Spatially Intensive Patterns of Water Clarity in Reservoirs Determined Rapidly With Sensor‐Equipped Boats and Satellites

Optical water quality affects the quantity and spectral composition of underwater light available to photoautotrophs along with light reflected off the water surface. In reservoirs, prominent gradients for optical water quality and clarity measures occur between inlets and outlets, allowing rapid surveys of ecosystem structure across diverse conditions along with tests of relationships between in situ and remotely sensed water indices. We used a sensor‐equipped boat in a clay‐rich subtropical drinking water reservoir to examine continuous spatial patterns of water clarity and the relationship between in‐lake turbidity and normalized difference turbidity index (ndti) from Sentinel 2 imagery acquired on the same date. Over the 62.3 km boat path, surface turbidity varied between 14 and 85 NTU (mean = 30), with highest values in shallower water near the main inlet. Results indicated a strong linear relationship between sensor turbidity and lab‐determined turbidity. A positive relationship between the satellite turbidity index and in‐lake turbidity, combined with a negative exponential relationship between Secchi depth and turbidity, provided a basis for predicting water clarity metrics continuously over the entire lake surface. In this turbid water body, predicted Secchi depth and turbidity over the whole lake had means of 0.22 m and 54 NTU, with 75% of the lake area <0.29 m and >26 NTU. Pursuit of general relationships involving optical properties of water from high‐speed, data‐intensive spatial surveys and remotely sensed surface reflectance facilitates further development of spatially explicit models of aquatic systems.

compared to open water, or shallow areas that supply resuspended sediment and organic matter during wind events.Reservoirs often possess particularly large gradients for turbidity and covarying factors between inlets and the dam (Thornton et al., 1982).Turbid conditions are frequently associated with seasonal runoff events, but systems with fine clays and shallow water depth may have elevated turbidity seasonally or year-round (Knowlton & Jones, 2000;Lind et al., 1992;Townsend, 2002).These water clarity differences influence the underwater light environment, determining how much photosynthetically active radiation reaches which locations and depths.
The most common apparatus for in situ measurement of water clarity in lakes, the Secchi disk, requires visual observation by humans over a fixed sampling point while remaining as stationary as possible.The simplicity and low cost of Secchi depth measurement has produced widespread water clarity data over many decades (Rose et al., 2016), but frequently for only one or a few sampling points in each lake on a given date.In contrast, turbidity sensors allow continuous and frequent measurements, producing high volume data.Using turbidity sensors placed on boats, researchers can obtain spatially continuous measurements of water clarity and other sensor variables logged at high frequency while moving (Crawford et al., 2015;Loken et al., 2018).Mobile sensor technologies are expected to grow as field crews increasingly use autonomous surface and subsurface crafts to generate water quality maps.
Remote sensing data from satellite missions such as Sentinel and Landsat can help track optical water quality at local to continental scales.Uncertainty about the strengths and limitations of surface reflectance indices such as normalized difference turbidity index (ndti, Kuhn et al., 2019) necessitates direct comparisons of remote and in situ data.Imagery from the Sentinel 2 multispectral instrument supports remote sensing of water at high spatial resolution (10 m grid), with frequent image acquisition (5 days return interval at equator, 2-3 days return interval at mid-latitudes) and near real-time data availability for bands required to compute indices of turbidity and water color.Challenges in analysis of surface reflectance include atmospheric interference from cloud cover or haze, cloud shadows, sun glint off the water surface, and inconsistent sun angles at different times or places.Amid these challenges, there have been few direct comparisons of spatially explicit reservoir turbidity data collected in high volume at the same time as satellite image acquisition.
Large spatial gradients in reservoirs provide promising opportunities to address the question: How is spatial structure linked to ecosystem function?To work towards this question, studies of the fundamental spatial structure of properties such as turbidity are still needed.The goals of this work are to understand (a) spatially intensive patterns of turbidity and associated water clarity measures over the entire surface of a reservoir, and (b) the relationship between in situ turbidity and satellite-derived turbidity (ndti) acquired on the same date.To address these goals, we conducted a rapid spatial survey using a sensor-equipped boat in a warm subtropical, southern plains drinking water reservoir (Lake Arrowhead, Texas, USA) that had a wide turbidity gradient between upper and lower sections.We timed our spatial survey to occur on an image acquisition date for the Sentinel 2 satellite mission.This system is one of >1,000 lakes and reservoirs that were sampled comprehensively in the 2022 National Lake Assessment (NLA) by national and state partners (Pollard et al., 2018), and we timed our survey within 48 hr of NLA sampling to allow future analyses.We expected to find a wide turbidity gradient along navigable sections of the reservoir, and a positive relationship between turbidity and satellite ndti.By integrating data from rapid spatial surveys and remote sensing, researchers may be able to generate rule sets that predict spatial heterogeneity of water quality in exquisite detail, over entire surfaces of diverse water bodies.

Site Description
Lake Arrowhead (33.75728 N,98.35888 W) was constructed in 1966 on the Little Wichita River as a water source for the city of Wichita Falls in northern Texas.The reservoir has a surface area of approximately 60.6 km 2 , maximum depth of 13.7 m, watershed area of 2140 km 2 , and watershed area: lake area ratio of 35.3.The watershed is mostly rural, with grassland as the dominant land cover type, followed by shrubland and cropland (Table 1), and soils in the watershed had a mean clay content of 37.6% dry weight.Mean annual air temperature (1981-2010) was 17.6 C, and mean annual precipitation (1981-2010) was 753 mm.These watershed variables were obtained using a custom web-based ShinyApp that returns records from the Streamcat data set (Hill et al., 2016) for National Hydrography Data set flowline segments located nearest to or downslope from user-specified points, aided by the nhdplusTools R package (Blodgett et al., 2022).This ShinyApp is available at https://waterfolk.shinyapps.io/streamcat/ and can retrieve StreamCat data for flowline segments nearest to any location in the conterminous United States.

Field Survey Methods
We conducted a rapid spatial survey of surface water in Lake Arrowhead on 2022 Aug 12 using methods similar to those of the Fast Limnology Automated Measurement platform used in Crawford et al. (2015) and Loken et al. (2018).Prioritizing water overlying the main river channel and navigable sections of reservoir arms, our sensor-equipped boat (Whaler Montawk 210) covered 62.3 km in 174 min.The mean boat speed over the entire survey, excluding times when the boat was stopped but including periods of slower travel through shallow (<1.5 m) reservoir arms, was 23 km/hr.In open water areas with sufficient depth for fast navigation, boat speeds ranged from 30 to 50 km/ hr.With this method, a diaphragm pump brings surface lake water from the top half meter through a transom-mounted intake pipe onto the boat deck in a continuous flow-through system containing a YSI Exo 2 sonde, flow cell, and probes for turbidity, specific conductivity, pH, and water temperature.Sensors were calibrated in the lab prior to the survey.A Garmin GPS 18X LVC was used to log latitude, longitude, and boat speed.All sensor and GPS data were logged in the field at 1 s intervals, generating >10,000 raw turbidity values over the survey.A Campbell CR6 data logger was used to store logged data and stream data in real time to mobile devices on the boat via a dedicated wireless connection.Real time monitoring of sensors facilitated adaptive water sampling in the field, ensuring that spatial contrasts were captured for supporting laboratory analyses of water.
Sensor data collection began at 9:49 a.m.local time and ended at 12:43 p.m., with the boat path and data collection times mapped in Figure S1.In this survey, we pumped water through the sonde flow cell at 3.44 L/min.Discrete 1 L water samples were collected from the top half meter of water at 5 different sites throughout the lake.Sites were sampled in the following order: A1 = arm 1, near the dam, a location that had 3 m of water depth; U = upper reservoir, near main inlet, 2 m depth; L = lower reservoir, near the dam, 9 m depth; M = middle reservoir, 5 m depth; A2 = arm 2, near main inlet, 4 m depth.Water samples were collected from the outflow of the sonde flow cell in an effort to capture water closest to the sensors.At water sampling sites, the boat speed was briefly decreased to zero km/hr to allow simultaneous measurement of Secchi depth.A 2 minute period of questionable survey data collected during an abrupt 180° turn (between 12:33 and 12:35 local time) was excluded from all analyses.
With a constantly moving boat, sensor readings do not reach equilibrium and instead reflect conditions of water integrated over recent seconds.Hydraulic lags between the water intake and turbidity sensor, and minimum response times for the sensor, were accounted for in analyses using methods described in Loken et al. (2019) and below.These corrections to raw sensor values provide more accurate estimates in space and time for in-lake conditions associated with each sensor reading.Before applying these corrections, we smoothed raw sensor data using a rolling mean of 3 observations to reduce noise in the 1 Hz data.To account for the physical distance water must travel through the intake pipe, tubing, and sonde flow cell before being sensed by the turbidity probe in this flow system, we used a hydraulic time constant (Hr) of 21 s.Consequently, logged sensor time was adjusted to lake time by subtracting 21 s, and then sensor values were assigned the Garmin GPS latitude and longitude values that had the same lake time.Xt-Hr refers to the sequence of hydraulically corrected sensor values.For this turbidity sensor, we used the parameter time constant (τs) of 5 s, where τs is the time required for a 63% response to a step-change input (Fofonoff et al., 1974;Loken et al., 2019).Reported sensor values (X0) were calculated as in Fofonoff et al. (1974) using X0 = Xt -Hr + τs dx/dt, where dx/dt is the instantaneous rate of change of sensor output based on a 3-point moving window around sensor values.After these corrections, sensor readings are resolved to <10 s, which corresponds to <100 m at medium boat speeds (35 km/hr) frequently experienced in this survey, <150 m at high speeds (55 km/hr), and <40 m at slower speeds (15 km/hr) that were sometimes required for safe navigation over shallower water depths (<1.5 m).

Lab Methods
Discrete water samples were brought back to the lab for turbidity analysis within 24 hr of collection on a Hach 2100N turbidimeter calibrated with five Formazin standards ranging from 0.01 to 4000 NTU.From each water Note.The lake has a surface area of 60.6 km 2 and maximum depth of 13.7 m.

Table 1
Features of the Lake Arrowhead Watershed sample, 200 mL was filtered onto a preweighed Whatman 47 mm glass fiber filter with an effective pore size of 0.7 μm (GF/F) and dried in a 100°C oven for a minimum of 24 hr.Dried filters were weighed to the nearest tenth mg and used to calculate the concentration of total suspended solids (TSS) in mg/L.Percent organic matter in suspended solids was measured as ash-free dry mass following ashing at 550°C for >1 hr.

Satellite Methods
We queried Sentinel 2 multispectral imagery from the Google Earth Engine data catalog (COPERNICUS/S2_SR_ HARMONIZED, Level 2A) using the rgee package (Aybar et al., 2020) in R (version 4.2.1, R Core Team, 2022).
For each point along the boat's path, surface reflectance band values and quality assurance values from the Sentinel 2A image acquired on the same date of the survey, 12 August 2022, were extracted into a tabular format.A second set of surface reflectance data for the whole lake on the nearest cloud-free date, 2022 July 28 (from Sentinel 2B), was separately queried using regular sampling of 25,000 evenly spaced points within an area of interest that encompassed the entire lake and immediately surrounding land areas.The scale parameter was set to 10 m in these Google Earth Engine queries.Normalized difference turbidity index (ndti) was calculated as ndti = (Red -Green)/(Red + Green) = (band 4 − band 3)/(band 4 + band 3) (Garg et al., 2020;Lacaux et al., 2007).Band three (560 nm on S2A, 559 nm on S2B) and band 4 (664.5 nm on S2A, 665 nm on S2B) have 10 m resolution.The Sentinel 2 SCL field (scene classification algorithm) obtained from Google Earth Engine was used to identify pixels obscured by clouds, cirrus, or cloud shadows.However, for the Sentinel 2 image acquired on 12 August 2022, the SCL field identified only a small proportion of the cloud shadows occurring on the lake surface.To more thoroughly identify dark portions of the lake surface corresponding to cloud shadows on this particular date and location, we used band 2 (496.6 nm on S2A, 492.1 nm on S2B, 10 m resolution), classifying pixels with band 2 values <500 as shaded.Neighbor pixels falling within a 0.0005° radius (45-55 m at this latitude) to the aforementioned cloudy, cirrus, or shadowed pixels were also identified, flagged, and excluded from our analysis of ndti in order to focus on the highest quality imagery.The different classes of pixels falling along the boat path were mapped in Figure S2 in Supporting Information S1, and these were represented in the following percentages: 48.9% water, 33.7% clouds, 17.1% cloud shadows, 0.29% cirrus.After subtracting out pixels that neighbored clouds, cloud shadows, or cirrus, 44.9% of the pixels remained for analysis of ndti.All pixels used in calculating ndti had a cloud probability of 0 according to the cloud probability mask field.For analysis of the cloud-free image acquired on 28 July 2022, we used the SCL field to limit the lake's geometry to pixels classified as water.

Data Analysis
We used R/Rstudio to conduct all analyses and create all figures and maps using the following packages: ggplot2, tidyr, Openstreetmaps, rgee, rshiny, nlstools (Aybar et al., 2020;Baty et al., 2015;Chang et al., 2022;Fellows, 2019;Wickham et al., 2023aWickham et al., , 2023b)).For the five lake locations where we collected discrete water samples, paired relationships between sensor turbidity, sample turbidity, Secchi depth, TSS, and ndti were examined.Linear regression and nonlinear least squares were used to fit expected relationships between turbidity, Secchi depth, and water sample data.For comparing sensor turbidity values with water sampling data, we took the mean of sensor data logged at locations falling within a 0.0003° radius (27-33 m at this latitude) from the water sampling location.
For sensor data logged from a boat at regular time intervals, changes in boat speed produce uneven sampling distances in space.To represent sensor values evenly in space we resampled the data at regular distance intervals of 10 m along the boat path using interpolation, which preserved the time sequence of changes and gave more than >6,000 turbidity values for analysis and visualization.Resampled sensor turbidity and supplemental sensor variables (conductivity, pH, water temperature) were mapped (Figure S3 in Supporting Information S1).Relationships between resampled sensor turbidity and satellite turbidity (ndti) were examined for the Sentinel 2 image acquired at 12:24 local time (17:24 UTC) on the same date as the spatial survey.We examined relationships between sensor turbidity and satellite turbidity using all available data (collection times between 9:49 and 12:43 local time, or 14:49 and 17:43 UTC), as well as more closely time-matched subsets of data collected within 90 and 20 min of the image acquisition time.

Results and Discussion
Water clarity is an important feature of lakes with broad consequences for underwater light availability, autotrophic communities, vision-oriented or light-sensing organisms, surface reflectance detected by satellites, and human perceptions of water quality.Our results provide an example showing wide gradients for turbidity, a key measure of optical water quality and water clarity, mapped rapidly at high resolution in a drinking water reservoir using a sensor-equipped boat and satellite reflectance data.Wide gradients such as those observed in Lake Arrowhead, Texas in August 2022 are useful for interrogating relationships between remotely sensed indices and in-lake measures of water clarity.Few other examples of spatially intensive surveys of lake or reservoir water clarity exist (but see Rodriguez et al., 2020).Other lake and reservoir properties such as chlorophyll, colored dissolved organic matter, or management-relevant factors such as dissolved oxygen and toxins may covary with turbidity or remotely sensed indicators (Handler et al., 2023), providing diverse motivations for understanding the spatial structure of these water constituents.The ability to conduct rapid snapshot surveys of in situ water quality over large distances (tens of km) allows relatively small research teams to match numerous new observations with satellite imagery closely, before spatial structure of the system has time to meaningfully change in association with factors such as wind, currents, or algal blooms.
Prior to the survey date, Sentinel 2 satellite imagery suggested a prominent longitudinal gradient with highest turbidity in the upper (southwest) part of this reservoir (Figure 1).In relation to our first goal of understanding water clarity over a whole reservoir, the initial 60 min of surveying with the sensor-equipped boat confirmed the large longitudinal gradient in turbidity between lower and upper sections of this system (Figure 2), with turbidity values ranging between 15 and 85 NTU over this approximately 15 km segment of boat path.Return trips between lower and upper sections of the reservoir on this survey date and excursions into three separate reservoir arms confirmed the spatial turbidity patterns multiple times in less than 3 hr of total time on the water along 62.3 km of boat path.The 2nd and 98th percentile turbidity values of 14 and 82 NTU from this survey indicated that spatial gradients in Lake Arrowhead spanned much of the range represented across human-constructed systems in Texas (n = 93) that were sampled in 2012 and 2017 for the National Lakes Assessment (NLA) (U.S. Environmental Protection Agency, 2022).More specifically, turbidity in human-constructed NLA systems of Texas had a log-normal distribution with 2nd and 98th percentiles of 1.02-138 NTU (mean = 17.5, median = 5.13), while NLA human-constructed systems nation-wide (n = 1,160) had 2nd and 98th percentiles of 0.42-77.1 NTU (mean = 13.2, median = 3.82).Uppermost parts of the reservoir at the Little Wichita River inlet could not be surveyed due to insufficient depth for safe navigation, but we still encountered high turbidity water near the inlet.In many reservoirs, turbidity values range widely over time.For example, Bayram and Kenanoğlu (2016) reported turbidity values from <10 to 848 NTUs over one year in the same system.Fewer studies have reported spatial turbidity differences within the same system on the same date over a similarly wide gradient as we observed in Lake Arrowhead.
While storms often cause suspended sediment plumes in reservoirs, especially near inlets, high turbidity in the upper section of this system was persistent   (Knowlton & Jones, 2000;Macías & Lind, 1990).In such shallow waters, regular cooling at night can also mix the water column and resuspend fine particulates.
Water samples and Secchi depths collected at five different sites supported the turbidity sensor readings from the spatial survey, and confirmed low water clarity over much of Lake Arrowhead on the survey date.Secchi depths were <0.7 m at all five sites, and similarly shallow Secchi depths are not uncommon in reservoirs of this region (Wagner et al., 2023).Sensor turbidity was strongly correlated with turbidity measured from water samples brought back to the lab (r 2 adj 0.999, Figure 3), with linear regression statistics indicating that sensor readings were biased slightly high (slope = 1.18 ± 0.0146, y-intercept = −0.547± 0.539, p = 0.39).There was a negative exponential relationship between Secchi depth and turbidity, as expected based on previous research (Jones & Knowlton, 2005;Rodrigues et al., 2020).Total suspended solids and turbidity were positively related (r 2 adj 0.906, Figure 3, slope = 0.425 ± 0.0676, y-intercept = 7.28 ± 0.0136, p < 0.01) (Davies-Colley & Smith, 2001).The lower site (L) closest to the dam had the lowest turbidity (11.8 NTU) and higher organic matter content (44% dry weight) compared to the other water sampling sites, as did the next closest site to the dam (site A1, 15.6 NTU, 42% of dry weight) reflecting a higher relative contribution from phytoplankton or other organic material.
In relation to our second goal of understanding the relationship between satellite turbidity (ndti) and in-lake turbidity, ndti from Sentinel 2 was positively related to sensor turbidity values along the boat path (Figure 4, r 2 adj = 0.94 using a log(x) model) when restricted to pixels free of clouds and cloud shadows.Despite considerable cloud cover in the Sentinel 2 image on this date (Figure 1, Figures S3 and S4 in Supporting Information S1), 49% of the boat path was free of clouds or cloud shadows, providing useful data for testing this relationship across the wide range of turbidity we encountered over the survey.Based on Figure 4, the observed satellite-sensor relationship was likely nonlinear, as expected based on previous work on satellite derived indicators (Unsalan & Boyer, 2004).A subset of survey data collected within 20 min of the Sentinel 2 image acquisition time ranged from 14 to 50 NTU, and thus represented a considerable fraction of the overall turbidity gradient.Figure S5 in Supporting Information S1 shows there was a strong linear relationship between ndti and in-lake sensor turbidity for both the subset of data collected within 20 min of image acquisition (r 2 adj > 0.95), and data collected within 90 min of image acquisition (r 2 adj > 0.94).Regression statistics indicated these 20 and 90 min matched data had similar slopes (0.0069 ± 5.1 × 10 −5 , p < 0.0001; 0.0075 ± 4.6 × 10 −5 , p < 0.0001) and intercepts (−0.30 ± 0.0019, p < 0.0001; −0.33 ± 0.0016, p < 0.0001).Capturing a wide gradient of turbidity was likely important for finding relationships between satellite reflectance and spatial survey data.Similar relationships between ndti, sensor turbidity, and Secchi depth have been found using less intensive spatial survey data in reservoirs of South-Central Chile (Rodríguez-López et al., 2021) and India (Bid & Siddique, 2019;Pompapathi et al., 2022).Spatial surveys over narrower gradients than those observed in Lake Arrowhead may not produce such strong relationships with satellite indices.Focusing on the lower end of the turbidity distribution (<25 NTU), there was considerable variation around the relationship between boat-based and satellite-based turbidity (Figure 4), suggesting that satellite indices of turbidity may not necessarily distinguish between conditions found in lakes and reservoirs with higher water clarity.The high volume data generated by this spatial survey approach facilitates testing of satellite ∼ sensor relationships by ensuring at least some survey data are collected at locations free of clouds and cloud shadows.In addition, because the entire boat path on this survey date was contained in one satellite scene, ndti values were derived for a common angle of observation, sun angle, and atmospheric condition, which may not be the case when multiple scenes are used to track differences in time or space.
Our work demonstrates the value of adaptive sampling informed by real-time sensor readings to capture gradients and transition points during spatial surveys of aquatic ecosystems.The locations for water sampling and lab determination of turbidity, TSS, and suspended organic matter were decided on the water during the survey, based on sensor readings relayed to mobile devices along with visual cues from observing water color change.While large contrasts in turbidity can often be discerned from a moving boat using visual cues, real time tracking with sensors opens opportunities for adaptive sampling of other variables such as oxygen and nitrate that are not directly visible to the human eye.Researchers often have a priori expectations of where higher/lower values or transition points for turbidity and other factors such as chlorophyll will be found, but reality does not always meet these expectations, especially for less familiar water bodies that may be encountered when field crews must visit many lakes.With such instantaneous feedback from sensors, the risks and costs of oversampling or undersampling with manual techniques can be reduced by prioritizing water sampling effort at locations that span gradients most relevant to objectives.Reservoirs are particularly useful systems for testing relationships between in-lake and remotely sensed optical water quality due to prominent gradients grounded in established theory (Thornton et al., 1982).The large spatial differences in optical water quality frequently encountered in just one or a few reservoirs, and differences between reservoir arms, may be used to accelerate development of spatially explicit models of diverse aquatic ecosystems, particularly for aspects of spatial structure that trend with surface reflectance.Integrating goals 1 and 2 using a combination of spatial survey data, remotely sensed surface reflectance, and water sampling at fixed points, statistical relationships such as those in Figures 3 and 4 provide a basis for predicting water clarity measures over unsampled areas of lakes and reservoirs.For example, Figure 5 shows maps of predicted turbidity, suspended solids, and Secchi depth over the entire surface of Lake Arrowhead on a cloud-free day (2022 July 28) 2 weeks prior to the boat survey date.To generate these maps, ndti values from 28 July 2022 were fed into the fitted ndti ∼ turbidity relationship determined along the boat path on 12 August 2022 in order to predict the complete turbidity surface in NTU.In turn, turbidity predictions were fed into Secchi ∼ turbidity and TSS∼turbidity equations determined from the five sampling sites (Figure 3) in order to predict the complete Secchi and TSS surfaces.Such complete maps allow us to understand the frequency distribution of conditions lake-wide in established units relevant to water research and management, and we can calculate whole lake summary statistics.For example, predicted Secchi depth and turbidity over the entire water surface had means of 0.21 m and 55 NTU (Table 2), with 75% of the lake area <0.29 m and >27 NTU.These kinds of whole system summary statistics have not been frequently available to lake and reservoir researchers and managers because field crews often prioritize sampling only near the deepest point or just a few sites in each system.Related questions include: How many surveys across systems and dates are necessary to generate robust, general, spatially explicit predictions of optical water quality at sampled or unsampled locations and times?How much surface area or volume do observations over the deepest point represent?More spatial surveys across a wider range of systems and dates would be necessary to 10.1029/2023JG007650 9 of 11 address these questions effectively, in part because photosynthetic pigments and the dissolved organic matter pool (Stadler et al., 2023) also influence satellite indices of optical water quality and relationships like the one shown Figure 4. Stated another way, environmental water is a complex mixture of particulate and dissolved substances that together affect surface reflectance, and more than one model may be useful or necessary to represent constituent gradients within different kinds of mixtures.We foresee researchers continuing to develop models that integrate satellite indices with in situ survey data to track and forecast optical water quality spatially intensively, across water bodies and over time, extending to both sampled and unsampled water bodies as training data accumulate.

Conclusions
Rapid spatial surveys with mobile sensor platforms provide a data-intensive approach for understanding gradients of water clarity and other optical water properties in lakes and reservoirs.We successfully captured spatially intensive patterns of surface turbidity in a subtropical reservoir using a sensor-equipped boat that continuously logged surface water data while frequently traveling at speeds >40 km/hr.Conducting surveys at this speed allows researchers to generate detailed snapshots of spatial structure before the system has time to change in association with factors such as wind, currents, or algal blooms.Consequently, when spatial surveys can be closely timed with satellite acquisition schedules, powerful and direct comparisons of in situ and remote sensing data are possible even for relatively small research teams.The wide gradient of turbidity we encountered in our survey of Lake Arrowhead, TX, was visible from space, contributing to a positive relationship with the satellite turbidity index from Sentinel 2 imagery acquired on the same date.Limitations of our current spatial survey approach include the restriction to variables that environmental sensors can measure, an inability to draw deep water while the boat is moving at high speed, and needs for a powered boat, trained field crews, and sufficient water depth for navigation.However, smaller autonomous and semi-autonomous rigs can address some of these limitations as they continue to become more feasible for a diversity of sensor technologies over time.Spatial surveys and remote sensing provide a basis to understand heterogeneity over entire surfaces of lakes and reservoirs.By integrating these sources of information, researchers may be able to generate rule sets that model and forecast optical water properties in aquatic ecosystems spatially explicitly.Reservoirs are particularly useful systems for capturing gradients in optical water quality due to the contrasting conditions that occur between inlets and the dam (Thornton et al., 1982) and between different arms of the same system.In the future, simultaneous deployments of both mobile and stationary sensors at multiple depths can help us understand reservoir ecosystems more three dimensionally.Rapid, data-intensive spatial surveys in lakes and reservoirs can increase understanding about ecosystem spatial structure, opening new avenues for future research on linkages between spatial structure and ecosystem function.

Figure 1 .
Figure 1.True color Sentinel 2 imagery of Lake Arrowhead on a cloud-free day (28 July 2022) compared to a partly cloudy image acquired on the boat survey date (12 August 2022).

Figure 2 .
Figure 2. A wide gradient of surface turbidity in Lake Arrowhead on 12 August 2022 collected by logging at high frequency on a sensor-equipped boat.White circles indicate water sampling stations where the boat was stopped.Dark gray rectangle near the top indicates location of the dam.Stations were visited in the following order: A1 = arm 1, U = upper, L = lower, M = middle, A2 = arm 2

Figure 5 .
Figure 5. Spatial patterns of Sentinel 2 normalized difference turbidity index (ndti), and predictions of in-lake turbidity, total suspended solids, and Secchi depth on a cloud free day (28 July 2022) in Lake Arrowhead.Predictions use statistical models of ndti shown in Figures 3 and 4.

Table 2
Summary Stats for Optical Water Quality Metrics Over the 62.3 km Boat Survey Path, and Predictions Over the Entire Lake Surface