Session: Methodology II

The verification of Sentinel-3 data calibration consistency with respect to previous missions is critical. In that context, Pseudo Invariant Calibration Sites (PICS) play an important role because there are suitable for sensor stability monitoring. The Committee on Earth Observation Satellites (CEOS) has identified 6 sites for their good spatial and temporal stability.  Among the 6 desert CEOS PICS, Libya-4 has proven to be the most stable though it has the most complicated topography. This site presents other decisive advantages: its large spatial extension covering an area of about 1x1 degree, the absence of vegetation and the existence of numerous data sets collected over Libya-4 as it has already been intensively used as PICS.

 

Libya-4 site is composed of long sand dune ridges that might impact surface Bidirectional Reflectance Factor (BRF) as a function of the sun azimuth angle. So far, only 1D Radiative Transfer Models (RTMs) have been used to simulate satellite signals over Libya-4. In order to further reduces modelling uncertainties, all possible source of errors needs to be analysed in details, especially concerning surface reflectance as it plays a dominant role above 600nm on the top-of-atmosphere BRF. This study addresses the impact on surface BRF of (i) the size of the selected region over Libya-4 and (ii) the effects of sand dune ridge alignment. Specifically, this work analyses surface BRF azimuthal dependencies due to sand dune organization for different Region-of-Interests size using a 3D Monte Carlo ray-tracing RTM. Such analysis is relevant when observations acquired by mid-morning and mid-afternoon sun-synchronous polar orbiting satellites are compared.

 

The topography is characterized with the 30m resolution ASTER Digital Elevation Model. Four different region-of-interest size, ranging from 10km up to 100km, are studied. Results show that sand dunes generate more backscattering than forward scattering at the surface. The mean surface reflectance averaged over different viewing and illumination angles is pretty much independent from the size of the selected area thought the standard deviation differs. Sun azimuth position has an effect on surface reflectance field that is more pronounced for high sun zenith angles. Such 3D azimuthal effects should be taken into account to decrease the simulated radiance uncertainty over Libya-4 below 3% for wavelengths larger than 600nm. 

ATSR-1 onboard ERS-1, ATSR-2 onboard ERS-2 and AATSR onboard ENVISAT are radiometers built specifically with the requirement for climate observations in mind, i.e. high accuracy and stability, in order to provide high quality Sea Surface Temperature (SST) observations from space. SST is an essential climate variable (ECV) with observations available since 1850, which makes it a very important parameter for climate monitoring, in addition to its importance as a boundary condition for climate and numerical weather prediction (NWP) models. The failure of ENVISAT in April 2012 brought a premature end to the long time series of SST observations made from the ATSRs since August 1991. Given that the next satellite instrument with similar characteristics and thus similar expected quality of observations, SLSTR, is planned for launch towards the end of 2015 on board Sentinel-3, there is a need to find a candidate satellite that can bridge the gap in SST measurements from space between AATSR and SLSTR. Due to restrictions relating to the orbit of ENVISAT and Sentinel-3 only MODIS on TERRA and AVHRR on MetOp-A are considered to potentially provide climate quality observations for a period that will overlap significantly with SLSTR. Here, the use of AVHRR is explored due to its collocation on the same platform with IASI, which is considered the reference instrument for calibration studies in the infrared spectrum. Previous studies have examined the calibration accuracy of the two longwave AVHRR channels using IASI, and so this study focuses mainly on the possible temporal trends, including the SWIR channel at 3.7 μm. The differences between single IASI channels and AVHRR channels show temperature and angle observation dependences, in accordance with previous studies and some of the channel pairs exhibit a nonlinear character. There is a spatial correlation of the IASI-AVHRR differences with the water vapour geographical distribution, which is also related itself to temperature. Results suggest that there is a day-night difference, which is not related to the temperature dependence and not examined by previously. The high noise of the shortwave IASI channels makes the comparison at low temperatures (i.e. over the Polar Regions) problematic. There are significant temporal trends in the differences between IASI and AVHRR channels being less or equal to 10 mK/yr. Given the fact that the required stability for climate monitoring is 0.01 K/decade and assuming that the trends are not related to IASI, this makes the use of AVHHR questionable regarding its quality to fill the gap between AATSR and SLSTR without a correction to the radiances applied using IASI.

The atmospheric correction of Sentinel-3 data over inland waters is a challenging issue. Typical ocean colour atmospheric correction algorithms are not appropriate for the correction of  inland waters as 1) they neglect the (often) non-zero  altitude of inland waters, 2) they assume a spatial homogenous background neglecting adjacency effects or 3) they make inadequate assumptions on the water leaving reflectance in NIR in order to retrieve aerosol information. On the other hand,  land atmospheric  corrections consider a Lambertian surface and therefore  don’t take into account  the contribution of the specular reflection at the air water interface.

In this paper we present a scene and sensor generic atmospheric correction scheme  OPERA  allowing to correct both land and water areas in the remote sensing image. The atmospheric correction scheme accounts for surface elevation variation, adjacency effects and for inland water targets also the non-Lambertian reflection of water surfaces. Through the use of a single atmospheric correction implementation discontinuities in the resulting reflectance between land and the highly dynamic water areas (such as turbid, tidal, shallow waters or macrophyte dominated waters) are reduced.

The atmospheric correction parameters are derived from pre-calculated MODTRAN-5 Look-Up-Tables (LUTs) in function of among others sun and view geometry, aerosol optical depth, ozone, water vapour and elevation. The latter allows for an accurate Rayleigh correction over higher altitude areas.

The atmospheric correction requires information on the aerosol and atmosphere. The aerosol optical thickness (AOT) is derived from land through a TOA radiance inversion of selected end-members in the scene. Over water the AOT is retrieved through spatial extension of the derived values over neighboring land pixels assuming local spatial invariability of the aerosol. Water vapour can optionally be derived on a pixel-by-pixel basis for sensors having at least one spectral band situated within the water vapour absorption feature, and reference bands across the feature.

For  inland water areas a fraction of the light reflected by the land can reach the sensor. As the contrast in reflectance between the dark waters and bright land is highest in the Near Infrared , the adjacency effect  is most visible in these NIR wavelengths. Over water the SIMEC (Sterckx et al., 2014) (SIMilarity Environment Correction) approach is used to calculate the contributing background or environment radiance assuming an invariant shape of the water leaving reflectance in the Near-Infrared.

Finally, over water surface an extra correction for the  reflected skylight is made to retrieve the water leaving reflectance.

Within the FP-7 SPACE INFORM project this atmospheric correction scheme is being evaluated for different current  (MERIS, Landsat-8) and future spaceborne sensors (Sentinel-3, Sentinel-2,  EnMap, PRISMA). 

The overarching objective of this research is to strengthen technical cooperation between the US and ESA programs such that the products forthcoming from a similar class of missions are as compatible as possible, in support of their combined use for science. In particular, we are focusing on the harmonization of the surface reflectance products from MODIS, VIIRS, and Sentinel 3. Surface reflectance is an important product as it provides the basis for the generation of some higher order land products.

Some of the objectives of the proposed research are to ensure:  (a) careful cross calibration between the MODIS-VIIRS suite and Sentinel 3/OLCI-SLSTR instrument   (b) consistency in methods for atmospheric correction to ensure the necessary level of quality in the surface reflectance Fundamental Climate Data Record, (c) effective cloud and shadow screening for the same reasons as above.

In terms of calibration, a variety of algorithms that have been developed for MODIS and VIIRS cross-calibration, that do not require absolute coincidence in time or in orbital track will be presented as well as their potential application to Sentinel-3. This method is applied on a routine basis to monitor the long-term performance of the NOAA VIIRS product as well as the quality of the NASA VIIRS Climate Data Record, with an accuracy of ~0.5%.

For atmospheric correction, we have developed several methods for early evaluation of surface reflectance that also will be of benefit to Sentinel-3.  Two methods have been developed (a) the   comparison   to   the   near   coincident VIIRS/MODIS reflectance   product, (b) the independent assessment at the AERONET sites,. We will show the results of that analysis and how these methods can be applied to Sentinel 3 data. Also an extensive amount of work has been done in the development of cloud/cloud shadow algorithms for the MODIS data that can be applied to Sentinel-3.  The algorithms will be presented as well as the validation results.

Suggestions on ways to increase technical cooperation will be presented.

The OLCI instrument on-board Sentinel-3 is in the continuity of the ENVISAT/MERIS instrument capability, with numerous improvements. The instrument is designed to incorporate extremely sensitive and stable radiometry, dedicated on-board calibration and a large number of spectral channels.

The OLCI radiometric calibration, as for MERIS, is based on in-flight measurements of sun-lit diffusers. Calibration relies on the diffusers on-ground characterisation, acting as on-board secondary reflectance standards: known diffuser BRDF and viewing/illumination geometry during acquisitions allow predicting spectral radiance at instrument entrance and, together with measured numerical counts appropriately corrected from instrumental effects, deriving calibration coefficients.

Radiometric performance is then validated by indirect methods using observation of natural targets.

Radiometric Calibration and performance monitoring

Radiometric calibration data processing and performance monitoring are mostly based on methodologies derived from MERIS.

Dark offset analysis and monitoring : with respect to spatial and spectral dimensions regularly over the mission, and with respect to time at orbital scale during commissioning phase, over orbits acquired with the calibration wheel in shutter position, and over the mission for long-term temporal trends from Calibration acquisitions.

Radiometric diffuser stability monitoring and ageing modelling: the reference radiometric diffuser being the secondary standard used for absolute radiometric calibration, its reflectivity must be monitored along the mission so that ground characterisation can be appropriately adjusted during instrument gain computations. This can be achieved through regular comparisons with the ageing monitoring diffuser, exposed about 10 times less frequently.

Sensor degradation monitoring and modelling: OLCI will acquire radiometric calibration data about every 2 weeks over the whole mission (routine frequency will be adjusted during Commissioning), allowing deriving instantaneous gain coefficients for each acquisition. Each set of gain coefficients is representative of the instrument at the time of acquisition and the series can be used to monitor the instrument sensitivity degradation. It can also be efficiently used to derive a long-term trend model, sufficiently robust to be used in extrapolation allowing to significantly lower the refreshing rate of the radiometric calibration auxiliary data as well as to smooth radiometric performance over time by using a continuous model instead of discrete changes. The degradation model also allows to correct for limitations of the diffuser BRDF model and to reduce the impact of the diffuser material inherent noise (speckle) by time averaging.

Spectral stability monitoring: the results of in-flight spectral calibration from the Erbium doped diffuser data – to be acquired every 3 to 6 months at 3 wavelengths – will be used to monitor the spectral stability of the instrument on the long-term. Erbium-doped diffuser spectral calibration data will also contribute to absolute spectral calibration of OLCI within a dedicated activity, not described here.

Some “single-shot” tools have also been developed to estimate Signal to Noise ratio:

Signal to Noise ratio estimation from short scale time variations of radiometric diffuser signal. This can be used routinely for OLCI as radiometric calibration data includes access to individual acquisitions at instrument nominal time sampling rate. The tool includes frequency filtering to eliminate short term BRDF variations and diffuser speckle so that the evaluated SNR accounts only for radiometric noise.

Signal to Noise ratio estimation from local spatial variation of homogeneous areas. Following the methodology of Hu et al. (“Dynamic range and sensitivity requirements of satellite ocean color sensors: learning from the past”, 2012) we assess signal-to-noise ratio (SNR) of ocean color sensors directly from acquired images. This methodology is based on the assumption that, at typical TOA radiances (Ltypical), a collection of neighboring pixels showing the smallest standard deviation are only deviating by instrumental noise. Neighboring pixels are found within macropixels of nxn adjacent pixels (n=3). A drastic selection of the macropixels showing the smallest standard deviation is made to eliminate those potentially impacted by geophysical noise. For each macropixel SNR is simply derived, per waveband, by Ltypical/σ with σ the standard deviation. Finally, the global SNR is obtained by the mean SNR of the selected macropixels. The training of this methodology is done on MERIS data covering the oligotrophic waters of the South Indian Ocean and South Pacific Gyre sites. Its results agree with SNR figures from other sources and its application is prepared for the estimation of the OLCI SNR during calibration/validation (CAL/VAL) activities

Radiometric Validation

In addition to S3 instrumental performance activities, analyses will be carried out to validate the OLCI radiometry using data acquired over natural sites. The verification of the in-flight calibration will be performed through different indirect approaches such as absolute calibration over Rayleigh scattering, cross-calibration and trending over desert and snowy sites, inter-band calibration over sunglint. These methods developed by ESA or CNES have been implemented in tools such as DIMITRI or SADE/MUSCLE. The verification of the OLCI calibration is based on the use of these tools which will ingest data and/or statistics (mean radiometry) extracted and calculated from L1 products acquired over natural areas.

The S3ETRAC toolbox, developed and operated by ACRI-ST, allows the extraction, pre-processing, filtering and statistics computation of OLCI data over a list of pre-defined regions of interest. The first version of S3ETRAC will perform data extraction and statistics calculation over deserts, snowy sites and oceanic sites (Rayleigh and sunglint areas). These sites have been selected for their appropriate optical properties to validate the radiometry of optical sensors: e.g. spatial and temporal radiometric stability for desert and snowy sites, maximization of the Rayleigh scattering over clear oceanic surface and under clear atmospheric conditions, etc. In a first step, S3ETRAC performs data pre-processing (normalization to reflectance, data quality verification, …) and filtering using a series of criteria minimizing disturbing effects such as clouds, bad atmospheric and meteo conditions (high aerosols optical thickness, high wind speed, …). These criteria are defined per type of target. Statistics are then estimated over the region of interest (desert, snow site) or per macro-pixel (over oceanic surface) and delivered to scientific experts for radiometric validation purposes.

Note that the functionalities of the S3ETRAC tool have also been extended to calculate SLSTR statistics in order to validate the radiometry of the reflective bands.

ODESA and MERMAID have proven to be efficient and practical tools for the implementation and quantitative validation against in situ measurements of new algorithms or simply for the improvement of existing ones. Both have been extensively used for MERIS third reprocessing as well as for the on-going MERIS fourth reprocessing. Both tools are currently being updated to support OLCI data processing and validation. Future ODESA releases will indeed provide OLCI processing source codes in addition to MERIS processing source codes already available. MERMAID will evolve to support OLCI data validation and will have face the challenge of providing rapidly OLCI to in situ matchups in support of the scientific community involved in the commissioning phase.

 

A fundamental evolution of ODESA and MERMAID on a user point of view is that they are now installed and delivered to users in the collaborative working environment of the Cal/Val Topic Centre (CTCP) operated by the Sentinel 3 MPC (Mission Performance Centre) team. Users will still be able to install ODESA and use MERMAID validation platform on their own computer but from a Sentinel-3 user point of view, the advantages of the CTCP are multiple:

Fast access to satellite data

Prompt an easier technical support

Regular data backup

Access to CTCP from anywhere providing an internet connection

No need of dedicated disk space, computing power or high speed internet connection at user premises

Easy data download to user machine through dedicated ftp access

 

The ODESA and MERMAID synergy has demonstrated its efficiency to the community. Their implementation as built in software in the CTCP will facilitate OLCI data analysis. The concept is now mature and ready to be extended to other fields if there is a need from the user community.