Data sources and site selection criteria

Stream temperature data were used to model the thermal regime for 77 stream sites throughout California. Data were downloaded from the United States Geological Survey (USGS) and California Data Exchange Center (CDEC), publicly accessible databases. Monitoring sites were initially filtered to exclude those in the Sacramento-San Joaquin Delta region to focus the analysis on freshwater thermal regimes and minimize the influence of tidal dynamics. Recent studies have recommended at least 8 to 12 years of continuous, daily average temperature data to reduce uncertainty from interannual variability [32,33]. Monitoring stations were filtered to identify those with at least 8 years of daily average stream temperature data to balance the desire for reduced uncertainty with sufficient spatial representation. Additional data were used from a long-term (10-year) monitoring network in the Shasta River watershed in Siskiyou County, northern California [46,50,51]. Final sites were located in 7 of 10 hydrologic regions of the state (as defined by the California Department of Water Resources: North Coast, North Lahontan, Sacramento River, San Francisco Bay, San Joaquin River, South Lahontan, Tulare Lake); no sites in the southern range of the state (Central Coast, South Coast, and Colorado River) had sufficient periods of record for our analysis. All data were reviewed to remove flagged data (per USGS and CDEC standards) and obvious outliers; the remaining years with a minimum of 8 daily average temperature observations for each day were used in the study. Any data gaps or missing values were not filled; the data was used as is. Daily average stream temperatures were calculated from sub-daily data sets using R (version 4.0) with RStudio [52,53].

Thermal regime modelling and classification

Thermal regime modelling used the mosaic package in R [54]. The reviewed datasets were used to calculate annual thermal regimes, defined by the daily mean temperature for each day of the water year (October 1 through September 30). Annual thermal regimes were modelled with a sine function [29,30]: where Tw is water temperature, n is the day of water year, and a, b, and n_o are coefficients that correspond to annual mean, annual amplitude, and phase (Fig 1). Coefficients a, b, and n_o were optimized using least square regression. Model fitness for each site was quantified using residual standard error; values closer to zero indicated better fit. Modelled thermal regimes were classified based on clustering and statistical analysis methods developed in Maheu et al. [31], which are briefly summarized here: mean, amplitude (i.e., the difference between the annual mean and annual maximum water temperature), and phase (i.e., day of water year when annual maximum occurs) metrics were calculated from each thermal regime model (Fig 1), then classified using Ward’s method, an agglomerative nesting algorithm. Each class comprises a cluster of individual sites, and is defined based on unique features of the clustering parameters (e.g, a cluster of sites in a class all show similar annual variability and amplitudes that are distinct from the other classes). The number of classes was determined using Calinski and Harabasz’s (CH) index and the sum of squares (“Elbow”) method [55]. In addition, we used the silhouette method to validate the appropriate number of classes [56]. Classes were examined for stability using the Jaccard coefficient, with stable clusters indicated by coefficients greater than 0.75 [57]. Clustering and statistical indices were computed using R packages cluster [58], factoextra [59], and fpc [60].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. A thermal regime model fit to observed data.

This example uses data from USGS site 11390500 (Sacramento River below Wilkins Slough near Grimes, CA). Cluster analysis is based on annual mean, amplitude, and phase metrics for each thermal regime model. Amplitude and phase are analogous to annual maximum and day of annual maximum metrics. Figure adapted with permission from Maheu et al. [31]: Fig 1.

https://doi.org/10.1371/journal.pone.0256286.g001

To further assess the relationship between each thermal metric and a given cluster, we used principal components analysis to describe the variation associated with each metric (annual mean, amplitude, and phase). We visualized the distribution of the clusters with an ordination plot of the first two principal components from the analysis, grouped by cluster. We examined relative contributions of each parameter to each principal component to determine which was more important to final clustering results. We also used the Principal Components Analysis (PCA) to identify the centroid of each cluster; then, we calculated the distance of each cluster member relative to the centroid of its respective cluster to identify weak members. For each thermal regime class, a histogram was made to examine the distribution of distance to centroid across all members. Weak members were defined as sites located furthest from the centroid. Additional clustering analysis was done using the same methods to assess whether weak thermal class members had fundamentally different dynamics that were lost in higher-order clustering, or were simply geographically distant from strong members in the same gradient or regime.

Influence of dam regulation

Dam regulation effects were examined by quantitatively and qualitatively assessing thermal regime patterns downstream of dams. The discontinuity distance downstream of a dam depends on many factors, including dam size, location on a river (e.g., headwaters versus lower reaches), operational objectives, and release strategy (e.g. hypolimnial release; [21]). Previous research showed that large dams in California’s Central Valley often influence thermal regimes 30–60 km downstream of release outlets [20,39]; Shasta Dam, impounding California’s largest reservoir, was shown to influence temperature patterns up to 250 km downstream [61]. Because varied dam sizes and reaches with multiple dams may show varied effects, histograms were generated for each thermal regime using sites that were within mainstem reaches 100 km or less downstream of a dam. Finally, thermal regimes in these regulated reaches were examined for member strength of each below-dam site to its respective regime class. We compared the distance of each site to its respective upstream dam to its strength as a member to its cluster as quantified by the PCA.

What constitutes a cold-water thermal regime?

In ecological terms, “cool” and “cold” are typically used to classify species based on temperatures that support optimal growth, and are often simplified to static thresholds. These thresholds vary by region: one study identified 10–15°C for cold-water fishes, 21°C for cool, and 30°C for warm in the Great Lakes region [62]; another suggested <20°C, 20–28°C, and >28°C for more general cold, cool, and warm-water optima [63]. Other studies use additional criteria to classify thermal regimes [6,31,41], but still rely on threshold-based definitions for cool and cold. Our results show that, based on criteria developed by Rahel and Olden [63], all but two of California’s thermal regimes would be considered cold; yet these supposedly equivalent cold regimes demonstrate a range of ecological performance related to targeted cold-water species objectives [64]. Regulatory guidance has trended toward more temporally refined thresholds based on target species and their life histories [65]. This analysis used three metrics to characterize annual trends and can be applied as a useful first step to identify desirable areas for cold-water management. Other studies show how additional metrics and short-term variability are useful to assess thermal regimes and their relationship to ecological function [6,41,42], which would be key factors in developing conservation strategies to support process-based thermal regime management. Refined classification, whether based on variability [31], geography [41], or some other feature, are important to distinguish thermal regimes that would otherwise be considered uniformly supportive of cold-water ecosystems using a threshold system.

In addition to developing a classification that captures the range of variability of California’s cool- and cold-water regimes, our objective was to provide a classification to support conservation decisions. To this end, the geographic scope of analysis showed important differences from other studies. Interestingly, our results showed more thermal classes for unregulated reaches when California was analyzed as a state as compared to national or regional analyses [31]. Five thermal classes were defined for unregulated sites, compared to three identified by Maheu et al. [31]. Two of those five classes were considered warm (both stable and variable), while the national analysis resulted in only cool and cold regimes in California. Some of these differences are explained by the inclusion of groundwater-dominated sites in this study, which accounted for two of the five classes. However, other sites that had been classified as cool (or stable) in the national analysis were reclassified as warm (or variable) in our analysis. Isaak et al. [41], which classified thermal regimes in California as part of the western U.S., identified five regime classes in California, though none represented groundwater and showed less diversty in hydrologic regions like the North Coast.

These results show the importance of geographic scope when developing a conservation strategy. Salmonids and other cold-water species have been documented as far south as Mexico [66], indicating that cool- and cold-water regimes extend further south than shown in this study. Though data gaps were not filled in this analysis, recent interpolation methods show promising utility for stream temperature records [67], which could potentially increase the number of sites included in the analysis and expand the geographic scope. Finally, while USGS and CDEC databases have many long-term temperature datasets, and temperature monitoring tends to focus on short-term summer periods, additional data may be available through other public, crowd-sourced sites as more comprehensive temperature data are collected [68].

The five-class system revealed nuanced differences between cool- and cold regimes, and highlighted the importance of groundwater-fed streams to support cold-water conservation. While agglomerative nesting showed comparable statistical strength of classifying California’s thermal regimes into three or five classes, fewer classes may be an oversimplification not necessarily useful for management decisions, which typically occur at state or local levels. Research by Null et al. [35] found that thermal regimes of California’s western Sierra Nevada rivers did not show the same shifts in desirable cold water habitat as in national-scale studies. Because California rests at the southern edge of the geographic range of many cold-water species [45], has diverse geographic and hydrologic streamscapes [34], and is strongly influenced by dam regulation [36], having high resolution of its thermal regimes and effects of regulation is desirable for conservation planning and investment. Thus, while three classes had slightly stronger statistical support, five classes provided more insightful differences between cool- and cold-water thermal regimes, particularly relative to groundwater and dam releases.

Warm and cold groundwater-fed springs accounted for two of California’s five thermal regimes: stable warm and stable cold. Although each class contained a single groundwater-fed site, these regimes illustrated a unique thermal pattern dominated by groundwater-fed spring sources. The stable cold regime, which included both a groundwater-fed spring and the outlet of Shasta Dam, was relatively unstable as indicated by its Jaccard coefficient and large spread of members from the cluster centroid. Additional data describing stable cold sources would improve understanding of this regime by indicating whether it is a stable class with high variability (which would account for the large spread of the initial two members) or better broken down into separate classes, possibly defined by large groundwater-fed springs and dams with large cold-water storage volumes. Despite studies that have classified thermal regimes in California [31,41], none explicitly identified thermal regimes for spring-fed sites in the state. However, the presence of “slightly thermal” groundwater-fed streams in California [50,69] suggests that such a class may be more prevalent than currently known. A separate, stable cold class dominated by releases from reservoirs, though, is unlikely given the dearth of reservoirs in California with the cold-water capacity comparable to Shasta Reservoir.

Do dams “reset” the longitudinal pattern of a stream’s thermal regimes?

Our study shows that dams do not reset thermal regimes: rather, they create discontinuities characterized by artificial regimes. These dynamics may persist for 10s-100s km downstream from a dam’s outlet [47,61], and tend to be compounded by multiple upstream dams (S2 Table). Differences between regulated and unregulated thermal regimes (excluding groundwater-dominated regimes) are illustrated by their annual magnitudes and variability. Variable cool regimes occurred exclusively in unregulated reaches, had more variable annual patterns (i.e., warmer annual maximums and cooler minimums), and had more predictable annual means, maximums, and day of annual maximum than stable cool regimes in regulated reaches. As a result of this variability, the sine model was a poorer fit for unregulated sites compared to regulated sites. Regulated thermal regimes showed the opposite trend. Stable cool regimes were strongly influenced by upstream dams and showed less annual variability, but higher variability among the three classifying metrics. Thus, although the overall annual temperature pattern was more stable in regulated reaches, the annual mean, annual maximum, and day of annual maximum varied more within this regime than in variable warm and cool regimes. This variability may relate to storage capacity, operational objectives, geography, or degree of regulation [36,41,48]. Stable cool regimes transitioned to variable warm regimes as the downstream distance from a dam increased. Variable warm regimes were generally at least 40 km downstream of dam outlets and may reflect a transition from regulated influences to dynamic equilibrium, when stream temperatures are dominated by heat flux due to ambient meteorological conditions [61].

The influence of dams and stable cool thermal regimes was further strengthed once factors like degree of regulation were taken into account. Grantham et al. [36] quantified the degree of regulation for California streams for both individual dams and cumulative number of dams above stream segments. The results for stable cool thermal regimes coincided with reaches that were strongly altered, specifically due to dams (alteration was defined as degree of regulation > = 1; see S2 Table). This extensive downstream effect of dam regulation begs the question of what a thermal regime’s natural evolution from headwater to lowland equilibrium may have looked like, and whether dams have eliminated an important transitional reach for temperature-sensitive ecosystem function. Additional analysis of thermal regimes given modeled, unregulated stream temperatures (e.g., [35]) could show the historic fate of thermal regimes over stream reaches currently dominated by dam releases and identify if the transition to equilibrium included regimes similar to those produced downstream of dams, or a more gradual shift in annual mean from variable cool to variable warm regimes.

Most notably, while stable cool regimes successfully mitigated elevated summer stream temperatures, they similarly constrained winter minimum temperatures and maintained artificially warm conditions. Research on the effects of dam regulation on stream temperatures tends to focus on the summer season [47], when elevated stream temperatures may lead to stress or increased mortality of cold-water species [12,45]. Fewer studies have focused on the negative effects of sustained periods of elevated winter temperatures on cold-water species [70]; we are unaware of studies that focus on the potential to replicate colder winter patterns with dam regulation.

Can dams be managed to replicate desirable cold-water regimes?

In stream reaches that lack a resilience to climate warming, cool- and cold-water habitat may be unachievable through dam regulation. In particular, the stable cool regime may present the greatest challenge to cold water conservation as it generally lacks the cooler winter temperatures of unregulated variable cool regimes. One notable result was the classification of the Shasta Dam outlet (site SHD)–the only reservoir to produce a stable cold thermal regime. At 4.6 million acre feet (MAF), Shasta Lake is California’s largest reservoir and maintains its cold pool through cold-water inflows, cooling that occurs during the winter, thermal stratification, and operational decisions [38]. Despite the large capacity of New Melones (2.4 MAF, 4^th largest reservoir in California), it, or any other dam included in this analysis, was unable to produce a stable cold regime at its outlet.

This study only considered the effect of the nearest upstream dam to a study site. Many streams have several dams, perhaps with compounded thermal effects [71]. For dams that lack both the capacity to produce a stable or variable cold regimes and lack passage above the dam, these barriers may be insurmountable for species’ recovery. While reservoir operation to support cold-water habitat has shown promise [39,72,73], our results suggest that improving passage or dam removal is likely needed to reunite species with the thermal regimes in which their life-history strategies originally evolved. Potential constraints are considerable, though, given the fundamental shift in underlying, unregulated thermal patterns as a result of climate warming, particularly in mid-elevation streams [35].

Finally, despite similarities between groundwater-fed and dam-influenced reaches, the conservation value of these reaches should not be conflated. Streams like the McCloud River, Pit River, Battle Creek, and Big Springs Creek are highly influenced by groundwater-fed springs, and were the only upstream-of-reservoir reaches to replicate the same stable cool thermal regimes found below reservoirs. The regulated reach downstream of Shasta Dam also illustrates a distinct antinode-node pattern that is characteristic of large-volume, groundwater-fed streams [50,61]. Despite similar thermal regimes, other research has shown how other aspects of these groundwater-dominated streams differ from runoff and regulated reaches [74]. Historically, groundwater-dominated streams have out-produced non-groundwater-dominated streams [74,75], and, for the streams still accessible to fish, are preferentially selected [73,75]. Previous studies have shown even in regions sensitive to climate warming, watersheds with larger flow volumes and groundwater contributions, like California’s Feather River, are less vulnerable to climate change [31,35]. Also, some Californian spring-fed streams have novel hydroecological feedbacks that drive their thermal regimes [51], influence reaches tens of kilometers downstream from spring sources [46,50], and support robust ecological productivity and conservation potential [11,74]. Thus, other factors, such as water quality (nutrients), physical habitat, flow regime, and novel ecohydrological feedbacks may still make spring-fed reaches more desirable habitat than regulated reaches despite their similar thermal regimes.