Contributors: C. Merchant (University of Reading), L. Carrea (University of Reading), B. Calmettes (Collecte Localisation Satellites), N. Taburet (Collecte Localisation Satellites), R. Kidd (EODC GmbH), C. Briese (EODC GmbH), A. Dostalova (EODC GmbH)

Issued by: EODC GmbH/Richard A Kidd

Date: 21/12/2022

Ref: C3S2_312a_Lot4.WP3-TRGAD-LK-v1_202204_LK_TR_GA_i1.1

Official reference number service contract: 2021/C3S2_312a_Lot4_EODC/SC1

<style>
.special_indent ul > li > ul {
    padding-left: 0;
}

.special_indent ul > li > ul > li > ul {
    padding-left: 1.5em;
}
</style>

History of modifications

Related documents 

Acronyms

General definitions

Level 2 pre-processed (L2P): this is a designation of satellite data processing level. "Level 2" means geophysical variables derived from Level 1 source data on the same grid (typically the satellite swath projection). "Pre-processed" means ancillary data and metadata added following GHRSST Data Specification, adopted in the case of LSWT.

Level 3 /uncollated/collated/super-collated (L3U/L3C/L3S): this is a designation of satellite data processing level. "Level 3" indicates that the satellite data is a geophysical quantity (retrieval) that has been averaged where data are available to a regular grid in time and space. "Uncollated" means L2 data granules have been remapped to a regular latitude/longitude grid without combining observations from multiple source files. L3U files will typically be "sparse" corresponding to a single satellite orbit. "Collated" means observations from multiple images/orbits from a single instrument combined into a space-time grid. A typical L3C file may contain all the observations from a single instrument in a 24-hour period. "Super-collated" indicates that (for those periods where more than one satellite data stream delivering the geophysical quantity has been available) the data from more than one satellite have been gridded together into a single grid-cell estimate, where relevant.

Target requirement: ideal requirement which would result in a considerable improvement for the target application.

Threshold requirement: minimum requirement to be met to ensure data are useful.

Brokered dataset: dataset that has been generated and validated generally with Research and Development (R&D) efforts in an external initiative/project, and then tailored to be made available via this Copernicus service.

LSWT: Lake Surface Water Temperature is the temperature of the water at the surface, within roughly 10-20 micrometers from the surface (skin temperature). The product for this service consists of one value of the temperature for each resolution cell of a 0.05deg regular grid. The temperature is in kelvin and it is equipped with uncertainty in kelvin and quality level.

LWL: Lake Water Level refers to the water level above the geoid. The product for this service consists of one value of the water level for each lake. The level is in metre.

Scope of the document

This document aims to provide users with the relevant information on requirements and gaps for each of the given products within the Land Hydrology and Cryosphere service. The gaps in this context refer to data availability to enable the ECV products to be produced, or in terms of scientific research required to enable the current ECV products to be developed in response to the specified user requirements.

The Lakes Service provides two products: a Lake Surface Water Temperature (LSWT) product, and a Lake Water Level (LWL) product.

Initially an overview of each product is provided, including the required input data and auxiliary products, a definition of the retrieval algorithms and processing algorithms versions; including, where relevant, a comment on the current methodology applied for uncertainty estimation. The target requirements for each product are then specified which generally reflect the Global Climate Observing System (GCOS) ECV requirements. The result of a gap analysis is provided that identifies the envisaged data availability for the next 10-15 years, the requirement for the further development of the processing algorithms, and the opportunities to take full advantage of current, external, research activities. Finally, where possible, areas of required missing fundamental research are highlighted, and a comment on the impact of future instrument missions is provided.

Executive Summary

The Lakes Service provides two Essential Climate Variable (ECV) products, specifically lake surface water temperature (LSWT) and lake water level (LWL). The LSWT climate data record (CDR) is a daily gridded product derived from observations of one or more satellites. It contains estimates of the daily mean surface temperature of the lake, from 1995 to 2020, and has been attempted for about 2000 lakes examined by the European Space Agency (ESA) Climate Change Initiative (CCI) Lakes¹ initiative. The LSWT CDR v4.0 product is composed of the brokered ESA CCI Lakes CDR extended within the Copernicus Climate Change Service (C3S) service up to October 2021. The satellites contributing to the time series are: ATSR-2, AATSR and AVHRR MetOp-A, AVHRR MetOp-B, MODIS Terra, SLSTR Sentinel3A and SLSTR Sentinel3B.

The LWL CDR, which is both brokered and generated in the ECV Lakes Service, is an estimate of the mean surface height of the lake, wherever at least three valid observations have been made within the intersect between the satellite ground track and a given lake. The LWL product targets 166 lakes worldwide, from September 1992 to December 2020, with daily to decadal monitoring (in CDR v3.1 with a status date of April 2021). The satellites contributing to the time series are: TOPEX-Poseidon, Jason-1/2/3 and Sentinel-3A/3B. The data format for both LSWT and LWL products are netCDF4 classic, adopting relevant CF (Climate Forecast) conventions². Initially CF was developed for gridded data from climate and forecast models (hence "CF") of the atmosphere and ocean, but its use has subsequently been extended to other geosciences, and to observations as well as numerical model outputs.

Continued reliance of the LSWT product on data from the AVHRR sensor is assured, due to the the MetOp and MetOp SG programmes, which guarantee a secure supply of AVHRR data up to 2042. The inclusion of data from VIIRS would have a relevant impact on the LSWT product quality, but research is needed for its exploitation, and none is presently planned or proposed.

The requirements for the Lakes Service products are largely reliant upon the statements from GCOS, published literature and experience from other CDR projects. For LWST, the threshold for user requirements are generally already reached. However, more in situ data are required in order to be able to provide reliable assessments of product stability. For LWL, either the target or the threshold target has already been reached.

Further development of the retrieval methodologies is required. For the LSWT product, improvements in pixel classification and in the optimal estimation (OE) retrieval algorithm are required. Adaptation to a 0.025^o gridding should be possible and useful if there is genuine user demand. For the LWL product, an automatic version of the geographic extraction zone for altimetry measurements is required, along with improvements to the geophysical corrections of the extracted data.

The uncertainty estimation within LSWT has been fully developed within CCI Sea Surface Temperature (SST)/Lakes activities and is considered to be mature. For LWL the uncertainty variable only estimates the precision of the measurements and not the accuracy, and this will be addressed in the CCI Lakes project.

In addition to LSWT and LWL, elements of lake surface reflectance, lake area and lake ice cover and thickness are included in the GCOS Lake ECV definition. A review of the opportunity to broker datasets addressing these gap areas is scheduled at a later stage of the project (Period 2 of the C3S activities ending in October 2023), and is not covered by the analysis reported here.

Section 1.2 briefly presents the Lake ECV products provided in the service - lake surface water temperature (LSWT) and lake water level (LWL) as background to the remainder of the report.

Section 2 presents known statements of requirements directly relevant to the products in the context of the C3S, in terms of definitional, coverage, resolution, uncertainty, format and timeliness requirements. The C3S team's view and interpretation of these statements of requirement and their relevance to the C3S service is stated.

Section 3 presents an analysis of gaps and opportunities:

• current observational constraints and additional/future sources of satellite data
• known areas for improvement of LSWT and LWL estimation methods
• known areas for improvement of LSWT and LWL uncertainty estimation methods
• lake ECV components not presently delivered by the Hydrology service within the C3S 312b LHC service

Reliance on External Research

Since the C3S programme only supports the implementation, development and operation of the CDR processor, any scientific advances of the C3S products entirely rely on funding provided by external programmes, (such as CCI-Lakes, H-SAF, and Horizon2020, amongst others). Thus, the implementation of new scientific improvements can only be implemented if external funding allows for it. This depends both on the availability of suitable programmes to support the R&D activities and the success of the C3S contractors in winning potential suitable calls.

Product description: Lake ECV Service

Introduction

The Lake ECV Service consists of two products – lake surface water temperature (LSWT) and lake water level (LWL) presented as a Climate Data Record (CDR) which is updated once a year. The CDRs are created as an adaptation of state-of-the-art products specifically for the C3S service where scientific advancements have been investigated and applied. For these versions of the CDRs, the latest ESA CCI Lakes datasets are employed. The C3S extends every year the CDR and if available includes new advancements of the datasets.

The Lake ECV products

Brokered and Generated LSWT CDR v4.0

The LSWT climate data record (CDR) brokered to the C3S is a daily gridded product derived from observations of one or more satellite sensors (L3S, level-3 super-collated). The reported LSWT is an estimate of the daily mean surface temperature of the lake, wherever at least one valid observation has been made within the spatial grid cell on a given day. The grid is a regular latitude-longitude one at 0.05 degree intervals.

In addition to the cell-mean LSWT data, the product contains (for more details see the Product User Guide and Specifications document, RD.4):

• an uncertainty estimate for the LSWT as an estimate of the daily cell-mean value;
• a quality level indicator for the LSWT between 0 (invalid) and 5 (excellent), the recommended quality levels for most applications being 4 and 5;
• the satellite/s and instrument/s from which LSWTs were combined to make the gridded estimate;
• a flag indicating whether a cross-sensor offset adjustment has been applied to the temperatures.
• metadata, including funder and citation instructions;
• the main lake ID for each cell (from ESA CCI Lakes).

The data format is netCDF4 classic model³, adopting relevant CF conventions⁴.

The CDR v4.0 covers the period 1995 to 2020. It has been brokered from the ESA CCI Lakes initiative, produced with LSWT v4.5. The sensors (and satellite missions) contributing to the time series were: ATSR-2 on ERS, AATSR on Envisat, MODIS on Terra, AVHRR on MetOp-A, AVHRR on MetOp-B, SLSTR on Sentinel3A and SLSTR on Sentinel3B.

The CDR v4.0 contains scientifically consistent time series since the same physics-based algorithm has been employed for all the sensors so that the brokered dataset can be used seamlessly with the extended one.

LWL V3.1: Brokered and Generated CDR

The LWL climate data record (CDR), brokered to the C3S, is a timeseries product derived from observations of one or more satellites. The reported LWL is an estimate of the mean surface height of the lake, wherever at least three valid observations have been made within the intersect between the satellite ground track and a given lake.

In addition to the lake-mean LWL data, the timeseries contains:

• the UTC time of acquisition;
• an uncertainty estimate for the mean LWL;
• metadata, including lake name in English, location, country, funder and citation instructions.

The data format is netCDF4 classic model⁵, adopting relevant CF conventions⁶.

The v3.1 CDR covers the period 1992 to 2020 under identical reprocessing, so there is no brokered/extended distinction in this case. The measurement sensors contributing to these time-series are radar altimeters onboard the following satellite platforms: TOPEX/Poseidon, Jason-1/2/3. Sentinel-3A, and Sentinel-3B.

User requirements

Based upon the precursor ESA Climate Change Initiative project addressing the Lake ECVs survey of user requirements for satellite-derived lake products, this section reports updates on the GCOS requirements which, however, are not yet publicly available yet. This section relies also on statements for the Lake ECV published literature, experience from other CDR projects, and requirements emerging from the definition of the service. The requirements are continuously updated using perspectives that emerge from users of the service and their feedback, and from any user requirements survey that is undertaken in the ESA CCI Lakes project.

The requirements involve different aspects of the product such as the definition, coverage, the spatial and temporal resolution, uncertainty, format and timeliness.

LSWT

Definitional requirements

Coverage

Spatial and temporal resolution

Uncertainty requirements

Communication of uncertainty

Data uncertainties

Format requirements

Timeliness requirements

LWL (V3.1)

Definitional requirements

Coverage

Spatial and temporal resolution

Data uncertainties

Format requirements

Timeliness requirements

Analysis of gaps and opportunities

Satellite observational constraints and opportunities

Lake surface water temperature (LSWT)

The LSWT observing system from space consists of ~1 km resolution infra-red imaging radiometers. In particular, the following sensors can be exploited for LSWT retrieval:

• Along-Track Scanning Radiometers, ATSRs (ATSR2 from 1995 to 2003, AATSR from 2002 to 2012): These were satellite-based sensor systems that had two-point brightness temperature calibration accuracy, mid-morning overpass time and low noise, delivering high LSWT sensitivity. They were dual-view sensors, but currently only the single view is used for LSWT retrieval because i) the spatial resolution of the forward view is lower and not useful except for lakes with widths exceeding ~10 km in both directions, and (ii) the current archives are not geolocated with respect to altitude differences, which is needed for lake processing. The ATSR2 and AATSR sensors (nadir view), which were more technologically advanced later additions to the ATSR family, have been used in the ESA CCI Lakes LSWT CDR. Due to some sensor problems and the eruption of Mt Pinatubo, further research and development (R&D) is needed to extend the CDR back using ATSR1.
• Sea and Land Surface Temperature Radiometers, SLSTRs (2016 to present day): similar to the AATSR but with a wider swath, and in-pair operational constellations planned for deployment through to ~2030, the SLSTR family offers a much-improved coverage compared to that which was available previously. The SLSTRs have now reached a stability of operation. SLSTR has been used in the ESA CCI Lakes LSWT CDR with data from 2016.
• Advanced Very High Resolution Radiometers, AVHRRs: these are satellite-based sensor systems that offer mid-morning single-view measurements and a larger swath with respect to the ATSRs. The global full resolution data (1 km, Full Resolution Area Coverage: FRAC) are capable of LSWT available from the MetOp-A and MetOp-B platforms (satellites). EUMETSAT will maintain MetOp AVHRR up to MetOp-C, and thereafter will provide MetImage⁹ . The MetOp-A and MetOp-B AVHRR have been used in the LSWT v4.5 CDR and, due to the large coverage of LSWT per lake, MetOp-A AVHRR is the reference sensor for the harmonised time series (for the time being: this may switch to SLSTR in future). MetOp-B AVHRR is additionally exploited for the C3S extension and MetOp-C will be added for the next C3S extension. Older AVHRR data are only available globally at reduced resolution, although a proposal to research the use of 1 km data collected over Europe to extend the series back in time regionally is under consideration by ESA.
• MODerate resolution Imaging Spectroradiometer, MODIS: the sensor on the Terra satellite is used for the CDR v4.0. MODIS is in a stable sun-synchronous orbit with local Equator crossing time of 10:30 am, which is relatively consistent with the ATSRs and MetOp AVHRRs. It has a viewing swath width of 2,330 km and views the entire surface of the Earth every one to two days. Its detectors measure 36 spectral bands between 0.405 and 14.385 µm, and it acquires data at three spatial resolutions – 250 m, 500 m, and 1,000 m, with the channels relevant to our algorithms being those with 1,000 m resolution. MODIS has been used in the ESA CCI Lakes LSWT CDR with data from 2000 until 2020.
• Visible Infrared Imaging Radiometer Suite, VIIRS: this sensor system on board the Suomi NPP satellite extends, and improves upon, the AVHRR and the MODIS instruments. It is single view, offers a non-traditional LSWT band that could reduce impacts of aerosols on retrievals, has 750 m nadir resolution that would enable better observation of small (few km) lakes, and has a day-night band that would facilitate use of night time data (particularly the cloud detection step). The opportunity afforded by VIIRS for LSWT is significant, but research is needed for exploitation and none is presently planned or proposed. The level 1 data access is a practical challenge, and also expensive.

In summary, with R&D, there are opportunities that would extend the LSWT CDR to earlier times (1991 globally, mid 1980s for Europe), with characteristics similar to the current resolution and quality. In the current extensions of the record, uncertainty decreases and coverage increases as AVHHR MetOp-C are brought into the service. To capture more small lakes, a better resolution instrument is required, and VIIRS is a possibility here, although presently no mechanism for the necessary R&D and practical measures can be identified to make the progress needed to take advantage of this opportunity. Against the targets, the gap analysis is as summarised, therefore, in Table 1.

Table 1: LSWT Gap Analysis Summary.

Lake water level

Table 2: LWL Gap Analysis Summary.

Improvement of retrieval algorithms

Lake surface water temperature

LSWT estimation has three steps:

1. pixel classification to ensure the viewed pixel is suitable for LSWT retrieval
2. LSWT retrieval
3. gridding across sensors to make the multi-sensor L3S product

The priorities for improvement in each area are described in the following:

Classification: (1) The day-time classification of a pixel is based on a combination of threshold tests on the visible (VIS), near-IR (NIR), and short-wave-IR (SWIR) channels. Since lakes present different optical properties, there are failures to detect water in certain cases, such as in situations where lakes are turbid, or shallow and salty. Lake-specific thresholds may improve this, although it is a considerable R&D task to achieve this for ~2000 lakes. (2) The day-time water detection is not applicable at night-time. Moreover, it is not applicable for the ATSR1 sensor since the VIS channels are not available. To include night-time LSWT observations requires thermal-only water/cloud/fog/ice discrimination, which could almost double the density of observations, and reduce uncertainties in gridded daily products. Bayesian methods used for SST have been used for lake observations from ATSRs, and this should be considered for future versions.

LSWT retrieval: The optimal estimation (OE) retrieval algorithm will continue to be the retrieval of choice for LSWT, because it is context specific. The main improvement to come will be to the source of prior information used in the radiative transfer model needed for OE, namely the switch to ERA-5. This is done only for MODIS for the current versions. The LSWT records from different sensors are adjusted using overlap periods to be unbiased in the lake mean compared to AVHRR MetOp-A. The better calibrated SLSTRs may be considered as a reference for the future (e.g., for LSWT v5.0).

L3 gridding: Current gridding is at a 0.05^o resolution. Adaptation to a 0.025^o gridding should be possible and useful, if there is genuine user demand. This may be addressed as a future development of the service after the priority tasks of bringing additional sensors into the data stream are successfully completed.
The context in which R&D can be pursued to underpin some service developments is, for LSWT, a future ESA Lake CCI project which may start by the end of 2022. The R&D elements for LSWT are limited by resources to a few weeks' effort on each of the following:

1. Preconditioning of water detection using a more dynamic water bodies mask – this is mainly retrospective;
2. Explore potential of context-sensitivity water detection thresholds;
3. Use of LSWT v4.5 to inform Bayesian cloud detection for night-time data;
4. Revisit method of harmonisation across sensors;
5. Improve the validation of the LSWT uncertainty estimates;
6. Explore retrieval benefits and limitations (mainly spatial resolution) of using dual-view from SLSTR for LSWT.

All R&D progress in the ESA Lake CCI will ultimately enter the C3S service, via the CCI-generated, and then brokered to C3S, dataset, and validated transition of the updated research code to generate future annual C3S time series extensions.

Lake water level

The current state-of-the-art R&D leading to the V3.1 CDR relies partly on a manual approach to estimate the geographic extraction zone of altimetry measurements. An automated version of this R&D has been implemented in the frame of the present project to ramp-up the products and be able to provide water level for a wider network of lakes. This has enabled a threefold increase in the number of lakes monitored between the first (V1.0) and the latest (v3.1) version of the C3S dataset. New lakes will be available in future versions. The method relies on a database of lake delineations and a land/water mask (from Global Surface Water Explorer, Pekel et al. 2016), intersected with the theoretical ground-track of the satellites and the lakes polygons defined in the CCI Lakes project.

Then, the extracted data must be corrected for various propagations (e.g. corrections for ionosphere, wet troposphere, and dry troposphere amongst others) and geophysical corrections (e.g. geoid, pole tide, solid earth tide amongst others) based on models and with limitations. The geoid model, in particular, does not include small wavelengths of the geoid, and this must be estimated based on altimetry data and a posteriori corrected. The algorithm has been improved to cover both simple (cf Figure 1, left panel) and complex (cf Figure 1, right panel) cases.

Figure 1: Automatic extraction of altimetry measures over specific lakes. The left panel shows a simple case of automatic intersection between satellite ground tracks and the polygon defining a lake. The right panel shows a more complex case including some land zones in the target lake that need to be excluded in the processing.

These two implementations are performed to improve the number of lakes monitored in the LWL product (see Section 3.1.2). Additionally, other R&D algorithms should be developed within the CCI-Lakes project and then be implemented for operational use to improve the quality of the product.

Improvement of uncertainty estimation

Lake Surface Water Temperature

L3C uncertainty: A comprehensive approach to estimate the LSWT uncertainty in L3 has been developed within the CCI SST work and it comprises the following components:

1. Propagation of instrument noise (uncorrelated effect): Values for independent random noise in the input satellite sensor brightness temperatures (BTs) are propagated through the retrieval at the level of full resolution. Arising from independent random errors, the resulting uncertainty in the L3 cell average LSWT is straightforward to calculate, and depends on i) the number of pixels in the cell average as well as (ii) the noise in each. For L3 cells at higher resolution this component of the uncertainty needs to be re-evaluated.
2. Retrieval uncertainty (locally correlated effect): The inverse solution is always an LSWT selected from a distribution of potential solution LSWTs, all of which could be compatible with the observed BTs and background information to within their uncertainties. The retrieval uncertainty is therefore the dispersion of those potential solution LSWTs. In the optimal estimation, this dispersion is evaluated using standard equations for estimating a posteriori error covariance. This error is highly correlated across pixels at the scale of 0.05°, and thus the uncertainty in this component does not reduce when forming a cell average.
3. Sampling uncertainty (uncorrelated effect). Generally, the LSWT in the cell is not fully observed in space because of partial cloud cover. The available information is the standard deviation of the LSWT in the pixels that are observed, and the fraction of possible pixels that were in reality clear-sky and retrieved. These two parameters have been shown to be able to give a good estimate of the uncertainty arising from subsampling the cell, and this source of uncertainty is included in the L3 uncertainty estimate.

The different uncertainties are then aggregated to generate total uncertainty, which is provided in the products. The uncertainty can be validated, and the various components can be further refined (parameters better estimated and better validated) over time. Furthermore, understanding of the spatial and temporal scales of the error correlations over lakes can be improved. Potentially, alternative methods of representing the uncertainties (i.e. ensembles) can be considered.

L3S uncertainty: The per-lake inter-satellite bias correction generates an uncertainty which is included in the estimation of the L3S LSWT uncertainty.

The uncertainty estimate for LSWT is mature, and the ongoing refinement should focus on determining appropriate parameters to use for additional sensor data streams, and updating such parameters for all sensors if reason to do so emerges.

Lake water level

The uncertainty variable distributed in the LWL product, associated to the Water Level variable, is currently estimated as the Median Absolute Deviation of selected water level measurements along-track (at level 2). The median value of the selected measurements at level 2 provides the median water level (level 3) . It estimates the precision of the measurements but not the accuracy part. The improvement of this uncertainty variable depends on achievements in the CCI Lakes project, but no strategy is currently foreseen to improve this variable.
The ongoing offline validation exercise will provide global statistics on the LWL product and a characterisation of the global uncertainty based on:

• Comparison to other altimetry products (e.g. Global Reservoirs and Lakes Monitor: Glili-REALM¹⁰)
• Comparison to in situ data (e.g. HYDROLARE¹¹)

Lake ECV components not presently in the service

The GCOS definition (RD.1) of the Lake ECV includes, in addition to the LSWT and LWL, the elements of lake surface reflectance, lake area and lake ice cover and thickness. A review of the opportunity to broker datasets addressing these gap areas is ongoing, and is not included in this report.

References

Pekel, J.F, Cottam A., Gorelick N. et al. High resolution mapping of global surface water and its long-term changes. Nature 540, 418-422 (2016).

Taburet et al. (2020) Lake and river water level, 300m, Version 2.1: Product User Manual. Copernicus Global Land Operations: Cryosphere and Water - CGLOPS-2 (JRC Framework Service Contract N° 199496). Available at:
https://land.copernicus.eu/global/sites/cgls.vito.be/files/products/CGLOPS2_PUM_LakeAndRiverWaterLevel-V2.1_I2.11_0.pdf (last accessed 10th December, 2022).

Related articles

Related articles appear here based on the labels you select. Click to edit the macro and add or change labels.