Inland Water Quality

The Sahelian region has experienced a dramatic rainfall deficit over the second half of the last century with severe droughts in the early seventies and eighties. In parallel and paradoxically, an increase in surface runoff has been observed over the same period of time in different areas of the Sahel. In particular, Sahelian ponds have shown a dramatically increase in their surfaces as well as an increase in turbidity. The hydrological processes responsible for this paradoxical situation, less rain but more surface water, still need to be fully understood. This is fundamental to predict the future evolution of these ponds, which provide water for cattle and people living in the area.

In that respect, small water bodies (< 250 m²) play a pivotal role for Sahelian societies, as opposed to the few rare large lakes and rivers (eg Niger river), but these small water bodies are far less studied, because it requires high spatial and temporal resolution that only recent satellites can provide.

At the same time it is important to monitor the water quality of these ponds in relation to health issues, particularly regarding diarrheic diseases that are still a major cause of childhood mortality in these areas.

This work aims at better monitoring both water quantity and quality of Sahelian ponds and lakes using remote sensing data. The methodologies developed to do this have been tested at different sites over which in-situ water height and/or water turbidity measurements are available

 Water surfaces derived by classification of optical remote sensing images from high resolution satellite (SPOT, FORMOSAT, LANDSAT) were combined to in-situ water level measurements to estimates water volume for the Agoufou pond in the Gourma region in Mali.  The total amount of water supplying the pond  was calculated by applying a  pond water balance equation which takes into account volume changes, evaporation, rainfall and infiltration. This methodology has  been applied to archive data by different remote sensing satellites (SPOT, LANDSAT, CORONA), and extended to other Sahelian ponds in Mauritania and Niger, to quantify the dramatic changes in the water amount supplying the ponds over the last 60 years This provides a basis to evaluate hydrological model approaches.

Concerning water quality, turbidity can be used as a proxy for fecal bacteria. In an area where environment survey is minimal or inexistent, water turbidity monitoring from space may prove very important. However, estimating turbidity by remote sensing in this region is quite challenging given the high aerosols content which complicate atmospheric corrections, the pond's high turbidity values, that are outside the range of most studies, and the small size of most ponds. In-situ turbidity, suspended particulate matter, water reflectance by spectroradiometer and water absorption properties have been measured at different sites (the Agoufou pond, the Bagré lake in Burkina Faso, the Guiers lake in Senegal) and used to evaluate different indexes to derive water turbidity  from the reflectance in the red and infrared bands of moderate and high resolution optical sensors (MODIS, SPOT5/Take5, LANDSAT). This allowed monitoring the temporal evolution of pond's turbidity at the seasonal, interannual and decadal scale and linking it to other hydrological parameters, such as ponds' water volume and rainfall that can be helpful to predict its future evolution.

The methodology developed here for the Agoufou pond could be easily applied to Sentinel-2 data and later to SWOT data which will provide at the same time water levels, pond's surface and turbidity and can therefore be used to derive large scale estimates of water amount and quality of the numerous and rather small ungauged Sahelian ponds.

Concern over toxins and public health threats resulting from Cyanobacterial Harmful Algal Blooms (CHABs) have gained attention as reoccurring and seasonal blooms persist in many waters. Recent work has suggested that inland CHABs are linked to clusters of Amyotrophic Lateral Sclerosis (ALS), a progressive neurodegenerative disease. Due to traditional barriers monitoring and assessment techniques are challenging considering the number, size, and distribution of inland freshwater lakes. The objectives of this work are to map inland lake water quality and carry out eco-epidemiological modeling to assess ALS incidence, lake water quality, and risk factors. BigData approaches for multi-scale satellite remote sensing were applied to map lake water quality metrics including chlorophyll-a (chla), cyanobacteria density, and Phycocyanin Concentration (PC) among others. Trophic status for all lakes over 6 hectares across the continental united states were mapped using Landsat archives, in situ observations, and randomforest. Automated processing of reflectance for more than 200,000 scenes was executed using 6S, a combination of MODIS and MERRA paramaterization, and criteria filtering. Withheld independent samples show a robust monitoring tool with Receiver Operating Characteristic curves generally above 0.80 and R2 of 0.65. Chlorophyll-a, cyanobacteria density, and Phycocyanin Concentration were mapped for the northeast and Great Lakes USA regions using fused semi-analytical and empirical algorithms with MERIS, MODIS, Proba, Rapid Eye, and Landsat. Operational masking for clouds and floating aquatic vegetation was integrated. Empirical and semi-analytical models achieved high overall accuracies (R2 0.62-0.88; RMSE 0.85-1.56 ug/L). An ALS case database with completed questionnaire surveys, including information on residential history and related risk factors, was integrated into our spatial analysis framework. Spatially-aware logistic regression and Bayesian Hierarchical Modeling were used to assess relationships among lake water quality and ALS cases. Generally, eco-epidemiological modeling shows that poorer lake water quality and elevated cyanobacteria levels were significantly associated with increased odds of belonging to an ALS cluster in the region. A broad impact of this study that benefits society is the development of methods and study findings that help understand how water quality and harmful algae impact human health. All map products and Python code are shared in an open source and transparent for the user community.

Recent high resolution land colour sensors such as Sentinel-2 Multi Spectral Instrument (MSI) have opened new possibilities for monitoring small inland water bodies, where moderate resolution sensors (e.g. Sentinel 3 OLCI) are not sufficiently spatially resolved. The spectral coverage (including SWIR bands for atmospheric correction) and short revisit time of S2-MSI make it an interesting platform for water quality retrieval in areas where remote sensing, including atmospheric correction, has proved particularly challenging, i.e. small inland water bodies and near coastal areas. Still, as MSI is mainly designed for terrestrial applications, it does not fulfil all requirements for water quality remote sensing applications such as narrow bands, a high dynamic range and radiometric resolution, and high signal-to-noise ratio (SNR). Therefore it is important to evaluate the radiometric performance of MSI over case 2 inland waters.
The work explores the use of MSI to retrieve inland water constituent concentrations (total suspended matter (TSM), chlorophyll (Chl-a), and coloured dissolved organic matter (CDOM)) over a variety of case 2 water types. It is important to evaluate both the range and sensitivity of constituents that can be detected by MSI using existing retrieval algorithms.
A modelling pipeline is set up using HydroLight to produce water-leaving radiance and subsequent 6S radiative transfer modelling to propagate the signal through the atmosphere to produce top-of-atmosphere radiance, L_TOA. MSI radiometric parameters (channel spectral responses, SNR and analogue-to-digital quantization) are used to simulate the signal recorded by MSI over water. Our preliminary results have shown that using simple one band peak values or band ratios, it is possible to use S2 MSI to retrieve TSM and Chl-a, but retrieval of CDOM is more challenging.
In this study, case 2 water-leaving radiance is simulated for wide concentration ranges of Chl-a, TSM and CDOM, thus representing water types from clear to extremely absorbing and scattering cases. The modelling is based on a standard four-component case 2 model of HydroLight with additional input from in situ measurements to provide more realistic water type scenarios. To take into account the effect of atmosphere in the MSI retrieval capability, several scenarios, representing varying aerosol load, are used in the atmospheric modelling. Changes in L_TOA driven by water constituent concentrations are compared with the noise equivalent radiance as specified for MSI to indicate the accuracy of the sensor in estimating those constituents. As the S2 MSI bands have high spatial resolution but relatively low SNR for water quality applications, the effect of spatial binning to improve the SNR is explored in more detail.
We will present the results of the predictive modelling study and show how current water constituent retrieval algorithms should perform on MSI imagery. Finally, the modelling results are compared to first data from Sentinel-2 MSI.

GloboLakes is a five-year NERC funded research programme investigating the state of lakes and their response to climatic and other environmental drivers of change through the realization of a near-real time satellite based observatory (Sentinel 3) with archive data processing (MERIS, SeaWiFS) to produce a ~20-year time series of observed ecological parameters and lake temperature for over 1000 lakes globally.  However, the diverse and complex optical properties of lakes means that algorithm performance often varies markedly between different lake types and as such global solutions cannot be found.  The GloboLakes project proposes to overcome this challenge by developing a processing chain whereby algorithms are dynamically selected at the pixel-level according to the optical properties of the lake under observation.

The LIMNADES data repository is an initiative developed under the auspices of GloboLakes and has so far collated measurements of in-situ water-leaving reflectance and inherent optical properties for more than 1500 sampling stations located over 80 lakes globally.  In this presentation we will show the results of a comprehensive clustering analysis performed on the LIMNADES water-leaving reflectance data and how this was used to identify different lake optical types and develop a framework for algorithm validation.  In addition, we will present our initial results from the testing of biogeochemical retrieval algorithms using the LIMNADES database and show how their performance varies across the different lake optical types.

Water Quality products from inland waters (rivers, lakes and reservoirs) have attained attention during the recent past. The EU WFD requires monitoring of all surface waters exceeding a certain size. Based on the experience to provide water quality information from satellites gained for coastal waters, research organisations and operational service providers have prepared themselves to provide such information also for inland waters. Ocean colour sensors, specifically Sentinel 3 OLCI will be fully applicable but their application is limited due to the comparable coarse spatial resolution. High spatial resolution data with good spectral characteristics and radiometric quality is a mandatory prerequisite, and Sentinel 2 is going to fulfil this in an excellent way. Landsat 8 is considered as precursor, already providing good turbidity like information but lacking spectral bands to provide chlorophyll / eutrophication products.

Aiming at global inland water products from Sentinel data requires not only appropriate algorithms but also newly designed data processing and storage infrastructure. Data flow and data volume will exceed what is in place until now. Within the FP7 GLaSS Project we have been addressing these challenges and opportunities and prepared for the uptake of Sentinel-2 and Sentinel-3 data for global inland water applications. One central element is the GLaSS CORE System which is designed to perform data download, pre-processing including subsetting, and data distribution to subsequent distributed lake specific thematic processing. From the pick-up point, the value-adding users are further processing the data with dedicated algorithms for serving end users with inland water quality products. The data stream from the ESA GSCDA to end-users (e.g. water authorities), the GLaSS core system is built for retrieval, subsetting and mosaiking of the EO data to regions of lakes, pre-processing to a standard product, archiving of the subsets, and distribution to the users, both systematically and on-demand. The system provides cloud statistics over subsets, if required exactly over the lakes or other user defined regions of interests which enables the users to select suitable products of their interest by queries. If lakes are covered by several tiles, they will be mosaicked to one product (along track) so that the area of interest is not divided into several products. All pre-processing steps are performed with the SNAP functionality which runs on Calvalus processing system.

The atmospheric correction is an important algorithmic step in the production chain. It is optionally offered by the pre-processing of the core system or will be part of the downstream lake specific thematic processing. The atmospheric correction over water of Sentinel 2 is a field of ongoing research. Methods currently available are ACOLIGHT (RBINS) and an adaption of the C2RCC (Case 2 Regional CoastColour) to Landsat 8 and Sentinel 2 (in preparation at the time of abstract writing), both available in the public domain, including ESA’s SNAP for C2RCC.

The in-water processing is also subject to ongoing research. A first global processing on inland waters has been achieved within the ESA DUE DIVERSITY II project, where the version 1 of the CaLimons chain has been developed, which is now being further developed by the UK GloboLakes project and applied to 1000 lakes globally. The inland water production chain at Brockmann Consult will consist of feeding the output of the GLaSS core system into the CaLimnos thematic processing.

Currently, the inland water processing chain comprising the GLaSS CORE System and the thematic processing is systematically serving Landsat-8 data. Thanks to the Sentinel Science Data HUB, the system could already be adapted and tested to download Sentinel-2 data systematically over different lakes. It will be prepared for full data ingestion as soon as data are officially available, which is expected to take place soon after submission of this abstract.

Sentinel-3 data will be included when available and we expect to report at the time of the LPS. As part of the S3VT team, we will have early access to data and will be able to be in place with a running system end of commissioning phase.

This presentation will give an overview of the algorithms deployed to generate inland water products, based on open source algorithm (mainly through the ESA SNAP toolbox), the infrastructure of the system, the data flows, pre-processing steps included and the distribution of data.