The Lake Water products within the Copernicus Global Land Service provide an optical and thermal characterization of more than 4000 (optical) and over 1000 (thermal) inland water bodies . The production is based on Sentinel-3 data (OLCI and SLSTR) and Sentinel-2 (MSI) data for the currently ongoing NRT service. The products contain four (sets) of parameters: lake water surface temperature, lake water reflectance (all wavebands that are available after atmospheric correction), turbidity and a trophic state index in 11 classes derived from chlorophyll-a concentration. Production and delivery of the parameters are over set 10-day intervals. While turbidity, trophic state index and temperature are provided as 10-day averages, the lake water reflectance product contain the best representative spectrum of the covered time span in order to preserve the spectral reflectance vector. The products are available at 300m and 100m for the water quality parameters based on Sentinel-3 OLCI and Sentinel-2 MSI, respectively. Lake Surface Water Temperature (LSWT) is provided at 1km resolution. In all cases, a dedicated production process is used starting from Level 1 source products. The algorithms used to derive the optical lake water products are implemented in the Calimnos processing chain which contains procedures to pre-classify water into 13 optical water types associated with dedicated, tuned algorithms for chlorophyll-a and turbidity. Both LSWT and water quality algorithms represent the state-of-the-art for operational processing, including developments in the ESA SST_cci, Lakes_cci and GloboLakes projects.
Besides the NRT services based on OLCI and SLSTR, the full archives from MERIS and (A)ATSR are available for the time span 2002 – 2012. All products are publicly available via the Copernicus Global Land Service Website and Viewing Services to be easily accessible at the individual user or lake level, or by countries for further analysis and reporting. It is possible to provide tailored, aggregated products such as statistical records of chlorophyll-a concentration, algal bloom occurrence, percentage of surface water effected by eutrophic classes, etc.
The evolution of the services foresees additional parameters (such as chlorophyll-a concentration and floating cyanobacteria information) as well as extended coverage by the temperature products.
A key aspect of a Copernicus Service is its long term, sustained operational character. However, this is only assured if the delivered products meet user requirements and find wide uptake. We will present the current status of the lake water products, their interoperability and latest validation results. We will demonstrate how the products are applied to water management issues and how the products can be used in global scale analyses. As a prominent example, we will show how UNEP currently uses the products to inform per-country statistics on indicators under the Sustainable Development Goal 6.3.2 to harmonize information for SDG assessment and reporting.

Lakes are sentinels and integrators of environmental and climatic changes occurring within their watershed. The influence of climate change on lakes is becoming increasingly concerning worldwide. Understanding the complex behavior of lakes in a changing environment is essential to effective water resource management and mitigation of climate change effects. The increase in summer temperatures has been estimated at 0.34 °C per decade with lake specific parameters like morphology contributing to the diversity of response at the regional level. The frequency of heatwave events in Europe is increasing and the recent IPPC report on climate change estimated an over 90% likelihood that there will continue to be an increase in the frequency of heat extremes over the 21st century in Europe, especially in southern regions. In July 2019, a heatwave occurred in Europe with record daily maximum temperatures over 40 °C observed in several places. Temperatures were locally 6 to 8 °C higher than the average warmest day of the year for the period 1981–2010. Heatwaves can have implications for the water quality and ecological functioning of aquatic systems, and for example in Europe several of these events have been associated with increased phytoplankton blooms. Of particular concern is the predicted increase in potentially harmful summer blooms of cyanobacteria with combined pressures of climate change and eutrophication.
The ESA Climate Change Initiative (CCI) Lakes ECV Project (https://climate.esa.int/en/projects/lakes/) combines multi-disciplinary expertise to exploit satellite Earth Observation data to create the largest and longest possible consistent, global record of five lake climate variables: lake water level, extent, temperature, surface-leaving reflectance (e.g., chlorophyll-a and suspended solid concentrations), and ice cover. The first version of the database covers 250 globally distributed lakes with temporal coverage, depending on parameter, ranging from 1992 up to 2019. This is expanded to 2000 lakes in version 2. The ESA Lakes_cci dataset was found to be a key resource for examining the implications of heatwave events on lakes. We examined heatwave events for European lakes, focusing on the 2019 event. The response of lake chlorophyll-a concentration, a proxy of phytoplankton abundance, was dependent on the lake type, especially lake depth, stratification and trophic state. However, the timing of the heatwave event itself was also important in the type of response observed. In many cases, the effects of resulting storms that ended the heatwave were more discernable than the heatwave itself. For example, in some shallow lakes, following the storm, chlorophyll-a concentrations increased markedly and remained high for the duration of the summer. Comparing the high frequency WISPstation data (2018-2020) with the CCI dataset allows for detailed cross validation. Some of the rapid fluctuations visible from the satellite record are supported by the in situ data. In addition, utilizing the phycocyanin pigment estimates from the WISPstation and microscopic counts, showed how cyanophytes played a key role in the sudden increases and declines in chlorophyll-a in mid to late summer. Heatwaves and subsequent storms appeared to play an important role in structuring the phenology of the primary producers, with wider implications for lake functioning.

Abstract: This study is undertaken as part of the Lakes_cci project (ESA Climate Change Initiative), which aims to provide a multi-decadal, multi-sensor and global (over 2000 lakes) climate data record of water-leaving reflectance (Rw) and optical-biogeochemical water quality products. This presentation describes how the first non-interrupted multi-decadal satellite observations of global inland water quality have been created using the Calimnos processing chain, including the first per-pixel uncertainty characterization in a dynamic algorithm processing selection scheme for inland water bodies.
In the first phase of the Lakes_cci, which includes an initial dataset released for 250 lakes and a more recent improvement of spatiotemporal coverage to > 2000 inland water bodies, the primary aims for the Lake Water-Leaving Reflectance (LWLR) thematic Essential Climate Variable were to (1) improve the characterization of algorithmic uncertainties and provide uncertainty estimates with each pixel, and (2) to fill the observation gap between the MERIS and OLCI medium resolution sensors using MODIS-Aqua, where feasible.
We developed a method to produce estimates of Chlorophyll-a (Chla) satellite product uncertainty on a pixel-by-pixel basis within an Optical Water Type (OWT) classification scheme. This scheme helps to dynamically select the most appropriate algorithms for each satellite pixel, whereas the associated uncertainty informs downstream use of the data (e.g., for trend detection or modelling) as well as the future direction of algorithm research. Observations of Chla were related to 13 previously established OWT classes based on their corresponding normalized water-leaving reflectance (Rw), each class corresponding to specific bio-optical characteristics. Uncertainty models corresponding to specific algorithm - OWT combinations for Chla were then expressed as a function of OWT class membership score. Embedding these uncertainty models into a fuzzy OWT classification approach for satellite imagery allows Chla and associated product uncertainty to be estimated without a priori knowledge of the biogeochemical characteristics of a water body. Following blending of Chla algorithm results according to per-pixel fuzzy OWT membership, Chla retrieval shows a generally robust response over a wide range of class memberships, indicating a wide application range (ranging from 0.01 to 362.5 mg/m3). Low OWT membership scores and high product uncertainty identify conditions where optical water types need further exploration, and where biogeochemical satellite retrieval algorithms require further improvement. This work was conducted using MERIS observations, due to its long operation (2002-2012) and coincidence with matching in situ data in the LIMNADES (Lake Bio-optical Measurements and Matchup Data for Remote Sensing) database, and had been transferred to OLCI (2016-now) on Sentinel-3 considering its similarities in radiometric performance and waveband configuration with MEIRS.
To fill the gap between the MERIS and OLCI observation periods, independent validation and tuning of bio-optical algorithms was performed to facilitate the application of this procedure to MODIS. Product continuities between MERIS, OLCI and MODIS were then evaluated, and first attempts were made to remove the inter-sensor bias.
This work accompanies version 2.0 of the ESA Lakes_cci Climate Research Data Package, released in spring 2022. Version 1.1 of the dataset, covering 250 lakes, was published in July 2021 (https://catalogue.ceda.ac.uk/uuid/ef1627f523764eae8bbb6b81bf1f7a0a). Using the dataset published and combining several use cases, our presentation will also briefly showcase the global application of our products on phytoplankton phenology, climate change and decision making of lakes.

Lakes play a critical role in global climate regulation, carbon cycling, fresh water supply, fishing, and tourism, yet they remain vulnerable to climate change and anthropogenic disturbance. It is now widely acknowledged that climate warming influences lake functioning and ecosystem processes. However, significant gaps exist in our understanding of the potentially confounding interactions between climate forcing, water temperature and other water quality parameters. This study investigates the relationships between water temperature with chlorophyll-a (Chl-a) and turbidity in 250 lakes by analyzing long-term (2002-2019) satellite-retrieved data. We also elucidate the factors affecting these relationships. Daily lake median Lake Surface Water Temperature (LSWT) from ATSR, AASTR and AVHRR as well as derived Lake Water-leaving Reflectance (LWLR) products from MERIS (2002-2012) and Sentinel 3 OLCI (2016-2019) were extracted from the European Space Agency (ESA) Climate Change Initiative (CCI) Lake project (https://climate.esa.int/en/projects/lakes/). Time series of LSWT, Chl-a and turbidity were firstly detrended using the generalized additive model (GAM) fitting. A cross-correlation analysis was then carried out to investigate their patterns of relationship. We defined a total of six relationship patterns between LSWT and Chl-a/turbidity. The globally dominant pattern between LSWT and Chl-a was one where LSWT showed a negative and lagged relationship with Chl-a. In contrast, the globally dominant pattern between LSWT and turbidity was one where LSWT exhibited a positive lagged relationship following changes in turbidity. In total, 58% of the lakes showed negative relationships between LSWT and Chl-a, while 64% of the lakes showed positive relationships between LSWT and turbidity. The analysis of the factors influencing the observed relationships included lake area, depth, elevation, latitude, precipitation, wind speed, LSWT, Chl-a and turbidity. A boosted regression tree was used to analyze the relative influence of each factor on these relationships. It was found that LSWT and turbidity are the main factors influencing the relationship between LSWT and Chl-a, with relative influences of 16% and 15% respectively. Higher LSWT and lower turbidity conditions tend to lead to positive relationships between LSWT and Chl-a. For the relationship between LSWT and turbidity, lake area and Chl-a concentration are the main factors, with relative influences of 18% and 14% respectively. Lakes with smaller surface areas and with lower Chl-a concentrations tend to lead to positive relationships between LSWT and turbidity. This study will discuss sensitivities of the relationship between lake temperature and Chl-a/turbidity to climate change.

This study was supported by the European Space Agency Lakes Climate Change Initiative (ESA Lakes CCI+) project.

Remote sensing of the water-leaving radiance from airborne or spaceborne platforms requires the compensation for the absorption and scattering effects of the intervening atmosphere between the sensor and the surface. This process is known as “atmospheric correction” or “atmospheric compensation” (AC). Typical methods for AC are based on the assumption that the surface reflectance is roughly spatially homogeneous at a large spatial scale (> 30 km). This assumption is appropriate over large bodies of water, in regions away from the shore and when the spatial variations in the signal have no large contrast per waveband. However, this assumption is no longer valid in proximity to floating mats of opaque objects (macrophytes, constructions) and land, as those surfaces present high contrast with the water-leaving signal, typically in the near-infrared due to the high molecular absorption by pure water. These spatial heterogeneity with strong contrast will contribute to the observed signal over water due to the diffuse transmission of the atmosphere. This process, know as adjacency effect, biases the estimation of surface reflection under the assumption of spatial homogeneity.

In this study, we draw upon the theoretical and practical applications from the fields of atmospheric correction and atmospheric visibility to implement a sensor-agnostic adjacency correction in the frequency domain to the atmospheric correction code ACOLITE. The frequency domain approach greatly speeds up the computation for any given solution, allowing to use an iterative scheme to find the appropriate aerosol model and optical thickness. The processing chain can be summarized as:

1. For each aerosol model, iterate until the difference between the dark spectrum and the path reflectance is minimal, while keeping the fraction of negative VNIR pixels at a chosen threshold (0.1% of the scene):

1.1. Initial estimate of the aerosol optical thickness, under the assumption of spatially homogeneous surface reflectance;
1.2. Calculate the atmospheric point spread function;
1.3. Deconvolve at-sensor reflectance imagery and the point spread function;
1.4. Reconstruct the TOA reflectance without the upward diffuse contribution and estimate new dark spectrum;
1.5. Fit the corrected dark spectrum to the modeled path reflectance;

2. Select the aerosol model and optical thickness giving the lowest difference between the dark spectrum and the path reflectance.

The atmospheric Point Spread Functions (PSFs), per sensor waveband, for each atmospheric scatterer (maritime, continental, urban aerosols and Rayleigh) were calculated with a backward Monte Carlo code. The aerosol models and vertical profile from the 6SV code were used to calculate the atmospheric lookup tables in ACOLITE and the PSFs. Calculations were performed for pure scatterers and at a fixed optical thickness (aerosols) of 1 (unitless) and at variable surface pressures (Rayleigh). Results were then fitted to a model allowing to approximate the atmospheric PSF for variable mixing of atmospheric scatterers and surface pressure. These approximate equations are used in step 1.2 of the pseudocode above.

We evaluated the method applied to the Multispectral Instrument (MSI) onboard Sentinel-2A and B and to the Operational Land Imager (OLI) onboard Landsat 8, using field measurements from several small Belgian lakes made between 2017 and 2019. The results show a large increase in performance for the atmospheric correction in all bands and sensors, though the bias in the NIR wavebands remains large for low turbidity systems. The method showed better performance for OLI than for MSI, and this might be a consequence of the different spatial resolution and signal to noise ratio of the NIR bands of each sensor.

The root mean squared error (RMSE) when using the adjacency correction was on average 4 times (OLI/Landsat 8) and 2 times (MSI/Sentinel-2) lower than when the AC assumes spatial homogeneity.

Creating multi-mission satellite-derived water quality (WQ) products in inland and nearshore coastal waters is a long-standing challenge due to the inherent differences in sensor spectral and spatial sampling as well as in their radiometric performance. This research extends a recently developed machine-learning (ML) model, i.e., Mixture Density Networks (MDNs) to the inverse problem of simultaneously retrieving WQ indicators, including chlorophyll-a (Chla), Total Suspended Solids (TSS), and the absorption by Colored Dissolved Organic Matter at 440 nm (a_cdom (440)), across a wide array of aquatic ecosystems. We use an in situ database to train and optimize MDN models developed for the relevant spectral measurements (400 – 800 nm) of the Operational Land Imager (OLI), MultiSpectral Instrument (MSI), and Ocean and Land Colour Instrument (OLCI) aboard the Landsat-8, Sentinel-2, and Sentinel-3 missions, respectively. Our performance assessments suggest varying degrees of improvements with respect to second-best algorithms, depending on the sensor and WQ indicator (e.g., 68%, 75%, 117% improvements for Chla, TSS, and a_cdom (440), respectively from MSI-like spectra). Map products are demonstrated for multiple OLI, MSI, and OLCI acquisitions to evaluate multi-mission product consistency across broad spatial scales. Overall, estimated TSS and a_cdom (440) from these three missions are consistent within the uncertainty of the model, but Chla maps from MSI and OLCI are more accurate than those from OLI. Through the application of two different atmospheric correction processors to OLI and MSI images, we also conduct matchup analyses to quantify the sensitivity of the MDN model and best-practice algorithms to uncertainties in remote sensing reflectance products. The analysis indicates our model is less or equally sensitive to these uncertainties compared to other algorithms. Recognizing their uncertainties, MDN models can be applied as a global algorithm to enable harmonized retrievals of Chla, TSS, and a_cdom (440) in various aquatic ecosystems.