Multiparameter Sonde

YSI Inc. EXO2 multiparameter sondes were equipped with sensors for measurement of water temperature in degrees Celsius (°C), specific conductance in microsiemens per centimeter at 25 degrees Celsius, turbidity in formazin nephelometric units, pH in pH units, dissolved oxygen in milligrams per liter (mg/L), fluorescent dissolved organic matter (fDOM) in micrograms per liter (μg/L) as quinine sulfate equivalents (QSE) and in relative fluorescence units (RFU), chlorophyll (fluorescence) in µg/L and RFU, and phycocyanin (fluorescence) in µg/L and RFU. RFU is a measure of sensor response based on a percentage of the operating range of the sensor (0–100 percent) and is not directly tied to a quantitative value like concentration units. A centralized antifouling wiper was programmed to clean each sensor before every measurement. This instrument was chosen because of its wide use throughout the USGS to collect data recommended for studies that focus on phytoplankton biomass and community composition (Foster and others, 2022). All parameters were measured and recorded at 15-minute intervals. The instruments were deployed free hanging from the platform at near surface, mid-, and near bottom depths (table 1; fig. 3; fig. 4). Specification for the EXO2 sondes and individual sensors are detailed in YSI Inc. (2021) and are abridged in table 4.

Nitrate Sensors

At different periods, two different types of nitrate sensors were used during the study: the s::can nitro::lyser II and the Hach Company Nitratax plus sc. The sensors were switched mid-study because of sensor malfunction and loss of data.

S::can nitro::lyser II.—The s::can nitro::lyser II is an ultraviolet nitrate sensor with a 15 mm optical pathlength. The measured value is reported in units of milligrams per liter as nitrogen (mg/L as N). The sensor has integrated compensations for turbidity and DOM. The pathlength is mechanically cleaned with an antifouling wiper before each measurement. This instrument was chosen because of its previous use by the USGS in low nitrogen systems (USGS, 2016). The instrument was deployed inside the sensor well, as described in the “Platform Infrastructure” section, at the near-surface depth (table 1; fig. 3). The sensor recorded measurements at 5-minute intervals. The lower detection limit of the sensor is 0.0 mg/L, and accuracy is 2 percent plus one optical pathlength. Complete specifications for the nitro::lyser II are detailed in s::can (2011) and abridged in table 4.

Hach Company Nitratax plus sc.—The Hach Nitratax plus sc is an ultraviolet nitrate sensor with a 2 mm optical pathlength. The measured value is reported in units of milligrams per liter as nitrogen (mg/L as N). The sensor contains a two-beam absorption photometer with integrated compensation for turbidity. The measuring window is mechanically cleaned with an antifouling wiper before each measurement. This instrument was chosen for its ease of use and durability, as demonstrated in other studies; its measuring range; and its availability to rent from the USGS Hydrologic Instrumentation Facility. The instrument was deployed inside the sensor well, as described in the “Platform Infrastructure” section, at the near-surface depth (table 1; fig. 3). Sensors recorded measurements at 15-minute intervals. The lower detection limit of this sensor is 0.1 mg/L, and the accuracy is 3 percent plus 0.5. Complete specifications for the Nitratax are detailed in Hach Company (2023) and abbreviated in table 4. It should be noted, as of April 30, 2023, the Nitratax plus sc was discontinued by the manufacturer with an updated version available.

Orthophosphate Sensor

The Sea-Bird Scientific HydroCycle PO₄ is a wet chemical sensor that measures soluble reactive phosphate. The measured value is reported in milligrams per liter as phosphorus (mg/L as P). A copper screen and a 5–10 micrometer (µm; 7.5 µm average) polyethylene filter are used to mitigate the fouling of the sensor’s internal optics. This instrument was chosen because of its ability to be deployed in place, its self-contained design, its ability to run on direct current power, and its previous use by the USGS. The instrument was deployed inside the sensor well, as described in the “Platform Infrastructure” section, at the near surface depth and sampled at 120-minute intervals (table 1; fig. 3). The waste generated by the HydroCycle-PO₄ was collected in a dedicated vessel on the monitoring platform. This approach ensured that none of the waste, which is a byproduct of the phosphate measurement process, was discharged into the surrounding environment. Although the total chemical reaction time of the sensor is about 12 minutes, a longer measurement interval was used to extend the life of reagent cartridges and intake filters, and to reduce power consumption. Specifications for the HydroCycle PO₄ are detailed in Sea-Bird Scientific (2019) and condensed in table 4. It should be noted that as of October 1, 2022, sale of and service for the HydroCycle PO₄ was discontinued by the manufacturer.

Phytoplankton Classification Sensor

The Turner Designs PhytoFind is a multichannel fluorometer that processes fluorescence data from three optical sensors to provide estimates of total chlorophyll concentration in micrograms per liter (µg/L). PhytoFind also estimates the percent contribution relative to the whole phytoplankton population sampled from three groups: phycoerythrin-containing phytoplankton (such as cryptophytes and some cyanobacteria; reported as “mixed phytoplankton group”), phycocyanin-containing phytoplankton (such as cyanobacteria; reported as “cyanobacteria group”), and all other chlorophyll-containing phytoplankton (such as chlorophytes and diatoms; reported as “green and brown phytoplankton group”). A fourth optical fluorescence sensor provides integrated compensation for DOM, which is known to interfere with optical measurements (Downing and others, 2012; Bertone and others, 2019). This instrument was chosen to provide information on phytoplankton population dynamics. Additionally, this instrument was chosen to gain insight into the efficacy of another multichannel fluorometer, similar to the bbe Moldaenke GmbH FluoroProbe (bbe Moldaenke GmbH, 2023), which has been used extensively apart from this study by the NYSDEC (Prestigiacomo and others, 2022; Gorney and others, 2023). The PhytoFind was deployed inside the sensor well, as described the “Platform Infrastructure” section, at the near-surface depth, and sampled at 5-minute intervals (table 1; fig. 3). Specifications for the PhytoFind are detailed in Turner Designs, Inc., (2021) and abbreviated in table 4.

Temperature and Illumination Sensor

The HOBO UA–002–64 is a small, two-channel temperature and relative light level data logger. Water temperature is reported in degrees Celsius (°C) and relative light level (wavelength range from about 150 to 1,200 nanometers) in lumens per square meter (lux). This instrument was chosen because of its affordability, self-contained power and logging, and proven ruggedness as demonstrated in other monitoring programs. The instruments were deployed for use in a “temperature string” (fig. 3). The temperature strings were deployed free hanging from the platform at 1-meter intervals below the water surface (fig. 5), totaling 30 sensors at Seneca Lake, 26 at Owasco Lake, and 31 at Skaneateles Lake. Sensors recorded measurements at 15-minute intervals. Specifications for the UA–002–64 are detailed in Onset Computer Corporation (2018) and are abridged in table 4.

Photosynthetically Active Radiation Sensor

The LI–COR, Inc., LI–190R measures photosynthetically active radiation (PAR; wavelength range from about 400 to 700 nanometers) in micromoles of photons per square meter per second (µmol/m²/s). This instrument was chosen because of its wide use throughout the USGS to collect ancillary PAR data. The instrument was attached to the platform handrail, about 1.2 m above the water surface (fig. 3). The sensor recorded measurements at 15-minute intervals. Specifications for the LI–190R are detailed in LI–COR, Inc. (2023) and are abbreviated in table 4.

Weather Sensor

The Vaisala Weather Transmitter WTX536 is a multiparameter weather sensor that measures air pressure in millibars (mbar), air temperature in degrees Celsius (°C), relative humidity in percent, precipitation in inches, wind speed in meters per second, and wind direction in degrees (°). This instrument was chosen to provide information about meteorological conditions to help contextualize changes in water quality. The instrument was set atop a mast on the platform, about 2.1 m above the water surface (fig. 3). The sensor recorded measurements at 15-minute intervals. Specifications for the WTX536 are detailed in Vaisala (2022) and are abbreviated in table 4.

Multiparameter Sonde Measurements—Water Temperature, Specific Conductance, Dissolved Oxygen, pH, Turbidity, Fluorescent Dissolved Organic Matter, Chlorophyll Fluorescence, and Phycocyanin Fluorescence

Multiparameter sonde data (water temperature, specific conductance, dissolved oxygen, pH, turbidity, fDOM, chlorophyll fluorescence, and phycocyanin fluorescence) were collected in accordance with USGS national protocols and guidelines, some of which were still in development during the data collection period (Wagner and others, 2006; Foster and others, 2022; Booth and others, 2023). All data collected using multiparameter sondes are available through the USGS National Water Information System (NWIS; USGS, 2016).

Nutrient Sensor Measurements

Nitrate sensor data were collected in accordance with USGS national protocols and guidelines (Wagner and others, 2006; Pellerin and others, 2013). At the Seneca Lake platform, the s::can nitro::lyser II was used from May 24 to September 5, 2019, and the Hach Company Nitratax plus sc from August 5 to October 20, 2020. At Owasco Lake platform, the nitro::lyser II was used from May 14 to June 19, 2019, and the Nitratax from June 11 to October 22, 2020. Data are available through NWIS (USGS, 2016). Skaneateles Lake did not have a nitrate sensor deployed.

The orthophosphate sensor used wet chemistry to measure orthophosphate concentration. Because of this approach, standard USGS protocols for determining fouling, calibration drift error, or both (Wagner and others, 2006) are not applicable to the orthophosphate sensor (Peake, 2022). Therefore, data collection methods were modified in accordance with manufacturer instructions (Sea-Bird Scientific, 2019). The instrument is factory calibrated but periodically runs an automated calibration-validation cycle using onboard standard. All readings from the orthophosphate sensor used in this study are associated with qualitative metadata, such as data quality indicators and other quality-control test values, that are used to aid in the determination of data validity. Data are available through NWIS (USGS, 2016).

Multichannel Fluorometer Measurements—Phytoplankton Community Composition and Total Chlorophyll

USGS protocols and guidance for determining fouling, calibration drift error, or both (Wagner and others, 2006; Foster and others, 2022) are not applicable to the multichannel fluorometer used in this study for two reasons. First, the instrument is factory-calibrated, so no calibration functions are available. Second, because of the use of deconvolution algorithms that process data from multiple fluorescence sensors, the interference(s) affecting one fluorescence channel could affect all parameters (Johnston and others, 2022). For these reasons, data collection methods were modified in accordance with manufacturer instructions (Turner Designs, Inc., 2021). In place of calibration-drift checks in standard(s) of known concentration(s), sensor function was verified by comparing instrument readings of chlorophyll in the solid reference standard cap (part numbers 2380–900, 2300–901 and 2300–902) from before and after the deployment. Interference from fouling was assessed by comparing instrument readings of chlorophyll in the solid reference standard cap from before and after the fluorescence sensors on the instrument were cleaned. The DOM blank correction feature of the instrument was enabled, when necessary, as described in Turner Designs, Inc., (2021).

The complex nature of the multichannel fluorometer led to methodological inconsistencies in 2019 related to the unique procedures described above. Moreover, the application and effects of site-specific compensations for DOM—even in clear systems—were not fully understood because the instrument was new. As a result, the data collected during 2019 did not meet the data-quality goals of this study and were not published or considered in this report. The 2020 data are available in Johnston and others (2023).

Water Temperature and Illumination Sensor Measurements

Currently (2024), no USGS guidance exists for using illumination sensors. Although standards exist for water-temperature data, measurements from the combined water temperature and illumination sensor were collected as ancillary data to provide estimates of water temperature and light conditions vertically through the water column at a finer resolution than the resolution of the multiparameter sondes. Furthermore, the water temperature accuracy of the HOBO UA–002–64 is outside the 0.2 °C criterion specified in Wagner and others (2006). Data were collected following manufacturer instructions (Onset Computer Corporation, 2018). The data are available in Johnston and others (2023).

Photosynthetically Active Radiation Sensor Measurements

Currently (2024), no USGS guidance exists for using PAR sensors. Measurements from the PAR sensor were collected as ancillary data to estimate light conditions—in the wavelength range used by plants for photosynthesis—above the water surface. Data were collected following manufacturer instructions (LI–COR, Inc., 2023). The data are available in Johnston and others (2023).

Weather Sensor Measurements—Air Temperature, Precipitation, Wind Speed and Direction

Currently (2024), no USGS guidance exists for using meteorological sensors. Although some guidance exists for precipitation data, data from many USGS weather gages are considered “operational” (that is, ancillary) and assist with the evaluation of other data measured at these locations (USGS, 2005, 2023). Data were collected following manufacturer instructions (Vaisala, 2022). The data are available in Johnston and others (2023).

Nutrients, Dissolved Organic Carbon, and Chlorophyll-a

Nitrate, orthophosphate, DOC, and extracted chlorophyll-a samples were collected according to methods detailed in the National Field Manual for the Collection of Water Quality Data (USGS, variously dated). Samples were analyzed by the USGS National Water Quality Laboratory in Denver, Colo., and data are available in NWIS (USGS, 2016).

Nitrate (mg/L as N; analyzed as “nitrate plus nitrite” and for all samples in this study, nitrite [determined separately according to Fishman (1993)] was 0.01 mg/L as N or less and therefore constituted a trivial part of the total relative to nitrate), orthophosphate (mg/L as P), and DOC (mg/L) samples were filtered using a 45-µm capsule filter. Nitrate and orthophosphate samples were collected into 125-milliliter high-density polyethylene amber bottles. DOC samples were collected into 125-milliliter baked amber glass bottles before preservation to <2.0 pH with sulfuric acid. Nutrient and DOC samples were shipped on water ice to the laboratory. Nitrate plus nitrite samples were analyzed colorimetrically by enzymatic reduction using an automated discrete analyzer, as described in Patton and Kryskalla (2011). Orthophosphate samples were analyzed by discrete analyzer phosphomolybdate formation and colorimetry, as described in Fishman (1993). DOC was analyzed by high-temperature combustion oxidation and nondispersive infrared spectroscopy as described in Standard Methods for the Examination of Water and Wastewater 5310–B (Standard Methods Committee, 2014).

For chlorophyll-a sample collection, water was transferred from the churn directly into 1-liter amber high-density polyethylene bottles then homogenized and filtered onto 47-mm diameter glass fiber filters with a 0.3 µm pore size (fig. 9). Filters were kept frozen in the dark and sent to the USGS National Water Quality Laboratory on dry ice. Chlorophyll-a samples were analyzed following the methods and procedures in Arar and Collins (1997) using the non-acidification approach.

QA and quality-control (QC; together, QA/QC) samples were collected, as sequential replicates, to evaluate the variability resulting from sample collection and processing techniques or temporal changes resulting from shifting environmental conditions (Mueller and others, 2015). Relative percent difference (RPD) was used to quantify differences in analytical concentrations and could not be calculated if one, or both, sample(s) from the replicate pair were not analyzed or were associated with a censored value (that is, less than the laboratory reporting limit) (Zar, 1999). RPD was calculated using the following equation:

Phytoplankton Identification and Enumeration

Phytoplankton samples were collected in 250-mL high-density polyethylene bottles and preserved with glutaraldehyde to a concentration of 0.25 percent by volume and analyzed for identification and enumeration by PhycoTech, Inc., Saint Joseph, Mich. Glutaraldehyde was used to preserve the fluorescence properties of the algae. Identification and enumeration using microscopy was conducted using N-(2-Hydroxypropyl)methacrylamide mounted slides with fluorescence enabled microscopy, a count threshold of 400 natural units, and a size range of detection from 0.9 µm to 25 mm. Phytoplankton were enumerated to the lowest possible taxonomic level, which was at least the genus level for most organisms. To correctly identify and enumerate the taxa present, which may vary by several orders of magnitude in size, counts were made at multiple magnifications ranging from 200 to 1000´ (PhycoTech, 2018). Random microscope fields were counted to a minimum of 400 natural units (colonies, filaments, or single cells) of all observed organisms (PhycoTech, 2018). Individuals from each species were then counted, and the average cell dimensions recorded for biovolume calculations. Phytoplankton data, sample processing and analysis methods, and QA protocols are detailed in PhycoTech (2018). The phytoplankton data are available in Perkins and others (2021).

Absolute value logarithmic difference (AVLD) was used to estimate differences in phytoplankton and cyanobacterial biovolume, in cubic micrometers per milliliter (µm³/mL), between replicate pairs. AVLD was used because RPD is sensitive to the rare taxa found in one of the samples and not the other. AVLD is calculated using the following equation:

Orthophosphate Sensor

All values measured by the orthophosphate sensor at Seneca Lake (2019–20), Owasco Lake (2019), and Skaneateles Lake (2019) were below the 0.002 mg/L as P minimum detection limit of the instrument. All measured values reported by the orthophosphate sensor at Owasco Lake in 2020 were below the 0.002 mg/L as P minimum detection limit of instrument, except for 11 time-series values. Using the Adjustable Trim edit tool in Aquarius, all values below the detection limit were replaced with “0.002” mg/L as P. Additionally, affected values were tagged with the qualifier, “Actual value is known to be less than reported value.”

In addition to time-series analysis in Aquarius, internal metadata were reviewed using the manufacturer’s software, Cycle Host. Explanations and guidance on QC review of the data and metadata provided by the HydroCycle PO₄ are described in the user manual (Sea-Bird Scientific, 2019). The following data were deleted because of abnormalities observed in the QC data plots (fig. 10), which indicated the reagents did not mix with the water sample (typically caused by pinching of the waste line of the instrument; 1/16-inch inner diameter tubing):

Overall, the orthophosphate sensor is complex to operate and maintain, and determining sensor’s performance is difficult. Nearly all sensor measurements were below the minimum detection limit; however, orthophosphate was rarely detected in laboratory-measured discrete samples, which aligns with the below minimum detection values reported by the HydroCycle PO₄. Loss of data was primarily attributed to damage (typically as pinching) of the waste line of the sensor, which required a site visit to identify and remedy.

Phytoplankton Classification Sensor

When analyzing data from the multichannel fluorometer, it is important to consider that the reported chlorophyll concentration and percent chlorophyll contributions per phytoplankton group are estimates and not associated with a specified accuracy by the manufacturer. The results from the instrument are assumed to demonstrate relative stability when measured inside the solid reference standard cap.

Seneca Lake platform (2020).—Fouling- and sensor-verification checks did not indicate changes in instrument readings of chlorophyll that exceeded the criteria described in Foster and others (2022). In addition to reasons described in the “Time Series Data Quality Assurance and Quality Control” section, the period August 24 to October 5, 2020, was removed from the time-series record because of a malfunction of the antifouling wiper parking outside the intended position, which resulted in variable interference to the fluorescence sensors. This interference was evident by irregular fluctuations in the time-series data (most pronounced in the integrated fDOM sensor data) and by documentation made at a site visit during which the issue was resolved (fig. 11).

Owasco Lake platform (2020).—Fouling checks did not indicate changes in instrument readings of chlorophyll that exceeded the criteria described in Foster and others (2022). Sensor verification checks indicated a change of 27.8 percent in instrument readings of chlorophyll during the deployment, which exceeds the calibration correction threshold described in Foster and others (2022); however, the recalibration or application of calibration drift corrections using standard methods are not applicable to this multichannel fluorometer as described above. In addition to reasons described in the “Time Series Data Quality Assurance and Quality Control” section, the period from August 10 to 24, 2020, was removed from the time-series record because of a malfunction with the antifouling wiper parking outside the intended position, resulting in variable interference to the fluorescence sensors. This interference was indicated by irregular fluctuations in the time-series data and was documented during a site visit during which the issue was resolved. During the deployments at Seneca and Owasco Lakes, frequent (that is, multiple occurrences per day) downward outliers were present in the time-series record of fDOM RFU measurements. Neither comparison with the fDOM sensor on the multiparameter sonde nor consultation with the manufacturer established confidence in the observed spikes. These outliers can affect the fDOM blank correction applied to all parameters reported by the multichannel fluorometer. Following the recommended best practices discussed in Foster and others (2022), specific guidelines for identifying erroneous spikes in the multichannel fluorometer data were developed for this project.

A variety of statistical methods for detecting outliers exist, and one method is to set an outlier bound at either two or three standard deviations from the mean (Thompson, 1935; Ahn and James, 1999). The outlier bound for these data was set at two standard deviations from the mean of a rolling 24-hour interval. This outlier detection was applied programmatically using R statistical software. In addition to the edits described above, all reported parameters from the multichannel fluorometer that were associated with flagged outliers in the time-series record of fDOM RFU measurements (2.3 percent for Seneca Lake and 1.7 percent for Owasco Lake) were also removed from the record.

Overall, the multichannel fluorometer is complex to operate and can require unique procedures to acquire integrated, site-specific corrections. Methods for verification and maintenance are also unique, and require a detailed understanding of the instrument, its function, and accessories or else risk reduced data quality. Moreover, outliers generated by the fDOM sensor on board the instrument can affect all reported measurements. Based on these data, the percentage of fDOM outliers was about 2 percent of the total data collected at Seneca Lake and Owasco Lake and may indicate a systematic error with the instrument rather than random error caused by environmental conditions. More data are required to determine the cause of fDOM outliers.

Water Temperature and Illumination Sensor

No fouling or calibration drift corrections were applicable to the water-temperature and illumination data measured by the combined sensor. Water-temperature data were verified by comparison with collocated water-temperature data from the near surface multiparameter sonde, such that the near surface combined sensor reading was verified and then used to verify the sensor reading at the next lower depth, iterating down the temperature string. Illumination data from the combined sensors were verified by comparison with collocated PAR data from the LI–COR LI–190R at near surface depths, then iterated down the string, as described for water temperature, assuming light reduction farther down the string. In addition to equipment malfunction, water temperature and illumination data from the combined sensor deployed at 15 m depth in Owasco Lake were removed from the time-series record for the periods from September 3 to 21, 2020, and October 5 to 22, 2020, because of an unexplained shift of about 0.5 °C for water temperature and about 1 lux for illumination, in comparison with adjacent sensors. Additional support for determining this period as related to sensor malfunction was provided by illumination overnight readings greater than 0 lux. Many other smaller periods of only illumination data at each study lake were removed from the record when comparisons with adjacent sensors indicated departures from expected trends (that is, decreasing illumination with increasing depth). Overall, the combined water-temperature and illumination sensors performed well; however, drift is more frequently observed in illumination measurements (as compared to adjacent combined sensors) than water-temperature measurements. Additionally, the lack of integration into the datalogger makes identifying issues less timely—requiring a site visit—than if data were telemetered.

Photosynthetically Active Radiation Sensor

No fouling or calibration drift corrections were applicable to PAR data measured by the sensor on the platform. PAR data were verified by comparison with illumination data from the combined sensors in the water near the surface, which were presumed to have relative correlation. Gaps in data were not frequent or long in duration; however, they did occur briefly at times across all lakes and years except for the data from Owasco Lake in 2020. Gaps in data were almost always associated with global power issues on the platform affecting the datalogger and telemetry equipment. Although the PAR sensor itself was not dependent on platform battery power, communication with the datalogger and (or) transmission of the data can be interrupted by power problems. Overall, the PAR sensor functioned well. PAR data were deleted on only two occasions: mid-August 2019 at Owasco Lake (about 24 hours) and Skaneateles Lake (about 40 hours), when PAR values were 0 µmol/m²/s despite near-surface illumination data from the combined sensor indicating fluctuations.

Weather Sensor

No fouling or calibration drift corrections were applicable to air temperature, precipitation, wind direction, and wind speed. Air temperature data were verified by comparison with collocated, near-surface water-temperature data from the multiparameter sonde and from the combined water-temperature and illumination sensor, which were presumed to have relative correlation. Precipitation, wind, and air temperature data were verified by comparison with corresponding precipitation, wind, and air temperature data recorded at the other lake platforms. Overall, the weather sensor functioned well, and gaps in data were almost always associated with global power issues on the monitoring-station platforms.

Chlorophyll

Chlorophyll was measured by two instruments in this study: the dual-channel fluorometer on a multiparameter sonde and a multichannel fluorometer. The relation between each sensor and related laboratory measurements is discussed; however, the sensors are not directly compared, in accordance with guidance from the USGS Office of Acquisitions and Grants (OAG)/Ethics when discussing instrument performance in publications by USGS authors. The reader should be able to discern similarities and differences between the two instruments based on the data and analyses provided.

The overall monotonic relation between the dual-channel chlorophyll fluorometer and laboratory-measured concentrations of chlorophyll-a was relatively strong (ρ=0.73, p-value<0.05; n=134); however, only 50 percent of the variation was explained with linear regression (r=0.71, p-value<0.05, R²=0.50). Sensor-measured chlorophyll concentrations ranged from −0.50 to 13.5 µg/L, with a median value of 0.90 µg/L (reported values less than 0 are possible from the EXO Total Algae sensor—typically a result of sensor calibration procedures); laboratory-measured chlorophyll-a concentrations ranged from 0.05 to 14.6 µg/L, with a median value of 1.30 µg/L (fig. 18). Generally, laboratory-measured chlorophyll-a values tended to be slightly higher than values from the dual-channel fluorometer. The laboratory-measured median was about 1.4 times higher than the sensor-measured median, and the laboratory maximum was about 1.1 times higher than the sensor maximum. The dual-channel fluorometer explained the most variance in laboratory-measured chlorophyll-a concentrations in Seneca Lake (R²=0.60) but did not explain as much in Owasco (R²=0.33) and Skaneateles (R²=0.38) Lakes. These differences by lake may be in large part attributed to the phytoplankton community composition or physicochemical conditions within the lakes, rather than variations in instrument function or calibration, as supported by the relative consistency across the depths at each lake.

The overall monotonic relation between chlorophyll concentration measurements from the multichannel fluorometer and laboratory-measured concentrations of chlorophyll-a was moderately strong (ρ=0.65, p-value<0.05; n=16). However, the multichannel fluorometer only explained 44 percent of the variation in laboratory-measured chlorophyll-a values (r=0.66, p-value<0.05, R²=0.44). Sensor-measured chlorophyll concentration ranged from 2.11 to 20.3 µg/L, with a median value of 5.46 µg/L, whereas laboratory-measured chlorophyll-a concentrations ranged from 0.05 to 14.6 µg/L, with a median value of 1.50 µg/L (fig. 18). Generally, the sensor measurements tended to be higher than measurements from the laboratory; sensor median was about 3.6 times higher than the laboratory median, and sensor maximum was about 1.4 times higher than the laboratory maximum. The relation between the sensor-measured and the corresponding laboratory-measured chlorophyll-a concentrations is different in Seneca Lake and Owasco Lakes. The multichannel fluorometer explained more than 2 times the variance in laboratory-measured chlorophyll-a values at Owasco Lake (R² = 0.70) than at Seneca Lake (R² = 0.31). Moreover, the multichannel fluorometer at Seneca Lake generally overestimated chlorophyll-a, relative to the fluorometer at Owasco Lake. The overestimation at Seneca Lake is likely attributed to the generally larger biovolumes of phytoplankton in the green/brown and mixed target groups of the sensor (Johnston and others, 2022; fig. 17; table 7). Although the number of observations is small, the multichannel fluorometer performed better in samples where cyanobacteria were dominant, as observed by Johnston and others (2022). For example, in Owasco Lake, sensor measurements of chlorophyll were within 10 percent of laboratory-measured values when cyanobacteria were dominant (n=2). More data are needed to establish additional confidence in these relationships.

Overall, the amount of variance in laboratory-measured chlorophyll-a explained by fluorescence sensor measurements differed among lakes and by sensor. Both the dual-channel and multichannel fluorometers demonstrated the ability to explain about 60 percent of the variance in laboratory-measured chlorophyll-a at Seneca Lake and Owasco Lake, respectively. However, the amount of variance explained ranged between about 30 and 60 percent, depending on lake and sensor. Although relatively good relations between sensor- and laboratory-measured chlorophyll-a values were observed in this study, the sample size and range are small, and using site-specific relations may increase the overall amount of variance explained by the sensor(s) (Rousso and others, 2020; Prestigiacomo and others, 2023).

Phytoplankton Biovolume

Total phytoplankton biovolume.—Although not a direct comparison, chlorophyll can be used as a proxy for phytoplankton biomass (Foster and others, 2022). Chlorophyll measured by the dual-channel and multichannel fluorometers were compared to laboratory-measured total phytoplankton biovolume. The overall monotonic relation between chlorophyll concentration measurements from the dual-channel fluorometer and laboratory-measured total phytoplankton biovolume was moderately strong (ρ=0.61, p-value<0.05; n=136); however, only 14 percent of the variation in phytoplankton biovolume was explained by sensor-measured chlorophyll (r=0.38, p-value<0.05, R²=0.14) (fig. 19). The linear relation with dual-channel fluorometer measurements was weaker for total phytoplankton biovolume (all r values about 0.20 less) than for chlorophyll-a overall and in individual lakes. By lake, the dual-channel fluorometer explained more variance in laboratory-measured total phytoplankton biovolume in Seneca Lake (R²=0.32) than Owasco (R²=0.09) and Skaneateles (R²=0.20) Lakes, similar to what was observed for chlorophyll-a.

The overall monotonic and linear relations between chlorophyll concentration measurements from the multichannel fluorometer and laboratory-measured total phytoplankton biovolume were not statistically significant (ρ=0.39, p-value=0.13; n=16); (r=0.42, p-value=0.09, R²=0.18) (fig. 19). Although not statistically significant, the multichannel fluorometer explained 18 percent of the variance in total phytoplankton biovolume values, similar to the amount of variance explained by the dual-channel fluorometer. As observed for chlorophyll, the relation between sensor- and laboratory-measures of phytoplankton biovolume in Seneca and Owasco Lakes were different. The multichannel fluorometer explained more than 2 times the variance in total phytoplankton biovolume at Owasco Lake (R²=0.67) than in Seneca Lake (R²=0.29). As explained above, the difference between lakes is likely attributed to differences in phytoplankton community composition, such as generally larger abundances of phytoplankton in the green/brown and mixed target groups of the sensor at Seneca Lake and more samples dominated by cyanobacteria at Owasco Lake (Johnston and others, 2022; fig. 17; table 7). As noted for the sensor- and laboratory-measured chlorophyll-a relations, more data are needed to establish additional confidence in the relations for both the dual-channel and multichannel fluorometers.

Cyanobacterial biovolume.—Although not a direct comparison, the cyanobacterial pigment phycocyanin can be used as a proxy for cyanobacterial biomass (Foster and others, 2022). Phycocyanin fluorescence measured by the dual-channel and multichannel fluorometers were compared to laboratory-measured cyanobacterial biovolume. The monotonic and linear relations between phycocyanin concentration measurements from the dual-channel fluorometer and laboratory-measured cyanobacterial biovolume were not statistically significant (ρ=0.10, p-value=0.27; r=0.04, p-value=0.34; n=134). The relations did not vary meaningfully by lake, which is different from the relations between the dual-channel fluorometer and chlorophyll-a and total phytoplankton biovolume. Phycocyanin concentration measurements from the dual-channel fluorometer ranged between −0.60 and 0.85 µg/L (reported values less than 0 are possible from the EXO2 Total Algae-Phycocyanin sensor), whereas laboratory-measured cyanobacterial biovolume ranged many orders of magnitude, from 145 and 1,636,996 µm³/mL (fig. 20). Based on these results, the phycocyanin concentration measurements from the dual-channel fluorometer may not be sensitive to changes in phycocyanin-containing phytoplankton (such as, cyanobacteria) across the range encountered in the study lakes, which may explain why differences in relations among lakes were not observed.

The overall monotonic relation between measurements of chlorophyll concentration contributed by cyanobacteria from the multichannel fluorometer and laboratory-measured cyanobacterial biovolume was moderate (ρ=0.58, p-value<0.05; n=16), but 88 percent of the variation in cyanobacterial biovolume was explained by the sensor (r=0.94, p-value<0.05, R²=0.88) (fig. 20). The amount of variance explained by the sensor is strongly influenced by 6 of the 16 observations in which the sensor measured chlorophyll concentration contributions by cyanobacteria. Ten of the 16 observations were zero, in contrast to the laboratory measurements of cyanobacterial biovolume, which ranged from 1,018 to 175,571 µm³/mL in those same 10 observations. The multichannel fluorometers did make non-zero measurements in conditions where cyanobacteria were dominant; one at each lake; Aphanocapsa (58 percent of total phytoplankton biovolume) at Seneca Lake, and Microcystis (88 percent of total phytoplankton biovolume) at Owasco Lake.

Identification of Important Variables

Variables from unique sensors (that is, when measurements in different units from the same individual sensor were highly correlated, only the variable with the strongest correlation is discussed herein) indicating the strongest monotonic relation were similar between laboratory-measured chlorophyll-a and total phytoplankton biovolume (table 8; table 9). Water temperature (collectively measured through to the full depth of the lakes by the multiparameter sonde and the combined water-temperature/illumination sensor) had the strongest monotonic relation for chlorophyll-a, total phytoplankton biovolume, and cyanobacterial biovolume (table 8; table 9; table 10). This is attributed to the effects of water temperature on the rate of photosynthesis in phytoplankton—warmer water generally leads to higher rates of photosynthesis and growth. As such, seasonal maxima in phytoplankton biomass are often observed during the warmest months of the year (fig. 17; Raven and Geider, 1988; Falkowski and Raven, 1997).

Linear associations varied depending on phytoplankton biomass indicator; however, one or more measurements from the multichannel fluorometer were the most strongly correlated variable for chlorophyll-a, total phytoplankton biovolume, and cyanobacterial biovolume. Although the range and number of observations is small, the instrument capability of detecting and measuring fluorescence signals from multiple phytoplankton pigments may be useful in improving estimations of phytoplankton biomass over the use of single-pigment instruments. Water temperature also had a strong linear association with chlorophyll-a and cyanobacterial biovolume but not total phytoplankton biovolume. The differences in linear relationships may be attributed to the phytoplankton biovolume peak occurring in the spring when the water is cooler, as opposed to the cyanobacterial peak, which typically occurs in the warmer summer months (fig. 17). For cyanobacterial biovolume, sensor-measured turbidity was also among the most strongly correlated variables (table 10). This relation is likely attributed to the very clear conditions in these lake systems and the cyanobacterial contribution to the total suspended particles in the water column. Strong relations between cyanobacteria and sensor-measured turbidity have been noted in New York lakes (Prestigiacomo and others, 2023) and elsewhere (Melendez-Pastor and others, 2019; Rome and others, 2021), including lakes with cyanobacterial communities dominated by Aphanocapsa (Magalhães and others, 2019). Some sensor measurements were collected at multiple depths throughout the water column, whereas others were collected at, near, or above the water surface; correlation analyses using only data from near the water surface did not substantially change the variables most strongly correlated with chlorophyll-a, total phytoplankton biovolume, or cyanobacterial biovolume.

Stepwise regression analysis.—Stepwise regression analysis for laboratory-measured chlorophyll-a indicated the inclusion of pH (p-value=<0.001) and relative fluorescence of chlorophyll (RFU) measured by the dual-channel fluorometer (p-value=<0.001) as independent variables. The resulting model was able to explain 49 percent of the variance in laboratory-measured chlorophyll-a (R²=0.49; n=65) (fig. 21). The multivariate model, which incorporated pH and relative chlorophyll fluorescence, explained about the same amount of variance as the single-variable model using the chlorophyll concentration variable from the dual-channel fluorometer (R²=0.50, n=134). Additionally, the multivariate model considering only pH, explained six percent more variance than the single-variable model using total chlorophyll concentration measurements from the multichannel fluorometer (R²=0.43, n=16). Since some sensors were only deployed at the near-surface location, an analysis of only data collected near the water surface was performed. The resulting model included relative fluorescence of chlorophyll (RFU) measured by the dual-channel fluorometer (p-value=<0.001), dissolved oxygen (p-value=<0.001), turbidity (p-value=0.001), illumination measured by the combined sensor (p-value=<0.001), and PAR (p-value=0.005), and explained 40 percent more of the variance in laboratory-measured chlorophyll-a (R²=0.89; n=26) than the overall model. However, the number of variables included may result in overfitting and poor performance with new data (Cawley and Talbot, 2010). When the two variables with a p-value greater than or equal to 0.001 were removed (PAR and turbidity), the resulting model explained about 10 percent less of the variance in laboratory-measured chlorophyll-a than the overall model. The variables that explained the most difference in laboratory-measured chlorophyll-a concentrations varied by lake. The Seneca Lake model included relative fluorescence of chlorophyll (RFU) measured by the dual-channel fluorometer (p-value=<0.001); this model explained 8 percent more variance than the overall model (fig. 21). The Owasco Lake model included fDOM measured by the multiparameter sonde (p-value=<0.001) and turbidity (p-value=0.008); this model explained 20 percent more variance than the overall model (fig. 21). Many studies have found nutrient measurements (many as laboratory measurements) to have the most explanatory power in chlorophyll-a models (Jones and Bachmann, 1978; Phillips and others, 2008; Jones and others, 2020; Yuan and Jones, 2020; Huang and others, 2022). In our study lakes, sensor-measured nitrate and orthophosphate were not significantly correlated with discrete observations of laboratory-measured chlorophyll-a. Sensors only measured bioavailable forms of nutrients, rather than total nutrients, which may explain the lack of relation; bioavailable nutrients may cycle very rapidly in healthy phytoplankton and microbial communities, which may account for the apparent lack of association.

Stepwise regression analysis for laboratory-measured total phytoplankton biovolume indicated the inclusion of relative fluorescence of chlorophyll (p-value=<0.001), illumination measured by the combined water temperature/illumination sensor (p-value = <0.001), and wind speed (p-value=0.04) as independent variables. The overall model explained 41 percent of the variance in laboratory-measured total phytoplankton biovolume (R²=0.41; n=66) (fig. 22). The multivariate model, which incorporated relative fluorescence of chlorophyll, illumination, and wind speed, explained about 25 percent more variance than the single-variable models using the chlorophyll concentration measurements from the dual channel fluorometer (R²=0.14, n=136) and the total chlorophyll concentration measurements from the multichannel fluorometer (R²=0.18, n=16). When only data from sensors deployed near the top of the water column were included in the analysis, the relative fluorescence of chlorophyll measured by the dual-channel fluorometer (p-value=0.002) and illumination measured by the combined water temperature/illumination sensor (p-value=0.03) were the only variables that entered into the model. The near-surface model explained two percent less of the variance in laboratory-measured total phytoplankton biovolume (R²=0.39; n=28) than the overall model. The variables that explained the most difference in laboratory-measured phytoplankton biovolume concentrations varied by lake. The Seneca Lake model included relative fluorescence of chlorophyll measured by the dual-channel fluorometer (p-value=<0.001); this model explained 6 percent more variance than the overall model (fig. 22). The resulting model for Owasco Lake included relative fluorescence of chlorophyll measured by the dual-channel fluorometer (p-value=0.01) and illumination measured by the combined water temperature/illumination sensor (p-value=<0.001); this model explained about 10 percent more variance than the overall model (fig. 22). Similar to chlorophyll-a., sensor-measured nitrate and orthophosphate were not significantly correlated with laboratory-measured total phytoplankton biovolume.

Stepwise regression analysis for laboratory-measured cyanobacterial biovolume indicated the inclusion of turbidity (p-value=<0.001), relative fluorescence of fDOM (p-value=0.04), wind direction (p-value = 0.01) and precipitation (p-value=0.01) as independent variables. The overall model explained 40 percent of the variance in laboratory-measured cyanobacterial biovolume (R²=0.40; n=66) (fig. 23). The overall model explained about 40 percent more variance than the model based on phycocyanin concentration measurements from the dual-channel fluorometer (R²=<0.01; n=136), and 48 percent less variance than the model based on chlorophyll concentration measurements from the multichannel fluorometer (R²=0.88; n=16). When only data from sensors deployed near the top of the water column were included in the analysis, turbidity was the only variable that entered the model (p-value=0.002). However, the model explained nine percent less of the variance in cyanobacterial biovolume (R²=0.31; n=28) than the overall model. The variables that explained the most difference in laboratory-measured phytoplankton biovolume concentrations varied by lake. The Seneca Lake model included water temperature measured by the multiparameter sondes (p-value=0.002); this model explained 15 percent less variance than the overall model (fig. 23). The Owasco Lake model included turbidity (p-value=0.002) and wind direction (p-value=0.04); this model explained 5 percent more variance than the overall model (fig. 23).

Many studies have found turbidity to be highly correlated with measures of cyanobacterial biovolume or abundance; however, it is not among the most commonly used variables in predictive models (Rousso and others, 2020). As observed for chlorophyll-a and phytoplankton biovolume, nutrient measurements were not statistically correlated with cyanobacterial biovolume, likely for many of the same reasons as described for chlorophyll-a. Similar to chlorophyll, total nutrient measurements tend to have a substantial amount of explanatory power in cyanobacteria models. When included with one or more other variables (such as, water temperature and secchi disk depth), more than 50 percent of the variation in cyanobacterial biovolume or abundance is typically explained (Rousso and others, 2020). Many of these models are based on discrete nutrient or other analyte measurements, rather than sensor-based measurements. While models with discrete measurements may have more explanatory power, estimates of cyanobacterial abundance or biovolume cannot be estimated in near-real time—an advantage afforded by sensors.

Overall, the explanatory power of basic regression models for laboratory-measured chlorophyll-a, total phytoplankton biovolume and cyanobacterial biovolume as described in this report were weak. Models derived from data from all lakes explained less than half of the variance in laboratory-measured values. For chlorophyll-a, the overall multivariate model did not explain any more variance than single-variable models based on fluorescence measurements. The overall multivariate model for total phytoplankton biovolume did have substantially more explanatory power (greater than 20 percent more variance explained) than single-variable models based on fluorescence measurements. For cyanobacterial biovolume, the overall multivariate model had much stronger (about 40 percent stronger) explanatory power than phycocyanin measurements from the dual-channel fluorometer. In fact, the analysis indicated that turbidity was a much more significant indicator of the variability in cyanobacterial biovolume than phycocyanin measurements from the dual-channel fluorometer, by several orders of magnitude. Measurements of chlorophyll concentration contributed from cyanobacteria from the multichannel fluorometer had strong explanatory power based on results from this study; however, the number of observations and range of conditions encountered were small, and results should be considered preliminary.

Limiting analyses to data collected from near the water surface only and lake-specific analyses did not indicate substantial differences (plus or minus about 20 percent) in explanatory power over the models developed with the overall dataset, suggesting regional models for the Finger Lakes may be useful. However, high explanatory power (that is, relatively high R²) does not necessarily represent a high level of model accuracy (Cawley and Talbot, 2010; Rousso and others, 2020), and additional research is needed to develop the best possible models to estimate chlorophyll-a, total phytoplankton biovolume, and cyanobacterial biovolume in the Finger Lakes. In addition, CyanoHABs generally did not occur at the open-water monitoring locations within Seneca, Owasco, and Skaneateles Lakes during this study, limiting the range of conditions encompassed by this study. Models developed for predicting CyanoHABs elsewhere typically have ranges in chlorophyll-a concentrations or measures of the cyanobacterial community that are orders of magnitude greater than those observed during this study. For example, the chlorophyll-a concentrations observed in this study never exceeded 15 µg/L, compared to other studies worldwide in which ranges can exceed 100,000 µg/L (Rousso and others, 2020; Chorus and Welker, 2021). Data collected over a wider range of conditions that represent a larger number of CyanoHAB events are required to better define the relations between sensor- and laboratory-measured data, and to develop more complex data-driven models for predicting CyanoHABs (such as time-lagged or artificial neural networks), which may provide greater insights than traditional linear modeling approaches (Rousso and others, 2020).