Evidence of Seasonal Changes in Evapotranspiration ... - AMS Journals

974

JOURNAL OF HYDROMETEOROLOGY

VOLUME 5

Evidence of Seasonal Changes in Evapotranspiration in Eastern U.S. Hydrological Records MATTHEW J. CZIKOWSKY

AND

DAVID R. FITZJARRALD

Atmospheric Sciences Research Center, The University at Albany, State University of New York, Albany, New York (Manuscript received 26 February 2004, in final form 10 May 2004) ABSTRACT Hydrologic records provide some of the most widespread, long-term data available that can be expected to contain signals of climate variation over time. Yet such data have not been used to identify the widespread effects of spring onset in the forested eastern United States, where streamflow is strongly influenced by vegetation. Three independent runoff characteristics are affected by the enhanced evapotranspiration (ET) that occurs with annual leaf emergence: (i) P 2 R, the difference between precipitation and runoff, returns to dormant season values and increases due to ET; (ii) the streamflow recession time constant (a measure of the time required after rainfall for streams to return to their baseflow levels) shortens; and (iii) the diurnal streamflow amplitude increases. The smallest watersheds in our dataset (area ,200 km 2 ) exhibit the diurnal streamflow signal most often and with the greatest amplitude, in accord with accepted relationships between channel length and watershed area. The dynamics of ET effects on runoff characteristics, illustrated using a simple model, suggest that the recession time constant and diurnal amplitude depend on bulk characteristics of a watershed. Using the P 2 R, streamflow recession, and diurnal streamflow signal methods, a spring onset date is obtained from the historical hydrologic datasets. The P 2 R and streamflow recession spring onset dates proceed more slowly than the diurnal amplitude spring onset date and those obtained by independent methods.

1. Introduction

E 5 P 2 R 2 DS,

(1)

Rainfall in the eastern United States occurs nearly uniformly throughout the year (Finkelstein and Truppi 1991). While there is a runoff pulse associated with snowmelt in the more northern states (e.g., Huntington 2003), the subsequent reduced streamflow during the growing season is a response to the onset of evapotranspiration (ET). The eastern United States contains some of the most extensive deciduous forests anywhere (U.S. Forest Inventory and Analysis; see http:// fia.fs.fed.us). This means that the onset of ET represents a widespread land cover change, one that occurs in as little as 10 days (Fitzjarrald et al. 2001). The abrupt change in ET withdraws some of the groundwater that feeds the streams, leading to a widespread springtime and summertime reduction in streamflow. Many small headwater streams in the region are intermittent, despite its overall moist climate (e.g., Hansen 2001; Paybins 2003). Once leaves are gone in autumn, streamflow increases even in the absence of precipitation (Doyle 1991). The surface water and energy balances are linked by their common dependence on evaporation:

A 5 2(Q* 2 G) 5 H 1 LE.

(2)

Corresponding author address: Matthew Czikowsky, Atmospheric Sciences Research Center, 251 Fuller Rd., Albany, NY 12203. E-mail: [email protected]

q 2004 American Meteorological Society

In the surface water balance equation (1), E, the total evaporation, includes ET, bare-soil evaporation, and evaporative losses through interception; P is precipitation; and R is the runoff. The storage term DS includes soil moisture storage, groundwater storage, snowpack, and intercepted water that has not yet evaporated. The sign convention arises by asserting that P 5 E 1 R 1 DS. Though not strictly correct, the storage term has commonly been assumed to be zero in the annual sum. Our analysis retains this assumption (see below). In the surface energy balance equation (2), A is available energy, Q* is net radiation (defined here as the sum of the upward longwave and shortwave radiation components minus the sum of the downward longwave and shortwave radiation components), G is the ground heat flux, H is the sensible heat flux, and LE is the latent heat flux (where L is the latent heat of vaporization of water). The sign convention is that upward fluxes (away from the surface) are positive. For example, an upward sensible heat flux H into the air is positive, and an upward ground heat flux G (from ground to air) is positive. The common term between the surface water and energy balance equations is evaporation (E) in the water balance and LE in the energy balance. Direct observations of ET are available for extended

OCTOBER 2004

CZIKOWSKY AND FITZJARRALD

FIG. 1. Observed Harvard Forest evaporation (bold) averaged from 1992 to 2000. Amherst, MA, P minus Tully River, MA, R averaged from 1928 to 1997 (solid). The BB notation refers to the average bud-break date; the 95% notation refers to the date of 95% leaf emergence; and LF and FF refer to the climatological last and first freeze dates, respectively.

periods at very few sites worldwide. Evaporation is often estimated as a residual once precipitation and runoff are accounted for, and storage is deemed to be small on an annual basis. This makes it impossible to estimate ET uncertainty. The connections among abrupt land cover change, evapotranspiration, and streamflow should be apparent in standard streamflow data. To illustrate our points, we focus first on one site where for over a decade evaporation has been measured at a 30m research tower at the Harvard Forest, a mixed deciduous forest in central Massachusetts, using the eddy covariance method (Fig. 1). Site and measurement details are presented in Moore et al. (1996) and Sakai et al. (1997). Directly observed annual evaporation totaled 481 mm, close to the long-term integrated P 2 R of 483 mm. Though the observed evaporation is an 8-yr average (among the longest series available anywhere) while the P 2 R values represent a 70-yr average, these data support the common assumption of zero annual net storage. The storage term DS, estimated by subtracting the evaporation curve from the P 2 R curve in Fig. 1, exhibits large seasonal variations. Negative storage values seen from days 80 to 200 reflect discharge occurring largely during the snowmelt period whose peak typically occurs near day 100. The positive storage after day 200 and before day 80 reflects accumulation. Points of zero storage, reflecting a near balance between discharge and accumulation, occur approximately at days 80 and 210. Near the time of the climatological last freeze (day 125), the forest tree bud break commences (Fig. 1), nearly coinciding with the latent heat flux increase. During leaf development and before the 95% leaf emergence date, hydrological- and energy-balance-related indicators of leaf emergence appear. The available energy is equally partitioned into sensible and latent heat flux at

975

FIG. 2. Harvard Forest daily averaged fluxes above the canopy for 1991–98. The available energy A, sensible heat flux H, and latent heat flux LE (dotted), where A 5 H 1 LE; BB and 95% notations as in Fig. 1; and H equals LE at day 154. (From Fitzjarrald et al. 2001.)

Harvard Forest on day 154 (Fig. 2; Fitzjarrald et al. 2001). After this date, latent heat flux dominates, as a consequence of the rapidly increasing evapotranspiration by the vegetation. The quantity P 2 R recovers to its pre-snowmelt dormant-season value on day 142, between the time of bud break and 95% leaf emergence at Harvard Forest. The increase in the hydrologically estimated evaporation, P 2 R, after this date is most likely the consequence of evapotranspiration by the growing vegetation. Even though the storage term is not zero on a seasonal time scale, the above example shows that hydrological measures such as P 2 R may be useful in determining the time of increased evapotranspiration in the spring. Current methods of detecting long-term shifts in the spring onset have limitations. Phenology networks, such as the lilac network of Schwartz and Reiter (2000), are important, but are limited to those regions where the indicator species grow. The Normalized Difference Vegetation Index (NDVI) or ‘‘greenness’’ index (Myneni et al. 1997; Jenkins et al. 2002; White et al. 1997) is of limited use in long-term analyses because of the short length of the satellite record. Energy partition approaches (Fitzjarrald et al. 2001) are limited by lack of longterm temperature and humidity data from climate stations. Hydrologic data, specifically daily runoff measurements, are some of the most widespread, long-term data available that can be used to identify spring onset or long-term shifts in spring. Daily runoff measurements have been used for this purpose in the western United States (Cayan et al. 2001) but have been neglected in the eastern United States. The potential to discern climate change via spring onset is motivation for studying relationships among streamflow, ET, and spring onset. We study three integrated bulk properties of streamflow that operate at three distinct time scales: (a) at the seasonal time scale, P 2 R can be used to track the

976


VOLUME 5

FIG. 3. Diagram illustrating the P 2 R method. Data used for the plot: East Branch Swift River, Hardwick, MA, runoff (1937–2000) and Hardwick precipitation (1948–2000).

onset of the growing season; (b) at the synoptic time scale (the time scale of frontal passages), recession times (times required for streamflows to return to baseflow, following rainfall events) show seasonal variation; and (c) at the shortest time scale examined here, the amplitude and phase of diurnal streamflow signals observed in some watersheds also exhibit seasonal variations. The dynamic behavior of properties (b) and (c) is most pronounced near the time of spring onset and leaf emergence. We address the following questions in this paper: • Are there detectable changes in streamflow characteristics that could be used as indicators of spring onset and leaf emergence? • Can these changes be tracked as indicators of spring leaf presence along the East Coast of the United States? In the process of addressing these questions, we offer an elementary model of how runoff characteristics may be linked to bulk watershed properties. We focus on three independent runoff characteristics that change as a consequence of enhanced evapotranspiration with spring onset and leaf emergence: • The date of the return of the average P 2 R record to pre-snowmelt pulse values, denoting the point at which ET becomes the dominant factor in the growing-season P 2 R increase. • Seasonal changes in the streamflow recession time constant following rainfall events. • Seasonal changes in the amplitude of the diurnal streamflow signal observed in some watersheds. For each characteristic we present one or more case studies before discussing spatial integrations.

FIG. 4. USGS daily value streamflow stations used in the study.

2. Background/methods a. P 2 R method A long-term averaged precipitation (P) minus runoff (R) curve was produced using a 30-day moving average filter, pairing each runoff station with the nearest available precipitation station. Data for the first 50 days of the year were averaged to give mean dormant-season P 2 R. This time period was chosen for averaging because the P 2 R values are stable during this period. Also, runoff during this period is not significantly affected by snowmelt or by direct vegetation effects. The peak of the snowmelt period represents a minimum in the longterm-averaged P 2 R curve. Following the snowmelt period, the day at which the P 2 R curve recovers to its day-1-to-50 dormant-season mean was deemed the transition point from the snowmelt portion of spring to the leaf-out portion (Fig. 3). Although this method is effective in identifying the spring transition date in regions with a strong snowmelt pulse signal (e.g., northeast United States), it is not useful in the southeast United States, where the snowmelt pulse is weak or nonexistent. Attempts to find a method to detect the onset of ET in the southeast United States from the P 2 R curve were unsuccessful. Also not useful are data from stations with excessive regulation from dam operation or power generation, since their P 2 R records no longer reflect natural occurrences. United States Geological Survey (USGS) descriptions of stream gauge sites and remarks on data quality were used to select appropriate stations. The 736 USGS daily value streamflow stations selected (Fig. 4) provided at least 30 yr of data ending with 2000. Precipitation data from 945 U.S. East Coast

OCTOBER 2004


977

FIG. 5. Schematic diagram illustrating the streamflow recession method.

stations were extracted from the National Climatic Data Center Summary of the Day Dataset (EarthInfo, Inc. 2001) using the same selection criteria as for the streamflow data. The mean distance between paired precipitation and runoff stations was 18 km, the standard deviation 16.3 km, and the minimum and maximum distances 0.2 and 51.6 km, respectively. The paired station distance can influence the reliability of the P 2 R method if the precipitation is highly localized. Precipitation in the northeastern and southeastern United States is most spatially coherent (mostly synoptically influenced) during the spring, fall, and winter seasons, with the least spatial coherence occurring during the summer with more convective precipitation (Walsh et al. 1982). For the purpose of long-term precipitation averages, we conclude that the density of the network stations is sufficient for valid statistical analysis. Application of the P 2 R method to data from a central Massachusetts watershed (Fig. 3) results in a spring-date estimate of day 136 (mid-May), which is consistent with the phenological and energy-balance measurements taken at nearby Harvard Forest. b. Streamflow recession method Streamflow recession, the decline in streamflow following a precipitation event, is the second runoff characteristic chosen for study because of its response to spring onset and leaf emergence. On a 42-ha upland watershed at Hubbard Brook, New Hampshire, Federer (1973) reported that streamflow recessions proceeded more quickly with the onset of tree transpiration in the spring and slowed with leaf drop in the fall. We report streamflow recession as a time constant, defined as the number of days required for the daily mean observed streamflow to reach 1/e of the peak value following a precipitation event. If ET removes some groundwater, we expect the streamflow recession time constant to decrease following leaf emergence (Fig. 5). The area

between curves (1) and (2) represents some of the water used by transpiring vegetation. Wittenberg (2003) has modeled and estimated ET losses in baseflow recession analysis considering the area between curves (1) and (2) mentioned above. We modeled this process (appendix) and found that the intrinsic time constant of a watershed need not vary seasonally for this recession period to shorten during the growing season. The USGS daily value streamflow dataset of 736 stations described in section 2a was used in calculating streamflow recessions. To use the 1/e criterion to determine the streamflow recession time constant, we assume that the falling hydrograph limb decays exponentially (e.g., Tallaksen 1995). To justify this assumption, we performed log-linear fits on a 29-station subset of the daily streamflow recession data that included 16 977 recessions. For this data subset, the median r 2 value was 0.95, and the median residual standard error was 0.12, indicating that the exponential decay assumption is adequate. Model results (appendix) indicate that the exponential decay assumption probably is adequate even in the presence of simultaneous diurnal streamflow fluctuations. The recession events used in the analysis were chosen in splitting the runoff time series into 7-day windows and identifying peaks. We selected only those peaks for which runoff declined to 1/e of the peak value without there being an interfering peak from another precipitation event. This may introduce a bias toward dry periods, but we are confident that precipitation is not interfering with the observed recessions. This stringent criterion reduced the total number of analyzed events to about five per year per station. Since stations with at least 30 years of data only are considered, a sufficient number of recession events were analyzed to detect the seasonal change in the recession time constant (Fig. 6). In the Wappinger Creek, New York, watershed (418399110N, 738529230W; area 5 469 km 2 ), the time constant reduction begins in late May with leaf emer-

978


FIG. 6. Medians of the streamflow recession time constant, Wappinger Creek, NY, 1929–2000.

gence and progresses through early June with leaf development (Fig. 6). We examined a 14-station subset of the data for differences in the recession time constants estimated using daily and 15-min data. The only available long-term (.10 yr duration) streamflow data are daily observations. We reasoned that if the seasonal change in the recession times is evident in these daily data, we could examine the entire East Coast station network. Recession times are longer for the daily data (Table 1). We identify an averaging bias that accounts for the longer recession times in the daily data. When averaged daily, a peak value from a precipitation event will be somewhat lower than the higher-resolution data. This causes the e-folding value to be lower, and consequently a longer period of time is required to reach the lower e-folding value. We conclude that daily data are generally adequate for our purpose. We examined whether the magnitude of a precipitation event affects the streamflow recession time constant and found the median streamflow peak did not vary among events with different streamflow recession time constants (see Czikowsky 2003, Fig. 3.7). c. Diurnal streamflow oscillation method A third runoff characteristic identified with spring onset and leaf presence is the amplitude of the diurnal streamflow signal. Under sufficiently dry conditions, a well-defined diurnal signal in streamflow is frequently

VOLUME 5

observed in small watersheds during the growing season. Decreases in streamflow can occur during the day when transpiring vegetation draws upon the groundwater supply that composes the baseflow of gaining streams, thereby reducing the stream inflow. Transpiration is at a minimum at night, resulting in increased stream inflow and total streamflow (Fig. 7). The inflow of groundwater occurs assuming the streams under consideration are gaining streams. This consideration is sufficient for our analysis because the streams in the studied network (Fig. 8) are all perennial streams, which are usually gaining streams sustained by groundwater inflow between precipitation input events (Dingman 2002). The Danish well-extraction study of Nyholm et al. (2003) presents direct observational evidence of riparian-zone ET-induced diurnal fluctuations in streamflow. Bren (1997) concluded that diurnal streamflow variations were due to transpiration by riparian and nearriparian vegetation only, with little contribution from vegetation outside this zone. The riparian zone is usually confined to a small area centered on the stream channel, often as small as 25 m out from each bank (Meyboom 1967). Goodrich et al. (2000) estimated the riparian zone to extend from 10 to 100 m from each bank. Using isotope analysis, Dawson and Ehleringer (1991) found that mature, deep-rooted riparian-zone trees do not use groundwater inflow into streams as their primary water source, but rather groundwater from a deeper source. They observed that it is primarily the younger, more shallow-rooted trees and herbaceous riparian vegetation whose transpiration affects streamflow. All previous studies of the diurnal streamflow signal in North America have been concentrated on arid or seasonally dry regions in the western part of the continent (e.g., White 1932; Troxell 1936; Meyboom 1967; Bond et al. 2002). Lundquist and Cayan (2002, their Fig. 1) claimed that the diurnal streamflow signal in the humid U.S. East Coast states could not be detected north of North Carolina and therefore confined their study to the western United States. However, to determine the presence of the diurnal streamflow signal they fitted 10day windows of data to a 24-h harmonic. This window length is clearly inappropriate for humid East Coast stations, which experience frequent frontal precipitation. The typical growing-season synoptic frontal passage frequency in this region is once to twice per week

TABLE 1. Streamflow recession time constant median, lower 95% confidence interval for the median, and upper 95% confidence interval for the median (days) for the 15-min and daily resolution data. The columns on the left-hand side are for the period 1 May–30 Jun, and those on the right-hand side are for 1 Jul–15 Sep). Yeardays 121–181 (1 May–30 Jun)

15 min Daily

Yeardays 182–259 (1 Jul–15 Sep)

Median

Lower

Upper

3.3 5.0

1.6 3.9

5.0 6.0

15 min Daily

Median

Lower

Upper

2.1 3.0

1.2 2.4

3.0 3.6

OCTOBER 2004


979

FIG. 7. Schematic diagram illustrating the diurnal streamflow signal.

(Freedman and Fitzjarrald 2001), so we chose a 3-day moving window for our analysis of diurnal streamflows. We looked for diurnal streamflow signals in a database created from the 15-min stream discharge records of 151 USGS stations (Fig. 8). We obtained 15-min discharge records in electronic form for most stations from 1998 to 2001; we obtained such data from select stations in New York back to 1989. The digital resolution of the streamflow data (gauge height) is 3 mm. Because of changing discharge–gauge-height relationships, discharge data resolution varies according to the discharge value. For the data shown in Fig. 9, the resolution is 0.005 m 3 s 21 . When normalized by watershed

FIG. 8. Map of 15-min USGS streamflow data stations used.

area, the resolution is 0.04 mm on watershed. This is one-third of the amplitude of the diurnal streamflow signal in this example, which is 0.12 mm on watershed. Our approach to these data was guided by modeling known mechanisms of water movement. Evapotranspiration removes groundwater only during the day. We examined the response of a simply modeled watershed to such diurnally asymmetrical forcing (appendix) and found that the diurnal streamflow signal amplitude and phase statistics could be adequately described for present purposes using a sine curve. Accordingly, we created 3-day data windows and then fitted the following sine curve to them: Ax 1 D 1 B sin[2p (x 1 E)], (3) where A is a trend, D is a bias, B is diurnal amplitude, and E is diurnal signal phase. The diurnal signal time fraction (the fraction of time the diurnal streamflow signal appears between precipitation events) was calculated by a least squares fit to the data, using (3). Data that exhibited the diurnal streamflow signal were chosen using a set of objective criteria that best matched a subjective selection of data that exhibited the diurnal streamflow signal. These objective criteria are described below. Precipitation or recession periods were removed by keeping only the data for which the sum of the residuals normalized by streamflow was less than one. Streamflows with diurnal variations greater than or equal to 2% of the streamflow during periods where the streamflow trend fitted from (3) was declining were considered to have the diurnal streamflow signal. The 2% threshold is near the limit of detectability for diurnal streamflow fluctuations. d. Daily and seasonal progression of the diurnal streamflow signal Streamflows begin to decline in the morning with the onset of incoming solar radiation (Fig. 9) and reach their

980


VOLUME 5

FIG. 9. Red Hill, NY, radiation (W m 22 ) and Biscuit Brook, NY, streamflow (mm on watershed 3 400) for 11–16 Jul 1998 (day of year 192–197). Note the limitations imposed by the coarse streamflow data resolution.

minimum shortly after sunset, indicative of diurnal streamflow fluctuations due to ET. There is an asymmetry in the diurnal streamflow signal identified by Lundquist and Cayan (2002). This characteristic is explained by the relatively short duration of ET forcing during the day (appendix). Diurnal streamflow fluctuations due to ET are characterized by a gradual rise and a sharp fall, while diurnal streamflow fluctuations due to snowmelt show a sharp rise and a gradual fall. The diurnal streamflow signals selected for this analysis were due to ET. We did observe a small number of cases in small northeastern mountain watersheds of diurnal streamflow fluctuations due to snowmelt in the month of April during some years. However, our objective analysis rejected those cases for one of three reasons. First, the overall streamflow trend was positive in many of the observed snowmelt-driven diurnal streamflow cases. Second, the amplitude of the snowmelt-driven diurnal streamflow signal did not exceed the 2% of total streamflow threshold. Third, the asymmetry of the signal was such that the sum of the residuals criterion was not met. Although the ET-driven diurnal streamflow signals were also asymmetric, the asymmetry of the snowmelt-driven signals was more pronounced. Data from the Biscuit Brook watershed in the Catskills of New York State (area 10 km 2 ) illustrate the seasonal progression of the diurnal streamflow signal (Fig. 10). A large increase in the amplitude occurs near day 150 (late May) around the time of leaf emergence. The amplitude remains high in June and July (Fig. 10a) but decreases in August as the groundwater supply be-

comes depleted. Normalizing the diurnal streamflow signal amplitude by total streamflow (Fig. 10b), we find that the diurnal streamflow variation, as fraction of total streamflow, reaches a maximum after day 200, near the time of maximum ET observed at Harvard Forest, which we take to be representative of the region. Within the resolution of the USGS streamflow and gauge-height measurements currently available, it is difficult to determine the exact hour of the streamflow minimum (Fig. 10c). For days 150 to 220, the minimum occurs between 2000 and 2300 eastern standard time (EST), lagging the solar radiation daily maximum by 8 to 11 h. Later in the growing season there is a trend toward earlier time of streamflow minimum, but with much greater variability. This variability may be indicative of the lower streamflow amplitudes observed late in the growing season (Fig. 10a). These estimates could be greatly improved were higher-resolution gauge-height data available. Since the amplitude of the diurnal streamflow signal is small, the possible effect of thermal expansion of water on the amplitude was examined. Typical growingseason values for Biscuit Brook, New York, are 0.03 mm on watershed per day for the diurnal streamflow signal amplitude (Fig. 10a) and 3 mm on watershed per day for the total streamflow (Fig. 10d). Expressing this variation as a thermal expansion gives 4.8 3 10 24 mm day 21 8C 21 . Such a signal would require the water to cycle through an unrealistic diurnal temperature range of about 508C. We conclude that effects of thermal ex-

OCTOBER 2004


981

FIG. 10. (a) Amplitude of the diurnal streamflow signal using stream discharge data (mm on watershed day 21 ) for Biscuit Brook, NY, 2001. (b) Amplitude of the diurnal streamflow signal normalized by total streamflow. (c) Time of streamflow minimum for those cases in which diurnal streamflow signal amplitude exceeded 2% of total streamflow. (d) Total streamflow (mm on watershed day 21 ).

pansion of water on the amplitude of the diurnal streamflow signal are unimportant. Constantz et al. (1994) identified a significant change in the viscosity of stream water, and thus the hydraulic conductivity of the streambed, as the primary mechanism driving diurnal streamflow fluctuations in losing streams in New Mexico and Colorado with large diurnal stream temperature ranges (;158C). Streams in forested regions of eastern North America that are gaining rarely experience such large diurnal temperature ranges (e.g., Bourque and Pomeroy 2001). Furthermore, Constantz (1998) reported that ET losses determined diurnal streamflow variations in gaining reaches. This mechanism is probably not relevant to our data, since all the streams considered here are perennial and are thus likely to be gaining streams. e. Watershed effects on the diurnal streamflow signal A detailed examination of the influence of forest cover fraction on the streamflow responses was made; details are presented in Czikowsky (2003). Little difference in diurnal streamflow time fraction or amplitude

occurs for forest cover fractions above 0.6 (Fig. 11), which encompasses the vast majority of watersheds considered here. We found that the smallest watersheds (less than 200 km 2 ) show the diurnal signal more frequently (Fig. 12a) and with greater amplitude (Fig. 12b) than do larger watersheds. A similar relationship relating watershed area to diurnal streamflow amplitude was found by Lundquist and Cayan (2002, their Fig. 9) for streams in the western United States. The relationships among watershed area, total stream length, and riparian area are important here because the diurnal streamflow signal has been attributed to transpiration in the riparian zone only (Bren 1997). Geometrically, mainstream length (L) scales to the watershed area (A w ) as L ; A hw, h ø 0.5 (Eagleson 1970). The exponent h has been determined empirically to fall between 0.5 and 0.6 (Dodds and Rothman 2000). Similarly, total stream length (Z) has been found to scale to the watershed area through the power law Z ; A bw 5 0.95 (Rosso and Bacchi 1991). Crave and Davy (1997) report that b lies between 0.7 and 0.9. Assuming that the riparian zone is of average width d,

982


FIG. 11. (a) Watershed forest cover vs diurnal signal time fraction, with the upper and lower 95% confidence intervals for the median. Values listed above the x axis indicate the number of station years included for each forest cover fraction bin. (b) As in (a), but for watershed forest cover vs maximum annual amplitude fraction of streamflow.

the riparian area (Arip ) is proportional to the total stream length: Zd ; Arip .

(4)

Combining the above expression with the relationship between total stream length and watershed area, and introducing the riparian-area fraction (Frip ), we find Frip 5

Arip 1 ; (1/b)21 ; Abw21 . Aw Z

(5)

For b , 1, the riparian-area fraction is inversely proportional to watershed area. Accordingly, the evapotranspiration of a higher percentage of the vegetation of small watersheds can be expected to affect streamflows in a matter of minutes or hours, thus creating a strong diurnal signal. A simple physical basis for relating the amplitude fraction of streamflow and diurnal signal time fraction to the riparian-area fraction can be deduced by considering the analytical model presented in the appendix.

VOLUME 5

FIG. 12. (a) Watershed area vs diurnal signal time fraction (solid line), with the upper and lower 95% confidence intervals for the median. Values listed above the x axis indicate the number of station years included for each watershed size range. Dashed curves represent the power-law relationship between watershed-area and riparian-area fraction from (5). Labels above the curves denote b values from 0.5 to 0.8. (b) As in (a), but for watershed area vs maximum annual amplitude fraction of streamflow.

From (A6), ET ; b sin10 (v/2)t; the large exponent of the sine term allows a realistic accounting of the day/ night asymmetry in ET typically observed. The illustrative model shows that ET forcing produces a streamflow perturbation S9 proportional to the riparian area, with the recession time constant t w predicted to be proportional to the riparian area (A11). The streamflow perturbation S9 also depends on the climate through P 2 E. The mean streamflow S is proportional to the watershed area, as well as to P 2 E. This allows us to express the diurnal streamflow signal amplitude fraction, S9/S : S9 Arip ; 5 Frip ; AbW21 . S Aw

(6)

The diurnal streamflow signal time fraction is related to the amplitude fraction since we defined the time fraction to be the fraction of time the amplitude fraction

OCTOBER 2004


exceeds 0.02. Within the existing data, the diurnal streamflow signal is near the limits of detectability. The relationship between time fraction and watershed area falls within the curves at b values between 0.5 and 0.6 (Fig. 12a). These exponents are within the range of the mainstream length scaling rather than the more plausible total stream length scaling. We speculate that this may be a consequence of the fact that the average riparian distance d is not constant. Note that even for large watersheds (;1000 km 2 ) the time fraction does not reach zero. The diurnal streamflow signal may still be observable even in data from large watersheds. Examining these details is left for future work. We found no relationship between the phase lag of the diurnal streamflow signal and watershed area (not shown), a result not unexpected since ET forcing is in phase for all sizes of watersheds, but this conclusion is provisional, owing to the low-amplitude resolution of the streamflow data. 3. Results a. Comparison of spring onset dates The oldest approach to inferring the date of spring onset in eastern North America is Hopkins’s (1918) geographical rule: spring is delayed 4 days per degree of latitude, 1.25 days per degree of longitude east of the origin, and a day for every 100 ft of elevation. Phenological networks, such as the lilac network of Schwartz and Reiter (2000), have been used more recently but are limited by the spatial and temporal distribution of the chosen species. The energy partition method of Fitzjarrald et al. (2001) allows for examination of long-term variability in the spring date, but the number of long-term climate stations with data before the 1950s is small. Satellite technology, in particular the NDVI, has been used to assess spring onset and its variability (Myneni et al. 1997; Jenkins et al. 2002). While satellite data are available at high spatial resolution, long-term climate inferences cannot yet be made because of the short duration of the datasets. We compare spring dates obtained from our three methods to the spring dates obtained by Fitzjarrald et al. (2001), who defined the spring date (d F ) as the date at which a daily temperature and humidity tendency Bowen ratio passes through 1.0. Note that each of these objective methods has an arbitrary offset date. The P 2 R spring-date data (d P2R ) and the d F spring-date data were interpolated onto a 18 3 18 grid covering the region from 388 to 468N latitude and 678 to 828W longitude. Since the P 2 R method is ineffective in regions with a weak snowmelt pulse, stations with an annual minimum P 2 R greater than 0.5 mm were excluded. The spring-date values from each corresponding grid cell were taken to produce the scatterplot shown in Fig. 13. There is a detectable increase in d P2R with an increase in d F . However, the relationship is not 1:1, and the slope

983

FIG. 13. Comparison of hydrological spring dates and the spring dates from Fitzjarrald et al. (2001), who define the spring date as the date the tendency Bowen ratio passes through 1.0. The scatterplot is of the Fitzjarrald et al. (2001) spring dates (d F ) against the P 2 R spring dates (d P2R ); the least squares fit is d P2R 5 1.9 d F 2 121; and r 2 5 0.59. The ‘‘R’’ notations refer to spring dates determined from the recession time constant method, defined as the date at which the streamflow recession time constant decreases to 4.5 days. Medians and ranges of R by latitude (388–448N), as well as the d F spring-date range for each corresponding latitude, are shown in the dashed crosses. The ‘‘A’’ notations refer to spring dates determined from the diurnal streamflow amplitude method, defined as the date the diurnal streamflow signal amplitude reaches 2% of the total streamflow. Values are given for the southeastern stations (328–368N), the mid-Atlantic stations (368–408N), and the northeastern stations (408–458N). The solid crosses denote the date ranges over those regions.

of the least squares fit (1.9) indicates that the progression of d P2R is much slower than that of the d F progression. A robust fit to the data did not change the observed slope appreciably. The relationship is not a result of the grid resolution; fitting finer grids yielded similar results. It is not a consequence of the grid interpolation, as a scatterplot of the d P2R and d F points nearest to each other did not change the relationship. A similar relationship exists between d P2R and the spring dates obtained from Hopkins’s (1918) geographical rule, the least squares fit of which is d F 5 1.55 dHOPKINS 2 91, r 2 5 0.46, as well as between d P2R and the ‘‘full bloom’’ dates reported in the lilac phenology network (Schwartz 1996; not shown). Why the hydrological measure of spring d P2R would proceed more slowly than energybalance, geographical, or phenological measures is not yet clear, though it may be related to the effect of overland flows and small streams in delaying ET effects on the recession. We plan to examine this in more detail in continuing work. A spring date was obtained from the streamflow recession time constant method by defining the spring date as the date the streamflow recession time constant passed through 4.5 days. Median values for 388 to 448N are shown (‘‘R’’; Fig. 13) with a 2-week range of recession spring dates for each point, a result of using 2week aggregates to obtain sufficient recession data. The relationship of the recession spring dates to d F

984


FIG. 14. Median streamflow recession time constant by latitude. Lines indicate periods from early May (top line) to Jul–Sep (bottom line).

is similar to that of the P 2 R spring dates; the recession spring estimates also proceed more slowly than does d F . Although the 4.5-day threshold for determining the recession spring date is arbitrary, changing the threshold from between 4 and 5 days did not change the observed slope. At a given site, the streamflow recession time constant gradually shortens from the time of bud break to full leaf emergence and development, which takes several weeks (Fig. 1). This finding is consistent with the observations of Sakai et al. (1997) who showed that leaf area index (LAI) slowly increases and canopy resistance to water vapor transport r c slowly decreases from the time of bud break to full leaf emergence and development. The magnitude of the spring-season decrease in streamflow recession time becomes greater as one heads north, peaking at about 428N (Fig. 14). U.S. forest inventory statistics for 2002 (http://www.ncrs.fs.fed.us/ 4801/FIADB/rpaptabler/2002prpapdraftptables.htm) indicate that the percentage of standing deciduous forest timber is greatest in the states from Pennsylvania to Connecticut (85%–90%), then decreases in the northern New England states north of 428N, to about 60% in New Hampshire. We speculate that the lower percentage of deciduous forest may help to explain the smaller decreases in streamflow recession times observed north of 428N. Testing this hypothesis is left for future work. Anomalously high values for the streamflow recession time constant are seen in the 338–358N latitude band (Fig. 14). This band includes the Carolina coastal plain region, whose watersheds are predominantly large, coniferous flatlands whose characteristics are in stark contrast to those of the primarily upland, deciduous, hardwood-dominated watersheds in most of the remainder of the East Coast network. The Carolina coastal plain region was specifically cited by Sun et al. (2002) as having different hydrological characteristics from those of the nearby hardwood, upland watersheds. These dif-

VOLUME 5

ferences included runoff/precipitation ratio, potential ET ratio, water yield, and stormflow peaks. The region’s wetland areas do not drain well and may account for its longer recession time constants. To infer the onset of spring from the third signal we studied, the amplitude of diurnal streamflow variation, we first used amplitude data fitted from (3), grouped the data into 7-day windows, and chose the upper quartile for each 7-day window. Watersheds less than 200 km 2 in area were selected, and the southeastern, mid-Atlantic, and northeastern regional data were compiled separately. The inferred spring date was taken to be the first date for which the diurnal streamflow signal amplitude (given by the upper quartile of each 7-day data window) exceeded 2% of the total streamflow. Two percent variation is near the limit of signal detectability. The 200-km 2 threshold was chosen since it includes the watersheds that are most likely to exhibit a marked diurnal streamflow signal (Fig. 12a), but even in those watersheds the diurnal streamflow is only observed between 10% and 40% of the time. As a result, choosing the median of the windows of fitted amplitude data yielded amplitudes below the 2% of the total streamflow threshold throughout the growing season. Therefore, the upper quartiles were chosen to examine the data for which the diurnal streamflow signal is expressed. The 7-day window was chosen because of sample-size constraints and allowed for a continuous dataset. Choosing windows less than 7 days in length resulted in periods containing no data exhibiting the diurnal streamflow signal during the growing season. The spring dates inferred by the diurnal streamflow amplitude method for the southeastern, mid-Atlantic, and northeastern stations (‘‘A’’; Fig. 13) exhibit a much smaller slope versus d F than do the P 2 R and recession spring dates. Changing the arbitrary 2% threshold value to 3% of the total streamflow does not change the slope appreciably. The faster progression in the diurnal amplitude spring date may be expected since the diurnal streamflow signal contains direct evidence of the time during the season when ET rises steeply. Furthermore, the effect of transpiring riparian-zone vegetation can be more quickly be detected in streamflow than observing the effect of ET on the spring-season recession time decrease or P 2 R increase, which require transpiration from a more mature vegetation over a greater part of a watershed. The results for the diurnal amplitude are provisional because of the lack of a long-term dataset as opposed to the P 2 R and recession data. For the southeastern stations, the spring date inferred from this diurnal amplitude method occurs at day 107 (Fig. 15). The spring dates for the mid-Atlantic and northeastern stations are 114 and 128, respectively. Note that the seasonal maximum diurnal amplitude is higher for the southeastern stations (.6% of streamflow) versus the northeastern stations (,4% of streamflow). Close inspection of the data revealed that diurnal

OCTOBER 2004

985


TABLE 2. Spring onset date change (mean and standard errors given) determined from the P 2 R method for 1976 to 2000 minus 1951 to 1975. The number of stations included in each region is noted by ‘‘n.’’ The stations were split into three regions using these latitude ranges: New England–New York (NE–NY): lat $ 418N; PA–NJ: 39.58N , lat , 418N; and VA: 36.58N # lat # 39.58N.

FIG. 15. Diurnal streamflow signal amplitude fraction of total streamflow, shown as upper quartiles for watersheds of size ,200 km 2 . Solid line includes southeastern stations (328–368N), dotted line the mid-Atlantic stations (368–408N), and the dashed line the northeastern stations (408–458N).

Region

Mean (1976–2000 2 1951–75)

n

NE–NY PA–NJ VA

26.6 6 0.6 1.6 6 0.7 25.0 6 1.6

236 130 112

research is needed to understand these relationships and explain why the spring dates are earlier in New England–New York and Virginia but not in Pennsylvania– New Jersey. However, the current analysis demonstrates the utility of using the long-term hydrological data to assess climate shifts. 4. Summary and conclusions

streamflow oscillations were not detected outside the period shown in Fig. 15. b. Is spring arriving earlier in the eastern United States? Since 15-min streamflow data have been collected in electronic form only in the past decade, long-term changes in spring dates inferred from diurnal streamflow signal amplitude were not determined. We note that the USGS has long-term records of strip charts of streamflow data that could be digitized and analyzed. This is left for future work. We constructed 2-week median plots of the streamflow recession time constant during the spring season, similar to Fig. 14, for the 1928–58 and 1970–2000 periods. These plots (not shown) are almost identical, indicating that if there were any changes in the timing of the decrease of the streamflow recession time constant in those years, they occurred during periods of less than 2 weeks. The P 2 R method, however, allowed us to assess long-term shifts in spring. We selected stations with at least 20 years of data for the 1951–75 and 1976–2000 periods and obtained a mid-to-late-twentieth-century spring-date difference. We observe a trend toward an earlier spring in the New England–New York region (Table 2) of more than 5 days between the middle and late decades of the century, consistent with results from this region reported by Fitzjarrald et al. (2001) and Schwartz and Reiter (2000). There is also a trend toward earlier springs in the last 50 years along the West Coast (Cayan et al. 2001). In the data for the Pennsylvania– New Jersey region we find almost no evidence of a change in the spring date. In Virginia there is evidence of a trend toward an earlier spring, contrary to the results of Fitzjarrald et al. (2001), who noted a trend toward later springs in Virginia during the last 50 years. More

We return to the questions put forth in the introduction to this paper. First, Are there detectable changes in streamflow characteristics that could be used as indicators of spring onset and leaf presence? Three streamflow characteristics change with and may be used as indicators of spring onset and leaf emergence. These are (a) the return to pre-snowmelt pulse values of long-term average precipitation minus runoff (P 2 R); (b) seasonal variation in the streamflow recession time constant; and (c) seasonal variation in the amplitude of the diurnal streamflow signal. The P 2 R method provides a point measure of the spring date. The streamflow recession time constant and diurnal streamflow signal amplitude methods can be tracked seasonally, and a single spring date can be obtained from both methods by choosing an appropriate threshold. The recession time constant and amplitude thresholds chosen in this study were 4.5 days and 2% of total streamflow, respectively. Second, can these changes be tracked as indicators of spring leaf presence along the East Coast of the United States? Springtime changes in the three identified streamflow characteristics can be tracked along the East Coast of the United States. The progression of the P 2 R and recession spring dates are much slower when compared to energy-balance, geographical, and phenological estimates of spring date. The progression of the diurnal amplitude spring date appears to be much faster than the P 2 R and recession date progressions. More research is needed to answer why these hydrological measures of spring exhibit this behavior. Applied to long-term data from numerous daily streamflow and precipitation stations, the P 2 R method can be used for assessing long-term shifts in the spring date. The most evident result from the P 2 R method is that spring was arriving in the New England–New York region by more than 5 days earlier in the late decades of the twentieth century than it had arrived in

986


the middle decades. This agrees with estimates made using with energy-balance and phenological estimates for this region and time span. We examined factors and watershed characteristics that may be associated with the presence and strength of the amplitude of the diurnal streamflow signal. Stations with the smallest watershed areas exhibited the diurnal streamflow signal with the greatest frequency and amplitude. The illustrative physical model (appendix) indicates that streamflow recession time and diurnal streamflow fluctuation amplitude depend on two bulk characteristics of a watershed: riparian area and hydraulic conductivity. In the model, increased evaporation results in more rapid streamflow recession and an increase in the amplitude of the diurnal streamflow fluctuation. In the model, the intrinsic recession time constant appropriate to a watershed need not vary seasonally for there to appear to be a change in the recession due to evapotranspiration. In the data, we observed behavior similar to that predicted by the model in both characteristics, during the time of increased evaporation in the growing season.

VOLUME 5

overland flow are neglected. Through suitable definition of mean and transient signals, it may be possible to generalize this method to address questions of baseflow recession (e.g., Tallaksen 1995, Wittenberg 2003), but that is not our present aim. Transpiration primarily by riparian-zone vegetation is responsible for diurnal in streamflow fluctuations (Bren 1997). Vegetation away from the riparian zone contributes little. The temporal behavior of the diurnal water table and stream inflow fluctuations can be written d dhgw (GW) 5 (ldrip r w f ) 5 P 2 E 1 I 2 D, dt dt

(A1)

d dh (RW) 5 (ldriver r w ) rw 5 1D, dt dt

(A2)

where (A1) is the groundwater budget (GW), and (A2) is the stream water budget (RW). The river length is l, the density of water is r w , the riparian-area width is drip, the river width is driver , the soil porosity is f, the groundwater height is hgw, the water level of the stream is hrw , P is the precipitation, E is the evaporation, I is infiltration, and D is the drainage (outflow) from groundwater to river water. The diurnal streamflow fluctuation mirrors that of the water table, with the groundwater outflow draining into the stream. Combining (A1) and (A2) results in

Acknowledgments. This work was done primarily with support from the National Science Foundation under Grant NSF ATM0085120, ‘‘Collaborative Research: Connecting Spring Phenology with Lower Atmosphere Energy-Mass Exchange, Phase Two’’ (70%). Our partner in this project was M. Schwartz, University of Wisconsin—Milwaukee. Additional support came from the Office of Science, Biological and Environmental Research Program (BER), U.S. Department of Energy, through the Northeast Regional Center of the National Institute for Global Environmental Change (NIGEC), through a subcontract with Harvard University under Cooperative Agreement DE-FC02-03ER63613 (30%). The United States Geological Survey (USGS) is acknowledged for providing the 15-min streamflow data used in the analysis. We thank David Eisenman, as well as the anonymous reviewers, for many helpful comments.

Assuming the soil porosity f is constant, we define hgwe 5 fhgw . Factoring out the riparian width drip from the left-hand side of (A3) yields

APPENDIX

dh9 5 p(t) 2 e(t) 1 i 2 2d, dt

Illustrative Model of Evapotranspiration Effects on Water Table and Stream Inflow Fluctuations

where the lowercase terms indicate area and density normalized water flux terms. Since the diurnal fluctuation is observed only during dry periods between precipitation events, we assume the precipitation and infiltration terms are zero. The evaporation term is approximated as

Our goal here is to present a first-order model using accepted hydrological relationships to seek signals of ET in streamflow. This elementary model, a version of the standard Boussinesq approach that incorporates a form of Darcy’s law is forced with diurnal ET forcing. The model illustrates connections among the storm recession time constant, the diurnal fluctuation amplitude, and bulk watershed properties. By concentrating on the intraseasonal events, this model only considers the interaction of the riparian zone, groundwater, and the stream. Other important hydrological processes such as

1

lr w drip f

Arip r w

2

dhgw dhrw 2 driver 5 P 2 E 1 I 2 2D. (A3) dt dt

1 dt

dhgwe

2a

2

dhrw 5 P 2 E 1 I 2 2D, dt

(A4)

where Arip is the riparian area, and the a is the ratio of river width to riparian-area width, driver /drip. Defining a height perturbation above baseflow h9 5 hgwe 2 ahrw, and dividing through by rw and the riparian area, yields

e(t) 5 b sin 2n

1 2 t2 , v

(A5)

(A6)

where b is the amplitude, and v the diurnal radial frequency. The quantity n 5 5 yields an adequate approximation for the observed evaporation (Fig. A1). This unusually high power allows for a sharp midday peak

OCTOBER 2004


FIG. A1. Harvard Forest ET and model forcing, 11–16 Jul 1998 (yeardays 192–197).

in ET and a long overnight period of evaporation values near zero. The drainage term is approximated as D 5 A d r w u,

(A7)

where A d is the drainage cross-sectional area, and u is the velocity. The velocity is given by Darcy’s law, expressed in one-dimensional form, u 5 2K

dh9 , dx

(A8)

where K stands for the hydraulic conductivity of the soil. Substituting u into D, and then dividing through by r w Arip yields the expression for the drainage term d: d 5 2K

A d dh9 A h9 ù 2K d , Arip dx Arip d w

(A9)

where d w is the watershed diameter (for the riparian zone and river). The drainage term d is rewritten as d5

2h9 , tw

(A10)

where t w is the watershed time constant, defined as

tw 5

d w Arip . K Ad

(A11)

Note t w is proportional to the riparian area and inversely proportional to the cross-sectional area of the drainage and the hydraulic conductivity. Substituting into (A5), the height perturbation above baseflow is

1 2

dh9 h9 v 1 2 5 2b sin 2n t , dt tw 2

(A12)

To write the above equation in dimensionless form, we introduce the scaling parameters for time (t 5 t sˆt), height (h 5 hs hˆ ), and evaporation [b 5 (h s /t s )bˆ ], with 

987

FIG. A2. Analytic solutions of dimensionless hˆ plotted against dimensionless time ˆt, with the amplitude of the dimensionless forcing bˆ varying from 0 (top solid line) to 3 (bottom solid line). The bottom curve is fitted with an exponential-type curve (bold) representing recession and a sine curve (bold) representing the diurnal response of hˆ .

representing dimensionless quantities. The characteristic time t s is taken to be 1 day. The height scale h s 5 B/a represents the drop in h rw caused by the total daily evaporation B over the riparian zone. The change in the perturbation height can be amplified by the ratio of the riparian area to river area, 1/a. For simplicity we assume a 5 1. Equation (A12) in dimensionless form is

1 2

dhˆ P 1 hˆ 5 2bˆ sin 2n (p tˆ ), (A13) dtˆ tw where bˆ 5 b(B/Pa), with the period P 5 1 day. The simple model has only two parameters. Realistic values for the range of bˆ are between 0 and 3, which correspond to evaporation rates of 0 and 100 W m 22 , respectively. The streamflow recession and diurnal amplitude bulk watershed characteristics are illustrated in the plot of the dimensionless solution (Fig. A2). The full analytic dimensional solution is given in Czikowsky (2003, 65–66). The model shows how increased evaporation results in a faster recession in hˆ as well as an increase in its subsequent diurnal amplitude. Note that the symmetric forcing produces an asymmetric diurnal response in hˆ , with the daily decline being more rapid than the daily increase, a feature observed in ET-driven diurnal streamflow fluctuations by Lundquist and Cayan (2002). Using the prescribed forcing in the model, an 18-h period of gradual rise in hˆ ending in the morning (0800 LT) is followed by a 6-h period of more rapid fall in hˆ ending in the afternoon (1600 LT). In reality, the period of time of the daily rise and fall in streamflow is not fixed and varies with changes in the length of day and cloudiness. Changing the model forcing accordingly would result in changes in the daily maximum and minimum streamflow times. The timing of the daily maxima and minima in hˆ varies slightly with the intrinsic model watershed time constant t w . Variations in t w also change the magnitude

988


of the recession in hˆ . Fitting a simple sine curve to the diurnal course of hˆ , similar to the analysis discussed in the main body of the paper, results in a correlation of 0.92, justifying that approach. At a large time ˆt, the analytical solution oscillates about an asymptotic value hˆ asym 5 20.246 t b/M, where M5

125 1 v t 2116 1 v t 219 1 v t 214 1 v t 2 1

2

2

3 (1 1 v 2t 2 )

1

2

2

1

for t . 3,

2

2

1

2

2

hˆ asym ø 0.246t b.

Thus, changes in the intrinsic time constant of the system t and the consequences of ET, through the forcing amplitude b, are largely equivalent. The unrealistic negative asymptotic value results from our definition of hˆ , which carries two embedded parameters, a and f [see Eqs. (A3) and (A4)]. REFERENCES Bond, B. J., J. A. Jones, G. Moore, N. Phillips, D. Post, and J. J. McDonnell, 2002: The zone of vegetation influence on baseflow revealed by diel patterns of streamflow and vegetation water use in a headwater basin. Hydrol. Processes, 16, 1671–1677. Bourque, C. P., and J. H. Pomeroy, 2001: Effects of forest harvesting on summer stream temperatures in New Brunswick, Canada: An inter-catchment, multiple-year comparison. Hydrol. Earth Syst. Sci., 5, 599–613. Bren, L. J., 1997: Effects of slope vegetation removal on the diurnal variations of a small mountain stream. Water Resour. Res., 33, 321–331. Cayan, D. R., S. A. Kemmerdiener, M. D. Dettinger, J. M. Caprio, and D. H. Peterson, 2001: Changes in the onset of spring in the western United States. Bull. Amer. Meteor. Soc., 82, 399–415. Constantz, J., 1998: Interaction between stream temperature, streamflow, and groundwater exchanges in alpine streams. Water Resour. Res., 34, 1609–1615. ——, C. L. Thomas, and G. Zellweger, 1994: Influence of diurnal variations in stream temperature on streamflow loss and groundwater recharge. Water Resour. Res., 30, 3253–3264. Crave, A., and P. Davy, 1997: Scaling relationships of channel networks at large scales: Examples from two large-magnitude watersheds in Brittany, France. Tectonophysics, 269, 91–111. Czikowsky, M. J., 2003: Seasonal and successional effects on evapotranspiration and streamflow. M.S. thesis, Dept. of Earth and Atmospheric Sciences, The University at Albany, State University of New York, 105 pp. [Available from M. Czikowsky in electronic form by request.] Dawson, T. E., and J. R. Ehleringer, 1991: Streamside trees that do not use stream water. Nature, 350, 335–337. Dingman, S. L., 2002: Physical Hydrology. Prentice-Hall, 646 pp. Dodds, P. S., and D. H. Rothman, 2000: Scaling, universality, and geomorphology. Annu. Rev. Earth Planet. Sci., 28, 571–610. Doyle, P. F., 1991: Documented autumnal streamflow increase without measurable precipitation. Water Resour. Bull., 27, 915–923. Eagleson, P. S., 1970: Dynamic Hydrology. McGraw-Hill, 462 pp. EarthInfo, Inc., 2001: 2001 NCDC summary of the day, east, Version 2.3 for Windows. EarthInfo, Inc. Federer, C. A., 1973: Forest transpiration greatly speeds streamflow recession. Water Resour. Res., 9, 1599–1604. Finkelstein, P. L., and L. E. Truppi, 1991: Spatial distribution of precipitation seasonality in the United States. J. Climate, 4, 373– 385. Fitzjarrald, D. R., O. C. Acevedo, and K. E. Moore, 2001: Climatic consequences of leaf presence in the eastern United States. J. Climate, 14, 598–613.

VOLUME 5

Freedman, J. M., and D. R. Fitzjarrald, 2001: Postfrontal airmass modification. J. Hydrometeor., 2, 419–437. Goodrich, D. C., and Coauthors, 2000: Seasonal estimates of riparian evapotranspiration using remote and in situ measurements. Agric. For. Meteor., 105, 281–309. Hansen, W. F., 2001: Identifying stream types and management implications. For. Ecol. Manage., 143, 39–46. Hopkins, A. D., 1918: Periodical events and natural law as guides to agricultural research and practice. Mon. Wea. Rev., 46 (Suppl. 9), 1–42. Huntington, T. G., 2003: Climate warming could reduce runoff significantly in New England, USA. Agric. For. Meteor., 117, 193– 201. Jenkins, J. P., B. H. Braswell, S. E. Frolking, and J. D. Aber, 2002: Detecting and predicting spatial and interannual patterns of temperate forest springtime phenology in the eastern U.S. Geophys. Res. Lett., 29, 2201, doi:10.1029/2001GL4008. Lundquist, J. D., and D. R. Cayan, 2002: Seasonal and spatial patterns in diurnal cycles in streamflow in the western United States. J. Hydrometeor., 3, 591–603. Meyboom, P., 1967: Groundwater studies in the Assiniboine River drainage basin. Part II: Hydrologic characteristics of phreatophytic vegetation in south-central Saskatchewan. Bulletin 139, Geological Survey of Canada, 62 pp. Moore, K. E., D. R. Fitzjarrald, R. K. Sakai, M. L. Goulden, J. W. Munger, and S. C. Wofsy, 1996: Seasonal variation in radiative and turbulent exchange at a deciduous forest in central Massachusetts. J. Appl. Meteor., 35, 122–134. Myneni, R. B., C. D. Keeling, C. J. Tucker, G. Asrar, and R. R. Nemani, 1997: Increased plant growth in the northern high latitudes from 1981 to 1991. Nature, 386, 698–702. Nyholm, T., K. R. Rasmussen, and S. Christensen, 2003: Estimation of stream flow depletion and uncertainty from discharge measurements in a small alluvial stream. J. Hydrol., 274, 129–144. Paybins, K. S., 2003: Flow origin, drainage area, and hydrologic characteristics for headwater streams in the mountaintop coal-mining region of southern West Virginia, 2000–01. U.S. Geological Survey Water-Resources Investigations Rep. 02-4300, 24 pp. Rosso, R., and B. Bacchi, 1991: Fractal relation of mainstream length to catchment area in river networks. Water Resour. Res., 27, 381–387. Sakai, R. K., D. R. Fitzjarrald, and K. E. Moore, 1997: Detecting leaf area and surface resistance during transition seasons. Agric. For. Meteor., 84, 273–284. Schwartz, M. D., 1996: Examining the spring discontinuity in daily temperature ranges. J. Climate, 9, 803–808. ——, and B. E. Reiter, 2000: Changes in North American spring. Int. J. Climatol., 20, 929–932. Sun, G., S. G. McNulty, D. M. Amatya, R. W. Skaggs, L. W. Swift Jr., J. P. Shepard, and H. Riekerk, 2002: A comparison of the watershed hydrology of coastal forested wetlands and the mountainous uplands in the southern U.S. J. Hydrol., 263, 92–104. Tallaksen, L. M., 1995: A review of baseflow recession analysis. J. Hydrol., 165, 349–370. Troxell, H. C., 1936: The diurnal fluctuation in the ground-water and flow of the Santa Ana River and its meaning. Trans. Amer. Geophys. Union, 17, 496–504. Walsh, J. E., M. B. Richman, and D. W. Allen, 1982: Spatial coherence of monthly precipitation in the United States. Mon. Wea. Rev., 110, 272–286. White, M. A., P. E. Thornton, and S. W. Running, 1997: A continental phenology model for monitoring vegetation responses to interannual climatic variability. Global Biogeochem. Cycles, 11, 217– 234. White, W. N., 1932: A method of estimating ground-water supplies based on discharge by plants and evaporation from soil. U.S. Geological Survey Water Supply Paper 659-A, 105 pp. Wittenberg, H., 2003: Effects of season and man-made changes on baseflow and flow recession: Case studies. Hydrol. Processes, 17, 2113–2123.