Spot5Take5: CNES-ESA simulate Sentinel-2 3

The purpose of this paper is to present the potential and expected contributions of Sentinel-2 data, for detecting and monitoring wetlands and inland water bodies and related applications.

Looking forward to the next Sentinel-2 mission, CNES and CESBIO with ESA support have carried twice Take5 experiment, to simulate the repetitiveness and associated data flow of Sentinel-2 satellite using, at the end of their life, SPOT4 and SPOT5 as a simulator of the time series that ESA's Sentinel-2 mission will provide. Satellite orbits were modified in order to insure a five days revisit, from February to June 2013 for the first Take5 experiment, and from April to September 2015 for the second one based on SPOT5. It has to be noticed also that SPOT data have of course similarity with the Sentinel-2 ones, in term of spectral coverage, thanks to its three Visible and one SWIR bands, as well as in term of spatial resolution, thanks to the 10m ground sampling of the multispectral bands of SPOT5.

Within this project, a monitoring program was established over small inland water bodies and wetlands, i.e. winter flooded areas of Muttersholtz and Rohrmatten, in the Alsatian flood plain (France), and over complex monsoon lakes in China, Poyang and Baidang lakes. For Poyang Lake (Jiangxi Province), Take5 experiment completes the long term surveillance done since 2000 within the framework of the DRAGON program which combines altimetry, in situ measurements, MR, HR and VHR data for both SAR and optical systems.

Unfortunately, due to cloudy weather, unfavorable for optical acquisitions, in 2013 only 28% of Sentinel-2-like data were exploitable over the Chinese area and 43% over the French study site. In the second experiment, the results are quite similar, over the 25 potential acquisitions, about 30% for Poyang lake of exploitable images were acquired, 36% over Baidang lake, and less over the Alsatian Plain, with only 6 images fully exploitable and 5 partially, given between 44 and 25% of success. This gives an idea of the actual frequency of exploitable images that could be reached by the Sentinel-2 sensor at these periods of the year.

However, despite the relative weak amount of data,  the obtained results confirm the high interest of SWIR bands for water bodies’ extraction, wetland identification, even if some precautions have to be  taken to avoid expected confusion, i.e. sand/mud banks which could cause overestimation of water surfaces. Moreover, thanks to the HR² (High Resolution and High Revisit) of Sentinel-2-like data, it was possible to monitor relative small hydrological systems of 2 to 10 ha, highlighting the important variability (25% to 35% in 5 to 10 days up and down; 87.5% of variation in 3 months;) of water surfaces within these areas. In regards to the SAR data, such as Sentinel-1, the Sentinel-2 like imagery, ie SPOT5 Take Five time series upgraded with 3 Landsat 8 images, allowed to characterize the invasive floating vegetation (Trappa, Zizania sp) and to monitor its extent from May to September.

Within the test sites representing two extremes in tem of size, the exploitation of these Take5 data provided a good overview of the potential of the Sentinel-2 system in term of spectral information or spatial resolution, but also in term of times-series analysis. The benefit of technical characteristics of Sentinel-2 will be:

-      The radiometric information will provide valuable information for applications in term of water detection, water quality assessment, as well as classical land cover applications.

-      A synoptic observation capacity over large and complex hydrological systems thanks to the relative wide swath of the Sentinel sensors.

-      The potential for small sensitive area thanks to its high resolution.

-      It highlights the synergy in term of revisit of SAR/optical data (i.e. S-1/S-2), this synergy is highly recommended for water bodies applications.

Finally the Sentinel-2, alone or associated synergistically with Sentinel-1, will be a powerful tool for mapping and monitoring rich, complex and sensitive large ecosystems such as the monsoon lakes in Asia, but also inland delta or smaller wetland and water bodies, all around the world. The DRAGON and Take5 experiments, illustrate the benefits and interests of these Sentinel data for global and local water bodies monitoring, but also for related applications such as biodiversity, epidemiology, or societal stakes (land/urban planning, risk management, etc). As the Sentinel-2 ramp-up phase will began soon, first acquisitions over Poyang Lake as well as over the Rhine flood plain will occur during winter 2015-2016, and first results of their exploitation should be presented, confirming the expected results obtained within the framework of the two Take5 experiments

There is significant interest in monitoring agricultural areas, in order to monitor food security, but also to understand the role of agriculture in the carbon and water cycles. The dynamic nature of agricultural landscapes, and the timely requirement for information on these areas make monitoring methods based on space-borne Earth Observation (EO) sensors very attractive. A major challenge when using optical data is that orbital choices and cloudiness can result in patchy coverage of the areas of interest, resulting in a poor estimation of fast evolving development periods. To mitigate this, observations from different sensors over the same area could be used, but this presents a challenge, as different sensors acquire data at different times, and with different spectral, spatial and angular characteristics. The use of physical model that describe the data acquisition process using radiative transfer (RT) theory are a way of consistently interpreting the observations in terms of biophysical parameters such as leaf area index (LAI), soil moisture or leaf chlorophyll content. The task of inferring the land surface parameters from the observations is however ill-posed, in the sense that the observations might not have enough information to constrain the parameters. Data assimilation (DA) techniques provide a consistent framework to supplement the observations with additional constraints that can result in a more robust estimate of the state of the land surface. A particularly interesting approach is that offered by variational DA schemes, that reduce the inference problem to one of minimising a cost function.

In this contribution, we apply the EO-LDAS concept [1] to combine a time series of medium resolution observations of surface reflectance from Landsat and SPOT sensors over agricultural areas in SW France and  Ukraine JECAM site. The observations are interpreted using an adequate RT model, and are supplemented by spatial and temporal regularisation to produce a spatially and temporally continuous estimation of the land surface state, fully qualified by uncertainties. In order to practically solve the problem, we make use of emulators (fast surrogate models) of the RT codes [2]. Emulators not only result in fast approximations to the RT models, but also in fast approximations to the gradient of these models, which can then be used to efficiently minimise the DA cost function. We present comparisons with ground measurements of time series of LAI as well as predictions of observations from other sensors as a form of validation.

Acknowledgements

Funding for this work was provided under the ESA OPTIRAD/SEOM SY-4Sci Synergy: S2-S3 Land projects. EOLDAS has been developed under funding from ESA (various awards) as well as under the NERC NCEO.

References

1. Lewis P, Gómez-Dans J, Kaminski T, Settle J, Quaife T, Gobron N, et al. An Earth Observation Land Data Assimilation System (EO-LDAS). Remote Sens Environ. 2012;120: 219–235. doi:10.1016/j.rse.2011.12.027

2. Gómez-Dans JL, Lewis PE. Efficient emulation of radiative transfer codes using Gaussian processes. Remote Sens Environ. In Prep;

Due to ESA Call in December 2014 the Remote Sensing Center of the Institute of Geodesy and Cartography has submitted the proposals for getting the SPOT Take5 images every five days since April 2015 for the following test sites: Biebrza National Park with unique in Europe wetlands and mountainous grassland located in the Western Carpathians.

The prime objective of the SPOT5 usage is to build a monitoring system based on remote sensing data for mountainous grasslands conditions. Using vegetation indices along with field measurements gives the opportunity to estimate the grassland growth conditions, and additionally estimate the Gross Primary Production (GPP). Based on these products, the biomass distribution maps and the biomass increase or decrease its variability during the growing season are provided.

The changes in NDVI from SPOT Take5 data reflected to the changes of biomass due to severe drought which occurred at the beginning of the season and later in July 2015. Figure 1 presents the distribution of NDVI at the wetlands area in the northern part of Poland. It was important to find the relationship between Sentinel-1 and the soil moisture for the whole basin of Biebrza wetlands taking into consideration NDVI values calculated applying Red and IR data but also considering the data in SWIR as the indicator of vegetation wetness.Satellite-based mapping to the growing season in Biebrza and Western Carpathians was developed to monitor the impacts of a changing climate. As the support of remote sensing data analysis two Sentinel-1 Soil Moisture validation sites with fully mobile meteorological stations covering grasslands and marshlands were installed. All the complex and continuous climate data records with in-situ measurements were applied to investigate local variation in phenology.Figure 2 presents the example of seasonal change in NDVI observations between April and June. The higher values of change imply the stronger increase of biomass within analyzed period. The vegetation maps are the basis to develop change detection algorithm for finding most sensitive locations responded to drought observed in Poland. SPOT Take5 NDVI time series data (Figure 3) are prepared to phenological monitoring with analyzing spatial variability of the growth.At 2 test sites proposed, the in-situ measurements were carried out during vegetation season including: vegetation moisture, LAI, radiant flux, NDVI, PAR, fAPAR, CO₂ exchange performed by a static chamber method, also height and volume of biomass. Meteorological stations located nearby supported the modeling approach.

The data of SPOT5 with a high spatial resolution of 10m and 5 day revisit come forward for monitoring hardly available areas. Between field campaigns every two weeks/ one month, we achieved 3/6 satellite images. The correlation is being done between the state of vegetation cover and its spectral reflectance obtained from SPOT5 observations and heat fluxes in order to assess the terrestrial carbon sequestration which is limited by rates of vegetation growth and its accumulation. The relation of the vegetation indices to the spatial estimations of potential carbon sequestration or release is going to be provided, then the carbon to PAR absorbed by the vegetation canopy and biomass measured at the ground. The terrestrial Net Primary Production (NPP) is the carbon related variable which plays important role between the plants and atmosphere. These relations obtained from correlation and comparison satellite data and results of field measurements allow performing the "scale-up" from in- situ measurements through the scale of SPOT5 to whole area of mountainous grasslands.

The article will present when and where the differences in NDVI and vegetation moisture occurred and how it was related to the differences in Sentinel-1 signal. We are expecting establish a method to quantify CO₂ exchange and vegetation of grassland condition.

Land surface temperature (LST) is a key input in the surface energy balance modeling. Spatial resolution of the thermal infrared (TIR) data is always lower than the visible and near infrared (VNIR) data onboard the same sensor. This is an issue in small-size-field agricultural areas, where satellite estimation of actual evapotranspiration (ET) is traditionally limited by the TIR pixel size. Several fusion and disaggregation techniques have been proposed to downscale the TIR data to the VNIR spatial resolution within a sensor. Beyond this, recent works have explored the possibility to downscale TIR data from low resolution sensors (SEVIRI, MODIS, AATSR, coming Sentinel-3,…) to the spatial scale of medium-high resolution sensors (ASTER, Landsat, Sentinel-2,…).

Sentinel-2 offers a 5-day revisit cycle with 10-30 m spatial resolutions, but is provided with no TIR band. In the framework of the Sentinel-2 preparatory activities ESA developed the SPOT5 take 5 experiment. From early April to the end of August 2015, SPOT-5 satellite was relocated in a 5-day orbit, before being decommissioned. Based on the spectral matching between both VNIR sensors, SPOT-5 was used to simulate Sentinel-2 products.

In this work we tested the potential of downscaling MODIS LSTs to Sentinel-2 spatial resolution, and using disaggregated LSTs as inputs in energy balance modeling to monitor high-resolution ET. Concurrent, or close in date, images from the SPOT5 take 5 experiment and Terra-MODIS were used for the LST disaggregation. The selection of these images was constrained to MODIS overpasses with a low viewing angle and cloud coverage. MOD09GQ and MOD09GA 250-m products were used for MODIS VNIR, and MOD11_L2 1-km for MODIS TIR. SPOT-5 L2A images were provided by CNES and ESA with 10-m spatial resolution. An agricultural area, located in the Barrax test site (Albacete, Spain), was selected for this study. LST transects, concurrent with MODIS overpass times, were performed using hand-held Apogee MI-210 TIR radiometers over 5 different fields, during summer 2015. All these measurements were corrected of emissivity and atmospheric effects. The Simplified Two Source Energy Balance Model (STSEB) was applied to the combination of MODIS/SPOT-5 images to estimate surface energy flux maps. A weighting lysimeter and a four-component net radiometer were installed at two of the five sites, a grass and a vineyard fields, allowing continuous ET and net radiation measurements. An eddy covariance system was also placed in the vineyard to register turbulent fluxes. Flux sensors, together with ancillary meteorological instruments, were installed 2 m above the canopy and registered data every 15 min. An average error lower than ±3 °C was observed in disaggregated LST, from a dataset of more than 30 measurements ranging 25-55 °C. An average relative error of 20% was obtained in the ET retrieval.

Even though further research is required, these results give confidence to the use of coming Sentinel-2 images, together with MODIS scenes, to derive disaggregated LSTs that can be used as inputs in energy balance models to downscale surface energy fluxes to the field-scale spatial resolution required for agricultural applications.

The integration process between agriculture and environment promotes the development of agricultural practices that preserves the nature and the rural heritage. EU directives (Water Framework or Nitrates) and the Rural Development Program define objectives and indicators to evaluate the agro-environmental impact of farm practices. The commercial goal to maximize returns and minimize costs increased the role of agro-technology, which is limited by the local resources. As business, agriculture aims at production and depends on the availability of local natural resources. For growing the production, it is necessary to implement a modern approach based on exogenous data, as satellite imagery, integrated in a thematic GIS, for evaluating the real need in water and fertilizers at parcel level. Based on these considerations, the European COPERNICUS initiative aims the development of EO based public services able to create benefit for the agricultural sector. EC transferred at member state level the responsibility to develop and sustain at national locally adapted COPERNICUS services. GEOFARM service contributes to the development of the Romanian agricultural sector (affected by the land redistribution process generating an unstructured rural landscape).

            The synergistic analysis through satellite images and agro-models (as support for the sustainable management of farms) allows the end-user to determine an optimal input for each affected area inside the plot, according to intra-parcel variability, by accessing a secured web interface, for visualize and locate the vulnerable areas.

            Benefiting from Earth Observation satellite data (SPOT, Landsat and Sentinel), the GEOFARM service will provide accurate harmonized geo-information regarding the phenologic status of crops by monitoring variables such as the vegetation state or the water cycle.

            A multi temporal database is being realized by using modern tools which describes in detail irrigated areas accepted as functional model for extrapolation at the entire irrigated area of the country. On the other hand, it raises the efficiency of the orientated web mapping, the number of spinoff products obtained by a facile interpretation employable also in other thematic applications. The materialization of this project is represented by the realization of a set of methods and rules for data processing and validation that allows to repetitively obtaining homogenous data on the irrigated plots.

            The implementation of a service for farmers needs efforts for disseminate the knowledge in view of implementing the research results, taking into account the specific landscape/ environment and socio- economic contexts. Such a solution needs an important scientific investment and a multi-disciplinary approach.