vertically integrated meridional moisture flux for the control experiment. Contour interval is 20 kg (ms)−1. Values greater than 60 kg (ms)−1 are shaded.

Figure 12.

(a) Daily P averaged over the Southwest (32–36°N, 105–115°W) from observations (open circles), the control experiment (solid line) and the dry experiment (crosses), (b) E every 6h from the control (solid line) and the dry experiment (crosses), (c) difference in daily moisture divergence between the control and the dry experiments, and (d) the cross section of the meridional moisture flux (qv) averaged over (107–115°W) at 32°N for the Control 1999 experiment.

[40] The mean E over AZNM on 1 July was about 5.2 mm d−1 and that was already above the 85th percentile calculated based on the 10-year RSM climatology. Increasing E another 125% will make E unreasonably high. Therefore a dry experiment with less soil moisture was performed. The 1999 dry experiment was performed with the same boundary conditions and initial surface conditions as the control experiment except soil moisture over AZNM. We decreased the model soil wetness at both layer 1 (0–10 cm) and layer 2 (10–200 cm) over AZNM (32–36°N, 105–115°W) to 75% of the values from the reanalysis II at day 1 (1 July 1999). Results are given as the differences between the 1999 control and 1999 dry experiments.

[41] The difference of E over AZNM between the control and dry experiments at day 1 is about 2 mm d−1 with a maximum over eastern New Mexico (Figure 11d). Within 8 days, the E difference over AZNM drops to about 1.5 mm d−1 but it remains positive through the whole month (Figure 12b). The E difference averaged over the month shows positive E over the Southwest with a maximum located in New Mexico. The mean P difference shows positive P anomalies over the area of positive E anomalies. After 10 days, there is an overall increase of [qv] associated with the GCLLJ, which contributes to the moisture convergence over the region (Figure 12c). Both E and the moisture convergence contribute to the increase of P (Figure 12a). In agreement with Kanamitsu and Mo [2003], the moisture surge events from the Gulf of California are important to rainfall over Arizona. Continuous inflow of moisture from the Gulf maintains E anomalies over the Southwest. That gives an opportunity for the direct impact through the convective instability or the indirect impact of E on P through the strengthening of the GCLLJ to occur.

[42] We presented two different cases: one dry case and one wet case over the Southwest. For the dry case, E anomalies were added through soil moisture over the Southwest during a dry period without strong surge events. E anomalies last for about 4–6 days and diminish. This is consistent with the small characteristic time To over the Southwest. When the next surge event occurred 6–8 days later, the E anomalies were too weak to induce changes in [qv]. Therefore the net impact is very small. During the wet event, less soil moisture was given in the beginning of a wet month. Surge events brought moisture to the region and maintained the E anomalies through the month. Both E and the strengthening of [qv] in the control experiment increase the moisture convergence. Therefore P increases.

6. Conclusions

[43] Relationships between soil moisture or E and P over the United States are examined using observations, the NCEP-NCAR reanalysis and 10-yr RSM summer simulations. Composites of P were formed based on monthly mean SMIs and daily E anomaly indices over the Great Plains and the Southwest. Because of uncertainty in both P and E, composites were computed using different data sets for different data period. Only the common features on composite maps are discussed. All composites indicate that the relationships between E and P anomalies are regionally dependent. The E anomalies over the Great Plains are associated with P downstream from the E anomalies and in the areas along the path of the GPLLJ. However, over the Southwest, there is no association between P and E anomalies statistically.

[44] The P events over the central United States and the Southwest are modulated by the GPLLJ and the GCLLJ respectively. The [qv] associated with the GPLLJ is much stronger and persistent than the [qv] associated with the GCLLJ. GPLLJ consistently brings moisture into the Great Plains and maintains the E anomalies. Therefore E is more persistent as indicated by large To. That gives more chances for the direct impact from E on P to occur. Soil moisture anomalies along the path of the GPLLJ can also alter the position or the strength of the GPLLJ and the changes in the moisture convergence influence P downstream. P anomalies wet the soil and have positive feedback to E. This may prolong the decaying time of E. Therefore the net impact on P is large.

[45] Over the Southwest, P is influenced by the moisture surges from the Gulf of California. The surge events are not controlled by E locally, but are influenced by the large-scale flow and easterly waves in the eastern Pacific [Stensrud et al., 1995]. During strong and continuous moisture surge periods, strong soil moisture or E anomalies over the Southwest are able to persist and impact P. When E anomalies occur during the period without moisture surges from the Gulf, E anomalies decay quickly and have no impact on P. Both E and [qv] anomalies over AZNM have T0 less than 4 days. The window that large E anomalies occur during the strong surge events is small. Therefore the net impact is also small.

Acknowledgments

[46] This project is supported by the grant from the OGP/GAPP program GC02-102.

Ancillary