Session: Land with Synergy of Proba-V and Sentinel-3

The PROBA-V satellite, developed by the European Space Agency and funded by Belgium, was launched on the 7^Th of May 2013. It is carrying a multispectral Earth observing instrument on a micro satellite platform. PROBA-V’s Vegetation Instrument continuously provides radiometrically and geometrically calibrated products at 1 km and 300 m spatial resolution to the large science users community on a daily basis. As such PROBA-V continues the time series of SPOT-VEGETATION (1998-2014) capturing the reflected sunlight in the same four spectral bands. Additionally, it bridges the gap between SPOT-VEGETATION and Sentinel-3. Hereto the 300 m products have been designed. The PROBA-V 1 km and 300 m products are being used in the Copernicus Global Land service to provide biophysical variables.

However, due to the design of the Vegetation Instrument, consisting of three separate camera’s, the central camera has an original spatial resolution of approximately 100 m in the three VNIR bands and 200 m in the SWIR band. Thanks to the outstanding geometric accuracy (the daily absolute location error is stable and stays below 100 m), PROBA-V is also able to deliver 100 m products. As they cover approximately the central 520 km of the entire swath, it takes five consecutive days to cover the entire globe. As such, users consider PROBA-V 100 m as a desired niche product between daily MODIS/Sentinel-3 and 16-day Landsat observations. Based on a user survey it was decided to produce 100 m products on an operational basis as from April 2015. Daily composites (S1) and 5-daily composites (S5) will be provided, both as Top-of-Atmosphere and Top-of-Canopy. This improved spatial resolution is a clear added value for applications like agriculture, deforestation and water body monitoring. The presentation will highlight the specifications of the new PROBA-V 100 m products and show the added value for several applications.

The SPOT-VEGETATION (VGT) sensor series ended its operation in May 2014 after a successful operation time of 16 years. The SYN-VGT products of Sentinel-3 were defined to elongate the time series of VGT. The SYN-VGT products are a spectral resampling of the OLCI and SLSTR bands toward the spectral bands of VGT. To fill the gap between VGT and Sentinel-3, data of the PROBA-V are currently used. PROBA-V is the direct successor mission to SPOT-VEGETATION, which provided near-daily global coverage of vegetation at 1km resolution for the past 16 years. As one of its main objectives, PROBA-V provides continuity of the SPOT-VEGETATION time series. To achieve this, its mission definition fulfills the same near-daily global coverage and provides products at 1km and 300m resolution in the same spectral bands (Blue, Red, NIR, SWIR). Although the Proba-V sensor was spectrally defined as similar as possible to SPOT-VEGETATION, there will nevertheless be differences to cope with. These relate to the differences in camera system, geometry, but also spectral characteristics. The objective of this study is to analyse the similarity and differences between the data of VGT and PROBA-V and attribute the differences to their causes. This study helps to anticipate on the transition from PROBA-V to the SYN-VGT products of Sentinel-3.

The impact of the difference in spectral response function (SRF) of the two missions (VGT and PROBA-V) was evaluated on a representative set of global land cover conditions, derived from standard spectral libraries and physically-based radiative transfer and atmosphere models (PROSPECT, SAIL and MODTRAN-4). Using the spectral responses for both missions, surface reflectances and NDVI are computed. Through linear regression analysis, spectral correction functions are estimated, which must be applied to PROBA-V datasets to obtain data similar to SPOT-VEGETATION. To evaluate the approach, the correction functions have been applied to PROBA-V ten-daily 1km composites and compared to corresponding SPOT-VEGETATION composites.

An evaluation on a limited period from 01/01/2014 to 01/03/2014 suggests that the spectral correction functions for the 4 spectral bands perform well. However, an additional offset factor must be introduced for the NDVI correction function, which can’t be explained from the differences in spectral response.

A final evaluation has been performed on the full period from 21/10/2013 to 21/05/2014, which confirms that applying the additional offset in the NDVI correction function reduces most of the systematic bias between PROBA-V and SPOT-VEGETATION surface NDVI. It further shows that the corrections improve slightly the consistency for the NIR, SWIR and NDVI datasets, but degrades it slightly for the Blue and Red datasets. As this can’t be explained by spectral response differences, other influencing factors are investigated and discussed, such as the difference in overpass time, calibration uncertainties and inconsistencies in the SPOT-VEGETATION products.

Apart from the spectral response functions, the impact of a different viewing angle selection in the compositing step led to large inconsistencies in the NDVI data set for large areas around the globe.

For the transition to the SYN-VGT products of Sentinel-3, the same methods to investigate the similarity and the differences between PROBA-V will be used. The evaluation is particularly important for the continuation of a number of data sets in the Copernicus Global Land Service (NDVI, VCI, VPI). 

The Copernicus programme is a European initiative for the implementation of information services dealing with environment and security, based on observation data received from Earth Observation (EO) satellites and ground based information. Within this context, ESA is responsible in particular, for the implementation of the Copernicus Space Component (CSC), consisting of Contributing Missions on the one hand on of dedicated Sentinel missions on the other hand, feeding the Copernicus services with operational EO data. The Sentinel-2 optical high-resolution imaging mission will be devoted to the operational monitoring of land and coastal areas.

To maximize the products suitability and readiness to downstream usage, the Sentinel-2 ground segment will systematically generate, archive and distribute Level-1C products, which will provide Top-of-Atmosphere (TOA) reflectance images (orthorectified using a global Digital Elevation Model (DEM) and projected on Universal Transverse Mercator (UTM)). A Level-1B product will also be available for expert users, providing radiance images in sensor geometry together with an appended geometric model.  Furthermore a Level-2A processor currently under development will allow to generate atmospherically corrected products.

The generation and access to the Sentinel-2 products will be guaranteed, within the overall CSC Coordinated Data Access system operated for Copernicus, by the Sentinel-2 Payload Data Ground Segment (PDGS). Within the Sentinel-2 Ground Segment, the PDGS is the system devoted to receiving the raw measurement data from the Sentinel-2 satellites, and transforming it into the final high-level products down to their archiving and availability for download by the users. The PDGS is also in charge of planning the observations for the Sentinel-2 constellation via the Sentinel-2 Flight Operation Segment (FOS) and to monitor the payload and overall mission and products performance along time.

Data from the Sentinel-2 mission, as any other Copernicus Sentinel EO data, will become available on a free and open basis to any user worldwide.

The presentation will provide a mission description, provide an update on the status of the mission, address the Copernicus user typologies and their Data access principles, as well as the high level operations plan and the ramp-up phase for Sentinel-2.

The presentation will furthermore focus on the complementarity of the Sentinel-2 to the Sentinel-3 mission.

Conditions in the atmosphere influence satellite measurements of the Earth's surface. The retrieval of accurate and physically-based surface properties is particularly important in vegetation monitoring and is still a remaining challenge. Various uncertainties may enter into the vegetation and other higher level products development from remotely-sensed data and its interpretation due to atmospheric and topographic impact. Atmospherically and data processing biased remote sensing data therefore needs a proper validation strategy in order to facilitate temporally comparable retrieval of surface properties from satellite measurements.

Background
Atmospheric effects are wavelength dependent, both additive and multiplicative in nature and include scattering, absorption, and refraction of light, as well as adjacency effects. Different methods were developed to remove the scattering and absorption component caused by path radiance and/or deal with other components, ranging from simple DOS calibration to complex BRDF retrievals. Atmospheric correction is important part of the pre-processing of satellite data, however atmospheric effects complexity in nature brings several challenges to atmospheric correction algorithms development and consequently their performance. Examination of the influence of atmospheric effects is thus not straightforward and needs to be systematically approached. There are basically three ways to approach: (a) by comparing the satellite reflectance values and the in-situ reflectance values acquired during the satellite overpass for a target area, (b) comparison with other sensor products and (c) by step-by-step examination of atmospheric correction algorithm effects, based on image properties changes (before and after correction comparison). In this paper we focus on the later approach, aiming to systematically investigate the performance and identify most important occurrences of correction effects in different settings.
 
The assessment was tested on twenty PROBA-V 100 m images with low cloud cover acquired from March to November 2014, covering the area of Slovenia and bordering countries, thus representing a variety of background and atmospheric conditions. The PROBA-V (Project for On-Board Autonomy – Vegetation) satellite mission is intended to ensure the continuation of the SPOT 5 VEGETATION products and is also serving as a demonstration and preparation for the European Space Agency Sentinel-3 satellite. The PROBA-V satellite is designed to provide a global coverage of land surfaces at a spatial resolution of 1/3 and 1 km with a daily revisit. Beside that central camera acquires data in 100 m resolution. Herein we evaluate the atmospheric impact on higher level products generation from PROBA-V 100 m data that is otherwise part of our working activities in ESA PECS project “Development of Proba-V 100 m Vegetation Products”.

Three topics will be addressed:
1. Comparison of two atmospheric correction approaches: SMAC and ATCOR,
2. Developing atmospheric correction validation strategy and
3. Assessment of the effects of atmospheric correction on products generation.

Methods
Application and comparison of two atmospheric correction approaches
Atmospheric correction transforms the TOA (Top of the Atmosphere) reflectance into TOC (Top of the Canopy) reflectance. To facilitate the retrieval of surface reflectance from PROBA-V 100 m data, we have explored the feasibility of two atmospheric correction approaches in order to compare their performance. The first algorithm is the Simplified Model for Atmospheric Correction (SMAC, Rahman and Dedieu 1994). This algorithm was selected as general PROBA-V data processing routine to ensure data continuity with the SPOT-VGT mission and thus represent a standard routine in PROBA-V data processing and products generation. The second algorithm used is  ATmospheric CORrection (ATCOR, Richter 1996). This algorithm is found to be suitable for satellite data of a variable range of spectral and spatial resolution.

In order to obtain TOC reflectance, both algorithms need information about the state of the atmosphere at the time of image acquisition. Whereas SMAC uses auxiliary information obtained from meteorological models and services, ATCOR tries to obtain them from remotely-sensed data. We found out that both algorithms can have problems with retrieving TOC reflectance in darker regions. Regions of cloud shadows and coniferous forests represent a problem for SMAC algorithm, and deep topographic shadows for ATCOR. Especially blue and red band reflectance values in those regions might be rendered to zero. Both routines were evaluated and essential results will be presented.

Developing atmospheric correction process validation strategy
Additionally, investigations were conducted to examine the performance of atmospheric correction with regard to what and how it should recover from atmospherically affected datasets. Statistical analyses were performed to embrace multi-sided validation settings:
- Analysis of relative gain, average reflectance and STD values per wavelength, land cover type, terrain and season characteristics between the uncorrected and corrected images,
- (In)consistency of the correction over the same land cover type (e.g. water, forests).

Assessment of the effect of atmospheric correction on products generation
Atmospheric correction is especially important in cases where multi-temporal images are to be compared and analysed. Not considering the effects of the atmosphere when calculating vegetation indices, may lead to discrepancies and misleading interpretation in the final outcome. We evaluate the influences of topography, haze and shadows on vegetation indices before and after applying atmospheric correction. Analysis in mountainous areas showed that atmospheric correction efficiently reduces the topography effect and helps to restore correct/expected NDVI values in shadowed regions.

Results and summary
The results reveal that atmospheric effects are variable and significant, and in several occasions might be reasonably measurable. Its possible validation approached from a systematic assessment with several identified relations is thus important for general remote sensing data analysis.

The main purpose of the study was to explore the atmospheric processing impact on the derivation of surface reflectance and higher level vegetation products from PROBA-V 100 m data over complex topography and land cover. A selection of best-performing validation metrics will be presented and discussed. An insight into characteristics of retrieved TOC reflectance in different wavelengths and vegetation products over specific land cover classes can be beneficial both for the improvement of atmospheric correction algorithms and comprehension of bio-physical products quality.