[Next Section] [Previous Section] [Table of Contents]

Methods

The minimal criteria for inclusion of urbanizing basins in this study were a flow record of more than 50 years, location near an urban or suburban area, and relatively minimal disruption of the natural flow by diversion or impoundment. The 50 year requirement was designed to ensure adequate years of pre- and post-urbanization records. The Tualatin and Newaukum Rivers and Johnson Creek were the only streams in Oregon and Washington found to fit these criteria.

In addition, a nonurbanized basin was needed as a reference to compare to urbanizing basins. The requirements for the reference basin were that it be in western Oregon or Washington, have no large city or metropolitan suburban area, have flow records of at least 50 years and have minimal disruption of the natural flow by diversion or impoundment. The Luckiamute River, gaged near Suver, was chosen as a nonurbanized reference basin. It meets the above criteria and is part of the lower Willamette basin, as are the other two Oregon streams.

Flow records were scrutinized for peak events higher than a 1 year recurrence interval (RI), defined from USGS data (Wellman and others, 1993; Williams and others, 1984). For each basin, patterns of peak flow rate (Q_p), partial event volume (Q_v) and Q_p/Q_v through time were analyzed. In addition, these variables were compared to periods of 15, 30, 60, 90 and 120 days of antecedent precipitation. The data sets for each basin are in Appendix A.

Measuring peak flows is one way of measuring changes in river flow. Peak flow is the highest flow of a period. Peak flows (Q_p) are a rate (m³/s), expressing a volume of water (m³) per unit of time (s) passing the gage. An increase in Q_p may indicate an increased flashiness of a stream. Because Q_p depends in part on the size of the drainage basin, it cannot be easily compared from stream to stream. To eliminate the effect of size, the flow rate can be divided by the area of the basin (km²). The resulting unit of measure, m³/s/km², was used in this study so that values are size independent.

Two main flow variables were used in this study, Q_p and Q_v. For the Luckiamute and Newaukum Rivers and Johnson Creek, the published instantaneous Q_p for each event was used as the peak value. However, for the Tualatin basin, only one instantaneous peak per year was published, no matter how many peak events occurred. When the published Tualatin instantaneous Q_p were compared to the mean daily flows of the same date, the daily mean flow averaged 4% less than the peak flow. Because this is within the 5% margin of error of the mean flow data, the mean data are an appropriate approximation to instantaneous peaks for the Tualatin basin (Williams and others, 1984).

Streamflow includes both surface runoff and baseflow (Figure 14). Baseflow is water that enters the stream from long term sources such as groundwater, rather than precipitation, and constitutes most of the streamflow found between precipitation events. Calculation of baseflow in peak events was not attempted. Laenen (1980) found that less than 10% of the total flow in peaks was baseflow and was not able to determine a well-defined relationship between baseflow and either basin parameters or antecedent conditions. Baseflow is often neglected in urban drainage because it is such a small percentage of total flow (Bedient and Huber, 1992).

An increase in volume can also indicate an increase in stream flashiness, since it indicates that more water is moving through the basin more quickly. However, the Q_v of a peak event is dependent on the definition of event duration. The same peak event can be viewed in a number of periods: a single day, several days, a week, or longer. The longer the time frame, the more individual differences among events are obscured. Conversely, if the defined event period is too short, there is little difference between Q_p and Q_v.

Potter (1991) used a three day summation of peak flows as a value for Q_v. The three days Potter used in his summation were the day with the peak flow and the day immediately preceding and following that day. By carefully defining the event period, a value for Q_v that is relatively objective can be obtained for events before and after potential urbanization.

For this study, the beginning of each event was defined as the first rise in peak flow. Defining the end of the event was more difficult. Because of the nearly daily rainfall during much of the winter in the Pacific Northwest, river flows in the study basins sometimes do not return to an event starting level for several days or weeks, even though several peak events might happen in the intervening period (Laenen, 1980). The definition of event duration is crucial in determining Q_v, so an existing baseflow separation method was used to help define a nonarbritary event duration in this study. There is, however, no single method of defining baseflow. Bedient and Huber (1992) call baseflow separation "more an art than a science," and simply counsel consistency in method. Dawdy and others (1972) defined baseflow as the area under a line projected from the amount of flow at the start of the peak event. When flow returned to that amount, the event was over. Dingman (1994) explains three methods.

One baseflow separation equation uses the size of the basin to determine the duration of an event (Dingman, 1994; McCuen, 1998). This equation, used by Clement (1984) to determine baseflow in his study of the Johnson Creek basin, was used in the present study to define the last day of the event:

N = A^0.2,

where N = number of days after peak and A = drainage area of gaged basin in square miles (Table 9).

Table 9. Periods used for partial event volume.

The partial event volume (Q_v) then, is the sum of mean daily peak flows from the event:

N

Q_v = å Q_p(i)

i

where Q_p(i) = peak flow in m³/s/km² for day i, and N = days of the event.

A ratio of Q_p/Q_v was calculated for each peak event as a measure of basin response. As the ratio of Q_p/Q_v increases, the basin is producing higher peaks in relation to similar volumes, an indication of flashiness.

The variables of Q_p and Q_v, and Q_p/Q_v were computed for each of the four streams for the duration of their flow records (Table 10). Then the events were categorized by season. Events over a 1 year RI were found only in the months November through April. Seasonal categories were defined as November-December, January-February, and March-April. Events were then categorized by size, based on recurrence intervals (RI). Size categories were defined as 1-1.9 year RI, 2-10 year RI, and events over 10 year RI.

Table 10. Number of events in each basin in period of record, by season (months), and by size (recurrence interval).

The same categories were compared to antecedent precipitation. Because precipitation records were only available beginning in 1948, there were smaller data sets for each basin (Table 11).

Table 11. Number of events in each basin in period of record coincident with precipitation records, by season (months), and by size (recurrence interval).

All tests were performed for significance at the 95% confidence level (a =0.05). For each basin, the data set of peak flow events over a 1 year RI was somewhat skewed. To compensate for this tendency, all peak and volume data were log transformed before statistical tests were performed (Figure 15). Log transformation creates a data set that more closely satisfies the requirement of many statistical tests for a normal distribution of data (Tabachnick and Fidell, 1989).

The data set for each basin was broken into two periods, early and late. Because the process of urbanization is gradual, there was no predetermined date used for when a basin became urbanized. The most effective contrast between the two periods was to have as long a period as possible between them (Ferguson and Suckling, 1990). To get the largest possible contrast in levels of development, but to still have large enough sample sizes to be valid, the early period for each basin was defined as the first 30 events, and the late period as the last 30 events. Because each basin has a different period of record and a different number of events, the dates for these periods varied for each basin (Table 12).

Table 12. Periods used for statistical tests on variables Q_p, Q_v, and Q_p/Q_v.

Because precipitation data for each basin were available for different dates than flow data, all tests involving APIs had to use different testing periods (Table 13).

Table 13. Periods used for statistical tests involving antecedent precipitation.

The first testing technique used was linear regression on values for peak flows (Q_p), partial event volumes (Q_v), and ratio of peak to volume (Q_p/Q_v) of each basin to determine the correlation coefficient (r) of these variables to the passage of time (Davis, 1986). The equation was:

r = Ö (SS_R/SS_T)

where SS_R = sum of squares due to regression and SS_T = total sum of squares. Although there are natural fluctuations in rainfall, through a 50 year period flow and volume rates should remain relatively constant, thus showing little or no correlation with the passage of time.

The significance of r was tested (Davis, 1986). The null hypothesis was that the two variables are independent. The equation was:

t = (rÖ (n-2))/(Ö 1-r²)

where r = the correlation coefficient and n = the number of data points.

Then F and t tests were performed on the data sets of each basin. If samples are normally distributed, each sample can be described by its mean and variance. Differences in these values represent samples that are significantly different from each other (Davis, 1986).

F tests determine the equality of variance; the null hypothesis was that the variances of two sample periods were equal. The equation was:

F = (s²_1/s²₂)

where s²₁ was the larger variance and s²₂ was the smaller variance.

If the F test did not cause the null hypothesis of equality of variance to be rejected, a t test was done to determine the equality of mean. The null hypothesis was that the means of two sample periods were equal (Davis, 1986). The equation was:

_ _

t = (X₁ – X₂)/s_e

_ _

where X₁ = average mean of sample 1, X₂ = average mean of sample 2, and
s_e = standard error of the mean.

The fourth technique was a one way ANOVA. The null hypothesis was that the means of two sample periods were equal (Davis, 1986). The equation was:

F =(MS_A/MS_W)

Where MS_A was the variance among samples and MS_W was the variance within samples.

Each of the above tests requires each sample point to be independent. Although these tests are commonly used in hydrological studies, it is not clear that flow data fit this requirement. For example, a large flow volume may leave such a high water table that even small precipitation events cause higher peak flows than normal, thus the events are not independent. A nonparametric test is acceptable for data that does not have independent events, and is common in peak flow analysis (Barringer and others, 1994, Lazaro, 1976). Clement (1984) used nonparametric tests in his study of the Johnson Creek basin. For this reason, the nonparametric Spearman-Conley test was performed (McCuen, 1998). A nonparametric test, like Spearman-Conley, assigns a rank to each event, and then tests the pattern of rank. Because it tests a second order value rather than the initial values themselves, it is generally regarded to be less robust than parametric tests (Gibbons, 1971).

The null hypothesis for the Spearman-Conley test was that adjacent values of peak flows were not correlated. If significant urbanization had occurred, peaks would increase in the later period, which would produce a positive correlation coefficient. The test equation was:

n

R_sc = 1 - ((6S d_i²)/(n³ - n))

i=1

where d is the rank of each event and n is the number of pairs, which is one less than the number of peak events.

Tabachnick and Fidell (1989) recommend a sample size producing at least 20 degrees of freedom for error in a univariate ANOVA to be robust to violations of normality. Sample sizes of 30 events in the early period and 30 events in the late period were chosen in this study, as a conservative estimator of behavior. However, when looking at the subsets of season and size, sample sizes were smaller. Only subsets with at least 10 events in the early period and 10 events in the late period are reported.

Daily rainfall totals were collected for each basin (WRCC, 1997). F tests, t tests, and ANOVAs were conducted on precipitation records to see if there was a significant difference between the early and late periods. Because of the importance of precipitation in creating stream flow, this was a key step to eliminate differences in precipitation between the two periods as the determining factor for differences in peak flows.

Antecedent precipitation indices (API) were also prepared for each basin: 15, 30, 60, 90, and 120 days, each measured backward from the date of Q_p. The equation was:

n

API = å P_i

i=1

where n = number of days and P_i = precipitation on day i.

These precipitation indices are gross indicators of soil moisture. Higher previous rainfall in the short or long term increases soil moisture and leaves less room for infiltration of the current precipitation. By using a combination of 15- to 120-day indices, effects from precipitation over a variety of periods can be measured.

The importance of APIs may be inferred from the fact that during the approximately 50 years of records for four river basins, not a single peak flow higher than a 1 year recurrence interval was seen during June through September. In western Oregon and Washington, 80 percent of the precipitation falls between October and April. Thus, a storm in June, past the end of the rainy season, does not produce flood conditions, because the soil has larger infiltration capacity. A similar storm in February, when the soil is close to saturation from several months of rain, may produce a quite different result.

[Next Section] [Previous Section] [Table of Contents]