Interannual variability of surface evaporative moisture sources of warm-season precipitation in the Mississippi River basin

Authors


Abstract

[1] Patterns of variability of the evaporative sources of Mississippi River basin precipitation are investigated. Time series of selected regions' evaporative contributions to warm-season precipitation in the Mississippi and its subbasins are produced on a 15-day basis for the years 1963 to 1998. Source region contributions are estimated by backward-trajectory analysis, which tracks water vapor backward in time from its site of condensation to its point of evaporation. The National Centers for Environmental Prediction (NCEP) reanalyses are used for the required atmospheric dynamics and observed data from NCEP Climate Prediction Center for precipitation. Correlation and principal component analyses of the source contribution and precipitation anomaly time series reveal a spatial pattern consisting of internal (Missouri, Red/Arkansas, and Lower Mississippi) and external (Texas/Mexico and Caribbean) sources, which explains statistically much of the interannual variability in warm-season precipitation. Further analysis of the pattern reveals that the external sources play a more important role than internal sources in the anomalies of precipitation. Examining a subset of the period, from 1979 to 1998, moisture flux analysis confirms the role of the external sources by showing the strong association between the precipitation anomalies and flux across the southern boundaries of the basin (which includes the contributions from Texas/Mexico and Caribbean). An index is introduced, based on measurable surface pressure differences between centers over the Atlantic Ocean and central United States (Mississippi watershed). The index is potentially useful in forecasting slow variations of precipitation supply in the basin.

1. Introduction

[2] This study analyzes the evaporative moisture sources of warm-season precipitation in the Mississippi River basin, with particular attention to the question of how interannual variability of precipitation may be related to variability of the sources. Patterns identified using multivariate statistical analysis are interpreted in terms of physical analysis of moisture transport on the hemispheric scale.

[3] Water acts as a means of heat transport and temperature regulation in the Earth system. The energetic role of the water cycle basically establishes the links among its components: the land phase (traditionally the focus of the science of hydrology), and the atmospheric phase (the focus of meteorology).

[4] Recent interdisciplinary scientific efforts such as the Global Energy and Water Experiment (GEWEX) Continental International Project (GCIP) and its successor, the GEWEX Americas Prediction Project (GAPP), have focused partly on understanding the mean state and variability of the water cycle and its potential response to climate change. Understanding the water cycle on continental to global scales is expected to provide better predictions that are useful for water resources management. The availability of global observational data sets and Global Circulation Models (GCMs) has fostered a growing understanding of the interactions and feedbacks among the water cycle and other biogeophysical processes and within the water cycle itself, although the data are incomplete and the models are imperfect. The most important implication of this growing understanding is that the analysis of the local and regional water cycles should be broadened spatially and temporally as no region of the globe is truly isolated hydrologically.

[5] A number of studies have attempted to identify and estimate the evaporative sources of regional precipitation supply. Brubaker et al. [1993] and Eltahir and Bras [1994] used models with moisture flux and evaporation data to estimate the precipitation recycling ratio, which is the ratio of precipitation originally from local evaporation to the total precipitation in a region. Dirmeyer and Brubaker [1999] used backward-trajectory analysis as a tool for identifying and estimating evaporative sources for precipitation supply in the Mississippi River basin. This method is an improvement over the extended Budyko's model [Budyko, 1958] used by Brubaker et al. [1993] because it can capture temporal variability of the moisture fluxes to, from, and within the region of interest, and their covariability with precipitation events, by using higher temporal resolution data sets of hydrometeorological parameters (rather than monthly mean values). Koster et al. [1986] and, more recently, Bosilovich and Schubert [2002] tagged evaporated water (as a tracer) and followed it within GCM simulations to identify and estimate surface evaporative sources for precipitation supply in a region. The results of these studies depend heavily on the physical parameterizations within the GCM.

[6] The studies of Brubaker et al. [1993] and Eltahir and Bras [1994] have been able to estimate climatological values of the local surface evaporative source contributions to precipitation in a region while the GCM studies have a larger spatial coverage (i.e., local and remote sources). However, neither set of studies did a very extensive analysis on the interannual variability of the contributions. This study extends the climatological analysis of Brubaker et al. [2001] by performing a spatiotemporal analysis of the interannual variability of the evaporative sources of warm-season precipitation in the Mississippi River basin (M) over a 36-year period, from 1963 to 1998.

2. Data and Methods

[7] This study analyzes results from the backward-trajectory analysis performed by Brubaker et al. [2001] for warm-season precipitation in the Mississippi River basin over the 36-year period. In short, backward-trajectory analysis tracks parcels of air contributing moisture to precipitation events in a (sink) region of interest during a time period t backward in time through the solutions of equations of mass, momentum, and energy to evaporation events in the source regions that contributed to the parcels' moisture. Hence the evaporative sources for the precipitation supply in the region of interest can be identified and their contributions can be estimated. (Note that the evaporation in source regions, local or remote, does not necessarily occur during period t because of the variable time lag between precipitation and evaporation events.) Details of the backward-trajectory analysis and its application in the Mississippi are given by Dirmeyer and Brubaker [1999] and Brubaker et al. [2001]. These studies use the hourly observed precipitation data of Higgins et al. [1996] and the National Centers for Environmental Prediction (NCEP) reanalysis of Kalnay et al. [1996] for other hydroclimatological fields. Using perturbation sensitivity analysis, Brubaker et al. [2001] concluded that source contribution estimates are somewhat sensitive to systematic errors but not to random errors in the NCEP evaporation field. Figure 1 shows the 19 regions of Dirmeyer and Brubaker [1999] and Brubaker et al. [2001] within the spatial domain of the present study. To be consistent with those studies, the regions in this study are delineated on the NCEP reanalysis' T62 Gaussian grid. The extent of the domain is based on the probable extent of the trajectories of the parcels of evaporated moisture supplying precipitation in the Mississippi River basin (outlined by a thick line). The definitions of the sink/source abbreviations are presented in Table 1.

Figure 1.

Sink/source regions for the Mississippi River basin (outlined in thick line) delineated on the NCEP reanalysis (T62 Gaussian) grid [Brubaker et al., 2001].

Table 1. Sink/Source Region Definitionsa
AbbreviationRegion
Subbasins of the Mississippi River Basin
LMLower Mississippi
MOMissouri
OHOhio
RARed/Arkansas
UMUpper Mississippi
 
The Mississippi River Basin
MLM + MO + OH + RA + UM
 
Other Land Source Regions
ARCArctic
NNANorthern North America
WESWestern N.A.
SOWSouthwest N.A.
TMXTexas/Mexico
SOESoutheast N.A.
CSACentral and South America
 
Ocean Source Regions
TEPTemperate Pacific
TRPTropical Pacific
BAJPacific in the North American monsoon region
TEATemperate Atlantic
TRATropical Atlantic
GOMGulf of Mexico
CARCaribbean

[8] The warm season in this study is defined as a period from 16 March to 17 August. The backward-trajectory analysis was done for a pentad-wise accumulated precipitation; there are 31 pentads during the warm season. Because the analysis accounted for evaporation events up to 20 days before the end of the pentad (hence a 20-day run period), the analysis actually covers a period from 1 March to 17 August. Table 2 shows the time indices used in this study.

Table 2. Time Indices Used in This Study
DatePentad15-Day (f)MonthSeason
2 (3) September to 1 (2) October1–61–2SepSON
2 (3) October to 31 October (1 November)7–123–4OctSON
1 (2) November to 30 November (1 December)13–185–6NovSON
1 (2) December to 30 (31) December19–247–8DecDJF
31 December (1 January) to 29 (30) January25–309–10JanDJF
30 (31) January to 28 (29) February31–3611–12FebDJF
1 March to 30 March37–4213–14MarMAM
31 March to 29 April43–4815–16AprMAM
30 April to 29 May49–5417–18MayMAM
30 May to 28 June55–6019–20JunJJA
29 June to 28 July61–6621–22JulJJA
29 July to 17 August67–7023AugJJA

[9] Using the backward-trajectory method, an evaporative source contribution time series was created for each of the regions identified in Figure 1, for each pentad in Table 2, for each year in the study. Summing the time series for all sources yields the time series of pentad-wise total precipitation in the basin. For the current study, the pentad-wise results have been further aggregated into 15-day time periods (three pentads), 10 time periods per year. The statistical portion of this study consists of analyzing this (19 source region)-by-(360 time period) array. Each column consists of the time series of precipitation contributed by the corresponding source in each of 10 warm-season periods for 36 years. In the results section, the data are summarized separately as time series, and analyzed collectively using cross-correlations and principal components analysis.

2.1. Principal Components Analysis

[10] Principal components analysis (PCA) is a statistical tool used to analyze intercorrelated multivariate data sets such as the source contribution time series presented here. Using PCA, we identify important spatiotemporal patterns in the interannual variability of source contributions in the Mississippi River basin during warm seasons over the 36-year period on a 15-day basis. In essence, principal components are orthogonal (uncorrelated) linear combinations of the variables that capture the shape of the multivariate data in hyperspace. Principal components were computed on the basis of the rules described by McCuen [1993]. First, 19 standardized anomaly time series were created by subtracting the respective mean intraannual cycle from each source contribution time series, and dividing by the respective standard deviation. The result is a (360 time period)-by-(19 source region) matrix of zero-mean, unit-variance anomalies, X. Pre-multiplying X by its transpose results in a symmetric 19 by 19 matrix of correlations between all pairs of source contributions, r:

equation image

[11] An orthonormal matrix v whose columns are the eigenvectors of r and a diagonal matrix e of corresponding real values can be found such that

equation image

Since e is a diagonal matrix, eT is equal to e, and the product eTe is the diagonal matrix of eigenvalues, λ. The matrix product of v and eT is a 19 by 19 matrix of factor loadings, each column corresponding to an eigenvector multiplied by its respective eigenvalue. The factor loading matrix transforms the original matrix of anomalies into a 19 by 19 matrix of principal components, U.

equation image

[12] Each column of U is known as a principal component (PC) of the system. Unlike the columns of X (the original standardized anomaly time series), the columns of U are orthogonal, and each column represents a statistically independent time series of combinations of the original variables. The eigenvalues indicate the respective importance of the PCs in explaining correlation in the data set. The first PC, or PC-1, is the column of U corresponding to the highest eigenvalue. PC-1 is parallel to the dominant axis of the data cloud in 19-dimensional space. Each principal component i is said to explain a share of the total system variance proportional to λ2. Each PC may be obtained separately by operating on the original data matrix with its respective column of the factor loading matrix,

equation image

2.2. Water Vapor Flux

[13] For physical interpretation of the statistical results, it is also of interest to analyze the flux of moisture across a boundary of the region of interest. The flux of moisture is the line integral of the perpendicular component of the moisture transport vector during a period of time. The vertically integrated moisture transport vector is defined as [Peixoto and Oort, 1992]

equation image

where p0 is the surface pressure, q is the specific humidity [kg of H2O per kg of air], V is the horizontal wind vector [m s−1], i and j are the zonal (west to east) and meridional (south to north) unit vectors, and g is the acceleration of gravity. The specific humidity (q) and the wind components vary with height in the atmosphere. Equation (5) uses the meteorological convention of using pressure rather than elevation as the vertical coordinate.

[14] The flux of moisture across the boundary of a region of interest can be computed as follows.

equation image

where A is the area of the region, nγ is the outward unit normal vector at any point on the boundary, and γ is the horizontal coordinate along the boundary of the region. Equation (6) uses the hydrologic convention of dividing by the total area of the region to express moisture import (or export) in terms of depth.

[15] Moisture flux analysis was available from a different, related study. The analysis is done for the entire year and covers a 19-year period from September 1979 to August 1998, with time indices as in Table 2. The moisture flux information allows us to seek physical explanations for the statistical findings of the precipitation source analysis.

3. Results

[16] Summary statistics of warm-season precipitation in the Mississippi River basin over the 36-year period, from 1963 to 1998, are presented as box and whisker plots in Figure 2. The values are given in kg m−2 d−1, averaged by the total number of grid cells in the basin (99). Each box-plot is a non-parametric statistical summary of 36 years' values during one of the ten 15-day warm-season periods in this study (from f14 through f23; see Table 2). Each plot shows the median (the line in the box), the lower and upper quartile (the lower and upper edges of the box), and the minimum and maximum values (the ends of the whiskers), excluding outliers and extremes. Outliers (the unfilled dots) are values lying between 1.5 to 3 times the interquartile range (IQR) from the edges of the box. Thus each box plot shows the central tendency and interannual variation for a particular 15-day period. The medians of the box plots in Figure 2 show the seasonal cycle of warm-season precipitation in the Mississippi River basin, with a maximum between the end of May (f18) and the beginning of June (f19). There are only four outliers in 15-day precipitation over the 36-year period identified in Figure 2. Two of the outliers, late June 1988 (f20) and early July 1993 (f21), correspond to the record drought and flood, and are within the periods of, respectively, La Niña and El Niño events as described by Trenberth [1997]. The other two high outliers, during 1973 and 1995, occurred after the end of El Niño events during those years.

Figure 2.

Box and whisker plots of warm-season precipitation, in kg m−2 d−1, in the Mississippi River basin (M) over the 36-year period on a 15-day basis. Box plots indicate upper and lower quartiles, the solid centerline is the median, and the ends of the whiskers are the minimum and maximum (excluding outliers and extremes). Outliers (the unfilled dots) are cases with values lying between 1.5 to 3 times the interquartile range (IQR) from the edges of the box. Extremes (the stars) are cases with values greater than 3 times the IQR from the edges of the box.

[17] Table 3 shows the correlation coefficient matrix of source contribution anomalies in the Mississippi River basin during the warm season over the 36-year period on a 15-day basis. The anomalies are the standardized departures from the interannual mean of each 15-day period. Because there are ten 15-day periods within the warm season, there are 360 data pairs (n = 360) for the 36-year period. The correlation coefficients in Table 3 are statistically significant (ν = 358; α = 1%; double-sided t-test).

Table 3. Correlation Coefficient Matrix of Source Contributions and Precipitation Anomalies in the Mississippi River Basin (M) During the Warm Seasons of the 36-year Period on a 15-day Basisa
Source/SinkMOGOMTMXCARWESBAJLMRANNATEPUMSOETEATRPOHSOWTRACSAARC[PwarmMy]
  • a

    The values shown are the correlation coefficients between source contribution anomalies and source contribution anomalies. The last column shows the correlation coefficients between source contribution anomalies and precipitation anomalies. Because the correlation coefficient matrix is a symmetric matrix, only the upper portion is shown. The sources are ranked in descending order on the basis of their warm-season mean contributions throughout the 36-year period [Brubaker et al., 2001]. Values greater than ∣0.6∣ are in bold.

MO1.000.340.510.330.680.470.440.720.470.440.570.190.130.260.260.300.250.290.160.72
GOM 1.000.610.510.320.500.460.420.040.430.150.410.410.430.100.370.180.190.070.73
TMX  1.000.540.520.480.420.640.070.400.190.230.300.380.050.670.360.480.060.78
CAR   1.000.300.450.490.320.130.400.250.210.420.360.210.270.670.520.120.75
WES    1.000.360.280.540.420.520.290.150.190.230.150.330.250.350.210.66
BAJ     1.000.330.460.110.570.180.160.100.470.100.340.270.220.050.66
LM      1.000.550.160.320.550.660.300.210.620.210.390.240.010.67
RA       1.000.180.320.410.280.220.280.250.390.290.26−0.010.68
NNA        1.000.230.490.03−0.050.030.330.080.100.180.450.34
TEP         1.000.250.170.120.350.190.250.230.300.140.64
UM          1.000.220.020.120.610.170.210.190.070.47
SOE           1.000.500.120.430.090.190.060.000.44
TEA            1.000.090.120.120.340.110.000.45
TRP             1.000.070.220.260.370.070.49
OH              1.000.050.190.080.040.34
SOW               1.000.160.190.000.49
TRA                1.000.580.050.54
CSA                 1.000.100.52
ARC                  1.000.18
[PwarmMy]                   1.00

[18] Table 3 shows that almost all of the correlation coefficients between pairs of source contribution anomalies are positive. A high positive correlation between two different sources suggests that their above-average (or below-average) contributions tend to occur during the same periods. A negative correlation would suggest that one source is enhanced when the other is reduced. The covariation may be due to similar rates of evapotranspiration in the source regions, and/or patterns of atmospheric transport that bring moisture from both regions and precipitate it in the Mississippi River basin. Given the complexity of the precipitation and moisture transport systems over the domain, a correlation coefficient of ∣0.6∣ and higher is considered high enough to indicate an important association. Correlation coefficient values greater than 0.6 exist between anomalies of (in descending order) Red/Arkansas (RA) and Missouri (MO), MO and Western North America (WES), Texas/Mexico (TMX) and Southwest North America (SOW), Tropical Atlantic (TRA) and Caribbean (CAR), Lower Mississippi (LM) and Southeast North America (SOE), TMX and RA, LM and Ohio (OH), TMX and Gulf of Mexico (GOM), and Upper Mississippi (UM) and OH. The members of each significant pair of regions are adjacent, suggesting that their correlations reflect the influence of the same circulation features [see also Bosilovich and Schubert, 2002], or regional-scale surface climate anomalies.

[19] Positive correlation coefficients between source contribution and total precipitation anomalies in Table 3 reflect the fact that total precipitation is mathematically the sum of all the source contributions. Above- (below-) average precipitation requires that one or more contributions be above (below) average. However, there is no mathematical reason that all the contributors should have the same degree of correlation with the total. The anomalies from Texas/Mexico (TMX) are most strongly correlated with total precipitation (0.78). Correlation coefficients greater than 0.7 (Table 3) are also obtained between the anomalies of the total warm-season precipitation and the contributions from Caribbean (CAR), Gulf of Mexico (GOM), and Missouri (MO).

[20] Even though a high correlation coefficient does not necessarily imply causality, the correlation coefficients between the contribution anomalies from Texas/Mexico, Caribbean, Gulf of Mexico, and Missouri and the total precipitation anomalies suggest the importance of the interannual variability of these source contributions to the interannual variability of the warm-season precipitation. Physically, a particular source's moisture contribution could be enhanced (decreased) by higher (lower) evapotranspiration in the region, by enhanced (decreased) transport of air from that region, by enhanced (decreased) precipitation mechanisms extracting moisture over the sink region, or by a combination of these mechanisms. The high correlation between the TMX source and M precipitation is consistent with suggestions by Bosilovich and Sun [1999] that the soil wetness affecting evaporation in the region also affects the intensity of the southerly low-level jet (LLJ) that brings moisture to the Mississippi River basin.

[21] Table 4 shows the elements of the first principal component, PC-1, the most important pattern of covariance in the data set. The elements of the pattern are the factor loadings, which are equivalent to correlation coefficients between each variable and the pattern. Figure 3 shows the factor loadings of PC-1 corresponding to each source as delineated in Figure 1. This spatial pattern accounts for about 34% of the total variance of source contribution anomalies during the warm season over the 36-year period on a 15-day basis.

Figure 3.

The first principal component, PC-1, from principal components analysis (PCA) as projected on Figure 1.

Table 4. First Principal Component, PC-1, From Principal Components Analysis (PCA) of the Source Contribution and Precipitation Anomalies During the Warm Seasons of the 36-year Period on a 15-day Basisa
VariablesI
  • a

    The left column shows the variables, which are the source contribution anomalies. The right column shows factor loadings between PC-1 and the variables. The factor loadings are equivalent to correlation coefficients between the variables and PC-1. Also shown is the explained variance (EV) associated with PC-1.

ARC0.16
TEP0.62
CSA0.53
BAJ0.64
GOM0.66
CAR0.71
TEA0.40
NNA0.35
MO0.74
LM0.71
UM0.53
RA0.74
OH0.41
TRA0.55
TRP0.50
TMX0.77
SOE0.45
WES0.66
SOW0.50
EV, %33.89

[22] The regions with high factor loadings are Texas/Mexico, Missouri, Red/Arkansas, Caribbean, and Lower Mississippi. Hence the intercorrelated ensemble of the TMX, MO, RA, CAR, and LM sources in Figure 3 appears to be an important spatial pattern in explaining the interannual variability of the warm-season precipitation. Figure 4 shows the time series of PC-1 (equation (4)) and precipitation anomalies in the Mississippi River basin (standardized) over the 36-year period on a warm-season basis. The correlation coefficient between PC-1 and precipitation is 0.98. This agreement demonstrates that the single variable, PC-1, a linear combination of the source contribution anomalies, captures nearly all of the interannual variability in Mississippi warm-season precipitation. It is of interest to further analyze the variables with a high factor loading in terms of their relationship with PC-1, and possibly with the interannual variability the warm-season precipitation in the Mississippi River basin. The standardized values of the contribution from the ensemble of Texas/Mexico, Missouri, Red/Arkansas, Caribbean, and Lower Mississippi and PC-1 have a correlation coefficient of 0.90. This correlation coefficient suggests a strong association between these five source regions and the warm-season precipitation, indicating that a higher (lower) contribution from this ensemble is expected to accompany a higher (lower) 15-day warm-season precipitation.

Figure 4.

Time series of the first principal component, PC-1, from principal components analysis (PCA) and precipitation (standardized) in the Mississippi River basin (M) over the 36-year period on a warm-season basis.

[23] Further categorization of the high factor loading sources of PC-1 into internal and external sources is done to compare their associations with the warm-season precipitation in the Mississippi River basin over the 36-year period. The standardized values of the contributions from sources external to the basin (Texas/Mexico and Caribbean, TMX and CAR) and the warm-season precipitation (standardized) have a correlation coefficient of 0.86. A lower correlation coefficient of 0.75, or a difference of about 18% in explained variation, exists between the contribution from internal sources (Missouri, Red/Arkansas, and Lower Mississippi), i.e., recycled precipitation, and the precipitation. The correlation coefficients suggest a stronger association between the external sources and warm-season precipitation than between the internal sources and warm-season precipitation.

[24] The above statistical exercises suggest that the importance of a particular source contribution to the mean does not necessarily imply the importance of the source to the interannual variability of precipitation in a region. Brubaker et al. [2001] found that the Gulf of Mexico was the second largest contributor in terms of the long-term mean, yet GOM is not among the five leading variables in PC-1. It should be noted that the pattern is constrained by subjective delineation of source/sink regions; a selection different from Figure 1 could result in different patterns. It should also be emphasized that the important spatiotemporal variability pattern of precipitation within a region is associated with larger-scale patterns of precipitation and moisture supply not reflected in this analysis.

[25] The structure of PC-1 suggests the influence of two distinct transport paths for moisture. The stronger is shown as the arc of high factor loadings leading from the Tropical Atlantic through the Caribbean, Gulf of Mexico, Texas/Mexico and into the Mississippi River basin itself. This reflects the circulation and associated moisture transport around the southern and western flanks of the North Atlantic subtropical ridge. The second path is indicated by the high loadings from the Temperate North Pacific eastward across western and southwestern North America. This is the signature of the mean midlatitude westerlies. Source regions that covary with these time-varying mechanisms of moisture supply are most likely to be associated with the interannual variability of precipitation in the Mississippi basin. This may explain why the Gulf of Mexico is excluded from the important sources of variability identified in PC-1.

[26] An examination of the NCEP reanalysis' moisture transport over the domain of interest during the warm season shows that the mean and variability of the moisture transport is more focused toward the eastern part of Texas/Mexico and the western part of the Gulf, lessening the impact of the Gulf in terms of moisture supply variability in the Mississippi basin. Berbery and Rasmusson [1999] and Roads et al. [1994] have observed the westward movement of the moisture transport over the Gulf during the Northern Hemisphere warm season. The arc pattern of PC-1 appears to be associated with the moisture transport vector field.

4. Moisture Flux Analysis

[27] Under a separate, related study, the moisture flux vector field was analyzed for the Mississippi River basin and its surroundings for a period from September 1979 to August 1998. Although the 19-year study was designed to address research questions that required a different definition of the temporal averaging window, the results are indexed to the same 15-day time periods defined in Table 2. Despite the differences in time period (19 versus 36 years) and averaging window, the moisture flux analysis from the 19-year study can be used to explore the physical reality behind this 36-year study's findings. However, direct comparison of the 19-year study's time series with those developed here would be inappropriate because of the different averaging windows.

[28] Correlation analysis between source contributions and warm-season precipitation was repeated for this 19-year period on the basis of the 20-day averaging window. Anomalies of precipitation in the Mississippi River basin and external source contributions have a correlation coefficient of 0.98 over the 19-year period. The correlation coefficient between the precipitation in the basin and local source contributions is 0.69. These suggest a stronger association between the imported moisture and the interannual variability of precipitation in the basin than the association between the recycled moisture and the variability.

[29] Moisture fluxes into the Mississippi River basin are used to analyze the interannual variability of moisture transport into the basin. The moisture transport, Q, was computed on a 15-day basis, using 16 sigma levels of the NCEP reanalysis (from the surface through approximately 300 mb). The 19-year mean Q vector field (Figure 5) shows the two major moisture transports importing evaporated moisture into the basin: from the southern and western boundaries of the basin [e.g., Benton and Estoque, 1954; Roads et al., 1994]. The moisture transports export moisture out of the basin through its northern and eastern boundaries. Because of the distance from the Pacific Ocean, and the Rocky Mountains' obstruction to flow, the moisture transport from the western boundary would be expected to import less moisture into the basin compared to the southern one.

Figure 5.

Mean of vertically integrated moisture transport, in kg m−2 d−1, during the warm season averaged over the 19-year period shown as vectors. The Mississippi River basin (M) and Tropical Atlantic (TRA) are outlined in bold. Also shown are contours of annual mean surface pressure, in mb, over the same period as proxies for surface orography.

[30] The transport of moisture from the south of the Mississippi River basin (south flux) is computed as the boundary integral of Q (equation (6)) over the southern boundaries of Red/Arkansas and Lower Mississippi. The boundary integral over the western boundaries of Missouri and Red/Arkansas is used to analyze the transport from the west (west flux). The anomalies of the south flux and precipitation in the basin have a correlation coefficient of 0.77 over the 19-year period on a 15-day basis. There is almost no common variation between the anomalies of the west flux and precipitation in the basin as suggested by their correlation coefficient of 0.01. The higher correlation coefficient between the south flux and precipitation in the basin is consistent with the suggestion of the importance of the LLJ from the Gulf of Mexico in importing moisture to the basin and the precipitation in the basin [Benton and Estoque, 1954; Giorgi et al., 1996; Helfand and Schubert, 1995; Higgins et al., 1997; Mo et al., 1995, 1997; Paegle et al., 1996; Rasmusson, 1967].

[31] Figures 6 and 7 show the box and whisker plots of the south and west fluxes, respectively, over the 19-year period on a 15-day basis in mm d−1. Because the fluxes are the component of Q that is perpendicular to the boundaries, negative values are expected to occur during the period. In both cases, a negative value indicates that moisture is actually leaving the basin across the boundary. Time indices in Table 2 are used for the figures. Figures 6 and 7 are plotted with the same y axis to compare the magnitude and variability of the fluxes both within each 15-day period and throughout the year.

Figure 6.

Box and whisker plots of meridional moisture flux, in mm d−1, across the southern boundaries of Red/Arkansas (RA) and Lower Mississippi (LM) over the 19-year period on a 15-day basis.

Figure 7.

Box and whisker plots of zonal moisture flux, in mm d−1, across the western boundaries of Missouri (MO) and Red/Arkansas (RA) over the 19-year period on a 15-day basis.

[32] Figure 6 shows a semi-annual cycle in moisture flux. A maximum in moisture transport occurs in late May (f18), dropping sharply in summer before rebounding somewhat by November. Minimum values are found during January–February. Variability is large during the seasons of maximum transport. Negative fluxes in Figure 6 occur mostly during DJF when the polar front (PF) over the domain of interest pushes moisture out of the Mississippi River basin over the southern boundaries of the basin. As expected, the south flux is the lowest during DJF following the seasonal position of PF and the seasonal decay of the anticyclone over Temperate Atlantic (TEA). Westerly transport into the basin (Figure 7) is much weaker and less variable than the southerly transport, but also possesses a semi-annual cycle that is largely in phase with Figure 6.

[33] Figure 8 shows the box and whisker plots of precipitation in the Mississippi River basin, in kg m−2 d−1 (or equal to mm d−1), over the 19-year period on a 15-day basis. The seasonal cycle of precipitation in the basin follows the seasonal cycle of the south flux in Figure 6. The beginning of July outlier in 1993 from Figure 2 is apparent in Figure 8. Again, because Figures 6, 7, and 8 are based on a 20-day averaging window, this outlier in Figure 2 is shown at the end of June of 1993 in Figure 8.

Figure 8.

Box and whisker plots of precipitation, in kg m−2 d−1, in the Mississippi River basin (M) over the 19-year period on a 15-day basis.

[34] Moisture fluxes are driven by transport that is strongly determined by large-scale winds. Because the winds are driven by surface pressure differences, it is of interest to examine the relationship between the south flux and surface pressure. Figure 9 shows the contours of correlation coefficients between anomalies of the south flux and surface pressure over the 19-year period on a monthly basis over the domain of interest. Figure 9 shows three distinct centers of action with cores over (40°–45°N, 140°–145°W) (positive correlation coefficients), (35°–40°N, 105°–110°W) (negative correlation coefficients), and (27.5°–32.5°N, 70°–75°W) (positive correlation coefficients).

Figure 9.

The correlation coefficient between anomalies of the south flux and surface pressure over the 19-year period on a monthly basis. The Mississippi River basin (M) and subbasins are outlined in bold.

[35] The cores of positive correlation coefficients (Pacific and Atlantic Oceans) suggest that positive anomalies of surface pressure over these regions create anticyclonic (clockwise) low-level circulation patterns and bring more moisture into the Mississippi River basin. The core of negative anomalies correlation coefficients (Southwest United States) suggests that negative anomalies of surface pressure over this region create a cyclonic (counter clockwise) pattern over the region and hence bring more moisture into the basin.

[36] The correlation coefficient between the anomalies of precipitation in the Mississippi River basin and the areal averages of surface pressure over the Pacific (40°–45°N, 140°–145°W) (positive correlation coefficients), North America (35°–40°N, 105°–110°W) (negative correlation coefficients), and the Atlantic (27.5°–32.5°N, 70°–75°W) (positive correlation coefficients) are 0.29, −0.64, 0.33, respectively, over the 19-year period on a 15-day basis. The correlation coefficients suggest the dominance of the pressure system over North America (negative correlation coefficients) over the other pressure systems on the interannual variability of the south flux. This is consistent with the suggestion that the intensity of the southerly LLJ to the basin, which plays a critical role in determining the basin's summer convective precipitation, is associated with the basin's cyclonic activity, indicated by the mean and variability of the basin's surface pressure [Bosilovich and Sun, 1999; Giorgi et al., 1996; Paegle et al., 1996].

[37] Because the North American core (negative correlation with the south flux) is geostrophically coupled with the other two cores, it is of interest to examine the coupled system to determine and examine the processes affecting the dominant moisture transport to the basin. The correlation coefficient between the anomalies of the areal averages of surface pressure over the Pacific center (positive correlation with the south flux) and Atlantic (positive correlation with the south flux) is 0.33, indicating a weak association between these two ocean-centered systems over the domain of interest. The pattern of surface pressure anomalies is not well described by simultaneous variation in all three cores.

[38] In order to examine the coupled dipole pressure systems, two indices are created in this study: Pacific-Mississippi (P-M) surface pressure (ps) difference and Atlantic-Mississippi (A-M) ps. The indices are the standardized values of the surface pressure differences between, respectively, the Pacific and Atlantic centers and the North American center.

[39] The A-M ps index has correlation coefficients of 0.80 and 0.71 with the anomalies of the south flux and precipitation in the Mississippi River basin, respectively, over the 19-year period on a 15-day basis (see Figure 10). Figure 10 shows the A-M ps, the south flux, and precipitation in the Mississippi River basin. The closeness of the values from the indices results in the high correlation coefficients among them. The correlation coefficients are statistically greater than the corresponding correlations with the negative North American core alone. The P-M ps index has correlation coefficients of 0.62 and 0.46 with the anomalies of, respectively, the south flux and precipitation in the basin. The correlation coefficients suggest the dominance of the Atlantic-Mississippi coupled pressure system on the interannual variability of the south flux and, hence, precipitation in the basin. This is consistent with the suggestion that the northward meridional wind anomalies (which are associated with the south flux, located between the cyclonic and anticyclonic dipole, which is represented by the A-M ps index) covary with the precipitation [Mo et al., 1997].

Figure 10.

The A-M ps, the south flux, and precipitation in the Mississippi River basin (M) indices over the 19-year period on a monthly basis.

5. Summary and Conclusions

[40] This study has analyzed the interannual variability of surface evaporative source contributions to precipitation supply in the Mississippi River basin. In doing so, the anomalies, which are the departures from the interannual means on a 15-day basis, of the source contributions were used. Statistical analysis on related moisture flux and surface pressure was also performed.

[41] On the basis of principal components analysis, this study identified the covariability among the Missouri, Red/Arkansas, Texas/Mexico, Lower Mississippi, and Caribbean contributions as an important factor in statistically explaining the interannual variability of warm-season precipitation in the Mississippi River basin. Further correlation analysis suggested the importance of contribution anomalies from two particular regions, Texas/Mexico and the Caribbean, in explaining the interannual variability of the warm-season precipitation. Over the 36-year period, the warm-season precipitation and the contribution from the Texas/Mexico and Caribbean have a correlation coefficient of 0.86.

[42] A new index, the A-M ps index, based on standardized values of measurable surface pressure difference over the 19-year period on a 15-day basis between the Atlantic (27.5°–32.5°N, 70°–75°W)-Mississippi (35°–40°N, 105°–110°W) regions, was introduced. The correlation coefficient of the A-M ps index with the moisture flux across the southern boundary of the Mississippi basin and the A-M ps index with precipitation in the basin anomalies are 0.80 and 0.71, respectively. The important contributions from the Texas/Mexico and Caribbean regions to precipitation in the basin are included in this southern boundary flux. Anomalies of the south flux and precipitation in the basin have a correlation coefficient of 0.77 suggesting the important association of the flux and the interannual variability of precipitation in the basin.

[43] The A-M ps index has a potential implication in monthly timescale forecasting of precipitation supply in the Mississippi River basin, especially in terms of the persistence of the ongoing precipitation system. Unchanged signs of the index may indicate that the ongoing precipitation system may persist for one to two months. The statistical associations of the index and the south flux and precipitation in the basin confirm this potential.

[44] The backward-trajectory analysis has been proven useful in the study of water cycle on a regional to hemispheric scale. The analysis tracks water vapor, backward in time, from precipitation to evaporation through the solutions of equations of mass, momentum, and energy using gridded data sets from the NCEP reanalysis in which observational data sets were assimilated. The direct use of reliable observational data sets with good spatial and temporal resolutions, as soon as they are available, will improve results from the analysis. The analysis provides a novel alternative to available GCM-based tracer simulations, which are forward in time (from evaporation to precipitation) and are strongly affected by the systematic errors in the GCM. However, as the analysis is also a modeling approach, it should be kept in mind that there are assumptions used in its algorithm. Also, the reanalysis data sets are not without biases. The backward-trajectory analysis assumes that precipitation at any time in each grid box is contributed from all layers in the lower troposphere and the contribution of each layer is proportional to precipitable water content in the layer. The analysis also assumes a well-mixed condition in the lower troposphere for evaporated moisture, which is generally a good assumption during daylight hours when most of evaporation occurs. The assumptions are discussed by Dirmeyer and Brubaker [1999]. Regarding the issue of biases in data sets, Brubaker et al. [2001] suggest that the analysis is somewhat sensitive to systematic errors but not to random errors in evaporation, which is the least reliable data set used in their study.

Acknowledgments

[45] The authors thank three anonymous reviewers for their thoughtful critiques and suggestions, which substantially improved the paper. The assistance from Jeffrey McCollum in proofreading the manuscript is greatly appreciated. This research is supported under the NOAA-NASA GCIP Program. Participation of A.S. and K.L.B. is supported by grant NA96-GP0268. P.A.D.'s participation is supported by grant NA96-GP0353. Access to the NCEP reanalysis data set is supported by NCAR Scientific Computer Division grant 35161040 to A.S.

Ancillary