4.1. Reconstruction Model
 Comparative statistics of reconstructed and instrumental streamflow and model calibration and validation statistics for the JCK and HEII gages are shown in Tables 2 and 3. The reconstruction models for the JCK and HEII gages incorporated the same three site chronologies, and the gages are highly correlated (r > 0.96, p < 0.01) over both the instrumental period and in their reconstructed flows. Because of this similarity and the fact that HEII incorporates a greater percentage of the overall flow in the upper Snake River, this lower gage will be the focus of the following results and discussion.
Table 2. Descriptive Statistics for Observed and Reconstructed Snake River Water Year Streamflow (MCM) at Jackson Lake Dam, Wyoming (JCK), and Near Heise, Idaho (HEII)
Table 3. Calibration and Validation Statistics for Streamflow Reconstruction Modelsa
|Gage||Predictor Sites (Lag)||r2||radj2||F Level||p Value||SE||RE||RMSE||Portmanteau Q (p)|
 The reconstruction model is fairly robust, explaining 62% of the variance in the instrumental record after adjustment for degrees of freedom (Table 3). While this is less than the predictive power of some southwestern streamflow reconstructions like the Colorado River at Lees Ferry, AZ (r2 up to 0.84) [Woodhouse et al., 2006], it compares favorably to northwestern reconstructions, including the Yellowstone River (radj2 = 0.52) [Graumlich et al., 2003], the Columbia River (r2 = 0.35) [Gedalof et al., 2004], the Oldman River (radj2 = 0.37) [Axelson et al., 2009], and the Wind River (radj2 = 0.38) [Watson et al., 2009]. Although the model is able to capture low flows and moderately high flows well, very high flow years such as 1997 (which is an extreme outlier in the instrumental record) are underpredicted by the tree ring model (Figure 4).
Figure 4. Modeled (gray) and observed (black) total water year streamflow (MCM) for the Snake River near Heise, Idaho, 1911–2005.
Download figure to PowerPoint
 The reduction of error (RE) statistic [Fritts, 1976], which can range from −∞ to 1, is greater than 0 (0.59), signifying that the model has predictive skill (Table 3). The root mean square error (RMSE) from the cross validation is close to the standard error (SE) of the calibration period, indicating that the model skill is not a product of overfitting. Regression residuals are normally distributed. Statistical tests for first-order autocorrelation (based on the Durbin-Watson test) and trend in the residuals indicate that neither is significantly present (p < 0.05).
4.2. Streamflow and Drought in the Upper Snake River
 The 415 year reconstruction of streamflow in the upper Snake River extends from A.D. 1591 to 2005 (Figure 5). This reconstruction lengthens the instrumental record by over 300 years, allowing a comparison of instrumental records with longer-term variability. The instrumental period is representative of individual extreme low-flow years. Years in the historical record such as 1977 and 2001 are quite severe even in the context of the longer-term record: with <60% of mean flow, these years rank among the top 10 low-flow years of the reconstruction. The more severe individual years in the reconstruction (Table 4) are not outside of the range of variability that would be expected from the instrumental record; however, regression models tend to bias reconstructed flows toward the calibration-period mean [Meko et al., 2007]. Low-flow individual years are distributed fairly evenly over each century of the reconstructed record (Table 4).
Figure 5. Reconstructed total water year streamflow (MCM), Snake River near Heise, Idaho, 1591–2005 (gray), with reconstruction 85th, 50th, and 15th percentiles (top to bottom; solid horizontal lines), RMSE (dotted horizontal lines), and 11 year moving average (thick black line).
Download figure to PowerPoint
Table 4. Upper Snake River Reconstructed n Year Means Two Standard Deviations or More Below Period Meana
|1 Year||5 Year||11 Year||21 Year|
|Year||Water Year Flow (MCM)||Period||Average Water Year Flow (MCM)||Period||Average Water Year Flow (MCM)||Period||Average Water Year Flow (MCM)|
|1695||3729|| || || || ||1639–1659||5694|
|1900||3943|| || || || ||1885–1905||5699|
| || || || || || ||1635–1655||5701|
| || || || || || ||1886–1906||5702|
| || || || || || ||1634–1654||5706|
 While individual extreme low-flow years are of concern, long-term periods of drought can be particularly taxing for water systems. Decadal variability in the reconstructed streamflow record was examined through the application of 5, 11, and 21 year moving average filters to the yearly time series (Table 4) and to standardized values of the yearly time series (Figure 6). The 1930s Dust Bowl drought is a severe period even in the context of the longer-term record: the 1930s rank high amongst the driest periods when examined over multiple time scales (Table 4).
Figure 6. Eleven year moving average of standardized (anomaly) values in reconstructed total water year streamflow, Snake River near Heise, Idaho. Gray shading represents periods above mean flow; black signifies periods below mean.
Download figure to PowerPoint
 Instrumental period droughts are eclipsed by conditions in the early to mid-1600s reconstructed flow. This period was severe in both magnitude and duration, dominating the record of drought periods (Table 4). Low-flow conditions were sustained over a long time period, with below-mean flow in 24 of the 34 years between 1626 and 1659 (Figure 6). This drought period has been documented in several other paleoclimate records in the West. Reconstructions of precipitation (A.D. 1226–2001) [Gray et al., 2004] and Palmer Drought Severity Index (A.D. 1405–2001) [MacDonald and Tingstad, 2007] in northeastern Utah identified this early to mid-1600s period as one of the most severe droughts on record in the Uinta Basin. LaMarche , in a study of tree ring records from California bristlecone pines, identified the beginning of this drought period (∼1620s) as the time of a climatic regime shift from cool moist to cool dry. Using tree ring reconstructions of the Palmer Drought Severity Index, Fye et al.  identified a “1930s Dust Bowl-like” drought from 1626 to 1634 that was centered over the Snake River headwaters and extended southwest across Utah, Nevada, and California.
 Unbroken periods of dry conditions can be particularly challenging for water resource management. Consecutive years of below-mean flow lasted in length from 2 to 7 years over the reconstructed record (Figure 7). The longest (7 year) periods occurred during the early 17th and 18th centuries. Examined this way, the 21st century drought would fall into the 6 year category (2000–2005), as it was interrupted by moderate flow in 2006. Over the past 415 years, there were six other periods in the record with consecutive low-flow periods as long as or longer than the early 21st century drought.
Figure 7. Consecutive year periods of below-mean flow in the reconstructed record of the Snake River near Heise, Idaho.
Download figure to PowerPoint
4.3. Synchronicity Between Western Rivers
 A dipole pattern of contrasting north-south anomalies in Western precipitation, streamflow, and snowpack records has been noted in previous research and attributed in part to ENSO-related modification of the prevailing storm track position [Redmond and Koch, 1991; Cayan, 1996; Dettinger et al., 1998]. Results from this study indicate that streamflow in the upper Snake River and the Verde River do have nearly equal, but opposite, correlations with SOI over the instrumental period (r = 0.35 and −0.32, respectively; p < 0.01), while the upper Colorado and Sacramento rivers do not have a significant ENSO signal. A north-south, out-of-phase relationship does not dominate the reconstructed streamflow records, however. Over the instrumental period, upper Snake River streamflow was significantly correlated with flow in both the upper Colorado and Sacramento rivers but had neither a positive nor negative correlation with the Verde River (Table 5). Over the longer period available from the reconstructed flows of the four rivers, the upper Snake River maintained a positive, significant relationship with the upper Colorado and Sacramento rivers and also has a low, but significantly positive, correlation with the Verde River (Table 5).
Table 5. Correlations Between Streamflow in the Snake River Near Heise, Idaho, and the Colorado, Sacramento, and Verde Riversa
| ||Instrumental 1911–2005||Reconstructed|
 All four rivers have had periods of synchronous and asynchronous flow (Figure 8). The strong and extended pluvial period across much of the West in the early 20th century is reflected in the records of all rivers in this study, though with an earlier peak in the Sacramento River record. This wet period coincided with much of the water development in the West, including the creation of the Colorado River Compact in 1922 [Stockton and Jacoby, 1976] and the initial construction of the Jackson Lake Dam in the headwaters of the Snake River in 1906. The 1930s Dust Bowl drought severely impacted northerly areas in the West and was the most extreme drought in the Snake and Sacramento rivers' instrumental records, but the upper Colorado and Verde rivers were more severely impacted by the 1950s drought. The upper Snake River was out of phase with the Verde River in the mid-1600s and mid-1800s (Figure 8). Correlation between flows in the upper Snake and the Sacramento and Verde rivers was highest in the 1700s (Table 5), a period characterized by low ENSO activity [Braganza et al., 2009] that was previously noted as a period of joint drought in the Sacramento and Blue river basins [Meko and Woodhouse, 2005]. In the latter half of the 20th century, a time characterized by unusually high ENSO activity [Gergis and Fowler, 2009], the Snake and Verde rivers have been particularly asynchronous (Figure 8c).
Figure 8. Eleven year moving averages of standardized reconstructed water year total streamflow for (a–c) the Snake River near Heise, Idaho (gray), (a) the Colorado River at Lees Ferry (black), (b) the Sacramento River four rivers index (black), and (c) the Verde River below Tanglewood Creek (black). Gage locations are shown in Figure 2.
Download figure to PowerPoint
 Over the 1591–2005 reconstruction period, the percentage of years with simultaneous above- or below-average flows in the Snake and Verde rivers (52%) was approximately equal to the percentage of years with flows of opposite signs in the two rivers (48%). There were more episodes of synchronous high flow (>+1 SD) or low flow (<−1 SD) in the Snake and Verde rivers (24 occurrences in the 415 year record) than events with high flow in one of the rivers and low flow in the other (15 occurrences) (Table 6). On the basis of a binomial model, the probability that these synchronous and asynchronous flows occurred by chance ranged from 0.03 to 0.14 (Table 6). The Snake and Verde rivers had a greater percentage of asynchronous flow than occurred between the Snake and Colorado rivers (7 of 415 years) or the Snake and Sacramento rivers (5 of 387 years). In the 387 year overlapping record of all four reconstructions, there were just nine occurrences of synchronous high- or low-flow conditions (based on 1 SD) in all four rivers (Table 6), but flow in the four rivers was of the same sign (above or below average) in 30% of individual years.
Table 6. High- and Low-Flow Years in the Snake and Verde Riversa
| ||Asynchronous Years||Synchronous Years|
|Snake River||>+1 SD||<−1 SD||>+1 SD||<−1 SD|
|Verde River||<−1 SD||>+1 SD||>+1 SD||<−1 SD|
| ||1971|| ||1787||1990|
| || || ||1838b||2002|
| || || ||1866b|| |
| || || ||1890|| |
| || || ||1907|| |
| || || ||1916|| |
| || || ||1965b|| |
| || || ||1978|| |
 The shifts between synchronous and dipole patterns in Western streamflow may be associated with two distinct modes of variation noted previously in research on instrumental precipitation and streamflow patterns [Sellers, 1968; Meko and Stockton, 1984]: (1) north-south fluctuations in storm track position that result in dominantly zonal flow and a dipole pattern of hydroclimatic variability and (2) east-west shifts in atmospheric pressure systems that result in meridional flow and West wide synchronicity. North-south zonal fluctuations have been attributed to ENSO-related shifts in the prevailing storm track position. Strong and persistent cool season meridional flow can result from atmospheric blocking caused by the failure of the Pacific subtropical high to retreat from its summer position [Carrera et al., 2004], as well as from changes in the strength and amplitude of the Pacific North American pattern (PNA) [Wallace and Gutzler, 1981]. A positive PNA pattern, characterized by a strong Aleutian low, ridging in the eastern Pacific and western United States, and a deep trough in the eastern United States, has been linked to snow deficits over the West in winter [Cayan, 1996] and low-flow conditions in western rivers over the instrumental period [Meko and Stockton, 1984].
 It is likely that both modes of variation have impacted drought in the Snake River Basin. Precipitation and streamflow in the headwaters of the Snake River have a modest but significant ENSO signal in the instrumental period, although this may also have been a time period of higher frequency and more extreme ENSO activity [Gergis and Fowler, 2009] relative to the previous three centuries. On the basis of spatial drought patterns in the four rivers, there are indications of a zonal flow pattern during two of the most severe droughts in the Snake River record (the 1630s and the 1930s), which were much less severe in the Verde River record. The Snake River's low-flow period in the early 1700s, which was long lasting but less severe in magnitude, is replicated in the flow of all four rivers and may be indicative of drought conditions resulting from persistent meridional flow.