1. Improving Sentinel-3 SAR mode processing over lake using numerical simulations
Access to fresh water is a key issue for the next decades in the context of the Global Warming. The water level of lakes is a fundamental variable which needs to be monitored for this purpose. The radar altimetry constellation brings a worldwide means to this question. Recent advances in radar altimeter on-board tracking modes have allowed in monitoring thousands of lakes and rivers. Now, measurements are widely available with better resolution: it is time to drastically improve the processing.
The altimetry waveforms over lakes are difficult to analyze, and very different from the ocean ones. We face a large variety of signals due to surface roughness, lake geometry and environment. The inversion process, named retracking, shall be able to describe all these components.
We propose here a retracking based on physical simulations taking as inputs the lake contour and the instrument characteristics. Fitting the simulation on the waveforms gives the water surface height. The algorithm has been tested on the Sentinel-3A and Sentinel-3B time series over Occitan reservoirs (France) and Swiss lakes and compared to in-situ references. Over small Occitan reservoirs (few ha to few km²), the UnBiased Root Mean-Square Error (ub-rmse) is better than 14 cm. Over the medium size Swiss lakes, the ub-rmse is better than 10 cm for most of them.
These performances often surpass by a factor of at least 2 those of the OCOG retracking (retracking available in operational products). It also even allows to measure water levels where it was unreachable before. This method, which we will described in detail in this presentation, is automated. This also proves that radar altimeters, even on very small lakes of few ha, allows reaching accuracy as good as laser altimetry (ICESAT2) which has been evaluated in [Cooley et al., 2021].
The impact of climate change on freshwater availability has been widely demonstrated to be severe. The capacity to timely and accurately detect, measure, monitor, and model volumetric changes in water reservoirs is therefore becoming more and more important for governments and citizens.
In fact, monitoring over time of the water volumes stored in reservoirs is mandatory to predict water availability for irrigation, civil and industrial uses, and hydroelectric power generation; this information is also useful to predict water depletion time with respect to various scenarios.
At present, water levels are usually monitored locally through traditional ground methods by a variety of administrations or companies managing the reservoirs, which are still not completely aware of the advantages of remote sensing applications.
The continuous monitoring of water reservoirs, which can be performed by satellite data without the need for direct access to reservoir sites and with an overall cost that is independent of the actual extent of the reservoir, can be a valuable asset nowadays: water shortage and perduring periods of droughts interspersed with extreme weather events (as it has been experienced across all Europe in the latest years) make the correct management of water resources a critical issue in any European country (and especially in Southern Europe).
The goal of this work is therefore to provide a methodology and to assess the feasibility of a service to routinely monitor and measure 3D (volumetric) changes in water reservoirs, exploiting the huge, various, and more and more increasing Earth Observation (EO) Sentinel big data.
However, to turn them into information and designing possible services useful for stakeholders, two main aspects must be considered: the computing infrastructure to store and handle the data and the models, and corresponding algorithms to extract the valuable information.
An experiment of the prototypal service is ongoing on two reservoirs in Italy providing the freshwater supply for nearly two million people. This experiment is based on HPC to process satellite data (including Sentinel-2 Level-0 data, that are not usually accessible to users, thanks to an agreement with ESA-ESRIN) and different monocular and stereo models to estimate the surface extent of reservoirs and its height variation; in addition, local information on water level are eventually considered for building an evolving 3D model of the reservoir itself. As a side objective, debris carried by tributary rivers (especially during the even more frequent extreme weather conditions) that can accumulate in the shallow sections of the reservoir and modify reservoir volume over time, could be detected.
Overall, the work addresses the following objectives (OBJ):
OBJ-1 [Scientific]: Investigation of the capabilities of EO Sentinel big data to provide timely monitoring of 3D changes in water reservoirs
OBJ-2 [Technical]: Development and implementation of a novel methodology in a free and open source software based on cloud computing infrastructure, exploiting 4D EO Data Cubes, to practically deploy new services for water reservoirs volumetric monitoring
OBJ-3 [Governance]: Application and validation of the services, in selected relevant cases of water reservoirs monitoring, where independent reference data are available
The work will have direct impacts directly connected to several United Nations Sustainable Development Goals: (6) Clean Water and Sanitation, (7) Affordable and Clean Energy, (9) Industry, Innovation and Infrastructure, (11) Sustainable Cities and Communities, (13) Climate Action, (15) Life On Land.
The retrieval of lake ice thickness (LIT, an Essential Climate Variable or ECV) from satellite remote sensing is a topic that has been gaining traction in recent years. As work on the retrieval of LIT intensifies in the coming years with the launch of new altimetry missions (Low-Resolution Mode: LRM and Synthetic Aperture Radar: SAR mode) and with growing interest in the production of climate data records of LIT through the processing of historical time series, there is a need to examine the impact of various ice and overlying snow properties on backscatter and brightness temperature measurements from various altimetry missions (1992-present). Our understanding of the interactions between ice/snow properties of frozen lakes and microwave radiation at frequencies operated aboard altimetry missions is in fact very limited. A project was therefore initiated by the European Space Agency (ESA) in 2020, under the name “Towards the retrieval of lake ice thickness from satellite altimetry missions (LIAM)”, to investigate the sensitivity of backscatter and brightness temperature measurements from ESA and non-ESA satellite radar altimetry missions to varying snow and ice properties on northern lakes.
This talk will provide a synthesis of key results from the LIAM project, notably: 1) forward modelling of brightness temperature (18-37 GHz) and backscatter/waveforms (3-36 GHz) from frozen lakes with varying ice (snow ice, bubbles, roughness at interfaces) and overlying snow (depth, density, wetness) properties using the Snow Microwave Radiative Transfer (SMRT) model linked to a 1-D thermodynamic lake ice model; 2) analysis of the impact of land contamination, snow on ice, and ice structure on radar backscatter and brightness temperature measurements, as well as the spatio-temporal variability of waveforms; 3) comparison of SMRT simulations with altimeter and radiometer measurements acquired over Great Slave Lake (Canada) and Lake Baikal (Russia); and 4) identification of ice and snow conditions, such as snow-free lake ice and snow wetness, that may limit the retrieval of LIT from the analysis of waveforms. Finally, the talk will conclude with a discussion of broader implications of the findings in light of the upcoming Surface Water and Ocean Topography (SWOT) and Copernicus Polar Ice and Snow Topography Altimeter (CRISTAL) missions.
Lake ice is a key component of the landscape in the northern hemisphere. The presence or absence of lake ice impacts local weather conditions and is a key factor to consider within weather forecasting models. Additionally, the presence of ice is important for local economies at northern latitudes, allowing for the establishment of ice roads that act as transportation and supply routes during winter months. However, with changing climate the thickness and length of ice seasons are trending towards thinner ice cover and shorter seasons. Observational records are an important part of understanding these changes, however, over the last four decades, there has been a decrease in the number of in situ observations of lake ice. The common solution to this decrease in observations is the use of satellite remote sensing. Remote sensing allows for the observation of large areas containing dense collections of lakes. Active microwave remote sensing, in particular synthetic aperture radar (SAR), has been the most popular remote sensing technology for the study of lake ice over the last 50 years. Advantages of this technology include the limited obstruction by clouds and the higher resolution provided by the imagery. The response of SAR backscatter to lake ice has been consistently reported, however, recent literature has shifted understanding of the mechanisms responsible for this response. Past observations focused on the role of tubular bubbles in the ice and the presence of a double bounce scattering mechanism. However, recent experiments using numerical modelling, polarimetric decomposition, and co-pol phase difference indicate that roughness of the ice-water interface and a single bounce scattering mechanism are more likely the dominant factor in the backscatter response observed from lake ice.
Forward modelling through radiative transfer models provides a unique opportunity to better understand how roughness of the ice-water interface and other lake ice properties contribute to the response of SAR imaging backscatter from lake ice. Several radiative transfer and numerical models have been developed to explore these contributions; however, each model presents individual limitations. For example, treating the ice column as a single layer, ignoring the presence of snow, or discounting the role of surface ice types. Additionally, experiments that have been performed use synthetic values for snow and ice properties (temperature) that influence the dielectric properties of the mediums. The ranges of ice properties tested have also been limited, focusing on small ranges or specific values. The recently published Snow Microwave Radiative Transfer (SMRT) model provides a modelling framework that can be used to address these limitations. The key advantages of SMRT are the allowance of multilayer ice and snow mediums, the ability to include roughness at different interfaces, and the inclusion of multiple electromagnetic and microstructure models. This research has two main objectives: 1) explore how changes in lake ice properties impact SAR imaging backscatter; and 2) investigate the use of SMRT for forward modelling of SAR imaging backscatter from lakes under varying conditions. Both shallow lakes that form tubular bubbles and deeper lakes where these bubbles are absent are compared in this research as deeper lakes have been largely ignored within the literature.
Initial experiments focus on exploring how changes in lake ice properties impact SAR imaging backscatter to identify the key properties for both shallow and deep lakes. To bring simulations closer to reality, a 1-D thermodynamic lake ice model is used to parameterize ice columns. Both a clear ice column and ice column with snow ice are developed based on lake ice model simulations and split into 4 layers to capture the temperature profile within the ice. In addition, for shallow lakes, the lower layer of the ice column includes spherical bubbles to serve as a representation of tubular bubbles. These experiments assume dry snow conditions to replicate conditions during ice growth. SMRT model simulations are run for three different microwave frequencies common for lake ice remote sensing from imaging SAR, 1.27 GHz (L-band), 5.4 GHz (C-band), 9.6 GHz (X-band) at three different incidence angles of 20°, 30°, and 40°. Properties tested include ice thickness, spherical bubble radius, ice porosity, RMS height, and correlation length. Each property is incrementally increased while the others are held constant; ranges of properties are based on previous field observations. One finding is that there was limited response across frequencies to changes in ice thickness and ice porosity. All frequencies show the highest response to RMS height supporting past conclusions that it is the key property impacting SAR backscatter from lake ice. Additionally, the response to RMS height was similar between shallow and deep lake scenarios. However, higher frequencies (X-band) show an increase in the response to surface bubble radius for the deep lake scenario. This response was lower for the shallow lake scenario due to the increased RMS height at the ice-water interface used as a baseline to replicate the extrusion of tubular bubbles. These results indicate that lower frequencies (L and C-band) are better suited to studying properties such as RMS height while higher frequencies (X-band) are better suited to studying surface ice properties.
These initial sensitivity experiments indicate the importance of RMS height in reproducing backscatter from lake ice within the SMRT framework. The next objective of the research is to conduct forward modelling to simulate backscatter using field data from both a shallow and deep lake located in Subarctic Canada. Malcolm Ramsay Lake (-93.78°, 58.72°) located near Churchill, MB is used as the shallow lake, and Noell Lake (68.53°, -133.56°) located near Inuvik, NWT is used as the deep lake. Field data collected on snow properties and ice structure are supplemented by lake ice model simulations to parameterize SMRT. To validate the results of these simulations multiple SAR images at different frequencies (L, C, and X-band) are acquired for the two lakes. Comparison of simulated and observed satellite backscatter indicates error values ranging from 1-3 dB and reasonable representation of the patterns in observed backscatter. Key limitations identified in the forward modelling were the representation of deformations of the ice surface early in the ice season and difficulties associated with modelling across different polarizations. Next steps for this research will include exploring the application of SMRT under surface melt conditions and the representation of different layers with varying water content. The forward modelling work conducted is an important initial step in progressing towards conducting inversion modelling using SMRT to retrieve key ice properties such as RMS height.
Lake ice thickness (LIT) is recognized as an Essential Climate Variable (ECV) by the Global Climate Observing System (GCOS). LIT is a sensitive indicator of weather and climate conditions through its dependency on changes in air temperature and on-ice snow depth. The monitoring of seasonal variations and trends in ice thickness is not only important from a climate change perspective, but it is also relevant for the operation of winter ice roads that northern communities rely on. Yet, field measurements tend to be sparse in both space and time, and many northern countries have seen an erosion of in situ observational networks over the last three decades. Therefore, there is a pressing need to develop retrieval algorithms from satellite remote sensing to provide consistent, broad-scale and regular monitoring of LIT at northern high latitudes in the face of climate change.
This talk presents a novel, physically-based retracking approach for the estimation of LIT by using conventional low-resolution mode (LRM) and synthetic aperture radar (SAR) Ku-band radar altimetry data. Details will be provided about the formalism of the LRM and SAR LIT retracking methods and assessment of retrieved ice thickness using thermodynamical simulations and in-situ data. Results will focus on LIT estimation obtained using Jason-2, Jason-3, and Sentinel-6 data over Great Slave Lake (Canada) for different winter seasons. Finally, the talk will highlight how these methods significantly improve the accuracy of the LIT estimations, paving the way towards regular and robust LIT monitoring with current and future LRM and SAR altimetry missions.
The LRM_LIT algorithm has been developed in the framework of the European Space Agency’s Climate Change Initiative (CCI+) Lakes project and is currently being implemented for the production of LIT time series from LRM data for Phase 2 of the project starting in March 2022. These data will be publicly available to the scientific community through a dedicated data platform, following the project schedule (2022-2025). The SAR_LIT algorithm is being developed within the ESA S6JTEX project that aims at enhancing the scientific return of the tandem phase between the Jason-3 and Sentinel-6 reference missions, allowing for continuity of observations across 30 years of conventional altimetry (from Topex or ERS in 1992) and SAR altimetry data, from Cryosat-2 to Sentinel-3 and now Sentinel-6 missions.