Over half a century record of limnology data from Lake Powell, desert southwest United States: From reservoir filling to present day (1964–2021)

Lake Powell is a large water storage reservoir in the arid southwestern United States. Here, we present a 58‐yr limnology dataset that captures water quality parameters from reservoir filling to present day (temperature, salinity, major ions, total suspended solids), as well as a 38‐yr record of Secchi depth, and a ~ 30‐yr record of nutrients, phytoplankton, and zooplankton assemblages. The dataset includes 5208 unique site visits spanning 258 unique sites of which 9 have been consistently visited. It also spans the establishment of an invasive bivalve (Dreissena bugensis, i.e. Quagga mussel) which was first detected in 2012. Given the general lack of long‐term data from lakes or reservoirs in arid regions, this dataset represents a unique contribution to regional, continental, and global‐scale limnology studies. As the hot drought in the desert southwest continues, we expect this dataset will inform water management decision‐making for this largest reservoir in the Upper Colorado River Basin.

. Locations of individual stations, broken into seven reservoir regions with frequently visited sites labeled (A). The data availability for six major data streams is summarized for frequently visited sites broken out by reservoir region (B). Frequently visited sites were defined as the two stations with the most water quality profile visits stratified by reservoir region. Tailwater data is also summarized even though no water quality profiles exist at these sites. Both profile and Secchi depth counts represent the number of discrete site visits with data. *Profiles from LPCR-249 are discrete water quality measurements from the surface of the river at the Lees Ferry dock. Additional water quality data from the Colorado River at Lees Ferry, Arizona (09380000 gaging station) are available at https://www.gcmrc.gov/discharge_qw_sediment/stations/GCDAMP. †Samples taken from the dam draft tubes are compiled here even though there are eight unique station codes that correspond to eight different turbines inside the dam (LPCRDT01-LPCRDT08). ‡Profiles from the draft tubes refer to discrete water quality measurements of water flowing within the power plant at Glen Canyon Dam, at one of eight draft tubes. Additional water quality data from Glen Canyon Dam near Page, Arizona (09379901 gaging station) is collected directly below the dam and is available at https://www.gcmrc.gov/discharge_qw_sediment/stations/GCDAMP. §Data availability here is combined for both the historical Bullfrog Bay sampling site (LPCR1692) and the current site (LPCR1679).

Background and motivation
Long-term datasets have provided unique insights into environmental processes, population and community dynamics, and emergent properties of aquatic ecosystems (Hampton et al. 2019). A few key datasets that span exceptionally long timeframes have supported diverse and interdisciplinary insights into changing lake conditions. Long-term datasets from artificial reservoirs are somewhat less common than in natural lakes. Still, understanding chemical and biological trends in these systems is especially important given that they are more highly managed systems that often serve multiple human uses (Hayes et al. 2017).
Here, we present a long-term water quality dataset from Lake Powell, a large arid reservoir (653 km 2 and 31 km 3 at full pool; Fig. 1). Lake Powell was formed by the construction of Glen Canyon Dam, which was completed in 1963. The Lake Powell Water Quality Monitoring program formally began in 1965 and was motivated by basin-wide concern over high salinity levels in surface waters across the Colorado River Basin. Although substantial reductions in salinity have occurred across the Upper Colorado River Basin since the Salinity Control Act (Rumsey et al. 2021), annual economic damages related to high salinity are still estimated at 380 million USD (U.S. Bureau of Reclamation 2017) and funding for the monitoring continues from the Salinity Control Program. Objectives of the monitoring program have evolved over time in response to emerging environmental concerns related to the Grand Canyon Protection Act and the establishment of the Glen Canyon Dam Adaptive Management Program among others.
The early programmatic focus on salinity has resulted in a substantial record of major ion concentrations in Lake Powell surface and bottom waters in addition to water temperature and dissolved oxygen profile data (Fig. 1). Interest in other limnological measurements like Secchi depth, nutrient concentrations, and phytoplankton and zooplankton community assemblage developed over time. The result is a long-term record which captures limnological change across a reservoir age gradient.

Deemer et al.
Limnology data from Lake Powell, desert southwest USA

Data description
This is an historical dataset that also includes primary data (-Andrews and Deemer 2022). Earlier descriptions of these data can be found in two U.S. Geological Survey reports (Vernieu 2015a,b). Data associated with these reports span from 1992 to 2009 for phytoplankton and zooplankton data and from 1964 to 2013 for physical and chemical data. Although older reports include spectrophotometric measurements of surface Chlorophyll a (Chl a) concentration, these data were excluded from this data release given the inconsistency in sample documentation and preservation which likely affected detection across storage methods (e.g., samples were either frozen, refrigerated, or desiccated). The dataset does, however, contain vertical profiles with fluorometric measurements of Chl a from 2010 to the present.
Data are served as a series of eight .csv flat files that link to each other based on common keys (Table 1; Fig. 2). These data can be downloaded at https://doi.org/10.5066/P9ZIKVYW (Andrews and Deemer 2022). We have improved the usability of previously published historical data by obtaining and reporting historical nutrient and major ion detection limits. We are also releasing historical and present-day QA/QC samples including reagent blanks, equipment blanks, duplicate, and spikes for nutrient and major ion data. For plankton data, this release contains standardized and normalized taxonomy that omits nomenclatural synonyms for taxa (Table 2). This dataset also contains historical water quality data from some infrequently visited sites that were not included in either of the original U.S. Geological Survey data series reports (Vernieu 2015a,b).

Monitoring sites and approach
Several monitoring phases have defined the sampling schedule for this dataset (Table 3) and are described in Vernieu (2015a). Overall, the program has conducted regular monthly sampling at three sites near the dam with reservoirwide sampling generally conducted quarterly or least annually at anywhere from 8 to 37 sites. Regular monthly sampling sites included one "forebay" station on the reservoir near the dam (usually "LPCR0024" but sometimes "LPCR0006") as well as two stations termed "tailwater" sites. The first of these stations is located 24.9 km below Glen Canyon Dam at the Lees Ferry boat dock "LPCR-249." This site is co-located with the long-term USGS Colorado at Lees Ferry gage (#09380000) and is meaningful in a water policy context given its proximity to the official dividing line between the Upper and the Lower Colorado River Basin (located approximately 3 km downstream). The second tailwater station is located within Glen Canyon Dam. Water was sampled from within the power plant at one of eight draft tubes (shortly after the water exits a turbine; "LPCRDT01" to "LPCRDT08").
At every on-reservoir site, Secchi depth was measured and a water quality profile was conducted. At a select number of sites additional samples were collected for nutrients, major ions, phytoplankton, and zooplankton. The nine sites with the longest data record for water quality are the three sites described above (LPCR0024, LPCR-249, and LPCRDT0X) as well as LPCR453, LPCR0905, LPCR1169, LPCR1679, LPCR2387, and LPSJR193. In three cases, historical sampling stations were relocated (moved by less than 3 km) to avoid boat traffic. For analysis, these stations can generally be considered interchangeable given that all sites are located within the historic main channel (LPCR1692 became LPCR1679, LPESC072 became LPESC069, and LPCR1772 became LPCR1799). Starting in 1982, inflow sites were also visited during reservoir-wide surveys. These sites were accessed by boating as far up-channel as could safely be navigated. Inflow sites may represent long-term stations or may be newly defined given water levels. They have unidirectional flow and are specified at the site visit level as "inflow." More detailed information about regularly visited sites and their historical representation can be found in Vernieu (2015a).

Profile data
Six different water quality instruments have been used to collect water quality profiles for this database. Before October of 2010, water quality instruments were deployed for manual sample collection from discrete depths. This was done using the "rule of 5s." That means a new discrete reading was recorded when one of the water quality parameters changed by a predetermined amount: temperature change of 0.5 C, conductivity change of 50 μs cm À1 , dissolved oxygen change of 0.5 mg L À1 , or turbidity change of 5 nephelometric turbidity units (NTU). Thus, the frequency of measurements generally declined with depth given that the hypolimnion is generally well mixed. After October of 2010, a 4-Hz Seabird Electronics 19 plus V2 SeaCAT conductivity, temperature and depth (CTD) profiler was used to collect measurements in continuous mode. The CTD has a built-in pumped depth, temperature, and conductivity recorder and was also equipped with an SBE 27 pH probe, a SBE 43 dissolved oxygen sensor with a 0.5-mil membrane for fast response time, and an ECO-FLNTU-RT combined fluorometer and turbidity sensor for turbidity and Chl a. Throughout the dataset, conductivity is reported as specific conductance. Seabird software was used to correct depth estimates using a set pressure offset based on the elevation of Lake Powell (offset was 1.258 calculated based on a pressure reading at the boat surface; see Application note 73 from Seabird). Only downcast measurements were saved to ensure that flow across the dissolved oxygen sensor was representative (and not backflush). Data were binned into 0.5 m increments from the surface to $ 1 m above the bed. At shallow inflow sites with well-mixed flowing water, the water quality sensor was allowed to equilibrate mid-channel and a single time averaged value is reported for each water quality parameter.

Deemer et al.
Limnology data from Lake Powell, desert southwest USA Secchi depth Secchi depth was measured beginning in August 1980. Secchi depths were taken by lowering the disk until it could no longer be seen and then raising the disk until it was visible again at which time a depth was recorded. In January 1995, a second Secchi measurement with a 1 m aqua-scope was added to each surface visit to control for visual disruptions due to surface turbulence. The aqua-scope is an underwater viewing device (also

Species level ID, abundance, biomass
SampleFk 2681 Deemer et al. Limnology data from Lake Powell, desert southwest USA called a bathyscope) and was only used when Secchi depths exceeded 1 m. Secchi depths were always measured from the shady side of the boat and field conditions were recorded in one of seven qualitative categories ("Excellent," "Excellent-Good," "Good," "Good-Fair," "Fair," "Fair-Poor," and "Poor"). These conditions describe the ease with which the Secchi measurement was made, which is influenced by factors like waves, rain, and wind.

Major ion and nutrient concentrations
Nutrient and major ion samples were collected from the surface (1 m) and bottom of the reservoir using Niskin-type samplers. For each sample site and depth, two bottles of unfiltered sample were collected-one with no preservative for electrical conductivity, pH, alkalinity, and total suspended solid (TSS) concentrations and one preserved to achieve 0.2 N H 2 SO 4 for total phosphorus (TP) and either total nitrogen  titles match the table descriptions in Table 1.

Deemer et al.
Limnology data from Lake Powell, desert southwest USA (TN) or total kjeldahl nitrogen (TKN). Two additional bottles were filled with filtered water (passed through a 0.45-μm mixed cellulose esters membrane filter)-one with no preservative for total dissolved solids (TDS) and major ions (SiO 2 , Fe, Ca 2+ , Mg 2+ , Na + , K + , SO 2À 4 , Cl À , HCO À 3 , and CO 2À 3 ) and one preserved to achieve 0.2 N H 2 SO 4 for soluble reactive phosphorus (SRP), nitrate + nitrite (NO À 3 + NO À 2 ), ammonium (NH þ 4 ), and more recently total dissolved phosphorus (TDP; starting October 2017). Filtering was done using a peristaltic pump (Geotech brand Geopump) and filter housings were equipped with vents to avoid over-pressurization. All samples were stored on ice until analysis. In the reservoir inflows, a different filter setup was used to accommodate much higher sediment loads. Historically, this consisted of a larger diameter 0.45-μm filter and prefilter setup. More recently, Geotech trace metal-grade high-capacity filters have been used. We conducted a methods test comparing the two filter types and found no detectable differences in concentrations of any of the filtered analytes (Fig. 3).
Nutrient and major ion samples were run according to standard United States Environmental Protection Agency (EPA) methods. Samples collected from 1990 to December 2004 were analyzed by the Bureau of Reclamation Lab at the Denver Federal Center. Samples collected from January 2005 through present were analyzed by the Bureau of Reclamation Lower Colorado Regional Laboratory. Laboratory holding times were 1 week for pH, electrical conductivity, TSS, and TDS, 2 weeks for alkalinity, 1 month for TN, TP, SRP, NO À 3 + NO À 2 , NH þ 4 , TDP, SiO 2 , Ca 2+ , Mg 2+ , Na + , K + , SO 2À 4 , Cl À , HCO À 3 , and CO 2À 3 , and 6 months for Fe. From February 1995 to November 2005, alkalinity was determined in the field concurrently with the collection of water samples (Vernieu 2015a).

Phytoplankton and zooplankton
Phytoplankton and zooplankton data have been collected at the same 11 stations (LPCR0024, LPCR-249, LPCRDT0X, LPCR0453, LPCR0905, LPSJR193, LPCR1169, LPCR1692, LPCR2085, LPCR2387, and LPESC119) consistently (if inundated) since 1993 (although some samples date back to 1992) as well as at additional sites during some time frames. At Bullfrog Bay, the historic sampling location (LPCR1692) was moved by 1.3 km to its current sampling location (LPCR1679) to avoid boat traffic. Likely due to its location (down-lake of the inflow plunge point, where the water is clear), the plankton community at this site is diverse and relatively abundant which may make it of interest in studies that aim to examine trends and drivers of reservoir plankton.
Phytoplankton samples were collected as discrete samples with a Niskin-type bottle-generally from a depth of 1 m. Five hundred milliliters to 1 liter whole water samples were preserved with acid Lugols and stored at room temperature in the dark before shipping to BSA Environmental Services, Inc. for analysis. For analysis, samples were prepared using standard membrane filtration techniques (APHA Standard Methods 10200) and a minimum of 400 natural units were enumerated to the lowest possible taxonomic level. Counts are reported to be accurate within 90% confidence limits as per Lund et al. (1958). Cell biovolumes were based on measurements of 10 organisms per taxon when possible and were estimated following Hillebrand et al. (1999).
Zooplankton data were collected via single vertical tows using a Wisconsin-style plankton sampler with a 12.7 cm diameter and an 80-μm nylon mesh. Samples were generally collected from 0 to 30 m (although 30-60 m tows were also conducted at some stations). Tows were conducted during daylight hours, a practice that can bias estimates of zooplankton density and biomass (Doubek et al. 2020). The approximate sampling time (AE 30 min) is reported with each tow to aid interpretation. For tailwater samples, the volume of water poured through the sampler is estimated via either flow rate (dam draft tube sites; LPCRDT0X) or number of 10 liter buckets filled (Lees Ferry; LPCR-249). For reservoir samples, the volume of water filtered is estimated based on the depth of the tow with $ 10.8 liter filtered per 1 m of tow depth. This approach assumes that each observer tows the net at a similar rate of 1 m per 3 s.
For analysis, samples were examined on Wilovert inverted microscopes using the Utermohl sedimentation technique. A minimum of 200 organisms were tallied per sample (or the Phytoplankton †, zooplankton *The "<" flag is contained in its own field outside the "flag" field. † There are no phytoplankton currently flagged as "D," but this will likely change in future versions of the data release (notes regarding sample degradation are only included from 22 January 2015 forward for phytoplankton and from 12 December 2016 forward for zooplankton).
entire sample was analyzed for samples with low zooplankton counts). For each identified taxa, up to 10 individual specimen length and width measurements were recorded for biomass estimates as in Beaver et al. (2018). In addition to zooplankton, invertebrate veligers of Dreissena bugensis (commonly known as Quagga mussels) were also counted, with the first individual identified in 2012. Phytoplankton and zooplankton analysis began in the year 2001, meaning that samples collected earlier (from 1992 to 2000) were often stored longer prior to analysis than more recent samples. Samples collected from 2002 to 2011 were analyzed in 2012. Funding gaps in more recent years (2012 forward) have resulted in shelf lives of 1-2 yr for some samples . Samples are analyzed following the methods in Beaver et al. (2018) and taxonomy is reported based on the Integrated Taxonomic Information System. Taxonomic designations reflect the nomenclature used at the time of analyses.

Technical validation
The following sections describe data quality assurance and quality checks (QA/QC) as well as data normalization procedures. The U.S. Geological Survey ScienceBase dataset is a versioned data release that is structured so that it can be regularly updated with new data as it becomes available and undergoes QA/QC and normalization procedures (Andrews and Deemer 2022). We plan to regularly update this repository.

Monitoring sites and approach
To aid data analysis efforts, one of four trip types is assigned to every site visit in the data release ("Forebay," "Whole Reservoir," "High Flow Experiment," or "Other"). The "Forebay" category comprises the regular sampling at both the reservoir forebay and in the tailwater, which has generally been conducted monthly. The "Whole Reservoir" sampling comprises the reservoir-wide trips that have generally been conducted quarterly. The "High Flow Experiment" sampling Table 3. Description of different phases of the Lake Powell water quality monitoring program. Frequency of sampling at various locations is described as either "M" for monthly, "Q" for quarterly, or "Y" for yearly. Adapted from table 1 in Vernieu (2015a). Colors correspond to data tables and match color scheme in Table 1  represents additional sampling conducted before, during, or after some (but not all) experimental high flow events. These experimental flows were conducted as maximum power plant releases or greater (via bypass) for 24-168 h in 1996, 1997, 2000 (twice), 2004, 2008, 2012, 2013, 2014, and 2018. There were also longer-term high steady flows conducted in the winter of 1997, spring of 2000, and in late spring to fall of 2011 (Melis et al. 2016). Profiles were sometimes conducted without accompanying chemistry or plankton sampling. These visits were assigned a "Forebay" or "Whole Reservoir" category if they represented the only trip in that month. Otherwise, these visits were assigned to the "Other" category.

Profile data
Profile data can be flagged at the individual measurement level or for an entire analyte. For analytes that were flagged as "do not use data" or "outside normal values", values were excluded from the data release. For less severe flags "other," values were still provided but were flagged and noted. In addition to flagging profile data when water quality sensors were not performing to specification, profile data were visually examined for inconsistencies. The combined fluorometer and turbidity sensor on the Seabird sometimes had anomalous readings near the surface, a phenomenon that was most pronounced for turbidity but also occurred for Chl a. To identify these instances, profiles taken with the Seabird (2010 and forward) with mean surface (0-0.5 m) turbidity readings > 1 NTU were selected. From this subset, profiles were flagged when their mean surface turbidity was > 200% of the mean turbidity directly below the surface (1-2 m). Given the importance of interflows and advective mixing on Lake Powell (Gloss et al. 1980), we looked for concomitant changes in other water quality variables across the same depth range as for turbidity. High surface turbidity readings with no concomitant changes in other variables were flagged outright. High surface turbidity readings that were accompanied by changes in other water quality variables were visually examined. If the data did not make intuitive sense, the binning and analysis of profile data was revisited to ensure that the averaging of surface values was not accidentally incorporating readings taken when the unit was outside of the water. Observations of small negative values in turbidity (À0.1 to À0.8 NTU) were relatively common prior to using the Seabird (n = 195) and these values were forced to zero.
Sensitivities for Seabird sensors are 0.005 C for temperature, and 0.0005 S m À1 for conductivity, 0.1 for pH, and 2% of saturation for dissolved oxygen. The typical stabilities of these data are 0.002 C per month, 0.003 S m À1 per month, and 0.5% per 1000 h of deployed time for dissolved oxygen. pH was not generally field calibrated in between factory calibrations so drift was likely substantial. Factory calibration reports show drift of up to 0.1 pH units, but drift was not reported consistently. More recent field checks of the pH against known pH 7 standard have shown drift of up to 0.2. The fluorometer has a sensitivity of 0.015 μg L À1 for Chl a and the turbidity sensor had a sensitivity of 0.005 NTU for low end turbidity measurements. The Seabird was sent in for factory recalibration every 1-2 yr.

Secchi depth data
To examine observer bias, observer initials were recorded alongside Secchi depth measurements for 91.2% of measurements (7784 out of 8535 measurements). Duplicate Secchi depth measurements from different observers make up 1464 out of 2956 site visits (49.5%) and can be used together with observer initials to quantify observer bias in Secchi depth measurements (Fig. 4).

Nutrient and major ion data
Method detection limits for nutrient and major ion data were estimated as either 4x the instrument detection or 3.14x the standard deviation of 7 samples run at the instrument detection limit. Instrument and lab analyst turnover have resulted in shifting detection limits over time for many of the analytes reported in this dataset. The number of decimal places reported for chemical concentration data reflect the estimated reporting limit, which we calculate as 3x the method detection limit. Sample flags were assigned to indicate various aspects of the sample treatment, analysis, and concentration (Table 2). Most flags have been used throughout the monitoring program's history, but the "E" flag is limited to samples run prior to 2005; more recently, samples that come back above the method calibration range are diluted and run a second time. There is also some historical variation in the way the "<" flag was assigned. For historical samples (before 2016), samples at the detection limit were assigned a "<" flag. After this point samples could be at the detection limit with no flag if the reporting limit was below the method detection limit. From March of 2012 for major ions and July of 2013 for nutrients, through 31 December 2021, method detection limits were calculated with each laboratory analysis set and are reported accordingly. For earlier data, method detection limits over longer time frames were estimated based on existing flagged values and the method detection those flagged values reported. pH values should be interpreted with caution since samples were stored with a headspace, allowing for carbon dioxide off-gassing. We report pH values < 1 as blank in this dataset.
Reagent and equipment blanks were collected beginning 01 December 1989 and have been collected on every trip starting in August of 2017. Duplicate and spike samples were collected as part of the historical sampling program (01 December 1989 through 18 December 2015 for duplicates; 21 July 1992 through 6 August 2014 for spikes). These QA/QC samples were matched with the original sample based on notes, depth, date, and station information. Spike amounts and expected concentrations were also derived from notes. Once matched, QA/QC samples were verified by comparing sample values to expected duplicate or spike values. We flagged cases where the measured value was more than 10% different from the expected value and when the absolute difference between expected and measured values was greater than the reporting limit (defined as 3x the detection limit). In cases where ≥ 50% of analytes came back flagged, the QA/QC sample was dropped from the database due to concern that it was mis-identified. The QA/QC information also allows the user to evaluate trends in the precision and accuracy of nutrient and major ion data (Fig. 5). Nutrient and major ion data for each trip were typically analyzed in their own batch. This means that reagent and equipment blank data can be linked back to samples based on Trip ID (located in the site visit  table).

Plankton data
Redundancies in the historic plankton database were eliminated so that there is only one instance of each genus and species, together with IDs that resembled a given genus species (cf.) reported (Table 4). This included basic normalization practices such as removing extraneous spaces from name strings. In addition, sample "fragments" from the old database were merged into single records. Historical cases where genus was reported with no "sp." or "spp." are reported as "genus sp." here.
Although some plankton samples underwent long holding times (up to 10 yr), older samples generally maintained the straw color indicative of good preservation (presumably via additional periodic spikes with Lugols). All samples that were held longer than 2 yr were flagged "H." No analysis notes were archived for samples collected prior to 22 January 2015 for phytoplankton and 12 December 2016 for zooplankton (with the exception of a small number of historic samples that were analyzed in 2021 and flagged accordingly). Analysis notes are available for more recent samples and a small number of compromised samples contain a note "specimen degraded" and are flagged "D."

Data use and recommendations for reuse
A key use for the water quality data collected from Lake Powell is to run a CE-QUAL-W2 model (Cole and Wells 2013) Fig. 4. Secchi depth measurement differences by observers at five longterm sites (without aqua-scope). Boxplots show the difference between the maximum and minimum Secchi depths recorded during the same site visit. Differences are reported as a percentage of the mean Secchi depth during that site visit. Colors correspond to the reservoir regions in Fig. 1. that predicts outflow temperature, dissolved oxygen and TDS concentrations to the Lower Colorado River (Williams 2007). Glen Canyon Dam generally releases colder hypolimnetic and metalimnetic water that is undersaturated with respect to dissolved oxygen, and water does not fully equilibrate with the atmosphere for about 50 river kilometers downstream (Hall et al. 2012). Because of declining water levels, the dam's fixed water intake structures are now drawing from the epilimnion of the reservoir (water intake depth was only 17 m below the surface in spring of 2022 vs. an historic range of 24-70 m below the surface).
As reservoir water levels decline, there are a number of concerns for downstream ecosystems. Releasing warm water from the epilimnion is a concern for the establishment of invasive warm water fishes in downstream river segments (Mihalevich Fig. 5. Comparison of analyte concentration values between original/expected samples and QA/QC samples for duplicate and spike QA/QC sample types. For samples that were duplicated, the comparison is between the original samples' concentration and their duplicates. For spike samples, the comparison is between the expected value as if the spike solution had been added to the original sample and the actual concentrations of the spike sample. The annotated text is the percent difference between the median of the original/expected samples and the QA/QC samples per analyte and QA/QC sample type.
Table 4. Genus species notation for phytoplankton samples.

Cf. genus species
Analyst thinks it is something similar to this genus species but is not sure of either the genus or the species Genus cf. species Analyst is confident of the genus and thinks the species is one similar to that reported Genus spp.
Analyst recognizes more than one distinguishable species of the given genus, but cannot ID them Genus sp.
Analyst recognizes one species of the given genus but it is not confident on the specific ID Genus species Analyst is certain of both genus and species Deemer et al. Limnology data from Lake Powell, desert southwest USA et al. 2020;Dibble et al. 2021;Bruckerhoff et al. 2022) and for the Blue Ribbon rainbow trout fishery downstream of the dam . Metalimnetic low dissolved oxygen events caused by the resuspension of deltaic sediments are also a concern for fish in the Glen Canyon Reach directly below the dam. Finally, the concentration of soluble reactive phosphorus in the dam outflow has been positively related to aquatic insect drift biomass and trout population dynamics in the Glen Canyon Reach (Korman et al. 2021;Yard et al. 2023) and reservoir outflows are considered an important piece of the riverine phosphorus budget in this phosphorus-limited system. Earlier versions of the Lake Powell database have been used to describe the water quality in the outflow from Glen Canyon Dam both during experimental flows (Hueftle and Stevens 2001) and as a result of storm-driven interflows (Wildman and Vernieu 2017). The dataset has also been used to describe the limnology of Lake Powell, including calcite precipitation dynamics (Deemer et al. 2020), greenhouse gas emissions (Waldo et al. 2021), and drivers of fish population dynamics (Pennock and Gido 2021). In the face of a hot drought and associated reservoir water level declines (Udall and Overpeck 2017), forebay temperature profiles have also been used to bound possible future outflow temperatures (Mihalevich et al. 2020;Dibble et al. 2021). The CE-QUAL-W2 model has also been used to predict how water consumption and storage decisions will impact temperature outcomes below Glen Canyon Dam, with implications for fishes' thermal constraints (Bruckerhoff et al. 2022).
As lakes and reservoirs are often considered sentinels of climate change (Williamson et al. 2009), we expect this dataset will be valuable for detecting climate-related trends in both physical limnology and reservoir food web structure. Although large lakes and reservoirs may be more buffered from climate change effects due to their greater water volume, 60+ yr of long-term data from the world's most voluminous lake, Lake Baikal, has documented significant lake warming and shifts in food web structure (Hampton et al. 2008). In the world's second-most voluminous lake, Lake Tanganyika, a long-term local record of surface water temperatures was combined with long-term wind data to show a trend of increasing lake stratification strength coupled to reductions in pelagic fishery production (O'Reilly et al. 2003). Thus, larger lentic waterbodies such as Lake Powell are still likely to reflect climatic changes with implications for whole ecosystem function.
Preliminary analyses of long-term trends reveal directional change in several parameters in Lake Powell in recent decades (Fig. 6). This sets the stage to pursue a range of research questions exploring the drivers of change and the ecosystem consequences of shifts in limnological conditions such as warmer water temperatures (Fig. 6A), seasonal changes in water clarity (Fig. 6B), and reduced calcium concentrations (Fig. 6C) to name a few.
Initial data analyses also suggest an increase in surface phytoplankton biovolume (Fig. 6D). This increase in algal biomass may be paired with changes in community composition that could have cascading effects on ecosystem function and water quality management. For example, cyanobacteria genera across all sites have appeared to shift considerably from the beginning of monitoring as compared to recent years (Fig. 7A). These community shifts could translate into changes in algal toxin formation and/or altered edibility for grazers. With growing occurrences of harmful algal blooms (HABs) in freshwater lakes (Paerl et al. 2020), recent reviews are suggesting researchers need to leverage studies undertaken to date to better predict and manage cyano HABs in a changing world (Burford 2020). With over 25 yr of phytoplankton and zooplankton data at multiple sites, some of which have monthly or seasonal collection frequency, the Lake Powell dataset is well positioned to address this growing global concern.
In addition, D. bugensis (Quagga mussels) were first recorded in Lake Powell in 2012 (Figs. 6E, 7B), so this water quality record spans the establishment of non-native bivalves in the reservoir. Plankton data could become extremely valuable in light of the Quagga mussel invasion. For example, zebra mussels have been implicated in promoting Microcystis in non-eutrophic waters elsewhere (Vanderploeg et al. 2001). Other invasive invertebrate and fish species may change the plankton community in the future, which cannot be documented without a well-established preinvasion data collection.
We urge caution when interpreting trends in phytoplankton samples as collection occurred solely at a depth of one meter. As a result, community data may not reflect what could occur at a chlorophyll maximum, likely located deeper into the lake due to the considerable water clarity and the potential for ultraviolet light avoidance at the surface. For zooplankton, there has not been a study to evaluate net clogging (and associated underreporting of biomass), however extremely high biomass volumes have not been qualitatively observed. Overall, the detailed plankton geometries collected in tandem with community counts represent another unique component of this dataset that could be explored further.
Finally, recent successful use of satellite data for estimating chemical (Olmanson et al. 2020) and biological (Kuhn et al. 2020;Topp et al. 2021) properties of lakes and reservoirs suggests that this dataset could support similar use of satellite data for Lake Powell. Lake Powell would be a particularly good Fig. 7. Example of reservoir biological community abundances (cyanobacteria communities (A) and zooplankton communities (B)). Pie charts represent average abundances during spring snowmelt (May/June) sampling at eight commonly visited on-reservoir sampling sites grouped into the earliest 5-yr time frame (1993-1997, left) and most recent 5-yr time frame (2017-2021, right). The eight commonly visited sites represented (LPCR0024, LPCR0453, LPCR0905, LPSJR193, LPCR1169, LPCR1692*, LPCR2085, LPCR2387, and LPESC119) were visited during most but not all years. Panel A shows cyanobacteria community abundances from surface samples only. Cyanobacteria genera that represented < 1% of the overall cyanobacteria abundance for a given time frame were represented as "other." The "other" category for the early time frame consists only of Pseudanabaena. The "other" category for the most recent time frame consists of Anabaenopsis, Aphanocapsa, Gomphosphaeria, Limnothrix, Pseudanabaena, Planktothrix, Spirulina, and Woronichinia. *While the historic StationID for Bullfrog Bay is LPCR1692, the current StationID is LPCR1679; data from both sites is included in this representation.
candidate for informing satellite-based water quality studies given the large number of cloud-free days that this reservoir experiences and its overall size.

Comparison with existing datasets
To our knowledge, this dataset is the longest record of water quality data from a human-made reservoir system. As such, it can fill a unique and important gap in our understanding of global limnology. Thus far, the lack of visibility of the historic version of this dataset has precluded its inclusion in some key syntheses of basic trends in global-and continentalscale limnology. These include trends in lake and reservoir surface water temperatures (O'Reilly et al. 2003;Sharma et al. 2015), mixing regimes (Woolway and Merchant 2019), and salinity (Dugan et al. 2017).
Long-term plankton datasets are particularly limited worldwide due to the technical expertise required for species identification, which increases the cost and time necessary for analysis. Even more rare are the datasets that include both phytoplankton and zooplankton data collected concurrently. The plankton monitoring in Lake Powell predates a similar monitoring approach in nearby Lake Mead by 15 yr (Beaver et al. 2018). Other notable lakes with long-term plankton data are rare, broadly dispersed, and largely contained within more northern latitudes, for example, Lake Baikal in Russia (Hampton et al. 2008), Lake Washington in the northwestern United States (Winder and Schindler 2004), Lake Muggelsee in Berlin, Germany (Gerten and Adrian 2000), and three Cumbrian lakes in the United Kingdom (Thackeray et al. 2015). This dataset and more like it would allow us to compare lentic waterbodies in northern latitudes to arid regions and to compare naturally formed lakes to human-constructed reservoirs.