What Is Relatively Permanent? Flow Regimes of Arizona Streams within the Context of the 2023 Conforming Rule on the Revised Definition of “Waters of the United States”

Abstract

The classification of stream flow regimes has been a subject of study for over a half century in the fields of hydrology, geomorphology, ecology, and water resources management. But with the most recent Supreme Court decision on jurisdictional Waters of the United States (WOTUS) and the 2023 Conforming Rule, the answer to the question of which waters are relatively permanent has increased in importance and urgency. One state where this question is salient is Arizona, where approximately 95% of its streams are nonperennial. In this study, we use long-term (>30 years) daily discharge records from Arizona to assess semi-natural flow regimes of arid streams within the context of the 2023 Conforming Rule. Using flow percentile distributions, we distinguished flow permanency—ephemeral vs. intermittent vs. perennial—for 70 stream reaches distributed throughout the state. Ephemeral streams had a median flow of 0 cms and a 75th percentile flow permanence less than 25% (i.e., less than 3 months of flow for every 7.5 out of 10 years). On the other end of the spectrum, perennial streams had a 90th percentile flow permanence of 100%. In the middle, intermittent streams had a 75th percentile flow permanence greater than 25% and a 90th percentile flow permanence less than 100%. We also assessed the effect of the recent megadrought (since 1994) on flow permanency. As a result of the megadrought, four perennial streams transitioned to intermittent, four intermittent streams transitioned to ephemeral, and one perennial stream became ephemeral. The flow classification we present here is specific to Arizona streams but could be useful to other arid regions seeking to answer the question of which streams are relatively permanent in a typical year.

1. Introduction

1.1. Purpose and Regulatory Context

On 8 September 2023, the Environmental Protection Agency (EPA) and the U.S. Army Corps of Engineers (Corps; collectively “the agencies”) published a Final Rule [1] that amended the definition of “Waters of the United States” (WOTUS) to conform to the U.S. Supreme Court’s (“the Court”) May 2023 decision in Sackett v. EPA and the Court’s interpretation of the Clean Water Act (CWA) [2]. In their consideration of the extent of the CWA’s geographical reach, the Court concluded that WOTUS encompasses “only those relatively permanent, standing or continuously flowing bodies of water ‘forming geographic[al] features’ that are described in ordinary parlance as ‘streams, oceans, rivers, and lakes.’” The Court also addressed the question: what are the “outer reaches” of the CWA? Most of their discussion focused on wetlands (which this article does not address); however, Justice Thomas, in his concurring opinion joined by Justice Gorsuch, referenced section 13 of the 1899 Rivers and Harbors Act and interpreted that a jurisdictional tributary (the focus of this article) “requires some form of surface water connection between a tributary and traditionally navigable waters” [2].

The major change to WOTUS in the Court’s May 2023 decision in Sackett v. EPA was the rejection of the “significant nexus” test, which the agencies had been using to assert jurisdiction over waterbodies that “significantly affect” the chemical, physical, or biological integrity of “traditional navigable waters” (i.e., (a)(1) waters). Accordingly, the agencies removed any mention of “significant nexus” from the revised Definition of Waters of the United States and their tributaries [1]. While the agencies revised the definition of jurisdictional wetlands to add “a continuous surface connection,” the same requirement was not added to the definition of a jurisdictional tributary. The revised definition of WOTUS establishes jurisdiction over “Tributaries of waters identified in paragraph (a)(1) or (2) of this section that are relatively permanent, standing or continuously flowing bodies of water” [1].

The January 2023 Final Rule (33 CFR 328) defined the ‘‘relatively permanent standard’’ as the test to identify relatively permanent, standing or continuously flowing waters connected to paragraph (a)(1) waters, and waters with a continuous surface connection to such relatively permanent waters or to traditional navigable waters, the territorial seas, or interstate waters. This definition does not appear in the September 2023 revised rule (or Conforming Rule), and thus moving forward, jurisdiction of tributaries will likely be based on how relatively permanent is interpreted.

In arid environments such as the southwestern United States, the interpretation of relatively permanent is particularly complex. The definition of WOTUS states that “Swales and erosional features (e.g., gullies, small washes) characterized by low volume, infrequent, or short duration flow” are not jurisdictional waters. However, the terms “low”, “infrequent”, and “short duration” are also relative terms (i.e., not absolute or independent). Merriam-Webster defines relative as “expressed as the ratio of the specified quantity to the total magnitude or to the mean of quantities involved” [3]. The quantity in question here is streamflow, or water discharge in units of cubic meters per second (cms).

In the May 2023 Sackett v. EPA decision, the Court reaffirmed the States’ traditional sovereignty over their intrastate surface waters. With the 2023 Conforming Rule limiting federal jurisdiction over certain waters, it is likely that future regulations of intrastate waters by the State will become more important, especially in arid states where relatively few waterbodies fall under CWA jurisdiction. In this article, we use one arid state—Arizona—as a case study to examine streamflow patterns over space and time within the context of the relatively permanent standard.

1.2. Overview of Nonperennial Streams in Arizona and the Broader Southwest

There are two types of nonperennial streams: ephemeral and intermittent. Ephemeral streams are those that flow briefly (typically hours to days) in direct response to precipitation [4]. According to the United States Geological Survey (USGS), an ephemeral stream channel is “a channel in which streamflow occurs inconsistently or infrequently and, except during periods of streamflow, is directly underlain by unsaturated alluvium or rock; ephemeral stream channels are most common in arid and semiarid regions and typically have a rectangular to steeply sided trapezoidal cross section, banks a meter or more in height formed of fine-grained, poorly consolidated over-bank sediment, and a nearly flat, sandy bed. Synonyms are dry wash, arroyo” [5]. Intermittent streams are those that flow weekly to seasonally during extended wet periods [4]. According to the USGS, “the flow of an intermittent stream typically is derived from wet-season runoff or snowmelt, and the surface of an intermittent stream, or the bed of the channel upon which flow occurs, typically is higher than the level of the zone of saturation in the adjacent water-bearing alluvium or rocks. This characteristic is fundamentally different from that of an ephemeral-stream channel, which at most times is separated from the zone of saturation by a variable thickness of unsaturated alluvium or rock” [5].

Nonperennial streams make up the majority of streams in the Southwest. Using the National Hydrography Dataset (NHD), Levick et al. reported that ephemeral and intermittent streams make up over 81% of all streams in the Southwest (Arizona, New Mexico, Nevada, Utah, Colorado, and California) [6]. Arizona has the highest percentage of nonperennial streams, with reports ranging from 94% [4] to 96% [7]. In the arid states with minimal snow accumulations—Arizona, Nevada, and New Mexico—the vast majority (~90%) of these nonperennial stream channels are ephemeral [6]. All of these percentages are likely higher due to the NHD not capturing all nonperennial streams [8,9].

In the field, nonperennial stream channels can be difficult to identify [10]. Historical indicators that have been used include discernible bed and banks and an ordinary high-water mark (OHWM), which was defined in the 1986 WOTUS regulations (33 CFR § 328.3) as “that line on the shore established by the fluctuations of water and indicated by physical characteristics such as clear, natural line impressed on the bank, shelving, changes in the character of the soil, destruction of terrestrial vegetation, the presence of litter and debris, or other appropriate means that consider the characteristics of the surrounding areas” [10]. As explained by Lichvar and Wakeley, the use of OHWM in the Southwest is problematic because arid channels are shaped by rare events and are not reshaped by subsequent smaller events [10]. Thus, the morphological features of southwestern landscapes inevitably represent rare, large, unordinary conditions. Or, as stated by the Corps, “Given that arid-region rivers generally respond to large floods by dramatically widening their banks, the limits of the geomorphically effective event will likely be much more extensive than the limits of a low-flow channel inset into a compound channel. Consequently, if the OHWM is set at the outer limits of the extreme event, the designated “waters” will encompass a much greater area than is occupied by more ordinary flows” [10] (p. 74). Given the limitations of this measure in arid landscapes as a snapshot feature, OHWM is not reliable by itself to establish the boundaries of jurisdictional tributaries.

Riparian vegetation can be a useful biological indicator in distinguishing flow regimes of stream channels [10,11]. Hydroriparian plant communities are characteristic of perennial reaches, Mesoriparian plant communities are characteristic of intermittent reaches, and Xeroriparian plant communities are characteristic of ephemeral reaches. However, upland species are often found along arid streams, and different vegetation mixtures/densities can be difficult to interpret. Further complicating the use of riparian vegetation as an indicator of flow regime are interdependent factors such as elevation, topography, geology, land use, and anthropogenic disturbance [11]. There is not yet a completely reliable indicator of stream channel presence or flow permanence other than streamflow data.

Three recent studies have conducted regional hydrological characterizations of gaged streams in the Southwest [12,13,14]. All three studies investigated the range of hydrologic conditions in their study areas and developed flow regime clusters/classes. The clusters/classes represent different responses to precipitation events and interactions with topography, geology, and land cover. Of these three studies, only Merritt et al. [14] did a statewide analysis of Arizona and thus is the focus of our summary here and later comparisons. They and others found that most ephemeral streams in Arizona and the rest of the Southwest had flow for only a short period following intense storms, like those associated with the summer monsoon [14,15]. Because precipitation in the Southwest is highly variable over both space and time (and usually localized, covering small areas), ephemeral streams have wide variability in flow patterns, with some streams exhibiting no surface flow for numerous years [14,16]. Intermittent streams were more difficult to classify, characterize, and separate from ephemeral streams because of the high variability in the annual number of zero-flow days and complex interactions between precipitation and landscape characteristics [14]. Nevertheless, Merritt et al. developed a flow regime classification and a modeled map (Figure 1) for streams in Arizona. Note that instead of using “intermittent” classes, they use “seasonal” or “weak seasonal” classes. Also noteworthy is that they acknowledge: “It appears that we can predict the ends of the ephemeral to perennial continuum reasonably well, but our models had difficulty distinguishing reaches that are more variable in the number of zero-flow days from year to year”.

If the classification, mapping, and characterization of intermittent streams is difficult, the determination of “relatively permanent” will be even more difficult. With the significant nexus test removed from the 2023 Conforming Rule, most—if not all—ephemeral streams will not be considered as WOTUS. The jurisdictional inclusion of intermittent streams will be nuanced and depend on guidance and approaches taken by various Corps Districts or states. The relative permanency of an intermittent stream will likely be considered within the context of regional precipitation patterns.

1.3. Precipitation Patterns in Arizona

The state of Arizona is situated in the southwestern United States (USA) at the intersection of North America’s four major deserts. This intersection is characterized by an arid climate, but also a diverse physical geography with different types of stream channels and flow regimes. Flow regimes in Arizona’s nonperennial streams are largely dictated by precipitation [14], and thus it is important to first understand these precipitation patterns.

Most areas in Arizona have a bimodal distribution of precipitation where precipitation is higher during the winter via frontal storms from the west and higher during the late summer via the North American Monsoon (NAM) from the south [17]. The NAM is responsible for most of the state’s annual precipitation, up to 75% in some years. NAM events are sporadic and intense, resulting in flashy streamflows with short duration. According to the National Weather Service [18], monsoon season officially occurs from 15 June to 30 September, which also marks the end of the hydrologic year. In northern Arizona, monsoon effects are usually not observed until mid-July. NAM precipitation tends to increase with topographic relief (Figure 2 and Figure 3). High relief features like the Mogollon Rim trigger thunderstorms during the monsoon season. Accordingly, many headwater streams with more reliable flows can be found along the Mogollon Rim and in the Central Arizona Highlands. Monsoon rainfall in Arizona has considerable spatial and temporal variability. A comparison between Flagstaff and Prescott—two cities in central Arizona only 100 km apart—provides a useful example [18]. These cities, respectively, received 277 and 340 mm of monsoon rainfall in 2021, but only 46 and 58 mm in 2020.

As one moves east and north in the state (and especially in the mountains), winter precipitation via frontal storms increases (Figure 2 and Figure 3) [17]. At higher elevations and below freezing temperatures, this winter precipitation contributes to snowpack, which may sustain springtime flows. Whether initiated by snow or by bands of rain, winter and spring streamflows tend to be less flashy than NAM streamflows. Frontal storms typically start arriving in mid-November and can occur through April; however, El Nino-Southern Oscillation (ENSO) cycles can have considerable impacts on this phenomenon. Indeed, ENSO is largely responsible for the high degree of inter-annual variation in Arizona precipitation [17]. El Nino phases typically increase low-intensity winter precipitation, and can also cause winter floods (e.g., 1993). La Nina leads to drier winters and can result in prolonged droughts (e.g., 1998–2001, 2020–2022).

Short-term droughts lasting months and long-term droughts lasting years are both common in Arizona [19,20]. Notable long-term droughts documented with meteorological data include 1898–1904, 1942–1956, and 1994–present. The most recent long-term drought started in 1994 with no end in sight [21], making it an exceptional megadrought that could last decades to centuries [22,23]. In sum, variations in precipitation over space and time are the primary influence on the flow regimes of Arizona’s nonperennial streams. Most Arizona streams exhibit (1) extreme intra-annual variability from the state’s bimodal precipitation patterns and short-term droughts, and (2) extreme inter-annual variability derived from ENSO cycles and long-term droughts.

2. Materials and Methods

2.1. Arizona Streamflow Data

There are various sources of streamflow data for Arizona. Cities like Flagstaff operate a small network of gages for early-flood warning. Multiple counties also have a network of early-flood warning gage sites, with Maricopa County (n = 217) being the largest. Pima County and Pinal County also have large networks, with about 50 gages each. While historical data exists for some cities and counties, very few sites had continuous long-term data. After exhaustive data collection and analyses, we found only 13 sites with at least 30 years of data; however, there were data reliability issues such as unrealistic repeating flow patterns and missing flow events for almost all the sites. Thus, we did not use any city or county gages for our analysis, save the one exception mentioned below.

The longest, most reliable, and most extensive streamflow data for Arizona is maintained by the United States Geological Survey (USGS). The USGS has streamflow data for 281 gaged sites in Arizona; 235 of these gages were active as of December 2023. Of all USGS sites in Arizona, only 84 gages were stream channels (i.e., not canals or diversions) with at least 30 years of daily discharge data. This 30-year minimum is important because it captures inter-decadal patterns and is the minimum that USGS uses for reporting temporal statistical distributions of streamflow. Fourteen of these gages were located immediately downstream of dams on the Colorado, Gila, Verde, Salt, and Bill Williams Rivers. These 14 dam-influenced stream gages were not assessed further, leaving 70 stream gages that we analyzed for semi-natural flow regimes (Table S1). One of the gages (USGS 9488650) was replaced with a redundant gage (MC 74507) at the same location by Maricopa County Flood District in 1990, and thus this record contains values from both agencies. The 6 years of data overlap, and consistency between datasets ensured confidence in using this aggregated dataset.

2.2. Flow Regime Variables and Analyses

The natural flow regime of a stream channel is defined by five attributes: magnitude, frequency, duration, timing, and rate of change [24]. Dhungel et al. [25] identified 16 flow variables (or metrics) that broadly characterize these five attributes. To make this classification more applicable to arid regions of the USA, Merritt et al. [14] used nine of Dhungel’s metrics and added three new ones to measure different aspects of zero-flow conditions. For this study, we selected metrics similar to Dhungel and Merritt but added variables that are particularly relevant to the flow regimes of arid streams and jurisdictional standards of the 2023 Conforming Rule. We used the Indicators of Hydrologic Alteration (IHA) software program [26] (v7.1) to calculate 12 flow regime variables (

Table 1. Flow regime variables relevant to arid streams and jurisdictional standards of the 2023 Conforming Rule. Year refers to Hydrologic Year (1 October–30 September), but Julian date starts on 1 January.

Flow Regime Variable	Definition
Mean annual Q (cms)	Average discharge (Q) of the river as measured by the sum of all individual daily flows divided by the number of days for the entire record. This value is sensitive to large floods. Its annual coefficient of variation (unitless) is a measure of flow variability, calculated as the standard deviation of all the daily flows divided by the mean annual flow.
Median Q (cms)	The average flow across the entire record as measured by the 50th percentile of ranked daily flows. This metric is not sensitive to large floods and thus is a better measure of the typical flow in a stream compared to mean annual Q. Median Q was also calculated at monthly and yearly intervals.
Intra-annual variability (%)	Percent of flow variability within an average year as measured by the difference between the maximum median monthly flow and the minimum median monthly flow, and then divided by the median Q.
Inter-annual variability (%)	Percent of flow variability across years as measured by the range (maximum minus minimum) of annual median flows divided by the median Q of the entire record.
Hydrologic reversals (annual count)	The number of times per year that flow switches from one type of period to another; also a measure of the number of changes between rising and falling periods.
Zero-flow days (annual count)	The average annual duration of a dry stream channel, reported as both mean and median.
Annual flow permanence (%)	Percent of the year there is flow in the channel, as measured by the annual distribution across the entire record: 10th, 25th, 50th, 75th, and 90th percentiles.
90-day minimum Q (cms)	Magnitude and duration of annual extreme low flow conditions, as measured by the annual distribution across the entire record.
90-day maximum Q (cms)	Magnitude and duration of annual extreme high flow conditions, as measured by the annual distribution across the entire record.
High-flow Q (cms)	Median number of ecologically significant high-flow pulses per year where flows rise above the low-flow channel but remain within channel banks (i.e., lower than a small flood). These typically occur during rainstorms or brief periods of snowmelt. IHA defines a high-flow pulse as any flow event that exceeds the 75th percentile of daily flows. The frequency (number of events/year) and duration (days) of high-flow Q were also measured.
Small-flood Q (cms)	The flow equivalent to bankfull discharge, which IHA defines as the discharge with a 2-year return interval. These floods are assumed to perform multiple biogeochemical functions. The duration (days) and timing (Julian date) of small-flood Q were also measured.
Large-flood Q (cms)	The flow equivalent to an extreme flood that will typically rearrange both the biological and physical structure of a river and its floodplain, which IHA defines as the discharge with a 10-year return interval. The duration (days) and timing (Julian date) of large-flood Q were also measured.

).

We also used IHA to assess the effects of the most recent megadrought (1994–current) by comparing flow regimes before and after 1994. Our record length requirement for this impact analysis was at least 15 years of flow data before 1994 and at least 15 years of flow data after 1994. The pre-impact period never started before 1957 on account of the 1942–1956 long-term drought. For this pre-drought vs. post-drought analysis, we focused on relative flow permanence and thus compared median flows, 90-day minimum flows, annual zero-flow days, and flow permanence percentile distributions. Given the common occurrence of zero median values and extreme outliers, we used Mood’s median test (nonparametric Chi-square) to test if flow distributions were statistically different between pre-drought and post-drought. All data distributions, analyses, and statistical tests were performed in JMP Pro (v17.2) Statistical Discovery software [27].

3. Results

3.1. Geography of Long-Term Stream Gages in Arizona

Most of the minimally disturbed 70 long-term USGS gages in Arizona were located along relatively large rivers such as the Little Colorado, Gila, Salt, and Verde Rivers (Figure 4). No gages on the Colorado River were analyzed because they were all dam-influenced. According to the high-resolution National Hydrograph Dataset Plus (NHDPlus-HR; published 30 August 2022), 37 of the 70 gaged stream channels were categorized as perennial (Table S1). Twenty-nine were categorized as intermittent, and only four were categorized as ephemeral. Using the NHDPlus-HR Strahler stream order as a metric for relative magnitude, 32 of the USGS gages monitored large stream channels (stream order: 7+) and 37 monitored medium stream channels (stream order: 4–6). Only one small stream channel (stream order: 1–3) in Arizona had a long-term USGS stream gage, the 2nd-order South Fork Parker Creek near Roosevelt (USGS 9498503). Contributing drainage areas for the 70 USGS gages range from 2.8 (South Fork Parker Creek) to 67,576 km², with the latter being the Little Colorado River near Cameron (9402000). The 25th, 50th (median), and 75th percentiles for contributing drainage area were 515, 1624, and 5596 km², respectively.

The distribution of USGS gages followed three other patterns (Figure 4). First, there were numerous gages around the major cities of Phoenix and Tucson. Second, there were many gages upstream of major reservoirs, likely to monitor inputs. Third, there was a large concentration of gages on streams draining the Central Arizona Highlands. All major ecoregions (Omernik Level 3) had at least two long-term gages except the Chihuahuan Desert, which is a relatively small area in the southeastern corner of the state. There was only one gage each in the northwestern and northeastern corners of the state; there were no gages in the southwestern corner of the state. By comparing Figure 4 to Figure 3, one can see that the driest parts of Arizona are not well-monitored for streamflow.

3.2. Flow Regime Analyses

We analyzed the flow regimes of 70 USGS stream gages in Arizona with long-term daily discharge records (>30 years) that were not located immediately downstream of a dam. The end date of our analysis was 30 September 2023, also known as the 2023 Hydrologic Year. All but two gages were current up to this date. The length of discharge records ranged from 30 to 112 years, with a median of 59 years (Table S1). We first organized the results by the five flow regime attributes: magnitude, frequency, duration, timing, and rate of change. We then addressed the relative permanence of these flow regimes (Section 3.3). Finally, we assessed the effect of the megadrought (since 1994) on relative permanence (Section 3.4).

3.2.1. Magnitude

Median Q (i.e., average flow, 50th percentile) of the 70 gage sites ranged from 0 to 9 cms. Eighteen sites had a median Q of 0 cms, and 18 additional ones had a median Q less than 0.1 cms. Mean annual Q, which is biased towards large floods, ranged from 0.01 to 23.8 cms. The maximum annual coefficient of variation (CV) was 28.8 and was highest for ephemeral streams. As is expected in arid environments with high transmission losses, there was not a significant relationship between discharge per unit area and average flow (mean or median). Also as expected, there was a high degree of variability of median flows within a year and among years. Intra-annual variability ranged from 17 for the Verde River near Paulden (9503700) to >1000 for five sites: Little Colorado River near Cameron (9402000), Sycamore Creek near Fort McDowell (9510200), Wet Bottom Creek near Childs (9508300), Whitewater Draw near Douglas (9537500), and Granite Creek near Prescott (9503000). Inter-annual variability was even more extreme, ranging from 49 for Verde River near Paulden to >10,000 for Sycamore Creek near Fort McDowell. Streams with high intra-annual variability tended to have high inter-annual variability (r = 0.73). Flow magnitude variability was not associated with drainage area, but streams draining the Central Arizona Highlands had relatively high intra-annual and inter-annual variability.

Small-flood Q (2-year return interval) ranged from 0.5 to 408 cms. Large-flood Q (10-year return interval) ranged from 1.4 to 1702 cms. Contributing drainage area was correlated with small floods (r = 0.38) and large floods (r = 0.46). Relative to watershed size, creeks draining the Central Arizona Highlands had the largest flood magnitudes, with notable examples being Oak Creek near Cornville (9504500), Dry Beaver Creek near Rimrock (9505350), Tonto Creek near Roosevelt (9499000), East Verde River near Childs (9507980), West Clear Creek near Camp Verde (9505800), and Black River near Fort Apache (9490500).

3.2.2. Frequency

The frequency attribute refers to how often a flow above a given magnitude recurs over a specified time interval. Given their ecological significance, the number of high-flow pulses per year is a common measure of frequency. According to IHA parameters, the 11 gage sites that were dry at least 75% of the year did not have a high-flow Q. For the other 59 sites, high-flow Q events ranged from 1 to 16 events per year, with a mean and median of 6 events per year. The typical duration of a high-flow Q event was 2–4 days.

3.2.3. Duration

Duration can be considered for high flows (e.g., high-flow Q events in previous section) and low flows (e.g., zero-flow days). Twenty-two gages did not have any zero-flow days in their record. That is, there was always some flow in these perennial stream channels. For the other 48 gages, mean zero-flow days ranged from 0.1 to 359 days/year. This last gage site—Vekol Wash near Stanfield (9488650)—only had 119 flow events (>0.01 cms) over its entire 34-year record, with a mean duration of 1.5 days per flow event. Annual mean and median zero-flow days were similar for most gages, indicating a normal distribution over time. However, there were a dozen gages where the mean was considerably higher than the median zero-flow days, potentially indicating drying of the stream channels over time. This temporal change is investigated further in Section 3.4. There was a lot of variability in the inter-annual and inter-decadal distribution of extended periods of zero-flow days for some gages. Almost half of the gage sites (n = 33) had a 90-day minimum flow of 0 cms for every 1 out of 10 years (i.e., 10th percentile). Twelve of these gage sites had a 90-day minimum flow of 0 cms essentially every year (i.e., 90th percentile).

Analyses of high flow durations had some interesting results. The 90-day maximum flow was above 0 cms for almost all sites for almost all years; however, it was below 0.1 cms for 24 gage sites for every 1 out of 10 years. For seven of these sites, 90-day maximum flow was 0 cms, meaning some Arizona streams do not have appreciable flows in some years. Small-flood durations had a wide range. Fifteen of the gage sites had small floods with a mean duration of 1 week or less, which is typical of many eastern perennial rivers. For the other 55 gage sites, mean small flood durations ranged from 9.5 to 100 days. The mean duration of small floods for all sites was 33 days. The year-to-year (or decade-to-decade) variability in small flood durations was extreme for some sites. The coefficient of dispersion (CD) was as high as 5 for three sites (all three relatively large intermittent streams) and greater than 2 for 17 other sites. Large flood durations also displayed extreme behavior, lasting longer than 100 days on average for half of the sites (n = 35). Twelve of these sites had a median Q less than 0.1 cms, five with 0 cms.

3.2.4. Timing

Large floods occurred throughout the year, but the average timing was winter (December-February) for most sites (n = 42). Most of these streams drained the Central Arizona Highlands. The monsoon season of mid-July to September was another common period of large floods and was the average timing for 16 sites. Small-flood timing generally agreed with large-flood timing for each site (p = 0.003); however, small floods were more likely to occur during the monsoon season compared to large floods. Months with the highest median Q were March (52% of sites), August (14%), February (12%), and April (10%). Zero-flows or other minimum flows occurred in summer (June or July) for 90% of the gage sites.

3.2.5. Rate of Change

The rate of change refers to how quickly flow changes from one magnitude to another, also known as flashiness. A common metric for rate of change is the number of flow reversals (i.e., the number of times that flow switches from one type of period to another). In arid streams that stay dry much of the year, this metric can vary widely from year to year. The Santa Maria River near Bagdad (9424900), for example, had a median of 19 flow reversals per year but ranged from 0 to 71. For all 70 sites, the annual median of flow reversals ranged from 6 (Vekol Wash near Stanfield, 9488650) to 126 (Virgin River at Littlefield, 9415000). By nature of their weather and environment, most arid streams are flashy over long time scales. Thus, we report the temporal distribution (by ranked percentile) of flow reversals among years (Table S1). Ephemeral streams with a median Q of 0 cfs had relatively few reversals per year, 23 per year on average. Intermittent streams had 78 reversals per year, and perennial streams had 98 reversals per year, on average.

3.3. Relative Permanence

In addition to the five standard attributes of flow regimes, we assessed the relative permanence of streams by calculating the percent of time during the year there was flow in the channel. This metric does not characterize continuous flow, although most extended zero-flow periods for non-ephemeral streams occurred during the summer months of May, June, and July. Annual flow permanence is useful in identifying stream types based on flow (Figure 5). Of the 70 gage sites, there were 17 ephemeral streams, which we define here as having a median Q of 0 cms and an annual flow permanence < 25% at the 75th percentile (i.e., less than 3 months of flow for most of the record). On the other end of the spectrum, there were 27 perennial streams, which we define as having 100% annual flow permanence at the 90th percentile (i.e., flow year-round for at least every 9 out of 10 years of record). The other 26 gage sites were intermittent streams, which we define as having an annual flow permanence > 25% at the 75th percentile, and an annual flow permanence < 100% at the 90th percentile. The annual median and mean zero-flow days were less than 200 days for all the intermittent sites. For ephemeral streams, the mean zero-flow days per gage site ranged from 205 to 359 days/year, with a collective mean of 301 days/year for the 17 gage sites.

When the empirical record of flow permanence was compared with the modeled flow regime classification of Merritt et al. [14], there was agreement with more than half of the sites (38 of 70; Figure 1). Merritt’s map correctly modeled two-thirds (11 of 17) of the “Ephemeral and flashy” channels (A1). Six of the other seven ephemeral channels were modeled as “Weak seasonal-NP, Wet winter-spring” (A2), and one was modeled as “Aseasonal-NP” (B2a). Only two of the 27 perennial channels were correctly modeled as perennial (B1b) by Merritt. Most of the perennial channels were modeled as intermittent, either as A2 (4 of 27) or B2 (20 of 27). The one perennial stream that was modeled as ephemeral (A1) by Merritt et al. was Pantano Wash near Vail (9484600), which is fed by a rare system of perennial springs. All of the intermittent channels were correctly modeled by Merritt et al. as intermittent, either as A2 (10 of 26) or B2 (16 of 26).

The modeled map of Merritt et al. had a higher accuracy for nonperennial streams compared to the NHD Plus-HR hydrographic classification. NHDPlus-HR did not identify any of the ephemeral streams correctly, instead classifying 16 of them as intermittent and one as perennial. Only 9 of 26 intermittent streams were classified correctly by NHDPlus-HR, with 4 being classified as ephemeral and 13 being classified as perennial. With regards to perennial streams, NHDPlus-HR proved to be more accurate, classifying 23 of 27 correctly; the other 4 were classified as intermittent. Overall, NHDPlus-HR overestimated perennial streams in Arizona, while the modeled map of Merritt et al. overestimated intermittent streams at the expense of both perennial and ephemeral streams.

3.4. Megadrought Effects on Flow Regimes and Relative Permanence

Fifty-three of the 70 long-term USGS gage sites qualified for the drought impact analyses because they had at least 15 years of flow data before 1994 and at least 15 years of flow data after 1994, which marks the beginning of the most recent megadrought. Of these 53 sites, 8 were ephemeral streams, 21 were intermittent streams, and 24 were perennial streams, as identified in Section 3.3 for their entire record. The median flow and 90-day minimum flow for ephemeral streams were 0 cms for both pre-drought and post-drought periods; however, the annual zero-flow days increased by 11 days during the drought for ephemeral streams (

Table 2. Mega-drought impacts on flow regimes and flow permanence. The percent change for median flow and 90-day minimum flow was calculated for each gage and then averaged by stream type.

	Pre-Drought Median Flow (cms)	Post-Drought Median Flow (cms)	Change (%)	Pre-Drought Annual Zero-Flow Days	Post-Drought Annual Zero-Flow Days	Change (Days)	Pre-Drought 90-Day Minimum Flow (cms)	Post-Drought 90-Day Minimum Flow (cms)	Change (%)
Ephemeral n = 8	0	0	0	273	284	+11	0	0	0
Intermittent n = 21	0.72	0.48	−55.3	31	81	+50	0.29	0.14	−62.0
Perennial n = 24	1.99	1.54	−22.8	0	0	0	1.25	0.94	−23.3

). Although the median value of 0 cms did not change for any of the ephemeral streams, Mood’s median test revealed that the number of daily flow values above the median significantly decreased during the drought at 6 of 8 ephemeral sites. For two of the ephemeral sites—Rincon Creek (9485000) and New River near Rock Springs (9513780)—the 75th percentile of flow permanence decreased from >25% before the drought to <25% after the drought. These two streams would have been classified as intermittent before the drought but have become ephemeral.

On the other end of the spectrum, there were no annual zero-flow days for perennial streams during pre-drought or post-drought periods (Table 2). Accordingly, flow permanence remained 100% at the 90th percentile for all perennial sites during the drought. Median flow and 90-day minimum flow both decreased by 23% during the drought for perennial streams. According to Mood’s median test, 20 of 24 perennial sites had a significant decrease in the number of daily flow values above the median during the drought, while one site had a significant increase—Gila River near Clifton (9442000). The other three perennial sites (9382000, 9504500, 9415000) did not have significant flow-distribution changes during the drought.

Intermittent streams were most impacted by the 1994–2023 drought (Table 2). During the drought, median flow decreased by 55% and 90-day minimum flow decreased by 62% for intermittent streams. All intermittent sites except for one (Gila River at Calva, 9466500) had significant decreases in the number of daily flow values above the median during the drought. Annual zero-flow days increased by 50 days, on average. Four of the intermittent streams experienced an increase of more than 100 annual zero-flow days during the 20-year drought. Using the definitions from Section 3.3., two of these intermittent streams became ephemeral streams: Whitewater Draw near Douglas (9537500) and Little Colorado River near St. Johns (9386030). For Little Colorado River near St. Johns, its annual median of zero-flow days increased from 0 (pre-drought) to 268 (post-drought), making it an ephemeral stream post-drought with a median Q of 0 cms and a 75th percentile flow permanence of 12.5%. Before the drought, it would have been a perennial stream with a 90th percentile flow permanence of 100%. For Whitewater Draw, its annual median of zero-flow days increased from 185 (pre-drought) to 299 (post-drought). Leslie Creek near McNeal (9537200) also transitioned to an ephemeral stream during the drought because its median flow decreased from 0.01 to 0 cms and its 75th percentile flow permanence decreased from 100% to 18%. Including the two previously mentioned ephemeral streams (9485000 and 9513780), there were four streams that changed from intermittent before the drought to ephemeral during the 1994–2023 megadrought. Four other intermittent streams would have been classified as perennial before the drought with a 90th percentile flow permanence of 100% but became intermittent after the drought with a 90th percentile flow permanence less than 100%. These four streams were Gila River at Kelvin (9474000), Agua Fria River near Mayer (9512500), Aravaipa Creek near Mammoth (9473000), and East Verde River near Childs (9507980).

4. Discussion

4.1. Data Limitations and Future Monitoring

The primary limitation of this study was the limited number of long-term stream gages in Arizona. We analyzed 70 stream reaches (0.1%) of the more than 70,000 different stream reaches that exist in Arizona. Thus, our analysis reveals the surface water hydrology of a relatively few nonrandom stream reaches in the state (Figure 4). Most long-term gages were located in the Central Arizona Highlands and near large cities. Notable areas missing from our dataset include the northeastern, northwestern, and southwestern corners of the state. In terms of drainage basins, the Little Colorado River—where the greatest increase in zero-flow days during the megadrought occurred—only had two of its tributaries monitored with a long-term gage. The combined drainage area of these two sites is 1698 km², which means the other ~70,000 km² of the basin is not being assessed for hydrologic changes. Similarly, the Bill Williams River Basin (13,900 km²) only had two tributaries with long-term gages. The Lower Gila and Lower Colorado River Basin (17,900 km²) had no tributaries monitored long-term.

Aside from drainage area and stream length, stream types (based on size and flow) were not proportionally represented. Our sample was biased towards large perennial streams because of the gage-selection criteria used by USGS. The purpose of the USGS network of stream gages, particularly for older gages with records longer than 30 years, was to meet five minimum federal streamflow information goals: (1) interstate and international agreements, (2) flow forecasts/flood warning, (3) river basin outflows, (4) long-term monitoring using benchmark (sentinel) watersheds, and (5) water quality [28]. Small stream channels (stream order: 1–3)—the ones with the most jurisdictional uncertainty—were virtually absent in Arizona’s USGS network of long-term stream gages. Only one met our criteria—South Fork Parker Creek near Roosevelt (9498503), a 2nd-order stream with a drainage area of 2.8 km². The USGS current network of 281 gages in Arizona includes several small streams, but the 30-year minimum for temporal statistical distributions was not met as of 2023. A score of USGS stream gages were installed around the turn of the millennium. Cities and counties across Arizona also installed many gages around the year 2000 and have been more rigorous with their data collection and historical records. Thus, there will be dozens more gages to analyze by 2030, including more small streams.

The perennial river bias was confirmed with our identification of 27 of the 70 gages (39%) as perennial (Figure 4 and Figure 5). Only about 5% of Arizona’s streams are perennial [4,7]. At the other end of the spectrum, approximately 86% of Arizona’s streams are ephemeral [6], but our sample only included 17 of 70 (24%). Roughly 9% of Arizona’s streams are intermittent, which means our sample of 26/70 (37%) also has a bias within nonperennial streams. Most of the new city and county stream gages installed in the past two decades have been on ephemeral streams (for early-flood warning), meaning that we will soon have better long-term assessments of ephemeral streams.

We used daily data for all our flow regime analyses, but many high-flow events on most streams in Arizona last less than a day, some only a few hours. We confirmed these durations with the minute-resolution data provided by Maricopa County for 13 gage sites. The consequences of only having daily data are: (1) peak discharges for flow events are underestimated, (2) the number of hydrologic reversals and thus rate of change are underestimated, (3) flow durations are overestimated, and (4) overall flashiness of the streams is underestimated. However, these issues are relatively minor within the context of relative permanence.

A more concerning issue is how “natural” are the flow regimes we analyzed? Our initial sample for this study was 84 long-term gages with at least 30 years of daily data; however, in our attempt to assess semi-natural flow regimes, we removed 14 gages that were immediately downstream of a dam. We use the term semi-natural in recognition that none of our study watersheds were exempt from human activities. These anthropogenic activities include, but are not limited to, irrigation withdraws, irrigation return flows, water supply pumping, wastewater treatment plant return flows, diversions, and dams farther upstream in the watershed. One estimate has 63% of Arizona’s streams being affected by land use and flow alterations [29]. In spite of that estimate, a comparison of this altered rivers geodatabase with our gage locations revealed that most of our studied streams had less than 11% alteration.

After acknowledging all our data limitations, we still believe that there is not yet a better substitute for characterizing flow regimes of arid streams than long-term daily discharge records. There are even more limitations and assumptions associated with snapshot indicators like the ordinary high-water mark (OHWM) [10] and snapshot assessments like the beta Streamflow Duration Assessment Method (SDAM), which the agencies are using in the arid West as “a scientific tool to provide a rapid assessment framework to distinguish between ephemeral, intermittent, and perennial stream flow at the reach scale” [30]. While hydrophytic plant species, aquatic invertebrates, and algal cover can be useful biological indicators of flow regimes, the month and year sampled will produce varying results for SDAM. As pointed out by Merritt et al. [14], trying to characterize flow regimes of arid region rivers without long-term flow records will likely lead to mischaracterizations.

In the absence of robust empirical data, we may have to rely on hydrologic models. Merritt et al. [14] highlight the importance of including the proper predictor variables in hydrologic models; however, they note that even the best models have limitations in simulating the unique climatic and geological controls of nonperennial streamflow in the region. Other studies have noted the limitations of coarse-scale models in simulating accurate streamflow conditions, especially in the Southwest, where controls on stream dynamics are complex and national datasets lack complete records of nonperennial streams [6,8,14,15]. More localized approaches, coupled with empirical methods to validate models, especially in understudied landscapes such as headwaters in low-slope areas [9], could provide more accurate methods for classifying stream network flow regimes and ultimately which nonperennial streams are relatively permanent.

4.2. Relatively Permanent Waters

Arizona streams exhibited wide ranges of intra-annual and inter-annual variability in their flow regimes and flow permanence, so much so that there was not a typical year for any of the gage sites. Therefore, we used percentile distributions of flow permanence to classify streams as ephemeral (median Q = 0 cms, 75th percentile flow permanence < 25%), intermittent (75th percentile flow permanence > 25%, 90th percentile flow permanence < 100%), and perennial (90th percentile flow permanence = 100%). These flow classifications are based on our definitions that leverage statistical distributions (Figure 5) and therefore may not match definitions from other studies. Indeed, there is a wide range of criteria used to distinguish perennial, intermittent, and ephemeral streams [31]. The dividing line between ephemeral and intermittent streams has the least consensus. For our definition of an ephemeral stream, we relied on the EPA and USGS definitions where ephemeral streams are those that flow briefly (typically hours to days) in direct response to precipitation [4,5]. This definition accords with Arizona Administrative Code for Water Quality Standards (Title 18, Chapter 11) [32].

Our flow permanence upper limit of 25% for ephemeral streams is the same one used by Merritt et al. [14], but was also selected to correspond to the 3-month flow threshold used in previous Corps documentation to designate Traditional Navigable Waters (TNW) tributaries that are “relatively permanent waters (RPWs) in a typical year”. With regards to this “relatively permanent” threshold of 3 months of flow (or 25% flow permanence), it depends which percentile is used or what is meant by a “typical year”. If the 50th percentile is used (i.e., every 5 out of 10 years), then all 26 of our intermittent streams and 5 of 17) ephemeral streams meet this standard. But if the 90th percentile is used (i.e., every 9 out of 10 years), then only 20 (of 26) intermittent streams meet this standard; none of the 17 ephemeral streams meet this standard. Thus, the “relatively permanent” standard will need to be specific with regards to inter-annual flow distributions and what is statistically meant by a “typical year”.

The purpose of this article is not to make jurisdictional determinations, nor set strict guidelines for policy. Rather, we are showing statistical flow distributions that can be considered for initial hydrologic thresholds. But we do recommend that jurisdictional definitions are explicit with regards to flow distributions. Following the 2023 Conforming Rule, the future protection of ephemeral and intermittent streams will likely be decided at the state level. While Arizona has not yet assumed control and administration of its own CWA section 404 permitting program, the Arizona Legislature established in 2021 a Surface Water Protection Program (SWPP) that created a Protected Surface Waters List (PSWL). This PSWL includes non-WOTUS nonperennial streams identified as in need of state-level protection, and the SWPP has a process in place to protect future nonperennial streams [32]. It is important to note that the state-level protection of many of these streams is not based solely on hydrological criteria. Ecological significance, water quality, aesthetic value, recreation potential, and wilderness characteristics are also taken into account.

The “relatively permanent” standard will also need to be placed within the context of climate change. In this study, we have reported historical ranges of variability and statistical averages that assume a stationary climate. But we know that climate is not stationary, particularly in the Southwest, where scientists have documented a megadrought that started around the turn of the millennium [21,22,23,33]. This megadrought means that some streams are likely to become drier over time. An excellent example is the Little Colorado River near St. Johns (9384000), which was a perennial stream before 1994 with a median Q of 0.16 cms and a 90th percentile flow permanence of 100% but has since become dry most of the time (268 annual zero-flow days) and is now an ephemeral stream with a median Q of 0 cms and a 75th percentile flow permanence of 12.5%. In addition to this drastic flow regime reversal, four other perennial streams shifted to intermittent during the megadrought and four intermittent streams became ephemeral. These nine flow regime changes (out of 53) mean that 17% of Arizona’s monitored stream reaches have changed flow types since the megadrought, which is comparable to a recent study of flow regime changes at 342 USGS stream gages in California [31], Arizona’s neighbor to the west.

Anthropogenic activities nearby or upstream, especially changes in irrigation, may have also influenced changes in flow permanence. In the Upper Gila River, we found that median flow increased by 7% in two of its reaches (9442000 and 9466500), which is very likely a result of the Arizona Water Settlements Act of 2004 (including the Gila River consent decree resolution) that significantly reduced groundwater pumping upstream and led to the retirement of thousands of acres of agricultural lands [34]. The Upper Gila River was an anomaly, however. Since 2000, hundreds of new groundwater wells have been installed across the state for irrigation, domestic, commercial, and industrial purposes [35]. The aforementioned California study found that “streams impacted by human activities had greater drying rates” [31]. We further acknowledge that human factors and climate processes can interact to exacerbate drying trends [31,36,37].

The next frontier in arid stream classifications will be dynamic maps that capture flow regimes over space and time. Since 2000, scientists and water managers have relied on the National Hydrography Dataset (NHD) for digital stream maps. While improvements have been made in resolution, accuracy, and value-added attributes (e.g., flow permanence) [38], the NHD is still not reliable for headwater channel locations [39], nonperennial channels [8,9], or flow permanence [40]. Our results confirmed the inaccurate representation of arid nonperennial streams in the NHDPlus-HR dataset. The NHDPlus-HR dataset incorrectly mapped two-thirds of the intermittent streams as either perennial (13 of 26) or ephemeral (4 of 26). None of the ephemeral streams were mapped correctly in NHDPlus-HR. Although this is a small sample of the state’s NHDPlus-HR stream segments, it appears that this dataset should not be relied upon for hydrographic classification in Arizona. The modeled dataset of Merritt et al. [14] was more accurate in mapping intermittent streams (all 26 correct), but incorrectly mapped half of the ephemeral streams as intermittent. We agree with Merritt et al.’s conclusion that their “results represent a promising step toward more effective assessment and management of streams in arid regions” [14], but their results (Figure 1) when compared to ours (Figure 4) also demonstrate that flow classifications do not follow geography precisely. The complex geomorphology of Arizona streams, combined with highly variable precipitation patterns, will make dynamic mapping of the state’s streams difficult; however, with advances in remote sensing and more stream gages coming online, we foresee an improved hydrographic future for Arizona and other arid regions.

5. Conclusions

This article provides an analysis of flow regimes of Arizona streams within the context of the 2023 Conforming Rule on the Revised Definition of Waters of the United States (WOTUS) and its relatively permanent flow standard for jurisdictional tributaries of traditional navigable waters. We analyzed empirical flow records of 70 USGS stream gages in Arizona with long-term daily discharge records (>30 years) that were not located immediately downstream of a dam. Our first important observation was the geographical bias of these 70 gaged stream reaches (Figure 4). The majority of gage sites were located on relatively large rivers, either perennial or intermittent, around major cities. Most medium-sized streams drained the relatively wet Central Arizona Highlands. The driest parts of the state and small streams were not well-represented in this network of long-term gages. We therefore recommend more streamflow data collection across the state, with a priority for small nonurban, nonperennial streams to advance a more representative statewide sample.

Our flow regime analyses revealed wide ranges of intra-annual and inter-annual variability, which is why we relied on percentile distributions of flow permanence to classify and distinguish streams as ephemeral (median Q = 0 cms, 75th percentile flow permanence < 25%), intermittent (75th percentile flow permanence > 25%, 90th percentile flow permanence < 100%), and perennial (90th percentile flow permanence = 100%). This classification uses the flow threshold of at least 3 months to be considered relatively permanent (or intermittent), which is in line with previous Corps documentation and other studies [13,14]. We used the 75th percentile for flow permanence (7.5 years of a decade) to represent a “typical year”. The median (50th percentile) only covers half the years, which we do not consider typical; while the 90th percentile includes almost all years (9 out of every 10) and does not allow flexibility for multi-year droughts or ENSO episodes. For perennial streams, use of the 90th percentile for flow permanence also allows flexibility for multi-year droughts and ENSO episodes. That is, a stream should still be considered perennial even if it has a few zero-flow days in a decade but should have continuous flow at least 90% of the time. In sum, the 75th and 90th percentiles capture the typical year while allowing for the extreme inter-annual variability that exists in arid southwestern streams.

As already mentioned, the purpose of this article is not to make jurisdictional determinations, nor set strict guidelines for policy. Rather, this study presents a classification of arid streams based on flow permanence in one state that can be used to inform hydrologic thresholds. Regardless of if this classification is used or another one, we recommend that jurisdictional definitions/standards are explicit with regards to flow distributions and climate regimes. Arizona’s megadrought since 1994 has impacted the flow regimes of almost all its streams. For nine streams, it has changed their flow type, including one from perennial to ephemeral. Empirical flow records are still our best indicators of flow permanence, but their period of analysis must be chosen methodically. If not, we could have serious water troubles like we do with the Colorado River Compact.