Satellite-derived cyanobacteria frequency and magnitude in headwaters & near-dam reservoir surface waters of the Southern U.S.

Abstract

Reservoirs are dominant features of the modern hydrologic landscape and provide vital services. However, the unique morphology of reservoirs can create suitable conditions for excessive algae growth and associated cyanobacteria blooms in shallow in-flow reservoir locations by providing warm water environments with relatively high nutrient inputs, deposition, and nutrient storage. Cyanobacteria harmful algal blooms (cyanoHAB) are costly water management issues and bloom recurrence is associated with economic costs and negative impacts to human, animal, and environmental health. As cyanoHAB occurrence varies substantially within different regions of a water body, understanding in-lake cyanoHAB spatial dynamics is essential to guide reservoir monitoring and mitigate potential public exposure to cyanotoxins. Cloud-based computational processing power and high temporal frequency of satellites enables advanced pixel-based spatial analysis of cyanoHAB frequency and quantitative assessment of reservoir headwater in-flows compared to near-dam surface waters of individual reservoirs. Additionally, extensive spatial coverage of satellite imagery allows for evaluation of spatial trends across many dozens of reservoir sites. Surface water cyanobacteria concentrations for sixty reservoirs in the southern U.S. were estimated using 300 m resolution European Space Agency (ESA) Ocean and Land Colour Instrument (OLCI) satellite sensor for a five year period (May 2016–April 2021). Of the reservoirs studied, spatial analysis of OLCI data revealed 98% had more frequent cyanoHAB occurrence above the concentration of >100,000 cells/mL in headwaters compared to near-dam surface waters (P < 0.001). Headwaters exhibited greater seasonal variability with more frequent and higher magnitude cyanoHABs occurring mid-summer to fall. Examination of reservoirs identified extremely high concentration cyanobacteria events (>1,000,000 cells/mL) occurring in 70% of headwater locations while only 30% of near-dam locations exceeded this threshold. Wilcoxon signed-rank tests of cyanoHAB magnitudes using paired-observations (dates with observations in both a reservoir’s headwater and near-dam locations) confirmed significantly higher concentrations in headwater versus near-dam locations (p < 0.001).

Graphical abstract

1. Introduction

Modern hydrologic networks include a combination of natural, modified, and artificial features. In the continental United States (CONUS), over 91,000 larger reservoirs and hundreds of thousands of smaller constructions are present throughout the landscape (Smith et al., 2002; Ignatius and Stallins, 2011; National Inventory of Dams (NID), 2018). These constructions were created for multiple purposes and provide important services such as water supply, flood control, navigation, and recreation (Zhou et al., 2016). Within the U.S., widespread reservoir construction is a relatively recent phenomenon with smaller mill dam creation of the 1800s followed by construction of much larger hydroelectric, flood control, and water supply dams during the mid 1900s (Ho et al., 2017). While creation of the largest dams has declined in the U.S. over the last few decades due to knowledge of environmental impacts and fewer available sites, proliferation of moderate-sized and small-sized reservoirs continues today (Grill et al., 2015). These artificial constructions function differently than natural lakes and create a series of unique engineering and ecological challenges including dam safety, sediment management, stream fragmentation, and water quality modification. Reservoirs contain approximately 10% of the total worldwide lake storage and given their abundance they are understudied ecosystems with vital importance for global sustainability goals (Guo et al., 2021).

Individual water bodies exhibit distinct limnological properties based on hydrology, geology, climate, land use, and other variables. When possible, lake and reservoir ecology should be independently examined for each unique water body based on local environmental conditions. However, there are several morphological and ecological characteristics that typically apply to artificially constructed reservoirs. Compared to natural lakes, reservoirs generally have a longer shoreline, shorter water residence time, larger catchment area, shallower photic zone, and warmer benthos (Havel and Pattinson, 2004; Hayes et al., 2017). As a result of sediment accumulation, shallow reservoir conditions allow light to reach benthic sediment, causing higher coverage of submerged aquatic vegetation and stronger sediment-water column interactions (Shivers et al., 2018). In addition, reservoirs have unique impacts on the hydrologic network. Reservoirs interrupt stream connectivity, impeding movement of aquatic species, isolating populations of aquatic communities, and limiting geneflow (Freeman and Marcinek, 2006; Freeman et al., 2007; Chappell et al., 2019). Reservoirs also modify evaporation rates and affect downstream water quality by sequestering nutrients and altering water temperature (Ignatius and Rasmussen, 2016; Ignatius and Jones, 2018).

Algal and cyanobacterial populations play an important ecological role as a primary producer, carbon sink, and oxygen source within lakes and reservoirs. However, large cyanobacteria blooms (cyanoHAB) significantly impact human health, the economy, and the environment (Clark et al., 2017). Human exposure to cyanobacteria is problematic because some of the more prevalent species may produce microcystins, anatoxins, saxitoxins, cylindrospermopsin, and other toxins (Merel et al., 2013). These toxins can affect the skin, liver, kidneys, reproductive system, and central nervous system (USEPA, 2014; Wilde et al., 2014; Breinlinger et al., 2021). The significant impact of cyanoHABs is exemplified by large-scale bloom events such as the drinking water supply crisis in Lake Erie (Obenour et al., 2014; Bertani et al., 2016) and Lake Okeechobee (Kramer et al., 2018). In addition, cyanobacteria blooms harm the economy by causing fishery closures, limiting access to recreation areas, and affecting housing values for shoreline communities (Graham et al., 2017).

Reservoirs may contain a combination of riverine and lacustrine limnological conditions located throughout different parts of the water body and provide unique environments for cyanobacteria growth (Graham et al., 2008). Shallow reservoir drawdown can rapidly decrease water depth, leading to warmer water temperatures and subsequent increases in biomass and cyanobacteria relative abundance (Bakker and Hilt, 2016). Some cyanobacteria, such as Anabaena and Microcystis, can preferentially alter buoyancy based on environmental conditions. Buoyancy regulation enables cyanobacteria movement up and down the water column to gain access to nutrients and optimize sunlight exposure (Reynolds et al., 1987; Mur et al., 1999). Reservoir headwaters are often extremely shallow due to sediment accumulation, restricting cyanobacteria vertical movement. In contrast, near-dam areas are typically deeper and cyanobacteria may move to lower depths, potentially diluting concentrations through dispersal throughout the water column. Compared to surface water blooms, cyanobacteria in deeper portions of the water column may pose a relatively lower exposure risk to some recreational users such as boaters. Cyanobacteria blooms can also impact water quality downstream from reservoirs. Dam releases can send cyanobacteria-laden waters downstream and the cyanobacteria can then travel long distances along rivers (Otten et al., 2015; Williamson et al., 2018). Reservoirs also often exhibit a longitudinal zone from river inflows to the deeper lacustrine zone near the dam, promoting cyanobacteria growth in the transition zone (Kortman, 2015). As the spatial distribution of cyanobacteria can vary substantially within a reservoir, additional analysis and knowledge about the spatial distribution of cyanoHABs is crucial to help focus management actions (Graham et al., 2008). Integration of cyanobacteria spatial heterogeneity has been long-recognized as essential to inform scientifically based reservoir management (Moreno-Ostos et al., 2006). In situ studies on individual lakes have found high spatial variance in the concentration of cyanobacteria across the surface of an individual water body (Havel and Pattinson, 2004).

Regional environmental conditions in the southern U.S. provide optimal conditions for cyanobacteria growth. The warm climate allows for year-round persistence of algae and cyanobacteria and a variety of cyanobacteria species are present in the region. For example, Lake Houston, Texas contains at least 26 known cyanobacterial genera (Beussink and Burnich, 2009). Cyanobacteria are monitored in regional reservoirs using seasonal field sampling methods, often at either one or two locations in the reservoir. Some select Texas reservoirs with documented cyanobacteria, such as Lake Waco, Lake Whitney, and Lake Houston, have been continuously monitored using deployed real-time in-lake sensors (Kiesling et al., 2008; Beussink and Graham, 2011). Mobile sensors can detect physicochemical properties and fluorescence monitoring with an in vivo fluorometer (IVF) sensor estimates chlorophyll concentrations in real-time. However, this real-time data collection is not deployed consistently and is not implemented at most reservoir sites due to the high financial cost (Ogashawara et al., 2013).

To better understand the spatial and temporal patterns of cyanoHAB occurrence, traditional cyanobacteria in situ monitoring may be augmented with remotely sensed satellite, airborne, and unmanned aerial systems (UAS) data to provide broader spatial coverage and higher temporal resolution (Vincent et al., 2004; Hunter et al., 2008; Li et al., 2012; Matthews et al., 2012; Ogashawara et al., 2013; Olmanson et al., 2015; Pyo et al., 2018; Kwon et al., 2020). There has been extensive demonstration of satellite-based methods to detect cyanobacteria in reservoirs and lakes (Simis et al., 2005; Hu et al., 2010; Wynne et al., 2010; Binding et al., 2012; Mishra et al., 2013; Kudela et al., 2015; Shi et al., 2017). These studies have shown that satellite sensors can be used to detect and quantify proxies of cyanobacteria biomass (Kutser, 2009). Satellites have been demonstrated to complement in situ methods with broader spatial and temporal monitoring in addition to potentially saving time, labor, and cost required to routinely monitor a large number of systems (Papenfus et al., 2020). As a result, satellites are now generally recommended for monitoring cyanobacteria in guidance documents from the World Health Organization (Chorus and Welker, 2021), Interstate Technology Regulatory Council (ITRC, 2021), and revised American Water Works Association M57 source to treatment manual. For example, estimates of cyanobacteria cell concentration in surface waters can be generated using the ESA MEdium Resolution Imaging Spectrometer (MERIS) and OLCI satellite sensor data (Schaeffer et al., 2019). While cyanobacteria estimates are limited to the first optical layer of the water column, the data provide a valuable assessment of surface water cyanobacteria concentrations. In addition, OLCI imagery is globally available at a 300 m pixel resolution with a temporal resolution of 2–3 days with a single satellite and 1–2 days with two satellites, allowing for broad application. When spatial distributions are examined, researchers typically investigate spatial dynamics within a single water body to gain insight about local cyanoHAB trends. Kwon et al. (2020) examined the spatial distribution and vertical migration of cyanobacteria using drone-based hyperspectral imaging within the Daechung Reservoir. Pyo et al. (2018) estimated the spatial distribution and dominant locations of phycocyanin in the Baekje Reservoir of South Korea using airborne hyperspectral imagery. However, use of satellite imagery to investigate small-scale variability generally remains limited for inland waters (Yan et al., 2018; Dev et al., 2022) and has not been comparatively analyzed across dozens of reservoir sites.

We analyzed sixty large reservoirs in the U.S. south (Karl and Koss, 1984) to examine trends within a climatologically consistent region experiencing active cyanoHABs. This research provides a unique perspective by comparing cyanoHAB spatial patterns within multiple reservoir sites. Research questions include:

• 1)Do cyanoHABs occur more frequently in reservoir headwaters compared to near-dam surface water locations?
• 2)Are cyanobacteria magnitudes higher in reservoir headwaters compared to near-dam surface water locations and does this vary seasonally?
• 3)Which of the study area reservoirs have the highest magnitude and most frequent cyanobacteria blooms?

2. Data and methods

2.1. Remote extraction of reservoir headwaters and near-dam surface waters

For the purposes of this study, analysis examined cyanoHABs in anthropogenic reservoirs in the southern U.S. states of Arkansas, Louisiana, Mississippi, Oklahoma, and Texas. OLCI imagery 300 m pixel size limited investigation to larger reservoirs and excludes smaller waterbodies below the minimum 900 m × 900 m surface area threshold. Small anthropogenic ponds in agricultural or urban areas are common sites for cyanoHABs due to relatively high nutrient loads, warm summer temperatures, and shallow waters (Kozak et al., 2019). However, small water bodies were not included in this analysis due to the spatial resolution limitations and potential for spectral mixing caused by terrestrial land cover at small reservoir sites. In addition, this research aimed to investigate patterns in headwater and near-dam reservoir locations and therefore excluded natural lakes. To eliminate natural lakes and identify larger reservoirs, we spatially intersected National Hydrography Dataset (NHD) polygons (Urquhart and Schaeffer, 2020) with NASA Global Reservoir and Dam (GRanD) database locations (Lehner et al., 2011). To compare cyanobacteria characteristics within reservoir headwaters and near-dam surface waters, water bodies with definable upstream and downstream regions were identified. Based on these criteria, a total of 60 reservoirs were selected (Fig. 1).

Fig. 1.

Map showing locations of sixty study area reservoirs in the U.S. South (left). Example of reservoir boundary creation for Lake Tawakoni Reservoir (right). Potential land interference was mitigated using a 450 m negative buffer. Potential ephemeral dry areas were excluded from analysis using JRC Water Occurrence data. Lastly, reservoir boundaries were divided into two equal-area polygons: headwaters and near-dam.

For remote sensing analysis, it is crucial to remove land from reservoir boundaries as reflectance from land surfaces produces an edge-effect and skews open water cyanobacteria concentration estimates at near-shore areas (Urquhart and Schaeffer, 2020). We used a 450 m negative buffer to exclude land from reservoir boundaries. In addition to removing land, ephemeral waters were excluded from the study, including periodically dry regions of a reservoir. To exclude these ephemerally dry areas, we utilized the Landsat-derived Joint Research Center (JRC) Global Surface Water Occurrence data, version 2 (Pekel et al., 2016). Lastly, the 60 reservoir boundaries were manually reviewed and edited in ArcGIS Pro 2.7 using ESRI World Imagery based on MAXAR Technologies 2021 imagery to ensure polygon quality and edit out any potential influence from small islands. The combined surface area of all 60 reservoirs totaled 2124 km². While this approach eliminated smaller water bodies and minimized the water surface area used in the study, the remaining reservoir boundaries were free from near-shore edge effects and allowed for robust analysis of cyanobacteria spatiotemporal patterns within each reservoir.

Reservoir boundaries were split into headwater and near-dam polygons based on proximity to the dam to quantify if cyanoHABs occur more frequently and in higher concentrations in reservoir headwaters compared to near-dam surface water locations. Dam locations were identified using the GRanD point data and manually verified. Segmenting reservoirs into near-dam and headwater areas based on proximity to the dam provides an easily deployable strategy to target potential cyanoHAB problem areas and can be rapidly applied to large study areas. We utilized the Subdivide Polygon tool in ArcGIS Pro 2.7 to divide each reservoir into two equal-area parts and categorized these segments into two groups: 1) Headwaters: areas closer to mainstem river inflows; and 2) Near-Dam: areas closer to the reservoir dam (Fig. 1). To examine the differences in cyanoHAB frequency and magnitude in headwater vs. near-dam reservoir surface waters, a simple method of dividing in half is applied in this study but other approaches of dividing into thirds or quarters would also be valid. The most straightforward method is applied here for ease of application and potential reuse. A simple strategy would be particularly useful in data-poor regions that lack the detailed bathymetric and limnological information that may otherwise guide subdivision of unique reservoir systems.

2.2. Satellite derived cyanobacteria

Representation of cyanobacteria biomass requires within week observations since phytoplankton dynamics occur on daily to weekly time scales (El Serafy et al., 2021). Therefore, the Sentinel-3 OLCI sensor was selected over other satellite sensors, such as Landsat, Sentinel-2 MSI, VIIRS or MODIS, because it was the most highly suitable combination of frequency of revisits, spatial resolution, necessary spectral bands, and radiometric sensitivity to detect cyanobacteria across inland reservoirs (IOCCG, 2018). Mission-long data records for OLCI are processed and distributed for the United States by the NASA Ocean Biology Processing Group at Goddard Space Flight Center through a data sharing agreement between NASA and ESA (Seegers et al., 2021). Briefly top-of-atmosphere (Level-1B) OLCI data were processed to Level-2 imagery by removing the contribution of spectral Rayleigh scattering from the top-of-atmosphere signal, where the Rayleigh-corrected top-of-atmosphere reflectance is $\rho \text{s}(\lambda )$. Quality control masks were applied to exclude erroneous Level-2 data including clouds and cloud shadow (Wynne et al., 2018). A high resolution (~60 m) land mask based on the NASA Shuttle Radar Topography Mission Water Body Data Shapefiles (NASA JPL, 2013) was used, with modifications by Urquhart and Schaeffer (2020) to correct for embedded inaccuracies in that data set, such as missing lakes and reservoirs.

The distinct spectral reflectance curve of cyanobacteria and green algae support robust remote sensing research applications. While cyanobacteria and green algae both exhibit similar high reflectance in near infrared, cyanobacteria exhibit unique scattering and absorption patterns in specific electromagnetic wavelengths such as high pigment absorption at 620 nm (Wynne et al., 2008; Mishra et al., 2009). Based on these reflectance properties, a spectral shape algorithm was developed to detect cyanobacteria biomass concentrations per image pixel. The cyanobacteria index (CI-cyano) was calculated using a spectral shape curvature method, as originally described in Wynne et al. (2008), updated in Lunetta et al. (2015), and algorithm progression fully detailed in Coffer et al. (2020).

The spectral shapes used in CI-cyano are calculated using the following equation:

$$SS(\lambda )={\rho }_s(\lambda )-{\rho }_s\left({\lambda }^{-}\right)-\left\{{\rho }_s\left({\lambda }^{+}\right)-{\rho }_s\left({\lambda }^{-}\right)\right\}*\frac{\left(\lambda -{\lambda }^{-}\right)}{\left({\lambda }^{+}-{\lambda }^{-}\right)}$$

where $\text{SS}$ is the spectral shape, $\text{ps}$ is Rayleigh-corrected surface reflectance, $\lambda$ is the central spectral band of interest, and ${\lambda }^{+}$ and ${\lambda }^{-}$ are the adjacent reference spectral bands above and below the central spectral band, respectively. Wynne et al. (2008) used this equation to assess cyanobacteria presence with $\lambda$ as 681 nm (nm), ${\lambda }^{+}$ as 709 nm, and ${\lambda }^{-}$ as 665 nm. If the fluorescence band (681) falls below the baseline between ${\lambda }^{+}$ and $\lambda$, cyanobacteria is likely present. Lunetta et al. (2015) updated the algorithm based on Matthews et al. (2012) to address potential false positives using a derivative that includes the 620 nm band, which is sensitive to phycocyanin, where SS(665) with $\lambda =665\text{nm}$, ${\lambda }^{+}=681\text{nm}$ nm, and ${\lambda }^{-}=620\text{nm}$. The CI-cyano uses the SS(681) to estimate biomass and the SS(665) as an exclusion criterion, where SS(665) < 0, cyanobacteria are presumed absent and SS (665) > 0 cyanobacteria are presumed present.

The CI-cyano algorithm may be converted to cyanobacteria abundance (cells/mL) as described in Lunetta et al. (2015). In addition to MERIS-derived estimates, the algorithm can be applied to provide cyanobacteria concentration estimates using the OLCI sensor (Coffer et al., 2021a). The data has been made publicly available to support public access and use in scientific analysis (https://oceancolor.gsfc.nasa.gov/projects/cyan/). This algorithm was previously validated across the U.S. (Lunetta et al., 2015; Tomlinson et al., 2016; Schaeffer et al., 2018; Coffer et al., 2020; Mishra et al., 2021). The CI-cyano algorithm was validated against state microcystin data with 84% accuracy with ~90% positives correctly classified (Mishra et al., 2021). CI-cyano was validated against national chlorophyll from WQP with a multiplicative bias of 1.11 (11%) and mean absolute error of 1.60 (60%) (Seegers et al., 2021). In addition, comparison against Unregulated Contaminant Monitoring Rule visual observations of blooms satellite-derived cyanobacteria and qualitative sampler responses had 94% agreement (Coffer et al., 2021b). National satellite-derived phenological patterns were confirmed against previously published cyanobacterial ecological studies (Coffer et al., 2020). The algorithm has also been documented across 25 state health advisories in California, Oregon, New York, Idaho, New Jersey, Utah, and Vermont (Schaeffer et al., 2018a, 2018b); and used by the Wyoming Department of Environmental Quality for recreational advisories (e.g. Wyoming, 2018a, 2018b, 2018c).

2.3. Google Earth Engine for cyanoHAB frequency & magnitude assessment

The OLCI CI-cyano version 3 data were processed for a five-year time period from May 1, 2016-April 30, 2021. This included 1826 daily images with a mix of cloud free, partially cloudy, and no-data images. The CI-cyano geotiffs were uploaded as image collection assets within Google Earth Engine (GEE) using the GEE python application programming interface (API). GEE provides a robust and rapid image processing computer infrastructure (Gorelick et al., 2017). The GEE integrated development environment (IDE) allows direct access to an easy to use API for executing computationally intensive GIS operations. API functions are grouped to generate computational graphs which are then executed in the Google Cloud infrastructure, where the data used by those graph functions already exists in distributed databases and any uploaded datasets have already been uploaded as cloud resources. These features provided rapid processing of the spatial aggregation functions required for this project. The reservoir boundary headwaters and near-dam shapefile polygons were also imported into GEE as vector assets.

Javascript code extracted pixels with valid cyanobacteria observation data and excluded land pixels, no-data pixels such as clouds. As some satellite images contained clouds or no data, the number of valid observation dates were counted for each pixel over the entire study period. For all dates with valid cyanobacteria observation data available, the frequency of high cyanobacteria concentrations was calculated for each pixel using the World Health Organization (WHO) threshold values of high concentration cyanobacteria equal to or greater than 100,000 cells/mL, medium concentrations cyanobacteria ranging from 20,000–99,999 cells/mL, and low concentrations below 20,000 cells/mL (Chorus and Bartram, 1999; WHO, 1999; Mishra et al., 2019) as described in Coffer et al. (2021b). The resulting raster geotiff output was the frequency of cyanoHAB occurrence of each pixel per observation period, with a value of 0 indicating no occurrence and a value of 1 indicating cyanoHAB occurrence in 100% of observations. Lastly, to evaluate seasonal fluctuations, mean temperature was calculated for meteorological summer months (June, July, August) and winter months (December, January, February) (Schaeffer et al., 2018a, 2018b; Cook et al., 2014).

To examine the difference in headwaters and near-dam cyanobacteria magnitude within each reservoir, a GEE map reduce percentile function was applied to reservoir boundaries to calculate minimum, second quartile, median, third quartile, and maximum cyanoHAB magnitude for headwaters and near-dam polygons each day. The daily time series were calculated for all sixty reservoirs using each headwater and near-dam polygon for a total of 120 polygon boundaries.

Finally, to summarize regional trends in headwaters vs. near-dam locations, all headwater polygons were dissolved to create a single “combined-headwaters” multipart polygon. Similarly, all near-dam polygons were dissolved into a single “combined-near-dam” multipart polygon. The two polygons (combined-headwaters and combined-near-dam) were equal in size and each included 1062 km2 of surface water, for a total surface area of 2124 km2 across the entire study area. The GEE map reduce percentile function was applied to calculate minimum, second quartile, median, third quartile, and maximum cyanoHAB concentrations for the combined-headwaters and combined-near-dam boundaries to provide a summary of trends throughout the region. The larger reservoirs contributed more pixels to the combined boundary datasets due to their larger surface area. However, the merged boundary statistics provide a simplified comparison of patterns in headwaters and near-dam surface waters across the study area.

2.4. Statistical analyses

Time series of percentiles were analyzed to compare frequency and magnitude of cyanobacteria between the headwaters and near-dam locations. Cyanobacteria median magnitude statistics (discussed in Section 2.3) were calculated for each reservoir and each observation date. Dates without observed data above threshold detection limits for both the headwater and near-dam location were omitted to compare cyanobacteria concentrations within each reservoir. For example, if cloud cover obstructed satellite observation for the water body headwaters, the associated near-dam reservoir data was not considered. As needed for statistical tests, a proxy value for non-detects of lowest observed concentration in the dataset divided by the square root of two was used to ensure that relevant pairwise daily differences could be calculated.

A frequency-based summary statistic was created by calculating the percent of high concentrations at each reservoir across all the days. For headwater and near-dam locations at each reservoir, this percentage is based on the number of observation dates with a detected concentration defined as >100,000 cells/mL divided by the total number of observation dates (Clark et al., 2017). A correlation analysis on these high concentration percentages was then performed across the 60 reservoirs. Frequency-based high concentration exceedance percentages were also calculated at the pixel level to support high-resolution mapping.

Magnitude-based analyses were performed by evaluating the pairwise cyanobacteria concentrations at the daily level for days where cyanobacteria concentrations were detected. For daily comparison purposes, pairwise maximum concentrations and quartile values were contrasted between headwater and near-dam locations. For significance testing purposes, non-parametric Wilcoxon signed-rank tests were performed by comparing daily median concentrations within each reservoir using the wilcox.test from the R stat package (R Core Team, 2021). One-sided tests were performed over the entire time period but also broken out by season. Since this is a paired test implementation, the null hypothesis is that the difference between the headwater and near-dam concentrations is zero or negative, with the alternative hypothesis being that the headwater concentrations are greater than the near-dam concentrations. Additionally, the clusrank package (Jiang et al., 2020) was used to perform the Wilcoxon signed-rank test across all locations, treating the reservoirs as clusters and providing inference across all the spatial locations included in the data set. This clusrank package implements a corrected variance formula (Rosner et al., 2006) that adjusts for the correlation structure between the headwater and near-dam time series within each of the reservoirs.

3. Results and discussion

Certain constraints and limitations must be noted for this study. Many species of cyanobacteria have preferential buoyancy and can float up and down the water column to seek optimal light conditions. This would lower the detectable concentration at the water surface, especially in deeper near-dam locations. The cyanobacteria concentration estimates provided here must be interpreted only for the surface water, or first optical layer, and does not capture the presence of potential cyanobacteria in deeper portions of the water column or along the shorelines. In addition, this study excluded multiple water bodies such as natural lakes and reservoirs below a minimum size based on OLCI 300 m pixel size.

3.1. CyanoHAB frequency

Of the 60 reservoirs examined, 59 experienced more frequent high concentration (>100,000 cells/mL) cyanoHABs in headwaters compared to near-dam surface waters (Fig. 2). Notably, all sixty reservoir sites had detectable high concentration blooms observed. The most active reservoir, Cedar Creek Reservoir, Texas, contained detectable high concentration cyanoHABs during >92% of observations. Additional Texas reservoirs with frequent high concentration cyanoHABs include Lake Palestine, Corder Lake, and Woodlands Golf Course Lake (Table 1); high concentration cyanoHABs were detected at each of these reservoirs on more than 84% of observation dates. Identification of these reservoir locations with higher cyanoHAB magnitudes can help guide monitoring efforts. For example, while some Texas reservoirs are regularly monitored (Kiesling et al., 2008; Gamez et al., 2019), they may not be sampled at headwater locations where more high magnitude cyanoHABs occur.

Fig. 2.

Scatterplot comparing the frequency of high concentration cyanoHABs in headwater vs. near-dam surface waters for 60 reservoirs (R² = 0.57). Frequency is the number of observation dates with high concentration cyanobacteria divided by the total number of observation dates. Reservoirs with more frequent high concentration cyanobacteria in headwater locations shown in dark grey (59 reservoirs). Reservoirs with more frequent high concentration cyanobacteria in near-dam locations shown in light grey (1 reservoir). 1:1 trendline included for reference.

Table 1.

Frequency of high concentration observations (>100,000 cell/ml), maximum cyanobacteria concentrations (cells/mL), and non-parametric Wilcoxon signed rank (WSR) test p values for paired daily data. Data are for headwaters and near-dam locations for sixty reservoirs for the five year study period (May 1, 2016-April 30, 2021). The last column indicates statistically significant elevation of headwater versus near-dam cyanobacteria concentrations when less than 0.05. Only N. Lake Texoma contradicts the trend of higher headwater cyanobacteria concentrations and only Arkabutla Lake shows a (slightly) higher frequency of observations greater than 100,000 cells/mL.

Reservoir Name	Headwaters Frequency > 100,000 cells/mL	Near-Dam Frequency > 100,000 cells/mL	Headwaters Max cells/mL	Near-Dam Max cells/mL	Paired headwater vs. near-dam cyanoHAB magnitude p value
Cedar Creek Reservoir	92%	75%	1,674,943	990,832	1.43E-91
Corder Lake	84%	31%	1,721,869	672,977	3.2E-120
Lake Palestine	84%	70%	1,106,624	862,979	7.56E-75
Lake Limestone	82%	34%	1,306,171	816,582	8.7E-98
Woodlands Golf Course Lake	82%	45%	2,754,229	2,398,833	5.73E-118
Cross Lake	81%	75%	1,076,465	963,829	1.88E-129
Falcon Reservoir	78%	70%	1,235,947	816,582	3.15E-84
Navarro Mills Lake	76%	73%	1,235,947	1,202,264	1.2E-09
S Lake Eufaula	74%	55%	1,629,296	1,047,129	3.08E-43
Lake Tawakoni	72%	29%	1,202,264	794,328	7.33E-104
Cypress Lake	71%	66%	1,629,296	1,458,814	0.00000156
Robert S Kerr Reservoir	71%	28%	1,270,574	619,441	2.73E-82
Lavon Lake	69%	50%	1,819,701	1,169,499	8.95E-92
Bardwell Lake	68%	53%	1,499,685	1,202,264	2.98E-76
Richland Chambers Reservoir	68%	25%	1,169,499	554,626	8.22E-94
Lake Conroe	67%	39%	887,156	496,592	1.68E-78
South Lake	66%	53%	990,832	887,156	1.11E-18
Toledo Bend Reservoir	66%	14%	1,047,129	2,831,392	1.66E-102
Somerville Lake	65%	38%	1,419,058	816,582	4.51E-73
N Lake Texoma	64%	50%	1,458,814	2,466,039	0.982
Wright Patman Lake	64%	60%	1,306,171	839,460	4.4E-29
Copan Lake	62%	50%	1,076,465	839,460	4.69E-24
Wister Lake	60%	51%	2,535,129	794,328	1.23E-7
Lake Thunderbird	58%	34%	1,137,627	862,979	1.36E-61
Lake Lewisville	57%	9%	1,380,384	636,796	1.87E-86
Lake O′ the Pines	57%	42%	751,623	672,977	3.58E-54
Ester Lake	56%	8%	816,582	444,631	3.66E-95
Case Lake	54%	14%	912,011	524,807	1.85E-70
Lake Houston	54%	40%	1,584,893	1,499,685	1.35E-33
Waco Lake	54%	43%	1,169,499	862,979	3.16E-24
Horseshoe Tank	53%	36%	3,250,873	794,328	4.4E-32
Lake Murvaul	53%	34%	963,829	816,582	1.44E-27
Bayou D’Arbonne Lake	52%	40%	1,076,465	751,623	6.44E-17
Fort Cobb Reservoir	51%	44%	1,923,092	2,089,296	1.01E-28
Oologah Lake	51%	16%	912,011	496,592	1.15E-59
Waurika Lake	49%	9%	1,584,893	963,829	1.19E-75
Arkabutla Lake	48%	50%	2,147,830	1,342,765	0.005
Eagle Mountain Lake	46%	16%	1,137,627	731,139	1.06E-56
Enid Lake	46%	36%	1,629,296	862,979	7.28E-38
Granada Reservoir	46%	27%	2,992,265	2,754,229	2.93E-45
Lake Erling	46%	26%	1,137,627	751,623	1E-15
Canton Lake	42%	36%	937,562	772,681	6.79E-36
Grand Lake O the Cherokees	42%	12%	1,076,465	816,582	5.77E-43
Keystone Lake	42%	14%	1,306,171	636,796	8.9E-45
Sardis Lake	41%	30%	5,970,353	6,309,573	9.21E-19
Joe Pool Lake	39%	9%	990,832	483,059	1.05E-58
Lake Dardanelle	35%	18%	963,829	554,626	1.68E-19
Pat Mayse Lake	33%	6%	751,623	539,511	2.92E-39
Sardis Lake	33%	22%	672,977	772,681	1.3E-17
Tom Steed Reservoir	33%	23%	839,460	602,560	4.88E-32
Martin Lake	31%	28%	772,681	772,681	2.23E-11
Grapevine Lake	30%	9%	937,562	496,592	9.99E-37
Kaw Lake	30%	3%	1,018,591	310,456	3.2E-58
Fort Gibson Lake	23%	7%	839,460	469,894	6.23E-28
Lake Arrowhead	22%	18%	1,819,701	1,976,970	0.034
Lake Claiborne	22%	4%	816,582	319,154	2.68E-26
Lake Whitney	19%	4%	937,562	2,089,296	1.14E-26
Ray Roberts Lake	19%	6%	751,623	539,511	7.37E-36
Foss Reservoir	18%	11%	691,831	570,164	6.67E-20
Monticello Reservoir	16%	4%	457,088	432,514	7.89E-26

Reservoirs had a positive correlation (R2 = 0.57) between cyanoHAB frequency in headwaters and near-dam areas (Fig. 2). However, some reservoirs showed more divergent patterns between headwaters and near-dam areas. For example, Ester Lake, Texas had relatively lower cyanobacteria concentrations overall, with detectable high concentration cyanobacteria in near-dam locations only 8% of observation dates. However, Ester Lake had high concentration blooms in small, isolated areas in the upper reaches of the headwaters 56% of the time.

Maps display the spatial distribution of high concentration cyanobacteria (>100,000 cells/mL) by indicating the frequency of cyanoHAB occurrence for each pixel cell (Fig. 3). The blue pixel colors indicate less frequent cyanoHABs while the yellow colors indicate more frequent cyanoHABs. It should be noted that cyanoHABs in near-shore regions are not captured in these maps as pixels outside of the reservoir polygon boundaries were excluded from analysis due to potential mixed-pixel effects at the shoreline. Future analysis of higher spatial resolution imagery and in situ monitoring of these near-shore locations is essential for full understanding of the spatial heterogeneity of cyanoHAB occurrence and frequency.

Fig. 3.

Maps showing the frequency of high concentration cyanobacteria for 12 of the 60 studied reser-voirs. The 12 reservoirs mapped here serve as a visual sample with shallower upstream portions of reser-voirs show more frequent high concentration blooms. The surface waters of deeper near-dam waters tend to have less frequent high concentration cyanobacteria observed.

3.2. CyanoHAB magnitude

A total of 1547 observations were recorded for dates with cyanobacteria magnitudes in both a reservoir’s headwater and near-dam polygon boundaries. Comparison of the paired-observation data (headwater and near-dam data for a single reservoir on a single date) demonstrates consistently higher magnitudes and quartile values in headwater locations (Fig. 5). Compared to near-dam locations, daily maximum concentrations were higher in headwater locations for 86% of observation dates (Fig. 5a). The same trend also held true for quartiles, with cyanobacteria concentrations in the headwaters higher 81% (Q3), 76% (median), and 72% (Q1) of observations (Fig. 5b, c, d).

Fig. 5.

Scatterplot comparing cyanobacteria concentrations for dates with paired headwaters-near dam observations for 60 reservoirs (total of 1547 observations): maximum value (a), 3rd quartile (b), median (c), and first quartile (d). Observation dates with greater cyanobacteria in headwater locations shown in dark grey. Observations dates with greater cyanobacteria in near-dam locations shown in light grey. 1:1 trendline included for reference.

Over the entire study period, 86% of reservoirs had higher maximum concentration values in headwaters. Table 1 shows that all reservoirs, except Lake Arkabutla, experienced more frequent high concentration cyanobacteria in headwaters. Additionally, Table 1 shows that all other reservoirs, except N Lake Texoma, demonstrated higher maximum concentrations and statistically significantly higher pairwise concentrations in headwaters via one-sided Wilcoxon signed-rank tests. Similarly, an aggregated Wilcoxon signed-rank test that treats the reservoirs as clusters also demonstrated higher headwater magnitudes (p < 0.001). The high concentrations of cyanobacteria in headwater locations may have important consequences for downstream aquatic health as cyanotoxins produced in headwater blooms can move downstream (Graham et al., 2012).

The spatiotemporal variability of high magnitude cyanoHABs is recognized in freshwater environments (Bowling et al., 2015; Brooks et al., 2016). For example, phycocyanin concentrations often exhibit extreme spatial heterogeneity within water bodies as wind and water currents can concentrate buoyant cyanoHAB patches (Dev et al., 2022). However, field-based water sampling efforts often rely on sample collections at a single reservoir location in the middle of the water body (Gamez et al., 2019). The identification of higher magnitude cyanoHABs in headwater locations conducted in this study can help prioritize future field-based monitoring efforts.

3.3. Environmental driving factors

Several variables likely contribute to the more frequent high concentration cyanoHABs in headwaters compared to dam locations. Most reservoirs are characterized by a longitudinal gradient with increased nutrient concentrations, turbidity, and algal biomass in headwaters and decreased concentrations near the dam (Conrad, 1986; Sneck-Fahrer et al., 2005; Beussink and Graham, 2011). This analysis demonstrated that cyanoHAB frequency also followed a longitudinal pattern with high concentration cyanobacteria detected more often in headwater locations. The spatial differences within reservoirs may occur due to nutrient inputs entering the reservoir from inflowing streams and becoming deposited in reservoir headwaters. The shallow depth in headwater locations also promotes algae and cyanobacteria growth (Scheffer et al., 1997; Shivers et al., 2018).

Previous studies have identified warm winter maximum temperatures as a predictor of cyanobacteria cell densities (Weber et al., 2019). For this study area, the warm southern U.S. climate allows all reservoir sites to remain above freezing throughout the winter season, promoting year-round cyanobacteria photosynthesis. This year-round pattern of cyanoHAB biomass was also identified in Coffer et al. (2020) for the southeastern U.S. and by USGS research in Lake Houston, Texas (Beussink and Graham, 2011). Temperature differences exist within the reservoirs, as well. Landsat 8 Level 2, Collection 2, Tier 1 surface temperature measurements over the five year study period were processed within GEE to calculate the mean temperature for each pixel within the study area. Reservoir temperatures fluctuate throughout the year with a headwater winter mean of 10 °C (near-dam winter mean 10.46 °C) and headwater summer mean of 30.46 °C (near-dam summer mean 29.92 °C). The range of annual surface temperatures in headwaters is likely caused by shallow water depths in these locations and cause headwaters to generally be hotter than near-dam areas in summer but colder than near-dam areas in winter (Fig. 4). More frequent cyanobacteria blooms and warmer summer temperatures in headwater locations have implications for cyanotoxins as higher temperatures may trigger microcystin release by toxin-producing cyanobacteria (Walls et al., 2018).

Fig. 4.

Landsat 8 Level 2, Collection 2, Tier 1 five year mean surface temperatures during summer months (June–August) and winter months (December–February), years 2016–2021. Lavon Lake (top) and Quarter Lake (bottom).

3.4. CyanoHAB regional time series

To summarize regional time series in headwaters vs. near-dam areas, all headwater polygons were combined into a single multipart polygon and compared with a multipart polygon containing all near-dam boundaries. For the five year study period (May 1, 2016-April 30, 2021), regional time series analysis of daily pixel values in the sixty study area reservoirs revealed different responses for headwater locations compared to near-dam locations (Fig. 6). For the combined-headwaters, observed cyanoHABs had a daily median cyanobacteria magnitude exceeding the concentration threshold of 100,000 cells/mL for 65% of observations. In contrast, when looking at the near-dam areas combined, only 40% of median cyanoHAB magnitudes exceeded this threshold.

Fig. 6.

Daily cyanobacteria concentrations (cells/mL) in combined headwaters locations and combined near-dam locations (log scale). The orange dashed line indicates the WHO high concentration threshold (>100,000 cells/mL) and the red line indicates extremely high concentration cyanobacteria concentrations (>1,000,000 cells/mL).

Daily maximum cyanobacteria magnitudes also showed elevated values in reservoir headwaters. Thirty percent of daily maximum observations in the headwaters exceeded the extremely high concentration of 1,000,000 cells/mL. This extremely high cyanobacteria concentration was only identified for 3% of the observations within near-dam locations.

Regionally, cyanoHABs exhibit expected seasonal trends based on previously established ecological patterns (Coffer et al., 2020). Freshwater cyanoHABs are often caused by diazotrophic cyanobacteria (Beversdorf et al., 2013) and cyanobacteria concentrations typically increase during the warm months as nitrogen depletion limits growth by other phytoplankton and nitrogen fixation becomes more advantageous (Havel and Rhodes, 2009). While combined headwaters and combined near-dam locations have similar concentrations in the winter, cyanobacteria concentrations increase more rapidly in headwater locations throughout the summer and remain elevated during the fall. An aggregated Wilcoxon signed-rank test of cyanoHAB magnitudes using 33,832 pixel-based paired observations (dates with unobscured satellite observations in both a reservoir’s headwater and near-dam locations) confirmed significantly higher median magnitudes in 59 of 60 headwater locations across all sampling dates. Additionally, seasonal Wilcoxon signed-rank hypothesis tests of paired daily data at the reservoir level indicate that the elevated headwater concentrations are robust, most pronounced in the summer (57/60 reservoirs with significantly higher headwater concentrations), similarly high in spring (55/60) and fall (54/60), and persisting through the winter (48/60). Fig. 7 depicts these differences in seasonal concentration distributions. These findings are consistent with USGS field monitoring and analysis of the Lake Houston surface-water-supply reservoir (Beussink and Graham, 2011). Beussink and Graham found consistent presence of cyanobacteria throughout the year and the largest cyanobacterial biovolume during summer (June– September) when temperatures exceeded 27 °C and water residence times increased beyond 100 days. Longer water residence times may increase cyanobacteria biovolume in summer months (Beussink and Graham, 2011). Cyanobacterial taxa such as Anabaena, Aphanizomenon, and Microcystis are generally slow growing and may benefit from the longer water residence times during summer (Wetzel, 2001). Additionally, researchers in Lake Houston also identified an association between high cyanobacteria concentrations and lower inflow stream discharge. Monthly sampling in Lake Lyndon B. Johnson and Lake Travis reservoirs also showed the highest concentration blooms in the summer months (Gamez et al., 2019).

Fig. 7.

Raincloud plots showing seasonal cyanobacteria magnitudes. This seasonal breakdown of cyanobacteria concentrations shows consistently elevated cyanobacteria concentrations in headwaters versus near dam areas. The half-violins at the top of each plot show the density distributions, the boxplot provide summary statistics, and the jittered points indicate data density across the concentration range. The probability distributions for headwater concentrations are noticeably higher (shifted to the right) compared to the near dam concentrations, particularly in the summer and fall.

The peak of cyanobacteria magnitude during warm seasonal conditions may indicate potential climate change impacts as global warming is predicted to extend the number of days experiencing warm seasonal temperatures. Cyanobacteria temporal frequency, spatial extent and magnitude is predicted to increase in inland lakes as a result of warming temperatures (Cayelan et al., 2011; O’Neil et al., 2012; Paerl and Paul, 2012; Chapra et al., 2017; Mishra et al., 2019). While the 5 year time period examined in this research does not provide an adequate temporal analysis to examine the influence of climate change, future examination of these trends within reservoir headwaters is strongly encouraged.

4. Conclusions

Study results confirm high concentration cyanoHABs (>100,000 cells/mL) occurred more frequently in reservoir headwaters compared to near-dam surface water locations in 59 of the 60 reservoirs evaluated. Results indicate that cyanobacteria magnitudes are higher in reservoir headwaters compared to near-dam surface water locations. For the 1547 paired headwaters and near-dam observations, maximum daily cyanobacteria concentrations were higher in reservoir headwaters on 86% of observation dates. Seasonal variation of high-concentration cyanoHABS was also more pronounced in headwaters with a clear increase in high concentration blooms during the summer months. Several of the study area reservoirs had high magnitude cyanoHAB concentrations and frequent cyanoHAB occurrence. Of the sixty reservoir sites examined, the twelve sites with the most frequent high concentration cyanobacteria occurrence were: Lake Palestine, Cedar Creek Reservoir, Fort Cobb Reservoir, Lake Limestone, Woodlands Golf Course Lake, Corder Lake, Somerville Lake, Navarro Mills Lake, Cross Lake, Falcon Reservoir, Lavon Lake, and Bardwell Lake. Ten of these reservoirs are located in the state of Texas, one of the states with the most cyanobacteria locations detected during the 2007 National Lakes Assessment (USEPA, 2009). Based on the repeated detection of concentrations exceeding 1,000,000 cells/mL in the headwaters of 30 study area reservoirs, results indicate additional research is required to better understand the spatial distribution of reservoir cyanoHABs throughout the U.S. south region.

The findings of this study contribute to our scientific understanding of cyanobacteria spatial distributions and temporal patterns within anthropogenic reservoirs. If the public were aware of locations with elevated cyanobacteria concentrations, more frequent blooms, and more pronounced seasonal variations of high concentration cyanoHABs, communities would benefit from more effective monitoring and mitigation and may reduce potential exposure by avoiding certain areas of the reservoir. The more frequent observations of cyanobacteria in reservoir headwaters may be a result of the shallow depth, warm summer temperatures, and nutrient rich environmental conditions in these sites. However, more research is required to better understand the drivers of headwater cyanoHABs. Reservoir headwaters are highly affected by land use conditions in the surrounding watershed as sediment and nutrient deposition occurs in these locations. To effectively mitigate frequent cyanoHAB occurrence, land use management practices must be considered to prevent excess sedimentation and nutrient loading in reservoir headwaters (Kozak et al., 2019). Future researchers should examine reservoir headwaters as important ecosystems and further explore which physicochemical and environmental factors drive cyanoHAB frequency in these systems. Additionally, these findings indicate water managers should also target headwater locations for water sampling efforts to monitor bloom events and prevent potential public exposure to cyanotoxins.

HIGHLIGHTS.

• Reservoir cyanobacteria varies spatially within the water body.

• CyanoHABS are more frequent near reservoir in-flow headwaters than near-dam surface waters.

• In reservoir surface waters, CyanoHAB magnitudes are higher near reservoir headwaters.

• Headwater and near-dam cyanoHAB magnitude differences are more pronounced in summer.

• Frequent high cyanoHAB concentrations indicate southern U.S. reservoirs require further study.