Ensemble machine learning and landsat observations reveal seasonal and spatial dynamics of water quality in a river-influenced estuarine system

Light reflected from lake surfaces can convey much information about water quality, especially algal abundance, humic content, turbidity and suspended solids. Light reflectance from lakes is complicated, and detailed spectra are needed for analysis of controlling factors. We obtained detailed reflectance spectra from the water surfaces of 15 lakes in east-central Minnesota and found patterns related to chlorophyll a (chl a), turbidity and humic matter (colored dissolved organic matter, CDOM). Increasing chl a and turbidity generally resulted in higher reflectance across the visible and near-infrared spectrum. Increasing CDOM led to low reflectance, especially below ~500 nm. Spectra of lakes with high chl a were distinguishable from those of lakes low in chl a, and lakes with low or high CDOM had readily distinguishable spectra. Several optical characteristics of lake water can be estimated from reflectance intensities measured over narrow wavelength bands. The ratio of reflectance at 700 nm to that at 670 nm was the best predictor of chl a over a wide range of conditions, including high turbidity and CDOM. Several relationships involving reflectance at 412, 443, 488, and 551 nm, the wavelengths used to calculate oceanic chl a from MODIS satellite data, also yielded a high R2. The ratio of reflectance at 670 nm to 571 nm provided the best estimates of humic color despite the low absorbance of CDOM at these wavelengths. Relationships involving reflectance for all 15 lakes in the range 400–500 nm, where CDOM absorbs light, had low r2 values; none was high enough for reliable estimates of lake color. For 10 lakes with low to medium chl a levels (≤10 mg m−3), regressions involving 412 and 443 nm yielded moderately good relationships. Airborne and satellite remote sensing thus might be used to identify lakes high in CDOM, and may provide reasonable estimates of humic color in lakes with low chl a levels.

Water masses, mixing, and the flow of dissolved organic carbon through the Irish Sea

Colored dissolved organic matter (CDOM) is one of the important fractions of dissolved organic matter (DOM) that controls the availability of light in water and plays a crucial role in the cycling of carbon. High CDOM absorption in the Mandovi Estuary (Goa) during spring inter-monsoon (SIM) is largely driven by both in-situ production and anthropogenic activities. Here we have presented the CDOM variation in the estuary during SIM of 2014–2018 and compared it with that of 2020 when the COVID-19 outbreak imposed lockdown was implemented. During 2020, low CDOM absorption was observed at the mid-stream of the estuary as compared to the previous years, which could be attributed to low autochthonous production and less input from anthropogenic activities. On the other hand, high CDOM observed at the mouth during 2020 is linked to autochthonous production, as seen from the high concentrations of chlorophyll a. High CDOM in the upstream region could be due to both autochthonous production and terrestrially derived organic matter. Sentinel-2 satellite data was also used to look at the variations of CDOM in the study region which is consistent with in-situ observations. Apart from this, the concentration of nutrients (NO3–, NH4+, and SiO44–) in 2020 was also low compared to the previous reports. Hence, our study clearly showed the impact of anthropogenic activities on CDOM build-up and nutrients, as the COVID-19 imposed lockdown drastically controlled such activities in the estuary.

Autonomous Surface Vessels (ASVs) are useful tools for monitoring and management of waterbodies to increase data capture rates and quantities within shorter time frames and at lower costs than manned methods. SeaTrac Systems Inc.’s SP- 4S ASV is an autonomous boat designed to provide a platform to collect water quality data on a long term (i.e., months) basis. Solar panels provide continuous power supply to the vessel and the instruments within. Autonomous steering and path tracking capability of the ASV allows users to predetermine a data collection path using Geographic Positioning System (GPS) waypoints. SP-48 is designed to collect water quality parameters that include Chlorophyll a (Chl-a), Phycocyanin (PC), Phycoerythrin (PE), Colored Dissolved Organic Matter (CDOM), Dissolved Oxygen, Temperature, Turbidity, Salinity, pH, Partial Pressure of Carbon Dioxide (pCO <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> ), and Backscattering. To provide collection of these parameters, six water quality sensors (SeaBird Scientific Inc.’s ECO-Triplet-FL3-B [Chl-a, PC, PE], ECO-Triplet-BB2FL [CDOM, Turbidity], ECO-Triplet-BB3 [Backscattering], SBE 63 [Dissolved Oxygen], ProOceanus Inc.’s CO <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> ProCV [pCO <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> ], and AML Oceanographic Inc.’s CT Xchange [pH, Salinity, Temperature]) were integrated into the ASV. Additionally, the ASV has integrated instruments that capture GPS, wind, and meteorological data. The GPS is captured by an Airmar GH2183 Network GPS Compass and the wind and meteorological data is captured by an Airmar 200WX WeatherStation. To receive the data from the water quality sensors, two options were considered. One possibility was to use sensor specific software to capture and store the data on the ASV onboard computer. This would require all software to run every time the ASV is deployed. In addition, the output files would not be accessible for real-time processing, preventing real-time data visualization and monitoring. The second option was to create a single interface to obtain data from all the sensors and send it to a server on a real-time basis. Ultimately, a connection tool (named Sensors Bridge) was developed to transmit water quality, GPS, and meteorology data by capturing information from communication (COM) ports and a LAN port onboard the ASV. The captured data was sent to a server located at Mississippi State University via a cellular network for storage and visualization. A Node.js server was created to provide the gateway to the database and the publicly available web application. The Node.js server processes the raw data, converts it to its final form, and saves it to the database. The data was stored in a PostgreSQL/ PostGIS relational spatial database. The web application (web app) named Water Quality Monitor was developed to visualize data in real-time as well as query historical data. The app contains four major components (Dashboard, Charts, Maps, and Add Location) that are accessible through a tabs interface in the app. The Dashboard component displays the last 30 records captured for each water quality parameter as a line graph depicting parameter magnitude as a function of time. Additionally, it displays the current location of the vessel and recently recorded points. There is also a bar graph showing all the parameters with the number of data points stored in the database, which can help monitor the quantity of records stored in the database for each parameter. The Chart tab assists in querying and visualization of historical data as a line or bar graph. Data can also be downloaded in image and spreadsheet formats. The Map tab provides the option of visualizing the water quality data spatially as a raster, vector, and heatmap. Users can download the data as a spatial file format. The Add Location tab enables system administrators to add a new study area. Once the boundary, name, and code of a study area are specified, database tables are automatically created. When the ASV starts capturing data from a study location, the data is saved to its respective locational database tables. Complete implementation of real-time data capture and visualization streamlines water quality monitoring. Additionally, captured data can be used for time series analysis. The current implementation lays a foundation for a decision support system.

Chromophoric Dissolved Organic Matter (CDOM) plays a critical role in the carbon and biogeochemical cycles within aquatic ecosystems. Satellite imagery can be employed to determine aquatic CDOM concentrations, highlighting the need for effective and precise algorithms for this task. In this study, a cruise survey dataset containing CDOM absorption coefficients and water-leaving radiances in the Pearl River estuary (PRE) was utilized to develop machine learning algorithms for CDOM retrieval from Landsat-8 Operational Land Imager (OLI) observations. Based on OLI wavelength bands, five bands and six band-ratios were chosen as input parameters for the machine learning models. Six machine learning models were trained to develop CDOM algorithms, including Support Vector Regression (SVR), Random Forest (RF), Extreme Gradient Boosting (XGBoost), Multi-Layer Perceptron (MLP), and Convolutional Neural Network (CNN). The results indicated that, among the six machine learning models, the XGBoost algorithm performed best, with the highest R2 value of 0.9 and the lowest CDOM root mean square error (RMSE) of 0.37 m−1, outperforming empirical algorithms. The XGBoost algorithm identified B4/B1 as the most critical input parameter, contributing 71%, followed by B3/B2 with a 16% contribution, where B1, B2, B3, and B4 are the wavelength bands of the OLI. These two band-ratios accounted for most of the contributions, suggesting their significant role in CDOM retrieval from Landsat OLI images. By employing the developed XGBoost algorithm, CDOM spatial patterns at six instances were derived from Landsat-8 OLI image reflectance, illustrating CDOM variations in the PRE influenced by various factors. Further analysis revealed that, in the PRE, tides and winds are the primary driving forces behind the spatial and temporal variability of CDOM. At present, the exploration of employing machine learning algorithms to infer CDOM concentrations in this region remains relatively limited; therefore, with a higher R2 value, the machine learning model we established unveils fresh and novel results.

The Fram Strait is a key region for investigating the exchange of Atlantic water with the Arctic Ocean. Uranium-236 (236U) from the two European nuclear reprocessing plants (NRPs) at La Hague and Sellafield provides a unique fingerprint in Atlantic water which can be used for studying its circulation patterns in the Arctic Ocean. And NRPs-derived 236U (236UNRPs) can be identified by its 233U/236U signature. In this study, we use colored dissolved organic matter (CDOM) absorption to constrain the selection of three Atlantic branch waters that carried different inputs of 236UNRPs in Fram Strait. This can potentially provide better estimates of transit times of Atlantic waters in the Arctic Ocean. High CDOM levels (a350≥0.35 m−1) in Fram Strait reflect the passage of Atlantic water transported to the Arctic by the Norwegian Coastal Current (NCC) and its extension and subsequently along the Siberian continental slope and shelf where the Ob, Yenisei and Lena rivers supply terrestrial organic matter with significantly high CDOM content. Conversely, low CDOM water represents Atlantic water that has remained off the shelf. Based on CDOM absorption, potential temperature (Θ), potential density (σΘ) and 236U concentration, the path of a given body of Atlantic water could be inferred and an appropriate NRPs input function constrained so that transit times could be estimated. Our results indicate that Arctic High CDOM Water (a350≥0.35 m−1) sourced from the NCC and Barents Sea Branch Water (BSBW) in the Barents Sea Opening has an average of 7–27 yrs transit time in the upper ∼200 m of the western shelf of the Fram Strait. Atlantic Low CDOM Water (a350<0.35 m−1, Θ>2°C) sourced from the Fram Strait Branch Water (FSBW) has a short pathway from the eastern Fram Strait. Arctic Low CDOM / High 236U Water (a350<0.35 m−1, Θ≤2°C, σΘ≤27.97 or 236U concentration ≥15 × 106 atom/L) sourced from the BSBW and the FSBW has an average of 22–28 yrs transit time. These findings demonstrate how combining measurements of CDOM with 236UNRPs can improve the robustness in estimation of transit times of different Atlantic water pathways in the Arctic Ocean. There are limited ways to empirically derive estimates and the values provided offer unique data for comparison with estimates from regional circulation models.

High spatial resolution inversion of chromophoric dissolved organic matter (CDOM) concentrations in Ebinur Lake of arid Xinjiang, China: Implications for surface water quality monitoring

Water quality monitoring is becoming increasingly important as human populations grow, industrial and agricultural activities expand, and climate change threatens to cause major alterations to the hydrologic cycle. With advent of new technology, autonomous surface vessels (ASVs) are able to provide data with high spatial and temporal resolution, which is critical for water quality monitoring and management. A suite of sensors has been integrated in a novel solar-powered ASV that can trave1 $\sim$ 9.26 kmph or can be stationed at a location collecting continuous data. The overarching objective of this paper is to present the efficacy of the ASV in collecting accurate water quality data by comparing these data with data from another set of independent sensors as well as laboratory analysis of water samples. In-situ water quality data from selected sites together with ASV data were collected from four study areas in Mississippi, USA. Salinity, temperature, pH, and dissolved oxygen (DO) measured by the ASV were compared with the measurements by a profiling sensor suite and ASV measured chlorophyll a, phycocyanin, colored dissolved organic matter (CDOM), turbidity, and partial pressure of carbon dioxide (pCO <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> ) were compared with the measurements from laboratory analysis of water samples collected from the same location and approximately at the same time as the ASV measurements. The comparisons produced correlation coefficients of 0.999, 0.985, 0.974, 0.755, 0.701, 0.633, 0.755, 0.839, and 0.999 for salinity, temperature, pH, DO, chlorophyll a, phycocyanin, CDOM, turbidity, and pCO <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> , respectively and the root-mean-square deviations for salinity, temperature, pH, DO, chlorophyll a, phycocyanin, and pCO <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> were 1.29 PSU, 0. $671{}^{\circ}\text{C}, 0.56,3.1\text{S}\text{m}\text{g}/\text{L}, 1.12\mu \text{g}/\text{L}$, 0.694 $\mu \text{g}/\text{L}$, and 18.8 $\mu \text{a}\text{t}\text{m}$, respectively. This ASV, along with its sensor suite, should be valuable for water quality modeling and management due to its potential to provide large amounts of accurate data.

Abstract. Light availability is of primary importance to the ecological function of shallow estuaries. For example, benthic primary production by submerged aquatic vegetation is contingent upon light penetration to the seabed. A major component that attenuates light in estuaries is colored dissolved organic matter (CDOM). CDOM is often measured via a proxy, fluorescing dissolved organic matter (fDOM), due to the ease of in situ fDOM sensor measurements. Fluorescence must be converted to CDOM absorbance for use in light attenuation calculations. However, this CDOM–fDOM relationship varies among and within estuaries. We quantified the variability in this relationship within three estuaries along the mid-Atlantic margin of the eastern United States: West Falmouth Harbor (MA), Barnegat Bay (NJ), and Chincoteague Bay (MD/VA). Land use surrounding these estuaries ranges from urban to developed, with varying sources of nutrients and organic matter. Measurements of fDOM (excitation and emission wavelengths of 365 nm (±5 nm) and 460 nm (±40 nm), respectively) and CDOM absorbance were taken along a terrestrial-to-marine gradient in all three estuaries. The ratio of the absorption coefficient at 340 nm (m−1) to fDOM (QSU) was higher in West Falmouth Harbor (1.22) than in Barnegat Bay (0.22) and Chincoteague Bay (0.17). The CDOM : fDOM absorption ratio was variable between sites within West Falmouth Harbor and Barnegat Bay, but consistent between sites within Chincoteague Bay. Stable carbon isotope analysis for constraining the source of dissolved organic matter (DOM) in West Falmouth Harbor and Barnegat Bay yielded δ13C values ranging from −19.7 to −26.1 ‰ and −20.8 to −26.7 ‰, respectively. Concentration and stable carbon isotope mixing models of DOC (dissolved organic carbon) indicate a contribution of 13C-enriched DOC in the estuaries. The most likely source of 13C-enriched DOC for the systems we investigated is Spartina cordgrass. Comparison of DOC source to CDOM : fDOM absorption ratios at each site demonstrates the relationship between source and optical properties. Samples with 13C-enriched carbon isotope values, indicating a greater contribution from marsh organic material, had higher CDOM : fDOM absorption ratios than samples with greater contribution from terrestrial organic material. Applying a uniform CDOM : fDOM absorption ratio and spectral slope within a given estuary yields errors in modeled light attenuation ranging from 11 to 33 % depending on estuary. The application of a uniform absorption ratio across all estuaries doubles this error. This study demonstrates that light attenuation coefficients for CDOM based on continuous fDOM records are highly dependent on the source of DOM present in the estuary. Thus, light attenuation models for estuaries would be improved by quantification of CDOM absorption and DOM source identification.

Colored dissolved organic matter (CDOM) is one of the important fractions of dissolved organic matter (DOM) that controls the availability of light in water and plays a crucial role in the cycling of carbon. High CDOM absorption in the Mandovi Estuary (Goa) during spring inter-monsoon (SIM) is largely driven by both in-situ production and anthropogenic activities. Here we presented the CDOM variation in the estuary during SIM of 2014 to 2018 and compared it with that of 2020 when the COVID-19 outbreak imposed lockdown was implemented. During 2020, low CDOM absorption was observed at the mid-stream of the estuary as compared to the previous years, which could be attributed to low autochthonous production and less input from anthropogenic activities. On the other hand, high CDOM observed at the mouth during 2020 is linked to the autochthonous production, as seen from the high concentrations of chlorophyll a. High CDOM in the upstream region could be due to both autochthonous production and terrestrially derived organic matter. Sentinel- 2 satellite data was also used to look at the variations of CDOM in the study region which is consistent with in-situ observations. Apart from this, the concentration of nutrients (NO3-, NH4+ and SiO44-) in 2020 was also low compared to the previous reports. Hence, our study clearly showed the impact of anthropogenic activities on CDOM build-up and nutrients, as the COVID-19 imposed lockdown drastically controlled such activities in the estuary.

Abstract. Carbon monoxide (CO) apparent quantum yields (AQYs) are reported for a suite of riverine, estuarine and sea water samples, spanning a range of coloured dissolved organic matter (CDOM) sources, diagenetic histories, and concentrations (absorption coefficients). CO AQYs were highest for high CDOM riverine samples and almost an order of magnitude lower for low CDOM coastal seawater samples. CO AQYs were between 47 and 80% lower at the mouth of the estuary than at its head. Whereas, a conservative mixing model predicted only 8 to 14% decreases in CO AQYs between the head and mouth of the estuary, indicating that a highly photoreactive pool of terrestrial CDOM is lost during estuarine transit. The CDOM absorption coefficient (a) at 412 nm was identified as a good proxy for CO AQYs (linear regression r2 &gt; 0.8; n = 12) at all CO AQY wavelengths studied (285, 295, 305, 325, 345, 365, and 423 nm) and across environments (high CDOM river, low CDOM river, estuary and coastal sea). These regressions are presented as empirical proxies suitable for the remote sensing of CO AQYs in natural waters, including open ocean water, and were used to estimate CO AQY spectra and CO photoproduction in the Tyne estuary based upon annually averaged estuarine CDOM absorption data. A minimum estimate of annual CO production was determined assuming that only light absorbed by CDOM leads to the formation of CO and a maximum limit was estimated assuming that all light entering the water column is absorbed by CO producing photoreactants (i.e. that particles are also photoreactive). In this way, annual CO photoproduction in the Tyne was estimated to be between 0.99 and 3.57 metric tons of carbon per year, or 0.004 to 0.014% of riverine dissolved organic carbon (DOC) inputs to the estuary. Extrapolation of CO photoproduction rates to estimate total DOC photomineralisation indicate that less than 1% of DOC inputs are removed via photochemical processes during transit through the Tyne estuary.

Remote sensing of CDOM and DOC in alpine lakes across the Qinghai-Tibet Plateau using Sentinel-2A imagery data

The ultraviolet absorption spectra of chromophoric dissolved organic matter (CDOM) can be used to trace its sources and to explore the dynamic of the CDOM pool. In previous studies, only the spectra above 240 nm can be used directly to characterize the CDOM in seawater, due to the overlapping of CDOM absorption spectra below 240 nm with inorganic chemicals such as NO3−, NO2−, Cl- and Br-. In this study, three different machine learning models, back propagation neural network (BPNN), random forest (RF) and extreme gradient boosting (XGBoost), were built to predict the CDOM ultraviolet absorption spectra between 215 and 350 nm after being trained with the raw absorption spectra of seawater. The optimal input wavelength range of the raw seawater spectra is 250-350 nm, and the optimal model parameters of machine learning algorithms were determined by using five-fold cross validation. The results show that the three models can well predict the CDOM absorption spectra. Comparatively, the XGBoost model gave the best prediction results. The reasons might be related to the fact that the XGBoost algorithm focuses on the residuals generated by the last iteration, which can reduce both variance and bias, especially for datasets with small sample sizes. Based on the predicted spectra by XGBoost algorithm, we calculated the spectra slopes of short wavelengths between 215 and 240 nm (S215-240) and between 215 and 275 nm (S215-275). The results show that the S215-240 and S215-275 are ~2 times the widely used spectra slopes between 275 and 295 nm (S275-295) obtained by traditional method based on the raw spectra. Moreover, the S215-240 and S215-275 are more relavant with salinity for marine CDOM than S275-295, suggesting spectra slopes of shorter wavelengths might be the better proxies for marine CDOM than that of longer wavelengths.

Inland waters, characterized by small scale but a large number, play an important role in the carbon budget and global carbon cycle. Colored dissolved organic matter (CDOM) is a significant indicator used for tracing dissolved organic carbon (DOC) in inland waters. Accurate remote-sensing estimation of CDOM concentration is still a challenge due to complex optical properties of inland waters. Many efforts have been made to estimate high concentration CDOM, leading to a knowledge gap in using remote sensing to estimate low concentration CDOM, which results in difficulty and uncertainty for estimating total carbon storage in global inland waters. Currently, few studies are devoted to estimating low concentration CDOM, while Landsat-8’s applicability for estimating low concentration CDOM is still unknown. In this study, two datasets, NRL_SFE and Lake_Erie, were collected to represent extremely low CDOM conditions that aCDOM (440) (the absorptions of CDOM at 440 nm) ranges 0.215–1.165 m–1, and 0.066–1.242 m–1, respectively. The best CDOM retrieval model (validation results: R2 = 0.78; RMSE = 0.161 m–1; MRE = 26.02%), aCDOM (440) = 0.483x –1.776, x = Rrs (B2)/Rrs (B4), was developed for monitoring CDOM in the two regions. Results show that Landsat-8 and the best model work well for estimating low concentration CDOM in inland waters. In addition, we have proved that Landsat-8 surface reflectance products, which are freely provided by USGS, are convenient and useful for developing remote sensing algorithm of CDOM estimation after correcting water surface reflectance. The image-derived CDOM’s spatial patterns in Lake Erie demonstrate the Landsat-8’s applicability to observe spatiotemporal variations of low concentration CDOM.

• A CDOM absorption coefficient was selected from seven reference wavelengths. • The optimal combinations of R rs wavelengths were determined by random forest. • The optimal random forest model with CDOM absorption coefficient was revealed. • Spatiotemporal distribution and characteristics of CDOM were confirmed. • Results can be used to reveal temporal and spatial variation in CDOM distributions. Colored dissolved organic matter (CDOM) in inland waters is used as a proxy to estimate dissolved organic carbon (DOC) and may be a key indicator of water quality and nutrient enrichment. CDOM is optically active fraction of DOC so that remote sensing techniques can remotely monitor CDOM with wide spatial coverage. However, to effectively retrieve CDOM using optical algorithms, it may be critical to select the absorption coefficient at an appropriate wavelength as an output variable and to optimize input reflectance wavelengths. In this study, we constructed a CDOM retrieval model using airborne hyperspectral reflectance data and a machine learning model such as random forest. We evaluated the best combination of input wavelength bands and the CDOM absorption coefficient at various wavelengths. Seven sampling events for airborne hyperspectral imagery and CDOM absorption coefficient data from 350 nm to 440 nm over two years (2016–2017) were used, and the collected data helped train and validate the random forest model in a freshwater reservoir. An absorption coefficient of 355 nm was selected to best represent the CDOM concentration. The random forest exhibited the best performance for CDOM estimation with an R 2 of 0.85, Nash-Sutcliffe efficiency of 0.77, and percent bias of 3.88, by using a combination of three reflectance bands: 475, 497, and 660 nm. The results show that our model can be utilized to construct a CDOM retrieving algorithm and evaluate its spatiotemporal variation across a reservoir.