Monitoring Coastal Water Body Health with Sentinel-2 MSI Imagery

Abstract

The Australian ‘New South Wales Estuary health assessment and biodiversity monitoring program’ has set state-wide targets for estuary health. A selection of water bodies is being monitored by in situ chlorophyll a concentration and turbidity measurements, indicators for water quality. We investigate whether the current monitoring program can benefit from the use of remote sensing derived data, analyzing chlorophyll a and water clarity estimates by the C2RCC and ACOLITE products based on Sentinel-2 MSI imagery for three lakes along the New South Wales coast. The C2RCC and ACOLITE products were partly successful in predicting chlorophyll a concentration and water clarity. Estimates based on Sentinel-2 MSI imagery were in the range of in situ measurements. However, results varied across years and lakes, and a significant correlation could not be found in every case. It is likely that the physical differences between the systems, such as nutrient input, tannins, and suspended algae/sediment matrix, influence the output of the algorithm. This may preclude the application of a ‘one size fits all’ monitoring approach, given the importance of local ecological phenomena in both influencing remote sensing observations and the nature of appropriate targets. However, the design of a monitoring program that incorporates remote sensing provides a way forward.

1. Introduction

Wetlands, such as coastal lakes and estuaries, are a major contributor to Australia’s biodiversity [1]. Australia is a signatory to the (Ramsar) Convention on Wetlands of International Importance [2] for biodiversity. Federal water and wetland monitoring programs are mainly focused on securing and managing water resources, following the Commonwealth Governments’ intervention under the Water Act [3,4]. As such, there is no overarching federal legislation for the monitoring of and reporting on estuarine conditions [5], but under the Australian Constitution, each State is responsible for natural resource management in its own territory, including water management and water quality monitoring. The maintenance of water quality parameters at a minimum threshold for biodiversity conservation is important for water quality management and biodiversity.

In the State of New South Wales (NSW), a selection of coastal water bodies is subject to annual monitoring and assessment of estuary health. These water bodies sit within the state-wide Monitoring, Evaluation, and Reporting Program (MER) developed by the NSW government [6]. The MER program contributes to the NSW Estuary Management Program, which provides technical and financial assistance to local councils in the management of coastal hazards, estuary health, and community uses [7]. The MER program tracks progress against the state-wide resource condition target set by the Natural Resources Commission (NRC 2005) and is based on a ‘pressure-stressor-outcome’ model [7]. Before 2015, the target for estuaries and coastal lakes was set at an ambitious level that “by 2015 there is an improvement in the condition of estuaries and coastal lakes ecosystems”. In recent years the focus has shifted towards monitoring values identified by local government authorities and reference values in minimally impacted systems.

There are many variables available to monitor water quality [8]. From an ecological point of view, among the most commonly used are chlorophyll a concentration, turbidity (NTU), oxygen concentration, dissolved organic matter content, mineral content, pH, salinity, Secchi depth, and temperature [8,9,10,11]. Currently, in situ chlorophyll, a concentration, and turbidity are monitored by the MER program as water quality variables that respond to pressures on the ecosystem (such as eutrophication, see also Box 1) [6]. According to the MER program, these two variables capture important aspects of the ecosystems and largely determine water quality. Chlorophyll a concentration may indicate the occurrence of algal blooms. Algae pose a hazard to a lake system as they may lower the availability of oxygen in the water, causing fish kills, and limiting light penetration to lake beds, which is detrimental for seagrass beds [12,13]. Turbidity provides a measure of the amount of sediment in the water. Too much sediment also limits light penetration and can smother aquatic fauna [13].

Box 1. Monitoring estuary health according to the MER program.

According to the MER program, estuary health is defined as an ecosystem with its various components (biological, physical, and chemical) operating effectively to maintain a functioning ecosystem within the limits of natural variability (OEH, 2016). Waterbody types differ in their natural physical, chemical, and biological properties [14]. Variations in geomorphic setting, estuary type and stage of infill, and biological processes influence the distribution of habitat types [15] by controlling water salinity, depth, extent, and nutrient status. Despite this variability, the NSW MER program framework and measured variables are similar for all water body types in the program [6,7]. Due to the local differences, not all indicators are applicable to all estuaries, but common variables to measure water properties in most water bodies are:

• Eutrophication: in situ microalgal abundance (phytoplankton chlorophyll a) and water clarity (turbidity)
• Habitat extent of seagrasses, mangroves, and saltmarshes
• Fish assemblages: species diversity, composition, abundance

Chlorophyll a

Based on lab analysis, chlorophyll is measured in mg/m³ in the MER program. In general, chlorophyll a concentration is considered low to moderate 0 to 10 mg/m³, moderate from 10 mg/m³), and high from 100 mg/m³ [16].

 

Water clarity

Turbidity is measured in Nephelometric Turbidity Units (NTU) in the MER program. NTU is a measure of the intensity of light that passes through a water sample at a certain depth. For reference, the turbidity of 5 NTU is just visible to the eye. River turbidity after a storm can increase towards 50 to 100 NTU [17].

The MER Program has set ‘trigger values’ as an important factor in determining water quality grades. Different trigger values are set for lakes and river estuaries. For lakes (used in this study), these trigger values are 2.5 µg/L for chlorophyll a and 6.7 NTU for turbidity [18]. Estuaries have a natural variation in chlorophyll a levels and turbidity. Above the trigger value, a variable is out of the ‘normal range’ and may call for further investigation.

Field sampling in the MER program is limited in time and budget and is usually conducted during the summer season when the risk of algal blooms is higher. The use of remote sensing (and derived models) that estimate chlorophyll a concentration (presence of algal blooms) and turbidity (water clarity) could aid traditional monitoring, extending both the spatial extent and the temporal resolution allowing early detection of trigger values. Remote sensing has the potential to provide additional information or expand the scope and extent of the MER program and is currently underutilized. Remote sensing biodiversity products (RS products) may provide more frequent and year-round data points. There are many algorithms available that estimate chlorophyll a and turbidity for various water body types, depending on water constituents [16,19,20,21,22,23,24,25]. Optical water properties are determined by water constituents. Typically, a distinction is made between case 1 and case 2 water. Case 1 water has a high concentration of phytoplankton compared to other chemicals. Thus, optical properties are determined primarily by phytoplankton. In case 2 water, the optical properties are influenced by other constituents such as mineral particles, colored dissolved organic matter (CDOM), and detritus degradation products [26,27,28]. This determines the choice of algorithm. Other environmental factors, such as open or closed water, the presence of tannins, other dissolved organic matter, and atmospheric interference, also influence the choice and outcome of remote sensing-based analysis. In case 1 water, algae are detected by peak absorption in blue/green spectra [29,30,31]. However, with the use of these spectra, an accurate estimation of chlorophyll a levels is especially difficult in oligotroph lakes with a high concentration of minerals or CDOM unrelated to phytoplankton. Therefore, other spectral bands, such as those in red/NIR wavelengths, are used for monitoring case 2 water, to detect the red absorption peak [25,32,33].

Water quality variables monitored with remote sensing that are most prominent in review papers are chlorophyll a, turbidity, suspended matter, and organic/mineral content, including algorithms to analyze different sources of imagery [9,11,34,35,36,37]. They have been reviewed by a number of authors, including Bierman et al. (2011), who provide an overview of the methods that can be used to analyze spatial and temporal patterns in coastal water quality and provide examples using MODIS imagery [35]; Mohseni et al. (2022) who review the various remote sensing methods to monitor ocean water quality and present a systematic analysis of the capabilities of different sensors which are commonly applied to ocean water quality assessment [37]; Ramadas & Samantary (2017) [9] provide a book chapter investigating the status of ongoing research for water quality monitoring and management, using remote sensing; as well as Wang & Yang (2019) who conclude that progress is being made with the use of remote sensing techniques, but propose that improvement is needed for temporal and spatial coverage, the presence of indicators lists and model generalization [36].

Many studies have compared algorithms using regression analyzes to determine their suitability for a specific environment. These studies often result in an adaptation of an algorithm to suit this environment, although often with the suggestion that they could be applied more broadly as well [20,21,23,28,38,39,40]. In the MER program, chlorophyll a and turbidity are key indicators that are monitored for all waterbodies. Existing remote sensing products such as chlorophyll a and turbidity algorithms could be a valid approach complementing existing in situ monitoring. However, it is unclear if these algorithms deliver consistent results across different aquatic environments that are subject to a single monitoring protocol such as the MER program. Ideally, for use in reporting on policy requirements and a more general application, a chosen RS product must be tailored to demonstrate an understanding of the ecological context yet not be too specific to only fit one environment. It is not currently known whether a remote sensing product designed for universal application in the monitoring of chlorophyll a and turbidity can be successfully applied to different NSW estuaries and reporting on policy requirements. Nor is it known if the in situ monitoring currently applied in NSW fits equally well across the different types of estuaries.

There is a need for an evolution from algorithm development towards the implementation of those algorithms that address specific scientific questions. The need for remote sensing to become further applied across multidisciplinary fields, such as the environmental impact of water quality and policy, is being increasingly highlighted [41]. Therefore, in this paper, we represent the results of remote sensing products that are suggested for universal use, tested amongst NSW coastal lakes. Remote sensing products used are the C2RCC (case 2 regional coast color) and ACOLITE (atmospheric correction for satellites) modules. These modules are ready to use (require no programming) and are intended for monitoring water quality. The goal of this study is to link satellite imagery and available algorithms to current policy requirements. The objectives of this study were to: (a) gather information on NSW coastal lakes from remote sensing ready-for-use products, (b) analyze this data according to the framework applied in the MER program and assess water quality grades compared to in situ methods and test whether results are similar, and (c) outline opportunities and challenges to adopting a one-size-fits-all approach and using ready-for-use remote sensing products for policy reporting purposes.

2. Materials and Methods

2.1. Study Area: Bombah Broadwater, Myall Lake, and Wallis Lake

Our study areas were Myall Lakes National Park (Myall NP) and Wallis Lake, New South Wales, Australia. These lakes are near each other (Figure 1). While all are shallow, their geomorphic settings and hydrological properties give rise to differences in geochemical properties. Even though conditions likely have changed towards higher nutrient inputs following European settlement, the lakes are considered to have relatively pristine and oligotroph water.

Myall Lake and Bombah Broadwater are both freshwater to brackish barrier lakes [15] located within the same catchment and Myall Lakes National Park. Surrounding land cover in the catchment largely consists of agricultural lands. Steeper slopes of the hinterland are largely protected by National Parks and Forestry reserves. There are a few urban areas surrounding the lakes. Myall Lakes NP is listed as an internationally significant wetland under the Ramsar Convention [42].

Despite being well-known tourist destinations, the lakes have not been greatly impacted by human activities [42]. Whilst still considered in pristine condition, Myall Lakes catchment has a few major environmental issues. Water quality is impacted by urban runoff, which is more prevalent closer to the ocean. Bombah Broadwater, which covers an area size of 2400 hectares, is somewhat more brackish than Myall Lake (5800 hectares) due to its closer proximity to the ocean [42,43]. Bombah Broadwater’s primary production is dominated by phytoplankton with significant inputs of terrestrial organic matter [43]. The water quality in Bombah Broadwater is especially responsive to inputs from the upper Myall River. Myall Lake has an organic fraction dominated by macrophyte inputs. A unique characteristic of the lake is the presence of ‘gyttja,’ large benthic accumulations that occur in the sheltered embayments of Myall Lake that do not affect the inner parts of the lake. This is not present in Bombah Broadwater [44]. Myall Lake has a higher water quality overall [45]. Although oligotrophic, both lakes suffer from seasonal algal blooms. The lakes are also characterized by dark colored water, a result of high tannin content from river runoff [42].

Wallis Lake, with an area size of 10.000 hectares, is a wave-dominated barrier estuary and has a strong marine influence because of its connection with the ocean. The lake has clear water with low nutrient, tannin, and chlorophyll a concentrations [45] (

Table 1. Waterbody type classification of Bombah Broadwater, Myall Lake, and Wallis Lake [15], and the monitored estuary health assessment indicators [6].

Lake	Waterbody Type	Monitored Indicators
Bombah Broadwater Myall Lake Wallis Lake	Brackish barrier lake Brackish barrier lake Wave dominated barrier estuary	Chlorophyll a Turbidity Chlorophyll a Turbidity Chlorophyll a Turbidity Seagrass presence

). The lake hosts salt marshes, seagrass beds, as well as macrophyte, sponge, and ascidian beds. The lake, therefore, is of great importance for marine fauna. In recent years, the number of sponges has been decreasing, stressing the importance of good management [46].

2.2. In Situ Data

We have received in situ data of the three lakes from the NSW governmental department concerned with water quality monitoring which samples the lakes yearly. During in situ sampling of Myall Lake, Bombah Broadwater, and Wallis Lake, the variables chlorophyll a concentration and turbidity (NTU) are measured, as these variables are used as indicators. Although more variables are measured and available, these are not calibrated for the different lakes and used in the MER program. Therefore, only chlorophyll a concentration and turbidity are used to assess water quality [6,13,47] over many years. Other variables are not part of the estuary health assessment as per the MER program. The aim of our research and paper is to focus on the variables that are currently relevant and used as indicators by the government for informing policymakers. Therefore, we used in situ chlorophyll a and turbidity measurements provided by the NSW government, as these data are used for policy reporting.

The in situ measurements were collected roughly every four weeks according to the (then) New South Wales Office of Environment and Heritage (OEH) during the summer seasons of the years 2017 and 2018. This amounts to 5 or 6 data points per season [47]. In situ probe measurements and water samples are taken from 2 or more zones in the lakes every few meters from multiple runs through the zones. Water samples of each lake were analyzed in a laboratory to determine chlorophyll a concentration (μg/L) with a spectrophotometer. Turbidity (NTU) was measured using a water quality probe that was lowered in the water column [6]. Sampling seasons ran from 1 September 2017 to 31 August 2018 for ‘season 2017′ and from 1 September 2018 to 31 August 2019 for ‘season 2018′. We used the average of the zones of the measurements for our analyzes, resulting in 5 or 6 data points per year. For an overview of data points, see Supplementary S1. A more technical description of the sampling and sample processing can be found in the technical report ‘Assessing the condition of estuaries and coastal lake ecosystems in NSW [18].

2.3. Imagery and Processing

The European Space Agency’s Sentinel-2 satellites are multispectral satellites with medium resolution [48]. We selected top-of-atmosphere satellite imagery from the Copernicus Access Datahub, the European Space Agency’s portal for imagery from the Sentinel missions. We used Sentinel-2 MSI imagery for 2017, ranging from 1 September 2017 to 31 August 2018, and for 2018, from 1 September 2018 to 31 August 2019 (for an overview of imagery, see Supplementary S1). Using ESA’s imagery processing software SNAP, all bands were resampled to a pixel size of 10 m. Spatial subsets were created of the lakes and surroundings to be processed by the C2RCC and ACOLITE modules.

The C2RCC processor is a ready-for-use remote sensing product, available as a module within ESA’s SNAP imagery analyzes software. Therefore, in practical terms, it is very suitable to be used in a policy context or when highly technical remote sensing knowledge is less available. Having been validated for different sensors (including Sentinel-2 MSI), it has good results for case 2 water [49,50,51]. Running the C2RCC algorithm in SNAP results in imagery products of chlorophyll a (mg/m³) and total suspended matter (g/m³) estimates. The ACOLITE processor is a freely available module is built with a graphical user interface (GUI), and so programming skills are not necessary to use this module. This allows for use by a broader audience, such as policy advisors. It provides many options for image processing, including multiple algorithms to calculate water quality variables, among which are chlorophyll a concentration (mg/m³), turbidity (FNU), and total suspended matter (g/m³). For this study, we selected algorithms for chlorophyll specified for oligotroph water (‘oceancolor 3′) or optimized for Sentinel-2 MSI redband (‘Moses’). For turbidity and total suspended matter, we chose algorithms optimized for Sentinel-2 MSI (‘Nechad 2016′).

There is evidence that chlorophyll a concentration might vary within the lakes, especially for Myall Lake, where an algae–sediment mix called ‘gyttja’ resides in the shallow embayments of the lake [43,44]. Therefore, we drew polygons around the inner lake, as well as the entire, to distinguish between gyttja presence or absence. Thus, for each lake, two polygons were drawn. One polygon includes the fringing borders of the lake, and another polygon excludes these shallow zones. The larger polygon was drawn following the outline of the whole lake, bordering land vegetation. To draw the smaller polygon (inner lake only), shallows defined by clearly visible sandy banks on the fringe of the lake were excluded (for an overview of the polygons, see Supplementary S4). Chlorophyll a, turbidity, and total suspended matter estimates were extracted from pixels captured in both polygons drawn over the lakes.

2.4. Comparing In Situ Data with Remote Sensing Product Outputs

Cloud-free satellite imagery was selected on or close to the dates on which in situ measurements were taken by the MER program (for an overview, see Supplementary S1). Preferably, the timeframe between in situ observations and captured imagery is short (around the same time) because water quality characteristics may change quickly. However, for this study, available in situ data points are limited to 5 or 6 observations during the summer season for each lake. Satellite imagery with little to no clouds covering the lakes was sparsely available as well. Due to these limitations, for regression analysis between in situ measurements and RS product outputs, we selected data points that were closest together, or at maximum 3 weeks apart. Sometimes multiple images fell within the three-week timeframe. In that case, we selected a maximum of 2 images to correlate with in situ measurements to allow for more correlations (see Supplementary S1 for an overview of the dates for which in situ measurements have been correlated with estimates based on Sentinel-2 MSI imagery)

In situ chlorophyll, a concentration is reported in µg/L, and C2RCC/ACOLITE outputs are in mg/m³. As both units (weight and volume) differ by three orders of magnitude in relation to each other, we standardized both to mg/m³. C2RCC and one ACOLITE product report turbidity as total suspended matter (g/m³). The total suspended matter is a measure of the concentration of suspended particles in a water sample. Light is scattered by suspended particles, but there is no predefined relationship between NTU and total suspended matter. However, when total suspended matter increases in general, so does turbidity [52]. The other water clarity variable estimated by ACOLITE was turbidity in FNU. FNU (Formazin Nephelometric Units) is a similar index to NTU but based on infrared light. For an overview of the work process, see Figure 2.

2.5. Policy Reporting

Chlorophyll a and turbidity in situ data are further categorized by the NSW government into water quality grades, which are reported yearly [18,45]. Water quality grades range from A (highest quality) to E (lowest quality). The MER program’s goal for all estuaries is to achieve grade A, which means very low turbidity and chlorophyll a concentration (pristine conditions). Water quality grades are calculated by counting the amount and values of measurements that were above the trigger value set for chlorophyll a (trigger value 2.5 µg/L) and turbidity (trigger value NTU 6.7). The values of chlorophyll a and turbidity are averaged. This average is compared to the benchmark, the average water quality across New South Wales, to obtain the final grade [18]. (for a summary of the calculation, see Box 2, for a full description, see Supplementary S2 [6,13]).

To determine whether data derived from satellite imagery would result in similar water quality grades, we ran this same calculation based on the algorithm outputs for chlorophyll a and turbidity (FNU). We only used estimates from imagery that was taken relatively close in time to in situ measurements and where the regression analysis produced significant results or close to significance (p-value max 0,10). We then compared the in situ and remote sensing-based measurements with the water quality grades listed in the Waterway and Catchment Report 2018 [45].

Box 2. Calculation water quality grades based on chlorophyll a and turbidity results.

A summary of the calculations originating from the MER program to calculate the water quality grades. The trigger value and the worst expected value are based on the averages found in Australia.

Water quality grades are calculated in five steps. These are:

(1)

Non-compliant score, based on the number of non-compliant samples. Non-compliant means the value of chlorophyll a or turbidity being above the trigger value.

• Trigger value chlorophyll a: 2.5 μg/L
• Trigger value turbidity: 6.7 NTU
• Expressed as: non-compliance = number of samples above trigger value/total number of samples (e.g., 0.6 = 3/5)

(2)

Distance of the non-compliant measured value to the worst expected value

• Worst expected value chlorophyll a: 30 μg/L
• Worst expected value turbidity: 20 NTU
• Expressed as: distance = (measured value − trigger value)/(worst expected value − trigger value)

(3)

Final indicator score (calculated both for chlorophyll and turbidity)

• Expressed as: final score = √(non-compliance score × distance score)
  • The final score is determined by the amount of non-compliant scores. If there are 0 non-compliant scores, then the water quality score is 0, which leads to water quality grade A (best score). The distance score is only calculated for data greater than the trigger value (non-compliant values).

(4)

Final zone score calculation

• Expressed as the average of the chlorophyll a and turbidity indicator score

(5)

Comparison against the benchmark and grading

• A: score between 0 and 0.07
• B: score between 0.07 and 0.27
• C: score between 0.27 and 0.44
• D: score between 0.44 and 0.60
• E: score between 0.60 and 1

3. Results

3.1. Correlation between Field Observations and RS Product Estimates

Generally, the accuracy of both chlorophyll a and turbidity estimates varied per year and lake. The algorithms were partly successful in estimating chlorophyll a concentration in Bombah Broadwater (2017, 2018) and for Bombah Broadwater and Myall Lake when seasons 2017 and 2018 were combined. A significant correlation could not be found for every season and every lake, and in some cases, overestimations occurred (especially for chlorophyll a estimates). For water clarity, the RS products were most successful for Myall Lake and Wallis Lake. See

Table 2. Significant correlations between in situ measurements and the RS products for both chlorophyll a and water clarity for both seasons, 2017 and 2018, for all three lakes.

Lake	Variable	Year	RS Product	R2	p-Value
Bombah Broadwater (inner lake)	chlorophyll a	2017	ACOLITE Moses	0.78	0.01
Bombah Broadwater (whole lake)	chlorophyll a	2017	ACOLITE Oceancolor 3 ACOLITE Moses	0.64 0.67	0.03 0.02
Bombah Broadwater (inner lake)	chlorophyll a	2018	ACOLITE Oceancolor 3	0.71	0.02
Bombah Broadwater (inner lake)	chlorophyll a	2017 + 2018	ACOLITE Moses	0.42	0.01
Bombah Broadwater (whole lake)	chlorophyll a	2017 + 2018	ACOLITE Oceancolor 3 ACOLITE Moses	0.35 0.41	0.02 0.01
Bombah Broadwater (inner lake)	Water clarity	2017	C2RCC suspended matter ACOLITE Nechad suspended matter ACOLITE Nechad turbidity	0.50 0.71 0.71	0.05 0.02 0.02
Bombah Broadwater (whole lake)	Water clarity	2017	ACOLITE Nechad suspended matter ACOLITE Nechad turbidity	0.56 0.56	0.05 0.05
Myall Lake (inner lake)	Water clarity	2018	C2RCC suspended matter	0.57	0.05
Myall Lake (inner lake)	chlorophyll a	2017 + 2018	ACOLITE Oceancolor 3	0.32	0.03
Myall Lake (inner lake)	Water clarity	2017 + 2018	C2RCC suspended matter	0.46	0.01
Myall Lake (whole lake)	Water clarity	2017 + 2018	C2RCC suspended matter	0.46	0.01
Wallis Lake (inner lake)	Water clarity	2017	C2RCC suspended matter ACOLITE Nechad suspended matter ACOLITE Nechad turbidity	0.77 0.96 0.96	0.05 0.00 0.00
Wallis Lake (whole lake)	Water clarity	2017	ACOLITE Nechad suspended matter ACOLITE Nechad turbidity	0.95 0.95	0.00 0.00
Wallis Lake (inner lake)	Water clarity	2017 + 2018	ACOLITE Nechad suspended matter ACOLITE Nechad turbidity	0.47 0.47	0.01 0.01
Wallis Lake (whole lake)	Water clarity	2017 + 2018	ACOLITE Nechad suspended matter ACOLITE Nechad turbidity	0.44 0.44	0.01 0.01

for all significant values, Supplementary Tables S6–S8 for all values, and Supplementary S5 for scatterplots plotting in situ measurements against the Sentinel-2 MSI-based estimates.

There was no clear difference in the RS product estimates when either the smaller polygons (inner lake) or larger polygons (inner lake and fringe) were used for pixel information extraction for both chlorophyll a and turbidity estimates. In some cases, estimates based on pixel data from the inner polygon were significantly correlated with in situ measurements. In other cases, estimates are based on the whole lake. A distinguishable pattern has not been observed, suggesting the presence of gyttja along the lake foreshores did not influence results overall. However, the Moses algorithm was unable to render full images of the lakes (see Section 3.1.1) and picked up only the reflectance from the fringe of the lakes (Figure 3)

R² and p-values were oftentimes similar when the results of the RS (Nechad) product’s turbidity and suspended matter were compared with in situ measurements (Table 2).

3.1.1. Chlorophyll a Concentration

Although the ACOLITE Moses RS product did produce a good correlation with in situ measurements for Bombah Broadwater in 2017 (R²: 0.78, p = 0.01), chlorophyll a concentration derived from Moses, was overestimated for all three lakes. Moreover, the product failed to render data for the whole lake but instead produced output for the fringe of the lakes (see Figure 3). Pixel data and results, therefore are from part of the polygons that were in the output areas only. The oceancolor 3 and C2RCC products did not have this issue. C2RCC was least successful in accurately monitoring chlorophyll a concentration, but in general, estimates were lower than those of Moses and more in range of the in situ measurements.

C2RCC was often overestimated for Bombah Broadwater and Wallis Lake but closer in range to in situ measurement for Myall Lake (see Supplementary S3 for all graphs). For Bombah Broadwater (Figure 4), estimates were above field observations (around 4 mg/m³) and the trigger value (2.5 mg/m³).

The oceancolor 3 algorithm produced estimates that were more in range with in situ measurements of chlorophyll a, and a significant correlation was found for Bombah Broadwater for the year 2018 (R²: 0.71, p = 0.02). For the year 2017 a significant inverse correlation was found (R²: 0.64, p = 0.03),

Combining the data points of both years (2017 and 2018 together) did not always improve the correlation results for chlorophyll a concentration. For example, the correlation of oceancolor 3 with in situ measurements for Bombah Broadwater decreased (R²: 0.35, p = 0.02), due to the inverse correlation for the year 2017. However, for Myall Lake, where none of the chlorophyll a estimates by RS products significantly related to in situ measurements, oceancolor 3 did correlate with in situ measurements when both seasons were combined, albeit poorly (R²: 0.32, p = 0.03).

3.1.2. Water Clarity

Water clarity estimated by all three RS water clarity products was high (low turbidity and low suspended matter), which agreed with the in situ turbidity probe measurements (NTU). Significant correlations were found for all three RS products in season 2017 for Bombah Broadwater (Nechad turbidity R²: 0.71, p = 0.02, Nechad suspended matter R²: 0.71, p = 0.02, C2RCC suspended matter R²: 0.50, p = 0.05) and Wallis Lake (Nechad turbidity R²: 0.96, p = 0.00, Nechad suspended matter R²: 0.96, p = 0.00, C2RCC suspended matter R²: 0.77, p = 0.05). For Myall Lake, a significant correlation was found for the C2RCC algorithm in 2018 (R²: 0.57, p = 0.05). Here also, combining years did not improve results.

3.2. Reporting Water Quality Grades

C2RCC and Nechad 2016 suspended matter estimates describe total suspended matter (mg/m³) and not NTU, which makes direct comparison and reporting about water clarity in a similar fashion as required by policy about difficult. However, turbidity is estimated to be very low for all three lakes, which concurs with the finding of the in situ measurements.

Table 3. Water quality grades based on in situ measurements and RS product outputs for Bombah Broadwater. Similar grades, when based on in situ measurements and RS product estimates. (* p-value is max 0.10).

Lake + Part	Season	Report Card Grade	RS Product		Grade Based on RS Product Estimates
			Chlorophyll a	Water Clarity	Inner Lake	Whole Lake
Bombah Broadwater	2017	B B B	Moses Oceancolor 3 * C2RCC	Nechad (FNU) Nechad (FNU) Nechad (FNU)	C B -	D C B

shows the water quality grades from the Midcoast Council Report Card [45] for Bombah Broadwater for reporting on season 2017. Bombah Broadwater was the only lake with significant values for both water clarity and chlorophyll a outputs. Even though the correlation between in situ measurements and chlorophyll a estimates was not always significant (for oceancolor 3 and C2RCC chlorophyll a, a maximum p-value = 0,10 was allowed for water quality grade calculation), similar water quality grades were found for either estimate involving the whole lake (inner lake & fringe) or just the inner lake.

4. Discussion

The advantage of remote sensing-based observations is that they can be conducted with a much higher frequency than currently conducted with in situ data collection [37]. The year-round availability of satellite imagery can provide early warning signals of deteriorating water quality and more timely use of limited budget and time for in situ measurements. Besides, in situ data collection is point-based, while remote sensing can provide full area covering information from the whole lake or parts of lakes [36]. However, for any remote sensing information product to be successful for use in monitoring estuary health targets and policy reporting, it should be as widely applicable as the traditional methods used in these policies. It should be well understood, and policymakers should be aware of the advantages. As a universal, ready-for-use solution for coastal estuaries and wetlands that are part of a single monitoring program, however, there are several impediments to the reliability and consistency of reporting using RS products in this context.

The C2RCC and oceancolor 3 products estimates of chlorophyll a concentration for brackish barrier lakes were similar to the long-term average values measured using traditional field sampling. The values of turbidity and suspended matter are very low, and comparing in situ measurements resulted in significant correlations for all three lakes. This suggests that these RS products could potentially be used for water quality monitoring and subsequent reporting.

4.1. ‘One-Size-Fits-All’ Solution and Ecological Differences

The lakes and estuaries are part of the MER program that measures the same variables (chlorophyll a and water clarity) across the NSW coast. Our results showed that the RS products have merit but are not accurate enough as a method to monitor water quality variables on their own. The outputs need interpretation, fine-tuning, and/or setting threshold values to make them fit for monitoring different ecosystems. A ‘generic’ or ‘one-size-fits-all’ approach, such as a single monitoring program or algorithm, might not fully capture local spatial, temporal, and ecological variation. This is in line with the finding of other studies that improvements are required in accuracy and generalized models [36].

When looking at water clarity, the three examined lakes have low turbidity (NTU) in general, and in Wallis Lake, both total suspended matter and turbidity were estimated to be low. Thus, both in situ measurements and RS products are indicating the water is very clear. All three water clarity algorithms estimated higher values for total suspended matter for Bombah Broadwater in comparison to Wallis Lake. This is in line with the finding that Bombah Broadwater is relatively turbid [43].

In an ecological context, the supposition is that low turbidity maintains lake ecosystem function, and the link between water clarity and seagrass growth depth and function is clear [53]. Thus, remote sensing could be a useful tool to monitor water clarity. However, because these systems already have relatively clear water, it might be difficult to pick up a clear signal from imagery, given that variability is toward lower detection limits. Ecological assessment may also vary substantially for different water bodies. For example, seagrass is but one part of an ecosystem to be considered in the mix of coastal lake management options, and the vulnerability of coastal wetlands to sea-level rise is directly related to suspended sediment concentrations and accretion rates [54,55,56]. For Bombah Broadwater and Myall Lake, seagrass (which suffers from turbidity) is not an indicator, but it is for Wallis Lake. Bombah Broadwater and Myall Lakes are surrounded by mangrove and salt marsh ecosystems. In this context, estimates from RS products may mean something different for Wallis Lake than for Myall Lake. Similarly, local phenomena, such as gyttja in Myall Lake, might warrant a different approach to monitoring which incorporates geographic variability in expected algal concentrations. For example, in relation to lake depth or proximity to shorelines, algae are often found in higher concentrations in the shallower parts of Myall Lake [44]. In our results, the Moses algorithm has picked up signals solely from the outer parts of the lake. As Australian coastal lakes are often oligotroph, it is likely that chlorophyll a will often be at the detection limits. However, the algorithm has been designed for turbid productive water, with a detection range between 1.09 mg/m³ and 107.82 mg/m³ [57]. It has been proposed as a possibly universal tool for chlorophyll a estimates but was only partial successful here. The oceancolor 3 algorithm produced more accurate results, despite being designed for case 1 (open) water. Thus, ecological knowledge and local finetuning remain a requirement for each system to make a correct interpretation of the outcomes.

4.2. Water Quality Grading for Policy Reporting

When we used the RS product estimates to calculate water quality grades according to current reporting methods, the grades were similar or one category apart when the oceancolor 3 outputs were used. In the three examined systems, turbidity does not cause major problems. Such as with in situ measurements, the estimates indicated that algae growth is more problematic than water clarity. Trigger values exceedance is a major factor in the calculation of the water quality grade. As for in situ measurements, chlorophyll a estimates were more frequently above trigger values than they were for water clarity. This suggests that RS products, even though the correlation is not always significant, do provide a means to calculate water quality grades.

4.3. Reliability and Consistency

For water clarity, the results (both field observations and RS product estimates) indicate high water clarity (low turbidity). However, over- and under-estimations occur, an issue also encountered in their application of Sentinel-2 data in European lakes.

Chlorophyll content is low for the three lakes in this study compared to eutrophic lakes (>100 mg/m³), even during algal blooms (+/−12 mg/m³). Trying to finesse a lower chlorophyll a concentration might be more difficult, as is shown by the results of the Moses algorithm. It is known that the C2RCC algorithm might not be highly accurate in smaller and shallower lake systems or systems closer to the coastline [51]. However, in our results, the outcomes of the C2RCC were more in range with in situ measurements than the results when the Moses algorithm was used, which often show a substantial overestimation. For Bombah Broadwater, an inverse correlation was found when applying the oceancolor 3 algorithm, implying over- and underestimations of chlorophyll a concentration. This might be due to the fact that Bombah Broadwater is a relatively turbid lake, and the oceancolor 3 algorithm is developed for more clear water, which hampers accurate processing of the images.

For water clarity, the Nechad algorithms that produce estimates in FNU and total suspended matter, turbidity is used as a proxy for suspended matter [58]. It should be noted that the remote sensing metric of total suspended matter, while analogous to water clarity as measured by the turbidity probe, is not always identical to turbidity and needs recalibration for local variations [59]. Because we used secondary data and in situ measurement of the total suspended matter was not available, no calibration between NTU and the total suspended matter was done. Therefore, the formulation of the trigger value based on NTU into a trigger value in the total suspended matter was unknown, and an assessment of over- and underestimation is hampered. Further research should be performed to establish the relationship between NTU and total suspended matter in these systems if algorithms that are suggested for universal application are to be used for regular monitoring. However, the results in Table 2 and Supplementary S5 suggest that a medium to strong correlation can be found for all three lakes and with all applied algorithms.

For optimal ground-truthing practices, it is best if remote sensing data and in situ measurements are collected around the same time. A high number of co-incident sampling points were precluded by the availability of cloud-free imagery and limited return periods of in situ sampling (due to logistical and financial constraints typical of environmental monitoring programs such as the government MER program). In the case of the NSW coastal lakes that are studied here, the monitoring program encompasses over 100 lakes for an area the size of France and Germany combined. Therefore, there are limitations as to how many times a lake can be sampled in situ during a summer season (typically 5 or 6 times, with a few weeks in between). In the case of the Sentinel-2 MSI satellites, fly-over frequency is about every 5 days, though due to cloud cover, usable images are not available every 5 days. In other words, combining in situ field observations with the satellite remote sensing imagery will not exactly match with field collection. This may affect the accuracy with which water quality indices are estimated by remote sensing. However, for similar applications where remote sensing is used to evaluate spatial and temporal variability and patterns (such as floods, droughts, pollution studies, and vegetation phenology), it is quite common that there are large time gaps in cross-calibration of field data with satellite images [60,61,62,63,64]. In this study, there were time gaps (maximum three weeks) between available imagery and in situ data collection, which may affect the accuracy of results if chlorophyll a concentration and turbidity change. However, there is evidence that algal blooms and suspended sediments take at least 10 days to reach a maximum [65], and these phenomena can take more than three weeks to dissipate [66]. The data points available in this study are within this combined minimum time period of 4.5 weeks from the start to the receding of algal blooms. In a review study [67], it has been found that while algal blooms are patchy in both space and time, and satellites miss some blooms due to cloud cover, monthly mean composites of chlorophyll-a were considered to be representative of the mean phytoplankton concentration. Designing a monitoring program that integrates traditional monitoring, ecological knowledge, and remote sensing can overcome time gaps between imagery and in situ data collection and, subsequently, improve the accuracy of remote sensing-based estimates.

Possibly due to the resource limitations of the MER program, the two indicators, chlorophyll a concentration and water clarity, are considered adequate to determine water quality. However, using other water quality variables, such as dissolved organic matter, mineral content, salinity, pH, and oxygen concentration, to assess water quality could improve the monitoring program. It could provide a more comprehensive overview of the water quality of the NSW coastal lakes, as well as provide additional opportunities for the use of remote sensings, such as the monitoring of dissolved organic matter content [34].

4.4. Recommendations for Further Development and Integration of Remote Sensing Data into Policy

With these constraints in mind, we make the following recommendations regarding ongoing monitoring of the aquatic systems under study as well as for broader applications:

(i).

Consideration must be given to carefully timing in-situ sampling in association with the fly-over times of satellites. This will improve the validity and statistical power of comparisons.

(ii).

Variables captured by in situ measurements and assessed via remote sensing must be calibrated for local application.

(iii).

It needs further testing if the remote sensing tools are sensitive enough to capture the local ecosystem variance [51]. Integration of remote sensing into policy may be challenging [68], but the benefits of having more frequent, transparent, and area-covering information are attractive.

(iv).

Even though the algorithm can process an image ‘as is’ (it processes atmospheric correction to above-water reflectance), some specialists perform additional processing steps. These steps include cloud and cloud shadow masking, the application of a land mask, sun glint removal, or (when shallow parts should be excluded) an automated process to determine reflectance at the bottom of the lake [69]. These are technical aspects that could be seen as hurdles by lay users.

(v).

The target set should be meaningful for the monitored ecosystem. As systems behave differently, different targets and approaches might be suitable for different water bodies. Similarly, for remote sensing and in situ measurements, monitoring chlorophyll a and water quality may be a solution for many coastal lakes, but it might miss local ecological details in different estuary types. So, while a state-wide policy is a good framework, it could also allow for ecological differences between systems and specified targets. It seems desirable to address the ecosystems with a more flexible target setting and apply stratified monitoring in which the water bodies along the coast are divided into more groups than only lakes and rivers. A water body classification [15] could be a starting point for dividing estuaries into more specified monitoring groups.

(vi).

For continuous use of remote sensing within a monitoring program, a standard method should be designed that is feasible from a policy perspective and sensitive enough to pick up ecological differences relevant to each system. Such a program should be set up in collaboration with remote sensing experts and ecologists.

5. Conclusions

There are opportunities to approach and report on water quality monitoring with remote sensing. The RS products are partly successful in predicting chlorophyll a concentration and water clarity, and estimates provide similar results in water quality grades according to policy targets. Currently, RS products could be used to detect changes rather than accurately predict water quality variables. It is worth further development and testing for (the limits of detection for) chlorophyll a and water clarity, as well as for the alignment of water clarity variables (total suspended matter or turbidity). For the application of remote sensing-derived data for monitoring and policy reporting, we recommend the setup of a coordinated target-focused monitoring program or experiment that utilizes in situ measurements in combination with remote sensing data. This may improve future monitoring and management of coastal ecosystem health as well as other ecosystems that are subject to regular monitoring.