2.1.

Study Area

The reservoirs sampled for this study include Lake Fork Reservoir, Grapevine Lake, Jim Chapman Lake, Lavon Lake, Lake Lewisville, Lake Palestine, Ray Hubbard Lake, Ray Roberts Lake, Lake Tawakoni, and Lake Texoma (Fig. 2). These reservoirs are within the Neches (Lake Palestine), Red (Lake Texoma), Sabine (Lake Fork Reservoir, and Lake Tawakoni), Sulphur (Lake Jim Chapman), and Trinity (Lakes Grapevine, Lavon, Lewisville, Ray Hubbard, and Ray Roberts) River Basins (Fig. 2). All reservoirs, except for Lake Texoma that forms the border between Texas and Oklahoma, are entirely in Texas.

Fig. 2

Location of reservoirs, sampling sites, and Sentinel-2 overpass footprints within the study area in north Texas (including a small part of southern Oklahoma for three of the sites on Lake Texoma).

These 10 reservoirs provide a collective capacity of more than $8.8\text{billion}{\mathrm{m}}^3$ of developed water⁴⁴ and range in surface area from 27.9 to $317{\mathrm{km}}^2$. Lake Fork Reservoir, Grapevine Lake, Lake Lewisville, Ray Hubbard Lake, Ray Roberts Lake, and Lake Tawakoni are used as sources of drinking water for the city of Dallas and surrounding areas. The City of Dallas has water rights for Lake Palestine, and this reservoir is expected to provide additional water resources to the Dallas area by 2027.⁴⁵

2.2.

Sampling Site Design

Thirty-one in-situ sampling sites were established across the 10 selected reservoirs (Fig. 2). At each reservoir, the sampling sites were divided into open-water (pelagic) sites and inlet sites. The Google Earth Engine LandTrendr Pixel Time Series Plotter was used to explore the spectral-reflectance histories of site locations, by plotting Landsat (1984 to 2022) time series of tasseled cap wetness and greenness spectral indices.³⁹ To account for the variability in spectral reflectance over the water surface of a given reservoir, different site selection criteria were used. At each reservoir, one or two open-water sites were selected near the center of the largest part of the water-surface area at deep locations where light reflection from the reservoir bottom would be considered negligible under normal reservoir capacity and where consistent tasseled-cap wetness and greenness time series spectral indices were obtained. In contrast to the open-water sites, inlet sites with interannual tasseled-cap wetness and greenness time series signals that varied were selected. As many as five inlet sites near tributaries, inlets, or coves per reservoir were selected. The sampling schedule and all of the water-quality data are provided in a companion USGS data release.⁴⁶

2.3.

Field Sampling

In-situ chlorophyll-a, phycocyanin, turbidity, water temperature, Secchi disk depth, phytoplankton taxonomy, and hyperspectral reflectance data were collected from boats during January to November 2020 to characterize a wide range of aquatic environments within each reservoir. The collection of in-situ chlorophyll-a samples was coordinated with Sentinel-2 overpasses. Wave action was avoided as much as possible for boat-operation safety and because surface roughness can increase light scattering, sun glint, and the occurrence of white caps, all of which can have negative effects on the spectra extracted from satellite imagery.¹⁶ Thus, lower wind speeds are more suitable for both sampling and imagery analysis. Waterbody sampling was delayed when wind speeds were greater than 25 kph.

Water-quality field properties were measured using standard USGS sampling protocols.⁴⁷ The in-situ chlorophyll-a samples were grab samples collected at a depth of 0.3 m. The samples were chilled to 4°C and shipped on ice to the Trinity River Authority Central Laboratory in Dallas, Texas, for analysis. Chlorophyll-a concentrations were determined by using standard spectrophotometric methods,⁴⁸ and an ambient water reporting limit of $3\mu \mathrm{g}/\mathrm{L}$ was set by the laboratory.⁴⁹ Reservoir water-clarity depth distributions across reservoir sampling sites were measured using a standard Secchi disk.⁴⁷ In-situ turbidity measurements were compared with the Secchi depth and chlorophyll-a concentration to aid in the interpretation of water-quality and water-clarity conditions on algorithm performance.²⁵ Phycocyanin, relative fluorescent units (RFU), turbidity, formazin nephelometric units (FNU), and temperature were measured in-situ with a YSI EXO2 multiparameter sonde.⁵⁰ A Kestrel 4500 pocket weather tracker was used to measure wind speed.⁵¹

Phytoplankton samples were collected and analyzed to identify dominant taxa across the reservoir open-water sampling sites and their spatiotemporal variability. The dominant algae and cyanobacteria taxonomy present at open-water sites in total biovolume (i.e., volume the individual algal organisms occupy in the water column), as well as the relative total biovolume (the percent total biovolume from the entire sample relative to all other taxa), were determined.^48,52–56 The phytoplankton samples were collected in 250-mL brown, opaque high-density polyethylene bottles and preserved on site with 2.5 mL of 0.25% glutaraldehyde solution. After collection, the phytoplankton samples were maintained at temperatures $<4°\mathrm{C}$ and shipped on ice to PhycoTech, Incorporated (PhycoTech in St. Joseph, Michigan) for analysis.

In-situ hyperspectral reflectance data were qualitatively compared with in-situ chlorophyll-a, phycocyanin, turbidity, and Secchi depth data to evaluate the hyperspectral reflectance observed for the Grapevine Lake sampling sites. Hyperspectral reflectance data were used to explore the optical signatures of algae and cyanobacteria pigments, chlorophyll-a, and phycocyanin, entrained in or accumulating on the water surface to aid in the interpretation of what portion of the spectrum might be most valuable for highly performing spectral indices.⁵⁷ Hyperspectral reflectance data of the water surface were obtained using an ASD HandHeld 2 visible NIR, 400 to 900 nm, spectroradiometer.⁵⁸ Boat shadows and sun glint were avoided during the collection of spectral reflectance signatures. The reflectance spectra were averaged during acquisition to minimize variations in brightness caused by movement of the water surface.

2.4.

Satellite Sensors

The first Sentinel-2 satellite was launched in 2015 and was designed for monitoring global lands and waters.²¹ The multispectral instrument (MSI) deployed on both the Sentinel-2a and Sentinel-2b satellites resulting in a combined revisit frequency of 5 days for the study area. The revisit frequency approximately doubles for reservoirs located where the 290 km overpass paths overlap (Fig. 2). The Sentinel-2 MSI sensor was selected because of its ability to detect the geophysical light reflectance of aquatic chlorophyll-a photosynthesis. Sentinel-2 offers expanded spectral resolution across the visible and NIR region of the electromagnetic spectrum compared with operational sensors; for example, the Landsat program was designed to monitor terrestrial vegetation that reflects brightly in NIR; however, the 800 nm (Landsat 8/9–band 5) frequency is absorbed by water. Sentinel-2 spectral resolution offers additional NIR bands capturing the “red edge” region of the spectrum. Sentinel-2 also provides advanced spatial resolution compared to operational inland (e.g., Landsat OLI) and ocean color sensors (e.g., Sentinel-3 OLCI). Sentinel-2 spatial resolution ranges from 10 to 60 m across the 13 spectral bands,⁵⁹ making it possible to obtain spectral data for narrow inlets with complex shoreline geometry. See Appendix Table 4 for central wavelength, bandwidth, and spatial resolution per spectral band.

2.5.

Description of Atmosphere Corrections

Three AC procedures, ACOLITE,³² MAIN,³² and Sen2Cor³³ were combined with spectral indices to identify optimal index-AC equations. These ACs were selected for previously demonstrated performance and because they can be implemented in Google Earth Engine. The ACOLITE dark spectrum fitting AC approach^31,34 has been demonstrated to achieve global-leading performance for optical water types with elevated chlorophyll-a and suspended solids concentrations in a comparison of state-of-the-art aquatic ACs.³⁰ Imagery processed with ACOLITE was produced by downloading imagery from the ESA Copernicus portal and processing with ACOLITE locally for analysis. The Modified Atmospheric correction for INland waters (MAIN) calculates AR and is designed to run within the Google Earth Engine environment.³² The Sentinel-2 Sen2Cor bottom-of-atmosphere reflectance image collection³³ is delivered by the ESA and is available in the Google Earth Engine data catalog as an analysis-ready dataset.³²

2.6.

Chlorophyll-a Detection Indices and Satellite Imagery Matches

The 17 chlorophyll-a detection spectral indices selected by King et al.²⁷ from the literature are applicable to Sentinel-2a/b based on the central wavelengths of spectral bands. Table 1 lists each index (also referred to as algorithms) that was tested herein and cites the paper that describes the equations represented by a given index name.^{22,24,60–66} Semi-analytical ratio indices are simplified empirical approximations of radiative transfer theory, whereas derivative spectral shape indices quantify the curvatures of spectral shape.²⁵ Semi-analytical indices are sensitive, and derivative indices are designed to be less sensitive to the effects of atmosphere.²⁵

The full in-situ chlorophyll-a dataset was initially matched with satellite imagery to maximize the sample size and to represent the range of conditions observed during this study. Filtering and masking logic were consistently applied across the three AC image collections to achieve a consistent comparison of index-AC performance.

Table 1

Spectral indices tested for the study area.27

A 90% cloud cover filter was applied to maximize the potential for matching in-situ water-quality data with satellite imagery across the regional study area. The Sentinel-2 operational cloud mask was applied for all image collections. Water detection and masking were applied using the normalized difference water index that utilizes Senintel-2 bands b3 and b8a (NDWI = (b3 - b8A)/(b3 + b8A)) to detect active water pixels and mask non-water pixels.⁶⁷ The chlorophyll-a spectral index value was retrieved using a 50 m radius for ACOLITE and 50 m diameter mean reducer for the MAIN and Sen2Cor image collections at each sampling site. In-situ chlorophyll-a samples were matched with satellite acquisitions nearest in time (±2 days) to the sample date.

The satellite imagery matches were manually reviewed across all three image collections for consistency and image quality at the sampling sites. Matchups were eliminated from the analysis in cases of excessive sun glint and when cloud or water masking did not remove all clouds or cloud shadows in proximity to a sampling site. In addition, the image collection matchups were manually reviewed to verify that there were no negative values in the blue wavelength at the sampling sites as low AR values have the potential to be overcorrected by AC procedures.²⁵

Because the ACOLITE AC is demonstrated to deliver high performance for waters with elevated concentrations of chlorophyll-a and suspended solids,³⁰ the chlorophyll-a detection indices were initially computed and ranked using ACOLITE in the R scripting environment.⁶⁸ If an index-ACOLITE combination achieved suitable performance metrics, then the index was tested across all three AC image collections.

2.7.

Estimating Chlorophyll-a Concentration with Spectral Indices

Univariate linear regression models were developed to relate the value of each spectral index-atmospheric correction combination to the observed chlorophyll-a concentration, each with the following form:

The coefficient of determination ($R^2$)⁷⁰ resulting from a regression between observed chlorophyll-a and index-AC values was used to identify the top performing index-AC combination. Coefficients of determination were computed and ranked for the 17 spectral indices combined with ACOLITE. Spectral indices in which at least a moderate fraction of the variance was explained by the regression model ($R^2>0.5$) using ACOLITE were then computed and ranked for their performance with the Sen2Cor SR and the MAIN-based AR image collections. The relation between in-situ chlorophyll-$a$ and in-situ turbidity was computed using the coefficient of determination. To explore and compare the spectral influence of turbid waters on the chlorophyll-$a$ estimator across the three AC procedures, the coefficient of determination was computed from the top performing chlorophyll-$a$ index and in-situ turbidity.

The linear regression models with the highest $R^2$ values for selected spectral indices that were used for chlorophyll-a concentration estimation were plotted with their corresponding prediction intervals and confidence intervals. The linear regression models were evaluated for performance across all sites, open-water sites, and inlet sites by comparing accuracy, precision, and bias.⁶⁹ Accuracy of predicted chlorophyll-a concentration values was assessed with both mean absolute error (MAE) and bias (systematic error), and root mean square error (RMSE) served as a measure of precision.⁷⁰ Although the RMSE and MAE metrics are similar, RMSE is influenced by outliers, whereas MAE is less sensitive to outliers.⁶⁹ Precision refers to the variability of the measurement error; the coefficient of variation (CV) was used to assess precision and is defined as the ratio of the standard deviation to the mean, thus expressing the extent of variability relative to the mean. Bias was measured as the average of the residuals, indicating under- or over-estimation of the true value.⁶⁹

The highest performing index-AC combinations were implemented in Google Earth Engine to produce reservoir-surface raster visualizations representing the spatial distribution of chlorophyll-a for visual comparison. Finally, a time series plot depicting the annual seasonal variation of chlorophyll-a was estimated for 50 m diameter mean values for all open-water and inlet sampling sites and filtered for cloud and sun-glint free satellite overpasses.

3.1.

Water-Quality and Reservoir Conditions

The chlorophyll-a concentrations measured in periodically collected in-situ water samples ranged from less than the reporting level of $<3$ to $87\mu \mathrm{g}/\mathrm{L}$; the turbidity ranged from 0.9 to 43.5 FNU, and the Secchi depth ranged from 0.06 to 2.83 m (Table 2). The phycocyanin fluorescence ranged from 0.18 to 5.25 RFU. The reservoir water temperature ranged from 10.1°C to 32.4°C.

Table 2

Descriptive statistics of in-situ chlorophyll-a (chl-a), turbidity, and Secchi depth collected for this study and compared to the satellite imagery matches subset.

The suitable satellite image overpasses ($n=49$) were matched with 44 field chlorophyll-a samples acquired at open-water and inlet sites. No satellite imagery matches were achieved for Lake Fork because the synchronized image acquisitions did not meet image quality thresholds at this reservoir. The resulting data are presented in tabular and graphical form. Summary statistics including the range, mean, and median values for chlorophyll-a, turbidity, and Secchi depth are provided (Table 2). A scatterplot (Fig. 3) depicts the seasonal variation of in-situ chlorophyll-a with corresponding satellite imagery matches. For the nine reservoirs with satellite imagery matches, a sunburst diagram (Fig. 4) depicts in-situ chlorophyll-a concentrations that were matched with satellite imagery by sampling date and site for each of the reservoirs.

Fig. 3

Scatterplot depicting seasonal variation of in-situ chlorophyll-a (Chl-a) concentration collected in the field with the subset of corresponding satellite imagery matches for open-water and inlet sites, January to November 2020.

Fig. 4

In-situ chlorophyll-a concentrations in micrograms per liter ($\mu \mathrm{g}/\mathrm{L}$) with the sample date (outer ring), sampling site (middle ring), and reservoir (inner ring) matched with satellite imagery selected within ±2 days of a satellite overpass, January to November 2020. The width of the star-burst plot is representative of the observed in-situ chlorophyll-a concentration sample value. The various shades of green differentiate individual reservoirs.

During the study period, all reservoirs were at some point dominated in relative biovolume by cyanobacteria (i.e., Cyanophyta); by contrast, Cryptomonads algae (i.e., Cryptophyta) and diatoms (i.e., Bacillariophyta) emerged as the dominant taxonomy less frequently for a subset of reservoirs.⁴⁶ Cyanobacteria from mid-June to mid-October dominated at most reservoirs except for the Jim Chapman Reservoir in August. Cyanobacteria had a larger dominance relative to other groups in the summer months (June to August) that declined in September and October. Bookending the cyanobacteria-dominated months were diatoms. Diatoms had the highest relative biovolume from January to early June and then again in late October to November. The shift in dominance may be from relatively high water temperatures in north Texas during the late spring to early autumn. Butterwick et al. observed that the diatom, Asterionella formosa, had slightly lower growth rates than the cyanophyte, Tychonem bourrellyi, at temperatures above 25°C, whereas the reverse is true for temperatures below 10°C. Cryptomonads were dominant at only one point in time (April) at only one reservoir (Grapevine). Figure 5 depicts dominant phytoplankton taxonomy and sampling calendar dates. The companion data release⁴⁶ provides water-quality, phytoplankton taxonomy, and hyperspectral data for the sampling locations. Water-quality and phytoplankton taxonomy data are also stored in the USGS National Water Information System database (NWIS).⁷¹

Fig. 5

Dominant phytoplankton taxonomy relative total biovolume (%) across reservoirs at open-water sampling sites, designated with sampling date, January to November 2020. Cyanobacteria are symbolized by a blue circle; cryptomonads are symbolized by a dark green diamond; and diatoms are symbolized by a green square.

3.2.

Satellite Chlorophyll-a Detection Performance

The Moses three-band spectral index (Moses3b) produced the highest ranked $R^2$ values across all index-AC combinations led by the Sen2Cor AC. The results for all index-AC combinations are provided in the Appendix [Fig. 11]. Moses3b applied to ACOLITE, MAIN, and Sen2Cor yielded $R^2$ values of 0.66, 0.62, and 0.70, respectively, for all sites (Table 3). The Moses3b-Sen2Cor combination was selected and implemented as the chlorophyll-a estimator and further evaluated for performance (Fig. 6). All Moses3b AC combinations at inlet sites outperformed open-water sites in terms of the $R^2$ metric. The Moses3b-Sen2Cor combination delivered the lowest overall measure of MAE and RMSE for all sites, as well as for the inlet sites (Table 3). All three Moses3b-ACs combinations produced similar variance magnitudes for all sites and the open-water and inlet classes. For the all sites group, the three combinations produced bias-free results (each delivered the same bias of 0 for all sites) and produced consistent positive bias for open-water sites and negative bias for inlet sites; refer to Table 3 for Moses3b performance comparison across ACs and site descriptions (all sites, inlet sites, and open-water sites). The correlation of in-situ chlorophyll-a and in-situ turbidity resulted in a low $R^2$ value ($R^2=0.35$) (Appendix Fig. 12). The influence of turbidity on the remotely sensed chlorophyll-a signal was evaluated by correlating the results of Moses3b values with in-situ turbidity. $R^2$ values ($R^2=0.11$ for ACOLITE, $R^2=0.33$ for MAIN, and $R^2=0.17$ for Sen2Cor) indicated that the influence of turbidity on the Moses3b index was consistently low (Appendix Fig. 13).

Table 3

Aggregated performance metrics for the Moses3b chlorophyll-a detection spectral index comparing ACOLITE, MAIN, and Sen2Cor performance for sampling sites.

Fig. 6

Comparison of in-situ and satellite estimated chlorophyll-a concentration ($\mu \mathrm{g}/\mathrm{L}$) using the Moses3b index across all matchups applied to (a) ACOLITE AR, (b) MAIN AR, and (c) Sen2Cor SR.

Grapevine Lake was used as an example for a qualitative comparison of index-AC combinations, in-situ water-quality data, and hyperspectral reflectance data because the August 19, 2020, site visit to this lake corresponded with high-quality satellite-imagery acquisition. The highest performing index-AC combination, Moses3b-Sen2Cor, was compared across the AC image collections and the highest-ranking indices to map and visualize aquatic chlorophyll-a concentrations (Figs. 7 and 8). In Fig. 7, chlorophyll-a concentration estimates obtained from the Moses3b index across the three ACs are compared {Sen2Cor [Fig. 7(a)], ACOLITE [Fig. 7(b)], and MAIN [Fig. 7(c)]}. In Fig. 8, the patterns of chlorophyll-a concentration estimates for the three highest performing chlorophyll-a indices applied to the Sen2Cor AC image collection are compared (Moses3b ($R^2=0.70$), S23BDA ($R^2=0.65$), and BR54 ($R^2=0.61$)] (Appendix Fig. 11).

Fig. 7

Images of chlorophyll-a concentration estimates obtained from the Moses3b index obtained by applying the (a) Sen2Cor, (b) ACOLITE, and (c) MAIN ACs for Grapevine Lake, August 19, 2020.

Fig. 8

Visual patterns of chlorophyll-a concentration estimates for the three highest-performing chlorophyll-a indices applied to the Sen2Cor AC image collection: (a) Moses3b, (b) S23BDA, and (c) BR54 for Grapevine Lake, north Texas, August 19, 2020.

Fig. 9

In-situ hyperspectral reflectance from Grapevine Lake with selected Sentinel-2 spectral bands showing the magnitude and shape of chlorophyll-a and phycocyanin spectral features acquired August 19, 2020, for the inlet and open-water sampling sites. Solid vertical lines indicate light measured by Sentinel-2 spectral bands for green (band 3), red (band 4), red edge (band 5), and NIR (band 6 and band 8a); dashed vertical lines correspond to the phycocyanin absorption trough at about 620 nm, the chlorophyll-a absorption trough at about 675 nm, peak chlorophyll-a reflectance at about 700 nm, and NIR absorption by water at about 740 nm. The “red edge” region of the spectrum extends from 670 nm to 700 nm and indicates the presence of chlorophyll-a.

Figure 9 depicts the relation between chlorophyll-a and phycocyanin spectral features acquired for the open-water and inlet sites compared with the Sentinel-2 spectral bands used in the chlorophyll-a detection indices (Table 1). The reflectance and absorption brightness magnitude of the spectrum indicate photosynthetically active waters with the unique spectral signature of cyanobacteria:⁵⁷ the green peak at about 560 nm (partially captured by Sentinel-2 band 3), the phycocyanin absorption trough at about 620 nm, the chlorophyll-a absorption trough at about 675 nm (partially captured by Sentinel-2 band 4), the NIR reflectance/scattering peak at about 700 nm (partially captured Sentinel-2 band 5), and NIR absorption by water at about 740 nm (partially captured Sentinel-2 band 6). Figure 9 shows there are more pronounced spectral features at the inlet site compared with the open-water site. The taxonomy observed at the open-water site was dominated by cyanobacteria.⁴⁶ In-situ chlorophyll-a concentrations, phycocyanin, and turbidity were concurrently higher at the inlet site than at the open-water site (chlorophyll-a was $19\mu \mathrm{g}/\mathrm{L}$ at the inlet site and $18\mu \mathrm{g}/\mathrm{L}$ at the open-water site, phycocyanin was 1.36 RFU at the inlet site and 1.16 RFU at the open-water site, and turbidity was 9.27 FNU at the inlet site and 3.26 FNU at the open-water site). The Secchi disk depth was observed to be greater at the open-water site (0.79 m), compared with the inlet site (0.39 m).

Finally, annual chlorophyll-a concentration was estimated across sampling sites from January to December of 2020. Chlorophyll-a concentrations were estimated from a 50 m diameter mean area for each open-water and inlet site from regional-scale cloud and sun-glint free satellite overpasses ($n=49$). Seasonal variations in chlorophyll-a concentrations were observed for open-water and inlet sites (Fig. 10).

Fig. 10

Moses3b-Sen2Cor mean chlorophyll-a concentration estimation simulation reveals seasonal variation for the open-water and inlet sites, January 2020 to December of 2020.

4.1.

Limitations and Areas for Improvement in Real-Time Satellite Monitoring of Water-Quality

Challenges associated with detection of algae and cyanobacteria from satellite sensors are not unique to this study and are well documented.²⁵ Real-time earth observation requires robust approaches that are resilient to changes in dynamic environmental conditions including water-quality, phytoplankton communities, and atmospheric conditions. Satellite sensor technology is currently (2023) undergoing rapid development that is enabling advanced spectral resolution, increased overpass frequency, imaging across a range of spatial scales, interoperability and harmonization of satellite sensors, and high performance cloud platforms that facilitate advancements in geoprocessing and spatial data science.^27,36,72

The chlorophyll-a concentration estimates described herein are limited to the in-situ matches between water-quality data and satellite imagery obtained during the study period. Because of the performance uncertainties described herein, chlorophyll-a concentrations estimated from satellite imagery would benefit from additional field observations for validation purposes. Deploying field crews across the region for in-situ water-quality monitoring coordinated with cloud-free satellite overpasses is logistically challenging. The detection and estimation process would benefit from the integration of additional data (e.g., from shore and buoy sensors⁷) to better inform and study the complexities of the dynamic reservoir systems observed across the north Texas region. Considering the relatively stronger performance at inlet sites, follow-up studies could benefit from calibration efforts that specifically target varying aquatic conditions observed within the reservoirs.

Chlorophyll-a estimations produced outliers for the highest chlorophyll-a observations in the in-situ dataset (chlorophyll-a: $87\mu \mathrm{g}/\mathrm{L}$; Secchi depth, 0.49 m at Lake Lavon, Site L1 on August 6, 2020) for the Moses3b approach across all ACs (Fig. 6). The Moses3b-ACOLITE combination yielded the second highest chlorophyll-a observation (chlorophyll-a: $83\mu \mathrm{g}/\mathrm{L}$; Secchi depth: 0.37 m at Lake Lavon, Site L1 on September 14, 2020). The Moses3b index combined with Sen2Cor and MAIN each generated a single negative estimate, when ACOLITE did not produce negative estimated values. The minimum reporting limit for chlorophyll-a of $3\mu \mathrm{g}/\mathrm{L}$⁴⁹ is used for calibration and presents an opportunity for improvement in chlorophyll estimation $<3\mu \mathrm{g}/\mathrm{L}$ because negative estimates are physically impossible. This study did not examine the intercomparison of the AC techniques and therefore cannot provide an explanation for why individual AC procedure performances achieve different outcomes; the use of AC factors is an ongoing topic of research in the remote sensing community.³⁰

The cloud-based remote sensing platform facilitates the seamless integration of additional satellite sensors which would improve the likelihood of obtaining high-quality cloud and sun-glint-free image acquisitions. Sensors with expanded spectral resolution that can detect phycocyanin (e.g., Sentinel 3) could be ingested into the real-time earth observation system to leverage sensor assets and capabilities across multiple satellite platforms.³⁰ Cyanobacteria genera reveal unique hyperspectral signatures,⁵⁷ and when applied to space-borne hyperspectral imagery, they have demonstrated potential to differentiate aquatic cyanobacteria genera⁷² advancing detection beyond chlorophyll-a and phycocyanin proxy estimators of algae and cyanobacteria.

Clouds, smoke, wind, and sun glint can impair the performance of satellite chlorophyll-a retrieval. Although AC routines and cloud cover masking are designed to correct for atmospheric effects on satellite detections, extensive cloud cover presents a physical limitation for real-time monitoring. Cloud cover is common in the study area. For example, in part of north Texas consistent with the study area, Wilson and Jetz⁷³ reported the presence of cloud cover in 47% of overpasses within the study area using high frequency daily MODIS satellite overpasses. As technology improves, cloud and cloud shadow detection and masking, as well as sun-glint correction techniques, will likely continue to advance the quality of near real-time earth imaging.