Predicting spring discharge of the Susquehanna River from a winter synoptic climatology for the eastern United States

Authors


Abstract

[1] Seasonal and interannual variations of freshwater flow strongly influence estuarine processes, exemplified by plankton biomass and productivity. The main tributary feeding Chesapeake Bay, the Susquehanna River, has shown threefold variability of spring flow in the last 52 years. The magnitude of spring discharge from the Susquehanna River is associated with the frequency and type of weather patterns transiting the eastern United States during winter and is related to the precipitation stored in the basin as snow and ice. Large-scale indices of climate variability, such as El Niño–Southern Oscillation and the North Atlantic Oscillation, have not proven to be strong predictors of freshwater flow in the Mid-Atlantic. We developed a synoptic climatology as an alternative way to quantify and classify regional weather, focusing on the types and frequency of occurrence of patterns we identified for winter. This approach was used to predict freshwater flow in spring and explained 54% of the variance of spring discharge after extreme outliers were removed. Responses of Chesapeake Bay plankton to contrasting years of weather pattern frequencies and associated freshwater flow were examined to illustrate ecosystem response to climatic forcing.

1. Introduction

[2] Freshwater flow into an estuary affects important physical and chemical processes, including circulation, stratification, sedimentation, nutrient loading, light attenuation, and dissolved oxygen [Schubel and Pritchard, 1986]. The distribution and abundance of many ecologically and economically important estuarine organisms, such as phytoplankton and zooplankton, are also strongly influenced by freshwater flow [Kimmerer, 2002]. Phytoplankton and zooplankton are key components of estuarine systems that are being used as indicators of ecosystem status [Paerl et al., 2003]. To detect change in ecosystems using plankton as indicators, we must understand how the indicators respond to environmental variability, driven largely by changes in freshwater flow. Ecosystem responses to freshwater flow have been documented for estuaries and coastal systems including San Francisco Bay [Cloern et al., 1983; Kimmerer, 2004], the Gulf of Mexico [Riley, 1937; Justić et al., 2003], and the Hudson River estuary [Malone, 1977; Howarth et al., 2000]. Freshwater flow into Chesapeake Bay has been related to dissolved oxygen [Boicourt, 1992; Hagy et al., 2004], phytoplankton biomass [Malone et al., 1988; Harding and Perry, 1997], zooplankton abundance [Kimmel and Roman, 2004], and larval fish recruitment [Wood, 2000; North and Houde, 2003; Jung and Houde, 2003].

[3] During the last 52 years, the Susquehanna River as the major source of freshwater to Chesapeake Bay has experienced threefold variability of spring flow (range = 988 – 3366 m3 s−1). Interannual differences in the types and frequencies of atmospheric circulation patterns that transit the Susquehanna River basin influence regional temperature and precipitation, and thereby strongly affect freshwater flow [Yarnal and Frakes, 1997]. Najjar [1999] showed that much of the streamflow increase that occurs during spring could be attributed to release of winter precipitation that is stored in the basin over winter as snow. A number of studies have addressed the relationship of atmospheric circulation to precipitation and freshwater flow [Peterson et al., 1989; Cayan and Peterson, 1989, 1993; McCabe and Ayers, 1989; Wilby, 1993], and specifically for the Susquehanna River [Crane and Hewitson, 1998; Lakhtakia et al., 1998; Yu et al., 1999; Najjar, 1999]. Despite the overriding influence of flow on ecosystem structure and function in this important estuary [Malone et al., 1988; Kimmel and Roman, 2004], a predictive link between freshwater flow and variability in atmospheric circulation has not been developed.

[4] Large-scale indices of climate variability, such as El Niño/Southern Oscillation (ENSO) and the North Atlantic Oscillation (NAO), have strong effects on marine ecosystems, such as the equatorial Pacific [Cane, 1983; Barber and Chavez, 1983; Stenseth et al., 2002] and the North Atlantic [Hurrell, 1995; Ottersen et al., 2001; Stenseth et al., 2002], but their influence in the Mid-Atlantic region is ambiguous [Read, 2002; Stenseth et al., 2003]. This is not to suggest that large-scale climate indices do not influence the Mid-Atlantic, but rather that these forcings are manifest through changes in regional-scale weather. An alternative way to characterize climate variability at smaller spatial (1000–2500 km) and temporal (interannual) scales is to create a synoptic climatology that is based on regional atmospheric circulation. Yarnal [1993] defines synoptic climatology as the relationship between atmospheric circulation and the surface environment. It is a statistical approach to classify and quantify predominant weather patterns in a region. This procedure condenses the large volume of data associated with atmospheric circulation into definable, commonly experienced weather patterns, and integrates the effects of the individual meteorological parameters related to each of the patterns.

[5] We developed a synoptic climatology for the eastern United States in order to describe the climatic variability in the region. We postulated that defining and quantifying the climatic drivers of freshwater flow would support analyses of the causes and scales of variability in estuarine ecosystems. This paper (1) describes and quantifies the predominant synoptic-scale weather patterns in the eastern United States over the last 52 years, (2) identifies anomalies in the frequency of occurrence of these synoptic-scale weather patterns that underlie interannual differences in spring discharge, (3) predicts spring flow from synoptic-scale weather patterns in winter, and (4) addresses how this predictive capability will improve our understanding of estuarine responses to climate variability, climate change, and anthropogenic perturbations, expressed in planktonic processes.

2. Methods

2.1. Data

[6] Twice daily (0 and 1200 h GMT), 5° latitude by 5° longitude gridded sea level pressure (SLP; mb) data were acquired from the National Center for Atmospheric Research (NCAR; http://dss.ucar.edu). These data were averaged to produce 19,358 daily maps of SLP for the study period, 1 January 1950 through 31 December 2002. Gridded data have biases that must be acknowledged [Reid et al., 2001], but they provide the best source of data for these analyses [Yarnal, 1993]. Daily and monthly data on temperature and precipitation for use in regression models and descriptions of synoptic-scale weather patterns were obtained from the National Climate Data Center (NCDC; http://cdo.ncdc.noaa.gov). Divisional data from the eight climatic regions within the Susquehanna River basin (Pennsylvania divisions 4, 5, 6, 7, and 8; Maryland division 6; New York divisions 1 and 2; Figure 1, inset) were weighted by area to produce a single estimate of temperature or precipitation for the basin. Climate division data were used because they provided comprehensive measures of temperature and precipitation from all stations in a division [Guttman and Quayle, 1996]. Freshwater flow (m3 s−1) for the Susquehanna River was obtained from the United States Geological Survey gauging station at Harrisburg, Pennsylvania (USGS-01570500; 100 km from mouth; http://waterdata.usgs.gov), and extrapolated to the entire watershed based on the relationship between flow at Harrisburg and the Conowingo Dam (USGS-01578310; 15 km from mouth; Harrisburg flow * 1.125 = Conowingo flow) to generate a continuous flow record for the entire period of analysis, as data for Conowingo only extended back to 1967. Data for the plankton analyses were obtained from the Chesapeake Bay Program monitoring cruises (CBP; http://www.chesapeakebay.net). Geographical regions for planktonic responses are defined as; upper > 38.8°N, middle 38.8°N–37.8°N, lower < 37.8°N.

Figure 1.

Map of synoptic climate region with inset of Susquehanna River Basin showing NOAA climate divisions.

2.2. Synoptic Climatology

[7] Surface SLP data were used to describe atmospheric circulation patterns following an eigenvector-based, map pattern classification procedure outlined by Yarnal [1993] and Wood [2000] (Figure 2). A 48-point (6 × 8) grid of SLP data covering the area 25° to 50°N latitude and 65° to 100°W longitude was identified as the region of interest (Figure 1). Next, an S mode eigenvector analysis (principal component analysis; PCA) was performed on a correlation matrix of SLP station data against time (days) to reduce spatial variability in the data set from the original 48 points to a smaller number of new, statistically independent (orthogonal) variables (PC scores) that explained 90% of the variance in the original data set. The number of variables to retain (7) was determined in two ways: 1) a “scree” test in which a major break in the plot of eigennumber versus eigenvalue establishes the number of variables to retain and 2) the N-rule test (eigenvalues > 1) [Yarnal, 1993]. Comparison of rotated and unrotated PC scores gave similar results and thereafter unrotated results were used for the analyses. The saved scores from the PCA were then submitted to a two-stage clustering procedure to identify similarly occurring modes of variance related to atmospheric circulation patterns. The first stage employed an agglomerative, hierarchical cluster analysis (average linkage) to maximize the between-cluster variance which was used to determine the number of clusters (10) comprising a significant fraction of the total number of days (>2%), and to provide “seed” values for a subsequent k means clustering procedure. This second clustering procedure regrouped the retained PC scores into one of 10 dominant “seed” clusters identified previously. Once all days were categorized into one of 10 clusters, the average SLP from each grid point within each cluster was determined, and average SLP maps were generated for visualization. These clusters represent the prevailing weather patterns experienced in the region. Monthly frequency of occurrence for each weather pattern was then determined for use in regression models.

Figure 2.

Flowchart of synoptic climatology methods, after Yarnal [1993].

2.3. Data Analyses

[8] All analyses were performed using S-PLUS 6.2 (Insightful Corp.) statistical software. Pearson's correlations were used to determine relationships between the frequency of occurrence of synoptic-scale weather patterns during winter (December–January–February) and the winter NAO index defined as the normalized SLP difference between the Azores and Iceland [Hurrell, 1995], the winter ENSO index defined as sea surface temperature anomaly in the Niño3.4 region (5°N–5°S, 170°–120°W [Trenberth and Stepaniak, 2001]), the winter Pacific Decadal Oscillation index (PDO) defined as the leading eigenvector of North Pacific sea surface temperature [Mantua et al., 1997], and the winter Pacific/North America pattern (PNA) defined as the dominant rotated empirical orthogonal function of 500 hPa geopotential height anomalies for the Northern Hemisphere [Barnston and Livezey, 1987]. Anomalies were calculated as the difference between monthly/seasonal conditions and the long-term average for weather patterns (1950–2002), temperature (1950–2002), precipitation (1950–2002) and planktonic responses (1985–2002). Simple linear regression models were used to determine the strength of relationships between average spring freshwater flow (March–April–May) and the winter climate indices for NAO, ENSO, PDO, PNA, and basin-wide temperature and precipitation. Using the complete 52 year data set, a multiple linear regression model was developed to predict spring Susquehanna River flow from winter cluster frequencies of occurrence with limited success. A robust least trimmed squares regression model [Rousseeuw, 1984] was used to determine outliers from the freshwater flow data set. These outliers were removed and a second regression model was developed from the modified data set. The ten clusters used as explanatory variables were not statistically independent from one another, violating an assumption of the regression model. However, this violation only affects the interpretation of the regression coefficients, not the r2, significance, or reliability of the predicted values, and therefore the model still provides valuable information [Shaw, 2003]. No interpretation of the coefficients was attempted in these analyses. Differences in planktonic response during example years and the long-term average were determined by t test [Zar, 1984].

3. Results

3.1. Climate Indices and Weather Variables

[9] Climate indices for NAO, ENSO, PDO, and PNA during winter explained less that 8% of the variance of Susquehanna River flow during spring and were not significant at the 0.05 level (Table 1). Regressions of regional average temperature and precipitation, individually and multiple regressions of temperature and precipitation combined, explained a maximum of 16.7% of the variance in spring flow, with precipitation and combined precipitation and temperature producing significant models (Table 1). Pearson's correlation coefficients between the frequency of occurrence of individual clusters and climate indices (NAO, ENSO, PDO, and PNA) revealed weak to moderate relationships for many of the variables, with strongest associations to the Niño3.4 index (Table 2). The correlations reached a maximum of 0.456 and were both positive and negative in sign.

Table 1. Statistics From the Linear Regression of Winter Niño3.4, NAO, PDO, PNA, Temperature, Precipitation, and Combined Temperature + Precipitation Variables Against Average Spring Flow From the Susquehanna River
Variablep Valuer2
Niño3.40.0680.073
NAO0.6210.006
PDO0.5780.007
PNA0.6250.005
Temperature0.1160.055
Precipitation0.0280.104
Temperature + Precipitation0.0190.167
Table 2. Pearson's Correlation Coefficients for Comparisons of Winter Indices for NAO, ENSO, PDO, and PNA Against Winter Frequency of Occurrence for Each Cluster
ClusterNAOENSOPDOPNA
  • a

    Significance at the p < 0.01 level.

1−0.031−0.235−0.0310.114
20.158−0.354a0.1350.109
3−0.2890.297−0.026−0.117
4−0.429a0.375−0.0340.072
50.380a−0.337a−0.086−0.209
6−0.1250.456a0.0660.042
7−0.418a0.397a−0.1010.045
8−0.1270.383a0.1450.036
90.452a−0.311−0.025−0.075
100.215−0.294−0.061−0.041

3.2. Synoptic Climatology

[10] We identified ten significant weather patterns (Figure 3), each occurring 3.9 to 16.8% of the days in the study period during winter (Table 3). Several maps showed very recognizable weather patterns, including the Bermuda High in cluster 1 and the nor'easter in cluster 4. The environmental conditions associated with each cluster are shown in Table 3, indicating whether the conditions are warm or cold, or wet or dry on days when the patterns occurred. For instance, when cluster 2 occurred during winter it was on average 3.0°C colder and received 0.9 mm d−1 less precipitation than the long-term average for December, January, and February in the Susquehanna River basin. Alternatively when cluster 4 occurred, conditions tended to be 2.0°C warmer and the basin received 2.7 mm d−1 more precipitation than average.

Figure 3.

Average sea level pressure maps for each cluster. Cluster number is in top left corner.

Table 3. Meteorological Characteristics for Clusters During Wintera
ClusterDays of OccurrencePercentTemperature Anomaly, °CPrecipitation Anomaly, mm d−1Wind DirectionWind Speed, m s−1Seasonality
  • a

    Standard deviations in parentheses.

  • b

    Bermuda High.

  • c

    Ohio Valley High.

  • d

    Nor'easter.

1b3537.43.2 (±4.6)0.9 (±4.9)W3.3 (±1.4)summer
2c80216.8−3.0 (±5.2)−0.9 (±3.2)NW3.8 (±1.5)winter
33467.2−0.6 (±4.3)0.7 (±4.9)NW3.2 (±1.8)spring/fall
4d2455.12.0 (±4.6)2.7 (±7.0)W3.9 (±1.7)winter/spring
569814.62.4 (±4.9)0.1 (±4.2)W2.9 (±1.2)summer
63447.2−0.4 (±4.8)−0.4 (±3.9)E2.2 (±1.5)fall
768114.2−1.9 (±4.6)−0.1 (±4.1)W4.9 (±1.7)winter
81883.92.9 (±4.7)1.5 (±4.9)E2.4 (±1.4)spr/sum/fall
94238.82.2 (±5.1)−0.1 (±3.9)S2.3 (±1.1)winter/spring
1069214.5−1.0 (±5.2)−0.9 (±2.9)S2.3 (±1.1)winter

[11] Each daily observation was associated with a map pattern, supporting computation of the frequency of occurrence for each weather pattern for specific time periods. Several map patterns were identified, using the monthly frequency of occurrence that had distinct seasonal signals (Table 3 and Figure 4). Weather patterns captured by clusters 1, 8, and 9 occurred commonly throughout the year, were predominant in summer, and comprised over 60% of June and July days, associated with positive temperature anomalies (Figure 5). Clusters more common in winter, including 2, 7, and 10, had a summer minimum and a winter maximum of up to 50% of January days, associated with negative temperature anomalies and low precipitation (Figures 5 and 6) . While cluster 4 did not occur commonly in any season, it may be disproportionately important due to potentially heavy precipitation that accompanies this pattern in winter and spring (Table 3 and Figures 4 and 6).

Figure 4.

Monthly average frequency of occurrence by cluster. Cluster number is in top left corner.

Figure 5.

Monthly temperature anomaly by cluster. Cluster number is in top left corner.

Figure 6.

Monthly precipitation anomaly by cluster. Cluster number is in top left corner.

[12] Deviations from the long-term frequency of occurrence during winter showed large changes for several clusters, including clusters 2, 7, and 10 (Figure 7), while others, such as clusters 1, 8, and 9, showed little variation (Figure 7), coinciding with winter and summer dominant clusters respectively. Trends in cluster frequency of occurrence during winter were tested with a Mann-Kendall trend test. There were no significant changes in cluster frequency of occurrence during winter (1950–2002), except for cluster 5 which had a small but significant positive increase.

Figure 7.

Time series of deviation (in days) from long-term average winter frequency of occurrence for each cluster. Cluster number is in top left corner.

3.3. Weather Pattern–Flow Relationship

[13] The multiple linear regression model developed using the complete 52 year data set produced a nonsignificant model (p = 0.32) with limited success in predicting spring freshwater flow (r2 = 0.22; RMSE = 520 m3 s−1) from winter weather patterns (Figure 8). A robust least trimmed squares regression model identified six points as being more than 2.5 standard deviations from the regression line of winter cluster frequency of occurrence and spring flow (Figure 9). These points were the first (1995), second (1981), and fifth (1969) driest and the first (1993), second (1994), and fourth (1972) wettest springs in the data set (1950–2002). After removal of these extreme points, the modified data set (n = 46) produced a new, highly significant model (p < 0.001) with a substantial improvement in variance explained (r2 = 0.54) and reduction in error (RMSE = 329 m3 s−1) (Figure 10).

Figure 8.

Regression of observed average spring freshwater flow from the Susquehanna River on modeled flow predicted from winter cluster frequency of occurrence using the complete (n = 52) data set.

Figure 9.

Plot of residuals by year from multiple linear regression model shown in Figure 8.

Figure 10.

Regression of observed average spring freshwater flow from the Susquehanna River on modeled flow, predicted from winter cluster frequency of occurrence using the modified data set (n = 46) with six outliers removed.

3.4. Planktonic Response

[14] Planktonic responses in Chesapeake Bay to strongly contrasting freshwater flow and associated weather patterns were exemplified by the conditions in 1985 and 1998 (Figures 11a–11f). During winter of 1984–1985, two of the patterns that occurred more frequently than average (clusters 2 and 10) were the driest (Table 3 and Figure 6), while the patterns that occurred less frequently than average (clusters 3 and 4) often produced high precipitation in winter (Table 3 and Figure 6); this led to spring flow 552 m3 s−1 (28.9%) below average (Figure 11a). Alternatively, the winter of 1997–1998 saw wet weather patterns (clusters 3, 4, and 8) occur 32 days more frequently than average, while drier patterns (clusters 2 and 10) occurred 23 days less frequently than average (Table 3); resulting in spring flow 559 m3 s−1 (29.0%) above average (Figure 11b).

Figure 11.

Planktonic response in spring to years of contrasting winter weather patterns. (a) Weather pattern anomalies for winter 1984–1985, (b) weather pattern anomalies for winter 1997–1998, (c) phytoplankton biomass anomalies for spring 1985 in three geographical regions, (d) phytoplankton biomass anomalies for spring 1998 in three geographical regions, (e) copepod abundance anomalies for spring 1985 in two geographical regions, and (f) copepod abundance anomalies for spring 1998 in two geographical regions. Error bars indicate standard error.

[15] Observations of phytoplankton biomass in 1985 and 1998 showed conditions significantly different from average, particularly in the middle portion of Chesapeake Bay (Figures 11c, 11d, and Figure 12). Long-term average biomass for the bay, in spring, reaches a maximum in the mid bay (9.5 mg m−3) with slightly lower concentrations in the upper (7.7 mg m−3) and lower bay (7.3 mg m−3; Figure 12). During the low-flow conditions of 1985, biomass was significantly (t test; p < 0.01) greater than average in both the upper (14.8 mg m−3) and mid bay (14.6 mg m−3), while the lower bay showed a nonsignificant 1.9 mg m−3 decrease (Figures 11c and 12). In 1998, high flow caused below average biomass in the upper bay (6.0 mg m−3), along with significantly above average biomass in the mid Chesapeake Bay (t test; p < 0.02; 13.8 mg m−3) and a modest increase of 1.0 mg m−3 in the lower bay (Figure 11d).

Figure 12.

Maps of spring phytoplankton biomass (mg m−3) for long-term average condition, dry year of 1985, and wet year of 1998. Maps are interpolated from Chesapeake Bay Program station data (n = 49). Black bars demarcate upper, mid, and lower Chesapeake Bay regions used in analyses.

[16] Eurytemora affinis, a dominant calanoid copepod and major food source for larval fish, responded strongly to differences in spring flow (Figures 11e and 11f). Average E. affinis abundance for spring is higher in the upper bay (mean = 18,159 m−3) relative to the mid bay (mean = 3684 m−3). Zooplankton abundance was significantly (t test; p < 0.001) below the long-term average for both the upper and middle regions of the bay in 1985 (Figure 11e). In the high-flow year of 1998, upper bay E. affinis abundance was close to the long-term average (16,005 m−3); while mid bay values were 10,444 m−3 above average (Figure 11f). Because of small sample size (n = 3) and high variance (standard deviation is 20582), the 1998 mid bay observations were not significantly different from the long-term average (t test; p > 0.20).

4. Discussion

[17] Climate interacts with ecology through local weather patterns [Stenseth et al., 2003]. Freshwater flow acts as an integrator of climate variability by reducing the short-term noise associated with local temperature and precipitation [Cayan and Peterson, 1989]. Hypothesized responses of estuarine ecosystems to climate change in the Mid-Atlantic are strongly coupled to changes in freshwater flow [Najjar et al., 2000; Neff et al., 2000; Gibson and Najjar, 2000]. Our primary goal of this paper was to describe a quantitative link between climate variability and estuarine plankton dynamics through freshwater flow. We have shown that large-scale climate indices are limited in predicting freshwater flow from the Susquehanna River. We developed an alternative methodology to classify and quantify regional climate variability with a synoptic climatology. This approach to quantifying climate variability provided us with a tool to predict spring freshwater flow with a reasonable degree of confidence. Finally we showed how these interannual variations in spring flow from the Susquehanna River influence plankton dynamics in Chesapeake Bay.

4.1. Climate Indices

[18] Large-scale climate indices, such as NAO and ENSO, provide an integrated measure of climate variability over broad spatial and temporal scales [Stenseth et al., 2002]. They are correlated to a limited extent with local weather patterns in the eastern United States (Table 2), but these indices do not support prediction of spring flow from the Susquehanna River (Table 1). Read [2002] identified “modest” correlations between NAO and flow for several smaller watersheds within the Susquehanna River basin during winter and spring, however, this analyses was limited to smaller watersheds with no anthropogenic impacts on flow (i.e., damns and urban development). In addition, Read [2002] looked at correlations between variables from the same season; we are interested in lagged flow in spring related to precipitation stored in the basin over winter as snowpack. Indices of ENSO have been used to successfully predict lagged flow in rivers of the western United States with contrasting patterns in the Pacific Northwest and southwestern United States [Redmond and Koch, 1991]. While the relationships are not as strong in the Mid-Atlantic, the positive correlation with ENSO (Table 1) may be related to increased storm frequency during El Niño years [Hirsch et al., 2001]. Stenseth et al. [2003] suggested that the lack of strong correlations between local weather patterns and large-scale climate indices can be related to a number of factors, including (1) variation in local response depending upon geographic location, (2) variation in the intensity of the index with season, (3) change in the relationship between local weather and climate indices over time, (4) nonlinear response of local weather to indices, or (5) simply that any given index may only explain a small fraction of the variance in a region's weather. Therefore our results of no strong correlation between large-scale climate indices and Susquehanna River flow are not unexpected.

4.2. Water Balance

[19] Another approach to predicting freshwater flow is to develop a water balance model that estimates flow from the difference between input and loss terms. Precipitation and temperature are two of the most important parameters for prediction of freshwater flow from water balance models [Thornthwaite, 1948]. Najjar [1999] developed a water balance model for the Susquehanna River basin using precipitation and temperature that successfully estimated monthly flow, however, the model used real-time precipitation and temperature to predict freshwater flow, providing limited forecasting ability. Our linear regression models using winter precipitation and temperature (alone and combined) to predict spring flow, while significant, did not explain a large portion of the variance in spring flow (Table 1). We believe this approach had limited success because, although important meteorological parameters, temperature and precipitation do not provide a comprehensive description of weather variability [Davis and Kalkstein, 1990].

4.3. Synoptic Climatology

[20] To address the limitations of climate indices and water balance models in predicting freshwater flow from the Susquehanna River, we developed a synoptic climatology of the region using maps of SLP. We quantified 52 years of synoptic-scale weather patterns affecting the eastern United States for the purpose of understanding how climate variation affects freshwater flow to Chesapeake Bay. These patterns agree well with literature descriptions of common weather patterns for the region in terms of map structure, seasonality in frequency of occurrence, and the weather conditions associated with each pattern [Hayden, 1981; Yarnal and Leathers, 1988; Davis et al., 1993, 1997]. High pressure patterns, such as clusters 1, 2, 7, and 9, and their average frequencies coincide well with seasonally distinct modes of the North Atlantic subtropical anticyclone described by Davis et al. [1997]. Interannual variations in the frequency of occurrence of these “summer” (warm and moist) or “winter” (cold and dry) modes, during winter, have implications for spring freshwater flow through changes in storage within the watershed. Because of their tendency to produce high wind, waves, and precipitation, much research has focused on the frequency, track, generation location, and path of Atlantic Coast “nor'easters” [Hayden, 1981; Davis et al., 1993; Zielinski, 2002]. Cluster 4 (Figure 3) represents the completion of a typical nor'easter track. Passage of this cluster is often associated with heavy precipitation in the Susquehanna River watershed (Figure 6). While relatively rare in frequency of occurrence these patterns are extremely important because of their potential to deposit significant amounts of snow over much of the watershed during winter. This snow often stays locked in the basin as “storage” until the warmer spring temperatures release the water as part of the spring freshet [Najjar, 1999].

4.4. Weather Pattern–Flow Relationship

[21] We have successfully downscaled from the frequency of occurrence of synoptic-scale weather patterns during the winter to spring Susquehanna River flow, explaining 54% of the variance in the modified data set. Removal of six “outliers” was necessary to obtain this result. The rationale for that decision is discussed below. First, the least trimmed squares regression identified these six points as being more than 2.5 standard deviations from the mean; these points were having a large influence on the regression results. Second, these points were hydrologic extremes as the first, second, and fifth driest (lowest flow) and first, second and fourth wettest (highest flow) springs. While the prediction of extremes is important, this model is better suited to forecast the more typical interannual variations in flow that still have significant impacts on Chesapeake Bay plankton. Finally, several of the wet extreme years (1993 and 1994) had exceptional events (blizzards) in March which were outside the time frame identified in these analyses for the climate forcing of freshwater flow. Similarly, during the dry years drought conditions prior to the winter time frame influenced the spring flow. Models developed with winter climate as the independent variables cannot be expected to predict flow that is dominated by events before or after that time frame. Because this model does not predict extremes well, inclusion of the outliers (using the entire data set) resulted in a substantial decrease (r2 = 0.22) in variance explained and a nonsignificant model (p = 0.32).

[22] There are several potential mechanistic explanations for why the winter weather patterns predict spring freshwater flow better than other variables. In large river basins, there is often a time lag between precipitation and basin flow, and that lag can often be as great as 50% of the precipitation on monthly timescales [Gleick, 1987]. Precipitation falling during the winter is often retained in the higher elevations of a basin as snow and is not released until spring temperatures melt it [Najjar, 1999]. The amount of water stored in this reservoir depends not only on the amount of precipitation falling, but also on winter temperature [Cayan and Peterson, 1993]. The synoptic-scale weather patterns used in these analyses take into account the cumulative effects of the weather associated with each pattern, including parameters such as wind speed and direction, cloud cover, and dew point, all of which influence storage [Davis and Kalkstein, 1990]. Finally, large-scale climate oscillations influence local environmental conditions through changes in local weather patterns [Stenseth et al., 2003]. Therefore use of the regional weather patterns, described by the synoptic climatology, to predict a local environmental response eliminates one potential source of variability in the linkage.

[23] Potential reasons for the unexplained variance in our regression include a lack of ability to address the magnitude of precipitation for certain weather patterns, disconnects between the artificial delineation of seasons used in the model, and large precipitation events in the spring that have immediate impacts on freshwater flow. For instance, small variations in the track of cluster 4 can produce large differences in the amount and type of precipitation the watershed experiences [Zielinski, 2002]. As mentioned previously, the forecasting ability of this model is largely related to the storage of winter precipitation as snowpack [Najjar, 1999]; therefore forecasting during other seasons is likely to be less successful. Because of the relationship between freshwater flow and Chesapeake Bay plankton dynamics, this model provides information that will be useful to managers of both water resources and estuarine ecosystems. This approach can be used to separate variability from trends in highly dynamic data sets by quantifying a climate signal that can be extracted. Future work will incorporate this technique and expand on the relationships between atmospheric circulation and Chesapeake Bay plankton discussed here.

4.5. Planktonic Response

[24] Chesapeake Bay phytoplankton dynamics in spring are described well by the interplay of light and nutrients, driven by variations in freshwater flow (Figure 12) [Malone et al., 1988; Harding, 1994]. Lower than average flow in 1985 resulted in reduced input of nutrients and sediment to the bay. Greater than average phytoplankton biomass was observed in the upper and mid bay during the spring due to increased photic depth (126% of LTA) associated with below average river-born sediment delivery [Fisher et al., 1988]. Negative phytoplankton anomalies were observed in the lower bay because lower than average flow exacerbated nutrient limitations (55% of LTA; Figure 11c). Alternatively, high nutrient and sediment loading associated with high flows in 1998 created shallower than average photic depth (87% of LTA) in the upper bay and concomitant decreased phytoplankton biomass. The mid bay saw positive biomass anomalies related to increased nutrient loading (130% of LTA), while the lower bay showed a slight biomass increase (Figure 11d) [Harding et al., 1986].

[25] Zooplankton, exemplified by the copepod E. affinis respond strongly to variations in freshwater input through changes in preferred low-salinity and low-temperature habitat, and changes in the size of the estuarine turbidity maximum (an area of plankton and fish aggregation located near the head of the estuary) [Kimmel and Roman, 2004]. During the spring of 1985 E. affinis abundance was well below average in the upper bay due to above average salinities (131% of LTA), low turbidity and reduced size of the estuarine turbidity maximum despite preferred below average temperatures [Roman et al., 2001], while above average salinities (119% of LTA) also reduced biomass in the mid bay region (Figure 11e). In 1998, estuarine conditions were favorable for E. affinis in the mid bay, where low salinities (61% of LTA) and high turbidity resulted in an expanded estuarine turbidity maximum in this region, while exceptionally high flows pushed favorable habitat conditions out of the upper bay resulting in below average E. affinis concentrations (Figure 11f).

5. Conclusions

[26] A large portion of the physical and biological variability in an estuary can be related to changes in freshwater flow [Schubel and Pritchard, 1986; Kimmerer, 2002]. Impacts of climate change on estuarine ecosystems are expected to be driven largely by changes in freshwater flow [Najjar et al., 2000]. Our ability to separate natural variability from anthropogenic trends in many key ecosystem indicators is strongly influenced by freshwater flow [Boicourt, 1992; Harding, 1994; Kimmel and Roman, 2004; Jung and Houde, 2003]. This paper has shown that the frequency of occurrence of winter weather patterns, described by a synoptic climatology, can be used to forecast spring freshwater flow with the caveat that extreme conditions may not be predicted well. Quantifying the link between regional climate and freshwater flow provides the information necessary to forecast ecosystem response to changing environmental conditions.

Acknowledgments

[27] The authors thank Bob Wood and Brent Yarnal for helpful discussions and Ming Li, Ed Houde, and Tom Malone for constructive reviews. W.D.M. was supported by a NASA Earth System Science Fellowship. Although the research described in this article has been funded in part by the U.S. Environmental Protection Agency through cooperative agreement R82867701 to Atlantic Coast Estuaries Indicators Consortium, it has not been subject to the agency's required peer and policy review and therefore does not necessarily reflect the views of the agency, and no official endorsement should be inferred. Contribution 3928 of Horn Point Laboratory, University of Maryland Center for Environmental Science.

Ancillary