Spot5Take5: CNES-ESA simulate Sentinel-2 1

In 2016, when Sentinel-2 constellation is complete, and for at least seven years, users will have access to high resolution time series of images acquired every 5 days, anywhere among the Earth land surfaces. This new data set will drastically change and enhance the way land surfaces are monitored using remote sensing. Sentinel-2 frequent revisit will ensure that a given surface will be observed at least once a month, except in the most cloudy periods and regions, enabling to develop operational applications. To help users get ready for this new type of data, simulation data sets were necessary.

This paper draws the lessons learned from the SPOT (Take 5) experiments, which were held in 2013 with SPOT4 and in 2015 with SPOT5. These experiments, operated by the Centre National d'Etudes Spatiales (CNES) with a large contribution of rhe European Space Agency (ESA) for the second one, aimed at providing time series of optical images simulating the repetitivity, resolution and large area of Sentinel-2 images, in order to help users set up and test their applications and methods, before Sentinel-2 launch.

To do these experiments, the orbit of the SPOT satellites was lowered by two kilometers, just after the end of their nominal operation [1]. In 2013, 45 sites were repeatedly observed by SPOT4, while in 2015, 150 sites were observed by SPOT5. In both cases, the sites were observed every fifth day during five months with constant observation angles. The obtained time series may be used for very diverse applications, such as land cover and land use, agriculture, phenology, hydrology, snow monitoring, coast monitoring, habitats characterization and biodiversity...

The SPOT (Take5) data set can be freely downloaded by any user from the following web site : https://spot-take5.org/. The THEIA Land Data Center, at CNES, produced and delivered the image time series with two different levels of pre-processing :

Level 1C ortho-rectified images expressed in reflectance at the Top of Atmosphere. These products are obtained with CNES Sigma processor.

Level 2A ortho-rectified images expressed in surface reflectance after atmospheric correction, provided with a cloud mask. These products are obtained thanks to the CESBIO/CNES MACCS [2,3] prototype processor.

These data sets were also very useful to test the pre-processors at a large scale and with a large diversity of sites and conditions. This communication will draw the lessons learned in terms of :

Data usage and interest in the community : more than 750 users for SPOT4 (Take5), already 250 users for SPOT5 (Take5),

User preferred products (75% of downloads concern Level 2A) and user feedback

Level 2A methods and analysis of the robustness of Level 2A processing methods. Suitability of using constant parameters in the L2A processing : only the aerosol model needs to be location dependent

Application potential of Sentinel-2 imagery in different domains (agriculture, forests, glaciology, coasts and inland waters, vulcanology...)

[1] Hagolle, O., Sylvander, S., Huc, M., Claverie, M., Clesse, D., Dechoz, C., ... & Poulain, V. (2015). SPOT-4 (Take 5): Simulation of Sentinel-2 Time Series on 45 Large Sites. Remote Sensing, 7(9), 12242-12264.

[2] Hagolle, O., Huc, M., Villa Pascual, D., & Dedieu, G. (2015). A Multi-Temporal and Multi-Spectral Method to Estimate Aerosol Optical Thickness over Land, for the Atmospheric Correction of FormoSat-2, LandSat, VENμS and Sentinel-2 Images. Remote Sensing, 7(3), 2668-2691.

[3] Hagolle, O., Huc, M., Pascual, D. V., & Dedieu, G. (2010). A multi-temporal method for cloud detection, applied to FORMOSAT-2, VENµS, LANDSAT and SENTINEL-2 images. Remote Sensing of Environment, 114(8), 1747-1755.

The presentation will demonstrate the use of time series analyses in the land cover mapping project CadasterENV. CadasterENV is an initiative to create a national consensus for a multi-purpose land cover mapping system at a national scale.

Data from SPOT5 Take 5 and Sentinel-2 will be used for several purposes:  to increase the opportunity to acquire cloudless images, to classify images based on time series and to conduct change analyses both within one growing season and between years. During late summer 2015, Sentinel-2 data was acquired over the test sites and will be integrated in the proposed method development as well as the SPOT5 Take 5 images from April to September 2015.

The objective of CadasterENV Sweden is to implement a multi-scale and multi-purpose land cover mapping and monitoring system for large areas of interest in Sweden according to national user specifications. The system will be comprised of two components:

a land cover (LC) mapping component based on HR (High Resolution) and VHR (Very High Resolution) satellite data, and

a land cover change (LCC) alert component, based on HR satellite data.

The project has been done in close cooperation with the major relevant Swedish organizations, e.g. Swedish Environmental Protection Agency, Statistics Sweden and the Swedish Board of Agriculture. It is being funded by the Data User Element program of the European Space Agency (ESA).

Sweden has a shorter vegetation season than most other European countries. This delimitates the possibility to acquire cloud free images during the vegetation season and, even more so, the optimal date (from a phenological perspective) or comparable dates from two different years. If time series from the vegetation period is available, this should provide an improved opportunity to identify vegetation types by analyzing how phenology develops over the vegetation period. On the plus side, Sweden’s high northern latitude corresponds to a relatively high revisit frequency from Sentinel, resulting in a large amount of data that can be used for analysis.

In order to stratify the different types of changes, the appropriate change detection tool and parameters for the selected types of land cover (stratum) should be selected.  Different change detection methods are used in order to handle different spectral change characteristics of a variety of land cover types. These analyses are all based on multi-temporal HR  and data are used together with the LC base map. The ultimate goal is to use the results to update the LC base map in a time and cost efficient way.

The CadasterENV Sweden project constitutes the baseline platform for an upcoming nationwide LC mapping and monitoring in Sweden. The aim is that the LC mapping and LCC Alert system will be fully integrated into a national monitoring system to continuously update the LC/LU data in Sweden.

Worldwide about 40% of all glacier-area can be classified as tidewater glaciers, where the terminus ends in water instead of land. At this front, ice is detached and produces icebergs. What drives this calving process is poorly understood. This lack of knowledge is in strong contrast to the other component of mass loss, surface melt, which can be modelled successfully to a high degree. In order to get a better understanding of tidewater glacier dynamics, better knowledge of the velocity structure is needed because these two aspects are highly correlated.
In this study we look at the velocity evolution of Kronebreen, an fast-flowing Arctic tidewater glacier, where half of the annual mass turnover stems from calving. The majority of calving flux occurs in the summer season. In spring/summer 2015 the glacier was sensed with a 5 days repetition rate by the Spot5-Take5 campaign.
The surface of Kronebreen is highly crevassed because all ice from the bigger Holtedahlfonna icefield is squeezed through this outlet. Consequently, visual contrast is high, but, due to its fast flow, its appearance changes heavily. This appearance change is problematic for multi-temporal cloud detection. Hence, the image matching methodology needs to be able to cope with partly obstructed cloud cover. Nonetheless, the multitemporal element supports the estimation of such a flow field, giving an insight into its temporal variation.
This testbed demonstrates the benefits of the new generation satellites, which monitor the Earth at high repetition rate. The potential of flow-field monitoring by Sentinel-2 becomes evident for Kronebreen. The satellite specifications bring an automatic seasonal velocity assessment into reach, even on a large scale.

The SPOT5-Take 5 initiative gave the opportunity to exploit the Sentinel-2 data for agriculture application even for the 2105 growing season. Starting in April 2015 the SPOT5-Take 5 data collection covers the beginning of the growing season of Northern hemisphere while Sentinel-2 took over the data acquisition from August 2015 when SPOT5-Take 5 ended. As for the first Take 5 experiment, placing the SPOT5 satellite on a 5 days repeat cycle orbit for five months provided a unique time series mimicking the data provision of Sentinel-2 a and b. This SPOT5-Take 5 data collection is of high value for the Sentinel-2 Agriculture (Sen2-Agri) project supported by the European Space Agency (ESA) and allow demonstrating on 2015 season for a set of sites.

The Sen2-Agri project anticipated this Sentinel-2 data availability two years ago by developing agriculture monitoring methods designed to exploit Sentinel-2 capabilities. For 2015 in-situ data were acquired during the growing season by 8 Joint Experiment of Crop Assessment and Monitoring (JECAM) site teams involved in the Sen2Agri project.

While the SPOT4-Take 5 was used as input for the methods benchmarking study, the SPOT5-Take 5 is used to apply the selected and implemented algorithms in the Sen2Agri open source tool box. This allows generating four different products relevant for agriculture monitoring, i.e. (i) cloud-free surface reflectance composites, (ii) dynamic cropland mask to be delivered every month during the growing season, (iii) cultivated crop type map and area indicator for main crop groups and (iv) vegetation indicators describing the vegetative development of crops. The Sen2-Agri processing system performances are assessed over 8 sites belonging to the JECAM network: Belgium, Burkina Faso, China, France, Mali, Russia, South Africa and Ukraine. The French and Malian sites were tentatively processed in Near-Real Time (NRT) while the products are generated off line.

For some sites the SPOT5-Take 5 and the Sentinel-2 data are complemented by Landsat 8 images which are pre-processed up to Level 2A. In parallel, in-situ data gathered during the 2015 growing season serve as ground truth for the accuracy assessment. Support and close interactions with the JECAM site managers are a cornerstone of this experiment, especially for the NRT sites which highlighted the operational requirements for a continuous delivery throughout the season. This paper also discussed the lessons learnt from this first year of agriculture exploitation of Sentinel-2 data for agriculture monitoring in terms of data flows, methods performances and operational constraints.

In sub-saharan Africa, smallholder farming systems face high inter-annual climatic risk, demographic pressure and soil fertility challenges. Crop production fluctuates strongly from metric to landscape scales due to variable management, soils, rainfall, and resulting abiotic and biotic stresses. Improving farmer access to inorganic and organic fertilizer is a key mechanism to achieving agricultural intensification, but cannot follow a one-size-fits-all approach in such heterogeneous settings. Efficient mapping of fertility status is then a crucial pre-requisite to spatial and temporal targeting of future investments, to improve agricultural productivity, food security and decrease farmer vulnerability.

This research aims to demonstrate the use of remote sensing technologies to map variations in fertility levels from field to landscape scales using a case study in the Koutiala district (Mali). There, the STARS project (Spurring a Transformation in Agriculture through Remote Sensing) provides an on-farm spatial database of biweekly measurements of crop height, LAI, FCOVER, chlorophyll content and biomass collected in 50 farmer fields (maize, cotton, sorghum, millet and peanut). In each field, five different mineral fertilization treatments are applied in 15x15m plots. Two concurrent seasonal stacks of SPOT-5 Take5, DigitalGlobe and SenseFly eBee multispectral time series (resolutions: 10, 2, 0.1m) for 2014 and 2015 growing seasons provide spatial assessments of canopy responses to variable soil and fertility conditions. The multi-seasonal data set allows assessing some part of the inter-annual variability in the measured vigor of plant canopy.  This approach enables mitigating the impact of the year-to-year variation on the fertility mapping. Prior crop type mapping and landscape stratification are required to reduce systematic variability in canopy responses due to soil properties and crop type. The crop type map is produced by a classification of SPOT-5 Take5 2015 time series with a reference ground dataset of 400 fields. Landscape stratification is achieved through a knowledge-based approach based on fertility gradients, function of catena position and distance to homesteads. Results show that fertility treatments are correlated to NDVI at 2-m resolution, and may be used to make valid and cost-effective inferences about variations in soil fertility at a landscape scale and beyond thanks to SPOT-5 Take5 images. This in turn may support the development of custom farm advisory and inputs procurement services for smallholders, informed by public domain satellite systems such as the recently-launched Sentinel-2 mission.