Session: Inland Water

The SIRAL altimeter on-board the ESA satellite CryoSat-2 is the first radar altimeter that is capable of operating in synthetic aperture radar (SAR) mode. The SAR technology provides a much higher along-track resolution compared to conventional altimetry. The higher resolution makes it possible to accurately monitor much smaller water bodies than previously.

In this study, which is part of the FP7 project Land and Ocean take up from Sentinel-3 (LOTUS), we examine and validate Cryosat-2 derived lake levels and as a reference we compare our results with lake levels based on Envisat data. As a test case we consider 5 lakes; Vänern (Sweden), Okeechobee (Florida,US), Arresø, Mossø, and Skanderborgsø (Denmark), which are ranging from approximately 9 km² to 5655 km² We estimate the along-track precision and the accuracy by validating the estimated lake levels with gauge data from Okeechobee and Vänern. We find that the precision based on CryoSat-2 increase significantly for small lakes compared to Envisat. Despite the spatially varying lake crossings, Cryosat-2 is able to provide lake level time series of a quality that is comparable to or better than Envisat. Hence, these results demonstrate the promising possibilities of the upcoming mission Sentinel-3, which potentially will be able to provide accurate time series for small lakes

This work aims at preparing for the exploitation of Sentinel-3 data in hydrology. As shown in Fig.1, the non homogeneous characteristic of inland water scenes leads to very different signals (in particular Multi-Look Data (MLD) or Stacks) than those obtained over open oceans [Bercher et al., Dresden 2013].

 

This work goes further. It is indeed a qualitative and quantitative characterization of the existing SAR Mode Altimetry data over the inland water domain. The main objective is to better understand the Individual Echoes, MLDs, SAR and RDSAR waveforms that are obtained over the main three areas found in Alti-Hydrology:

 

water area: defined by a Doppler Ground Cell mainly inside the high resolution water mask,

 

 

transition area: defined by a Doppler Ground Cell in the immediate vicinity outside the water mask but that still may intersect it,

 

 

non-water area: Doppler Ground Cell far enough from the water mask so that none of the Individual Echoes that built the MLD and therefore the SAR Waveform can intercept the high resolution water mask with its pulse limited footprint (useful to check a potential impact of the side-lobes).

 

In practice, CryoSat-2 tracks are collocated with high resolution water masks and SAR images over well studied or well gauged hydrological areas. The SAR images serve both to generate the most up-to-date water masks and to analyse and understand complex scenarios themselves. For each of the three aforementioned areas of interest, the following physical and statistical indicators rare computed:

 

the Mean Individual Echoes, Stacks and Waveforms (SAR and RDSAR) as reference signals,

 

 

the Kurtosis (peakiness), Skewness (asymmetry), Standard Deviation for Stacks and Waveforms (SAR and RDSAR),

 

 

Specifically for the MLD: Stack centre (mispointing and/or mean surface slope), Range Integrated Power (RIP) statistics, MLD static & animated plots with a crossing over the non-water area, transition area and water area categories.

 

The outcome of this work may impact future retrackers and ease the mapping of water bodies through water detection criteria. A preliminary study over the Amazon forest showed that the combination of a low RIP's Standard Deviation combined to high RIP's Kurtosis is characteristic of the presence of still water (cf. left column of Fig.2) [Bercher et al., Constance 2014].

 

The European collaborative project GLaSS aims to prepare for the use of the data streams from Sentinel 2 and Sentinel 3. Its focus is on inland waters, since these are considered to be sentinels for land-use- and climate change and need to be monitored closely. One of the objectives of the project is to compare existing water quality algorithms and parameterizations on the basis of in-situ spectral observations and Hydrolight simulations. A first achievement of the project is the collection of over 400 Rrs spectra with accompanying data on CHL, TSM, a_CDOM and Secchi depth. Especially the dataset on Lake CDOM measurements represents a rather unique reference dataset.

Comparison of algorithms is done for two rather clear lakes in Italy (Garda, Maggiore), shallow and eutrophic large lakes Peipsi and Võrtsjärv in Estonia, a set of Finnish small lakes, four stations in the Betuwe area in the Netherlands (a river station with high TSM and low CHL, and three clear lakes dominated by CHL) and very turbid Lake Taihu in China. As the dataset from Lake Taihu is large (twice as much as all other data together), analyses were performed separately on a set with and without Lake Taihu.

In expansion to the Hydrolight simulation approaches by CoastColour and IOCCG report 5 studies, our simulated datasets were created based on permutations of known SIOPs and concentration ranges per lake.  

We have tested altogether 6 different sets of algorithms within the framework of this exercise. Most of them solve together CHL, TSM and a_CDOM (BOMBER, BOREAL SIOCS, WISP and WASI algorithms); for Secchi transparency two band ratio algorithms were used (BG and TO). These algorithms are applied to the in-situ and simulated reflectance spectra which have been aggregated to MERIS, S2 MSI, S3 OLCI and L8 OLI bands.

A general finding of the exercise was that algorithms performed best for the lakes they were originally designed and calibrated for. Especially CHL and TSM proved to be very difficult to retrieve accurately with one algorithm/parameterization in all separate lakes. Within a particular lake, most algorithms produced consistent results. This would lead to the hypothesis that inter-lake variability of optical properties is much larger than the intra-lake variability. Of course this result has to be checked against the variability in measurement devices and protocols since each lake was sampled by a different institute. This activity, where protocols are documented and compared, is going on and will need continuous attention in the future in the form of instrument intercomparison / intercalibration and round robin activities.

An interesting positive finding of the exercise is that some algorithms are quite effective to retrieve CDOM over a large range of concentrations and geographic scales. Also empirical algorithms for Secchi-disk retrieval perform quite well.

For inland lakes the conclusion was clear that the Sentinel 3 band set is so similar to MERIS that algorithms can be ported 1:1. There will be differences, though, resulting from different signal to noise ratios between MERIS and OLCI that need to be evaluated in a later stage.

From an earlier activity in the project (intercomparison of atmospheric correction approaches above lakes) it has become clear that standard approaches over coastal waters do not apply in all cases to inland waters.

Water quality is a critical component of global fresh water security and ecosystem health, yet existing data are scarce and declining, have poor geographic and temporal coverage, and may be of questionable accuracy. The value of remote sensing to improving understanding of water quality is recognised [1]. These methodologies may provide a complementary data stream in the water quality monitoring toolkit, including detection of the formation of algal blooms.

In Australia the occurrence of cyanobacterial blooms is increasing and results in AU$180-240 million of costs to affected communities per annum  [2].  Increases in potentially toxic cyanobacteria species in blooms is a particular concern, increasing threats to both human and animal health [3]. Traditional field monitoring for algal bloom detection involves identification and cell counting which, whilst reliable, imparts a lag time and can only be limited in spatial extent. In the Australian context, overcoming the vast geographic distances to be covered is a particular challenge; satellite remote sensing thus provides an appealing complementary, spatially continuous alternative to water managers for water quality monitoring, albeit for a limited set of water quality parameters [4]

Whilst the underlying physics of optical inland water quality is the same as that for ocean colour, remote sensing of inland waters is complicated by greater variability in optical properties [4].  However, whilst inversion algorithms are sufficiently mature to cope with the variability of optical properties in inland waters, they are primarily limited by reduced knowledge of the bio-optical properties of the inland waters.  This is particularly the case for Australian inland waters [4].

The OLCI sensor on Sentinel 3 offers the potential of wide scale, frequent monitoring of water quality in large inland water bodies in support of monitoring and the development of early bloom alerts for water managers. Recent systematic studies over several years have increased the knowledge of the optical properties of Australia’s inland waters [e.g. 5]. We further present the results of an assessment of the OLCI’s potential for monitoring inland optical water quality dynamics in Australia over a range of optical water types found in New South Wales, Australia. The research is part of a joint CSIRO/ New South Wales Office of Water project into the development of an earth observation-based algal alert system built around a range of satellite platforms (e.g. Landsat, MODIS, Sentinel 2, Sentinel 3).  The aim of the project is to develop procedures that use proximal and remote sensing technologies to support cyanobacterial monitoring, including real time data for immediate management use, and a wider spatial coverage of major inland waters.

Using both modeled and real spectral data, in the initial stages established semi-empirical water quality algorithms for both suspended sediment concentrations and algal greening were developed to form the basis of an algal alerting system for NSW water managers. The algorithms were developed and tuned using in situ measurements of reflectance and water constituent concentrations taken during 2012 and 2013 over two seasons in these NSW water bodies. The results of the empirical analysis will be reported here. Later stages of the project will investigate the value of physics-based inverstion algorithms potentially coupled to alert algorithms for pigments specific in cyanobacteria.

References:

The Group on Earth Observations (GEO) Inland and Near-Coastal Water Quality Remote Sensing Working Group

Davis, J.R. and Koop, K. (2006) Eutrophication in Australian rivers, reservoirs and estuaries: A southern hemisphere perspective on the science and its implications, Hydrobiologia, 559:3–76.

Carmichael, W. W. (2001) Health effects of toxin-producing cyanobacteria: “The 37 CyanoHABs”. Human and ecological risk assessment: An International Journal, 7:1393-1407.

Dekker, A.D. and Hestir, E.L., (2012). Evaluating the Feasibility of Systematic Inland Water Quality Monitoring with Satellite Remote Sensing. CSIRO: Water for a Healthy Country National Research Flagship, Canberra, p. 123.

Hestir et al. (2015). The relationship between dissolved organic matter absorption and dissolved organic carbon in reservoirs along a temperate to tropical gradient. Remote Sensing of Environment, 156:395–402.

To enhance the inland water monitoring capability from space observations, two main ideas can be explored. The first one consists in increasing the temporal/spatial coverage of the radar altimeter measurements by enlargement of the altimeter constellation or densification of the observation points. Secondly, improvements in the water level estimation availability and precision are also required. The new generation of radar altimeter featuring a delay/doppler mode (currently on board the Cryosat-2 satellite and in the upcoming Sentinel-3 and future Jason-CS missions) will clearly fulfill this second point offering better performances than those of conventional altimeters. The high along-track spatial resolution of the delay/doppler altimeter may enable the measurement of high-resolution water level transects across rivers and other water basins (where conventional altimeters echoes are impacted by land contamination due to their larger footprint size). Such dataset would also permit to extract spatial and temporal derivatives of water level, which are a characteristic of major interest in hydrology. It is widely acknowledged that delay/doppler radar altimeter (Sentinel-3 will be first mission to cover the entire hydrological surfaces with a delay/doppler mode) would allow to investigate hydrological sites in more depth than ever before, providing new insights into the hydrological processes and phenomena occurring in such systems.

However, the delay/doppler altimeter capability is a new feature and no data products based on this measurement mode are provided yet or used operationally over inland water bodies. Actually, new innovative methods can be developed with respect to those implemented in the CryoSat-2 ground segment and in the on-going Sentinel-3 mission. It is of high importance for Sentinel-3 mission, and also for SWOT at a later time, to provide a significant increase in the monitoring capabilities of continental water bodies even before the launch of these satellites and to develop and validate suitable processing methods for improving satellite radar altimetry products for the end-users.

In this context, CLS is conducting a study to optimize the water level estimation performances obtained with delay/doppler measurements in hydrological conditions, by developing new methodologies in data processing and new analysis methods.

In this talk, we propose to present the problematic, the measurement principle together with our processing and validation methods.