Final Report Summary - IMAGINES (Implementation of Multi-scale Agricultural Indicators Exploiting Sentinels) Executive Summary:The Copernicus program is the EU response to the increasing demand for reliable environmental data. The Global component of the Copernicus Land Service, known as Copernicus Global Land Service (CGLS), provides bio-geophysical variables, derived from the imagery of low and medium resolution satellite sensors, describing the state of the vegetation, the energy budget at the surface and the water and cryosphere cycle to understand ecosystem dynamics, estimate crop yields, manage natural resources, assess carbon budget and characterize environmental conditions and changes for climate modelling. IMAGINES has performed innovation and development activities to support the operations of the Copernicus Global Land service, preparing the use of the Sentinels missions data in an operational context. Moreover, IMAGINES has favored the emergence of new downstream activities dedicated to the monitoring of crop and fodder production that are key for the implementation of the EU Common Agricultural Policy, of the food security policy, and could contribute to the Global Agricultural Geo-Monitoring Initiative (GEOGLAM) coordinated by the intergovernmental Group on Earth Observations (GEO).The main objectives of IMAGINES were to (i) improve the retrieval of basic biophysical variables, mainly LAI, FAPAR and the surface albedo, identified as Terrestrial Essential Climate Variables, by merging the information coming from different sensors (PROBA-V and Landsat-8) in view to prepare the use of Sentinel missions data; (ii) develop qualified software able to process multi-sensor data at the global scale on a fully automatic basis; (iii) complement and contribute to the existing or future agricultural services by providing new data streams relying upon an original method to assess the above-ground biomass, based on the assimilation of satellite products in a Land Data Assimilation System (LDAS) in order to monitor the crop/fodder biomass production together with the carbon and water fluxes; (iv) demonstrate the added value of this contribution for a community of users acting at global, European, national, and regional scales. IMAGINES (http://www.fp7-imagines.eu) has gathered 8 partners, 6 public bodies and two SMEs, from 4 European countries and one international organization (ECMWF). The duration of the project was 44 months, starting on 1st November 2012. IMAGINES has delivered (i) operational processing lines interoperable with the existing CGLS infrastructure and able to run automatically at the global scale to generate global biophysical products disseminated by the CGLS (ii regional high resolution biophysical variables derived from multi-sensor satellite data (iii agricultural indicators, including the above-ground biomass, the carbon and water fluxes, and drought indices resulting of the assimilation of the biophysical variables in the LDAS (iv maps of crop group and crop types updated along the season (v in-situ measurements collected during 64 field campaigns over 23 different sites from 2013 till June 2016 resulting in 40 high resolution ground-based maps of LAI, FAPAR and FCover (http://www.fp7-imagines.eu/pages/services-and-products/ground-data.php) used, in the CGLS, for the validation of moderate resolution biophysical products.Project Context and Objectives:The Copernicus program is the EU response to the increasing demand for reliable environmental data. The objective of the Copernicus Land Service is to continuously monitor and forecast the status of land territories and to supply reliable geo-information based upon (i) Earth observation data provided by the Copernicus Space Component and (ii) ground measurements collected by the Copernicus In-situ Component (Figure 1). The Global component of the Copernicus Land Service, known as Copernicus Global Land Service (CGLS), provides bio-geophysical variables, derived from the imagery of low and medium resolution satellite sensors, describing the state of the vegetation, the energy budget at the surface and the water and cryosphere cycle to understand ecosystem dynamics, estimate crop yields, manage natural resources, assess carbon budget and characterize environmental conditions and changes for climate modelling. The Copernicus Global Land Service was built in the framework of the FP7 Geoland2 project, which has set-up pre-operational services that have continued operations in the framework of the Copernicus Initial Operation phase.Figure 1: Overview of the Copernicus componentsIn the general Copernicus context, IMAGINES intended to: • Consolidate and continue the research and development efforts conducted in the framework of the geoland2 project to support the Copernicus Global Land Service (Figure 2), preparing the exploitation of the Sentinels missions data in an operational context to guarantee the continuity of the operations of the CGLS beyond 2014. Figure 2: IMAGINES in the Copernicus context• Favor the emergence of an original agriculture service using a new method to assess the crop and fodder production based on the assimilation into a Land Data Assimilation System (LDAS) of basic multi-scale biophysical variables, mainly LAI, FAPAR and the surface albedo identified as Terrestrial Essential Climate Variables, derived by merging the information coming from different Sentinel sensors and other GMES contributing missions.Thus, IMAGINES wanted to serve the growing needs of international (e.g. FAO and NGOs), European (e.g. DG AGRI, EUROSTATS, DG RELEX), and national users (e.g. national services in agro-meteorology, ministries, group of producers, traders) on accurate and reliable information for the implementation of the EU Common Agricultural Policy, of the food security policy, for early warning systems, and trading issues. It also aimed to contribute to the Global Agricultural Geo-Monitoring Initiative launched by the G20 Agriculture Ministers Council through links with the JECAM (Joint Experiment for Crop Assessment and Monitoring) developed in the framework of the intergovernmental Group on Earth Observations (GEO) Global Agricultural Monitoring (GEOSS Task AG0703 a) and Agricultural Risk Management (GEOSS Task AG0703 b). The G20 ministerial declaration stated that “we recognize the importance of timely, accurate and transparent information to address food price volatility and agree on the need to improve the quality, reliability accuracy, timeliness and comparability of data on agricultural markets (production, consumption and stocks)”. IMAGINES intended to contribute to address the “production” issue by developing tools able to analyse Sentinel data and integrate them into models across scales (from the regional to the global scales). IMAGINES targeted also to contribute to the monitoring of the carbon and water cycles at the global scale, in continuity with the activities developed within the geoland2 project. In addition, the higher spatial resolution products envisaged in IMAGINES aimed providing the relevant information needed within ICOS to understand the scaling up issues associated to the carbon and water tower fluxes measurements.The geoland2 FP7 project (2008-2012) had built:• pre-operational services providing biophysical variables able to describe the state of continental ecosystems at a 10-day frequency;• Land Data Assimilation Systems (LDAS) able to integrate, jointly, these EO-derived biophysical variables into land surface models, in order to monitor the vegetation biomass, together with fully consistent soil moisture, carbon and water fluxes estimates, at global and regional scales. To deal with the non-availability of Sentinel-2 and Sentinel-3 data during the project life, IMAGINES has performed all the analysis with PROBA-V data (available since 4th December 2013 after a successful launch on 6th May 2013) as proxy of Sentinel-3 data, and Landsat-8 data as proxy of Sentinel-2 data.Capitalizing on the achievements of geoland2, and considering the above mentioned data availability, the objectives of IMAGINES were:1. Investigating the retrieval of multi-sensor and multi-scale biophysical variables (LAI, FAPAR, FCover, Albedo) in view to exploit Sentinel sensors data2. Developing qualified software able to ingest and process multi-sensor data at the global scale on a fully automatic and reliable basis.3. Complement and contribute to the existing or future agricultural services by providing new data streams relying upon an original method to assess the vegetation above-ground biomass based on the assimilation of the above-mentioned satellite products, using either regional or global LDAS infrastructures, in order to monitor the crop/fodder production together with the carbon and water fluxes.4. Demonstrating the added value of this contribution for a community of users acting at global, European, national, and regional scales.Finally, IMAGINES aimed at contributing to the continuity of the Copernicus Global Land Services, set-up in geoland2, and implemented in the Initial Operations phase (2012-2013). In particular, IMAGINES intended to develop qualified processing lines able to process at global scale multi-sensor data (Sentinel-3 and PROBA-V), and demonstrate that they can run in a pre-operational environment. Such processing lines were dedicated to be integrated into a full operational production unit so that the global multi-sensor biophysical products (LAI, FAPAR, FCover, Albedo) ensure the continuity of existing ones.Project Results:The scientific and technical results of IMAGINES includes: i) the methodologies and the processing chains of biophysical variables (LAI/FAPAR/FCover and Albedo); ii) the use of the Land Data Assimilation System for crop and fodder monitoring; iii) the methodology for combining optical and microwave input data for crop land discrimination; iv) the database of in-situ measurements and their reference maps.1.1.3.1 Biophysical variablesThe objective was to develop a methodology for deriving land surface properties from Sentinels missions data. Since no Sentinel-3 neither Sentinel-2 data were available during the project life time, studies were carried out with proxy data i.e. PROBA-V for Sentinel-3, Landsat-8 for Sentinel-2. Innovative algorithm has been defined to retrieve some land surface properties from 300m PROBA-V input data. The corresponding biophysical variables are: the Leaf Area Index (LAI), the Fraction of Photosynthetically Active Radiation absorbed by the vegetation (FAPAR), the fraction of vegetation cover (FCover) and the albedo. INRA set-up a two-step approach, known as GEOV3: 1) the daily PROBA-V top of the atmosphere reflectance products are first transformed into instantaneous estimates of LAI, FAPAR, FCover (Step A in Figure 3); 2) smoothing and gap filling is achieved over a compositing temporal window that may be dissymmetric as in the case of the near-real time situation or at the beginning of the time series (Step B in Figure 3). Figure 3: Flow chart showing the two processing steps of the GEOV3 approachIn parallel, Meteo-France has defined the methodology to derive normalize surface reflectance and albedo from the 300 PROBA-V data. The approach is depicted in Figure 4 and comprises three successive steps: 1) the spectral Top-of-Canopy (TOC) reflectance values serve as the input quantities for the inversion of a linear kernel-driven BRDF model, which allows us to take into account the angular dependence of the reflectance factor; 2) the spectral albedo values in the sensor channels are determined from the angular integrals of the model functions with the retrieved parameter values; 3) the narrow- to broadband conversion is performed with a linear regression formula. Figure 4: Flow chart of the albedo determinationHYGEOS has developed the corresponding processing chains according to the specifications of the Copernicus Global Land Service. Once fully qualified, the processing chains have been delivered to VITO which has integrated them into the CGLS production facilities. The dekadal global LAI, FAPAR (Figure 5) and FCover and Albedo at 300m resolution are disseminated in near real time through the CGLS portal (http://land.copernicus.vgt.vito.be/PDF/portal/Application.htm).An exhaustive quality assessment exercise of the GEOV3 and Albedo products is being performed in the CGLS framework. This analysis uses, in particular, the in-situ reference database collected in IMAGINES to assess the accuracy of the products. The results will be presented in public reports available on the CGLS website (http://land.copernicus.eu/global/).While the albedo algorithm is fully interoperable with the Sentinel-3 data, the GEOV3 methodology, applied to PROBA-V sensor data, has been adapted, in theory, to be applied on Sentinel-3 data. This preliminary work will be continued as an evolution task in the framework of the CGLS.Figure 5: FAPAR map over Europe on 30th April 2014 derived from 300m PROBA-V dataIn the second part of the project, INRA has worked on the retrieval of biophysical variables at decametric resolution to prepare the use of Sentinel-2 data (Li et al., 2015). Then, fusion between 300m GEOV3 variables and decametric Landsat-8 products has been investigated to generate DHF (decametric-hectometric fused) products every 10 days. The algorithm has been evaluated over a demonstration site (South-West, in France) where a time series of Landsat-8 FAPAR products at 30m spatial resolution and dekadal PROBA-V GEOV3 FAPAR at 300m spatial resolution were available. The estimated DHF FAPAR products are consistent with the original Landsat-8 FAPAR data but improve their temporal smoothness (Figure 6).Figure 6: Temporal distribution of DHF FAPAR, corrected Landsat-8 FAPARa nd original Landsat-8 FAPAR over four sample pixels. The correction allows accounting for the inherent differences between Landsat-8 and PROBA-V FAPAR products.Finally, in collaboration with UCL which set-up a crop type classification map of the area, INRA derived FAPAR maps per crop types over a demonstration site. The FAPAR averaged over all pixels belonging to each crop at a given dekadal date is calculated. Figure 7 presents the variation of seasonal FAPAR profile of each crop type. It captures well the seasonal characteristic of each crop.Figure 7: Seasonal variation of FAPAR per crop type over Southwest site in 2014All decametric products are available on the IMAGINES web site: (http://www.fp7-imagines.eu/pages/services-and-products/landsat-8-biophysical-products.php)1.1.3.2 Land Data Assimilation SystemThe objective was to assess the utility of the LDAS tools to monitor the events that can cause anomalies in crop production.The possibility of improving the performance of land surface models (LSMs) using remotely sensed observations is a field of active research. The mechanism of integrating observations, in a statistically optimal way, into a numerical model is called “data assimilation”. The latter permits improving the representation of the dynamical behavior of a bio-geophysical system. Land data assimilation systems (LDAS) are needed to integrate satellite data providing information about land state variables such as the surface soil moisture (SSM) and leaf area index (LAI) into LSMs.Meteo-France has improved the LDAS chain over France domain (LDAS-France) including a new radiative transfer model within the vegetation, and extending the assimilation to the new soil multilayer model. The interoperability of LDAS-France with the Numerical Weather Prediction (NWP) assimilation system (SODA program) was assessed. This work was presented in two papers (Barbu et al., 2014, Parrens et al., 2014). In addition, a huge technical work was achieved: the LDAS-France chain was entirely translated from Matlab language in PYTHON and FORTRAN languages. This consolidates the chain and enables the automatic generation of figures and tables. The upgrade from a simplified extended Kalman filter to an ensemble Kalman filter (EnKF) was initiated. The EnKF was coded in SURFEX (simulation platform) and tested at a local scale for sites in southern France. A bias-correction scheme was implemented in the LDAS in order to mitigate the persistence in deep soil layers of soil moisture biases. This work was presented in Fairbairn et al., (2015). The validation focused on the above-ground biomass and the LAI, which are directly affected by the LDAS system. They were compared with output of a crop growth model (WOFOST) and the agricultural statistics (Agreste), provided over France per administrative units. Figure 8 and Figure 9 show a significant improvement after assimilating satellite observations. The assimilation triggers a much better performance in terms of correlation with Agreste data for both LAI and above-ground biomass.Figure 8: Simulated water limited LAI of straw cereals vs. observed (Agreste) yield, over 45 administrative units in France for a 5-year period from 2007 to 2011, representing 225 observations. From left to right: LDAS-France (No assimilation), LDAS-France (Assimilation), WOFOST output.Figure 9: Simulated water limited above-ground biomass (kg/m2) of straw cereals vs. observed (Agreste) yield, over 45 administrative units in France for a 5-year period from 2007 to 2011, representing 225 observations. From left to right: LDAS-France (No assimilation), LDAS-France (Assimilation), WOFOST output. A first version of the LDAS able to work on the Euro-Mediterranean area at 0.5° x 0.5° was implemented (Figure 10). The first tests to operate the system at a global scale were performed at 1° x 1°.The LDAS is currently used, every six months, to perform the cross-cutting monitoring of the Copernicus Global Land Service products over France and, very soon, over the Euro-Mediterranean basin. Figure 10: LAI standard deviation of differences in 2007 over the Euro-Mediterranean area, (left) before and, (right) after integrating LAI and surface soil moisture observations into the model, at a spatial resolution of 0.5° x 0.5° (atmospheric forcing data from ERA-Interim).OMSZ has also improved the regional LDAS over Hungary: a new version of the land surface model including a soil multi-layer diffusion scheme was implemented and validated against CGLS LAI and Soil Water Index (SWI) products and in-situ measurements of CO2 and water fluxes. The assimilation of SWI was also upgraded using a seasonal-based Cumulative Distribution Function (CDF) matching technique which improves the temporal correlation between the satellite products and the model. The ability of the modelling system to simulate inter-annual variability has also been validated in 2D over Hungary. Monthly anomaly maps of LAI and SWI were calculated from 2008 to 2015. Figure 11 illustrates the time series soil moisture anomaly at Hegyhatsal. The wet years of 2010 and 2014 and the drought in 2012 are very well represented by the simulations and the observations.Figure 11: Temporal profile of soil moisture anomaly from 2008 to 2015 at Hegyhatsal site (Blue: model output without assimilation, Green: model output without assimilation, Red: CGLS SWI product, purple: ground measurements)ECMWF works with a LDAS, running at global scale (global LDAS), based upon the CHTESSEL land surface model. A simplified soil moisture stress parameterization, discriminating high and low vegetation and offensive/defensive behavior of the plant, has been introduced into the modelling component of the global LDAS. This results in an improvement of surface energy fluxes simulation, especially for forest, and Gross Primary Production (GPP) assessment, mainly in case of water stress conditions. Then, a pre-processing chain, ingesting the 1km resolution CGLS LAI and Albedo products has been developed to derive the climatologies of these products over the period 1999-2012. Simulations showed that negative/positive anomaly of LAI is transmitted to the CO2 flux as a decrease/increase of the Net Ecosystem Exchange (NEE), as an increase/decrease of the sensible heat flux, and as a decrease/increase of the latent heat flux. This direct impact of LAI anomaly on the energy fluxes demonstrates the potential of the system for drought monitoring using NRT LAI product.A new LAI observation operator has been implemented: it is based upon a conservative disaggregation of the total LAI into high and low vegetation components. This characteristic allows the new operator to be scalable and permits a flexible usage of LAI observations from different sources. Following this step, an assessment of the impact of the combined assimilation of NRT LAI and albedo on the surface and near-surface atmosphere was carried out. Both offline surface runs and forecast experiments confirm the benefit coming from a more realistic treatment of vegetation by the use of NRT LAI and albedo analysis. With NRT products, anomalous year can be detected and surface fluxes are directly affected by their inter-annual variability. This demonstrates the potential of the assimilation system of NRT satellite products in detecting and monitoring extreme events, and in improving the near surface weather parameters assessment accounting for near real time issues such as rapid changes in the LAI due to fast growth or harvest as well as inter-annual variability due to an extreme drought or an extensive snow season that may inhibit growth. The details of this analysis are available in Boussetta et al. (2014).In collaboration with Meteo-France, a prognostic LAI scheme was implemented in CTESSEL and a modified version was introduced by ECMWF to consider a relaxation toward climatological and observation data. This modification was shown to prevent the model from actual drift that can be caused by non-constrained model parameters.Figure 12: Above-ground biomass relative anomaly [%] for November 2010. Relative anomaly is shown as difference between simulations using NRT LAI and simulations using LAI climatology (100*(NRT-CLIM)/CLIM). The land surface model of the LDAS was also adapted to generate above-ground biomass maps so that to assess the assimilation system outputs as drought indicators. Figure 12 shows a global biomass anomaly ratio map for November 2010. The anomaly is computed as relative difference between simulation using climatological data and simulation with NRT LAI assimilation. This plot shows the potential of the biomass output to detect extreme conditions as can be clearly seen for the drought over the Horn of Africa where the decrease of biomass was by more than 50% and for the wet spell over central and east Australia where the increase in biomass reached 60%.The regional LDAS (France and Hungary) and the global LDAS simulations of above ground biomass, water and carbon fluxes over a series of crops and grasslands site in France, Hungary and around the globe are available on the IMAGINES website (http://www.fp7-imagines.eu/pages/services-and-products/ldas-products.php).1.1.3.3 Crop type discriminationThe objective was to investigate the potential of merging optical and micro-wave imagery to improve the crop discrimination. UCL has defined a methodology to achieve multi-sensor crop classification along the season. This method produces three products:1. At the beginning of the cropping season, a pre-seasonal cropland extent map is produced using a dedicated land cover change algorithm based on metrics of the previous year at moderate resolution. The best local land cover map available helps this early diagnosis. If none is available, a global land cover map is used. Since those land cover map do not target agricultural land and since changes may have occurred, an image-to-map discrepancy has been developed to adjust it to the current regional conditions. 2. At the end of the winter, a knowledge-based crop group layer distinguishing winter crops from summer crops is delivered thanks to an automated phenological object-based time-series classification of medium resolution data. A first step relies on a harmonic components analysis to spatially group pixels with the same temporal trajectories. As multi-date segmentation is known to perform better than single-date but requires the prior identification of key-dates, segmenting on the harmonic components is an alternative to overcome this constraint. Second, the automated adaptive recognition decision rule relies on the presence or absence of an observable winter growth peak. 3. Along the growing season, a multi-sensor crop specific classification is achieved. The crop type layer (Figure 13) is updated as data acquisition progresses, taking into account the agricultural calendar and the crop rotation systems. Crops type maps are produced through an iterative segmentation, classification and fusion scheme of the moderate and high resolution optical and radar images. The updates take place at each new acquisition of high resolution imagery.The methodology was first developed over the Tula region, in Russia, and then adjusted for the Free State Province, South Africa, to take into account the differences in the crop calendar and agricultural practices. It was demonstrated that the combination of optical and radar data improves the field delineation and the spatial details at the cost of some commission, e.g. small tree patches confused with winter crops.Figure 13: example of Crop type layer over Tula region, Russia.The various crop maps over Tula and Free State Province are available on the IMAGINES website (http://www.fp7-imagines.eu/pages/services-and-products/crop-maps.php).1.1.3.4 In-situ databaseThe objective is to collect and process ground data according to CEOS Land Product Validation group recommendations so that they can be useful to assess the accuracy of the CGLS biophysical products (i.e. in terms of spatial representativeness).EOLAB conducted the following field campaigns in collaboration with local users:• 25 Mayo - La Pampa in Argentina (February 2014), in collaboration with INTA.• La Reina - Córdoba in Spain (May 2014), in collaboration with IFAPA.• Las Tiesas - Barrax in Spain (May 2014), in collaboration with ITAP.• La Albufera - Valencia in Spain (June-September 2014), in collaboration with the University of Valencia.• San Fernando, in Chile (January 2015), in collaboration with University of Chile.• Barrax, in Spain (May and July, 2015), in collaboration with ITAP.• Pshenichne, in Ukraine (multi-temporal, April- August 2015). A Service Agreement was signed with Integration-Plus, Ltd. for the collection of multi-temporal biophysical measurements using DHP. • Maragua Upper Tana, in Kenya (March, 2016), in collaboration with CIAT.• EUFAR AHSPECT multi-site campaign (June, 2015) in collaboration with Meteo-France. EOLAB participated in the first Agriculture Health SPECTrometry (AHSPECT) campaign from 22-25 June 2015. EOLAB conducted biophysical (LAI, FAPAR, FCover) measurements with several devices over seven sub-sites were ground stations are maintained by CESBIO, Meteo-France (SMOSMANIA) and INRA. The seven sub-sites (Meteopole, Peyrousse, Urgons, Sabres, Creón d’Armagnac, Condom, Savenès) are located between Toulouse and the Atlantic ocean in the South-West of France. • Collelongo forest site, in Italy (July and September 2015), in collaboration with CNR-IBAF. Collelongo is a FLUXNET site where collaboration was established for collection of continuous PAR data (installation of PASTIS-PAR systems) and collection of DHP acquisitions. Moreover, additional field campaigns were conducted by the local users who shared the data for processing:• SouthWest in France (June-September 2013) performed by CESBIO.• Merguellil in Tunisia (January-December 2013), performed by CESBIO.• Ottawa in Canada (June –September 2014), performed by AgriFood Canada.• Merguellil in Tunisia (January-May 2014), conducted by CESBIO.• Tula in Russia (April-September 2014). For this multi-temporal campaign, a sub-contract was signed with Centre of Ecological Expertise (AGROPROEKT).• Rosasco in Italy (July, 2014), performed by IREA-CNR.Finally, thanks to the collaboration with JECAM program, the ground data collected over Capitanata in Italy (March and April, 2015) was provided by Agricultural Research Council.In addition, autonomous systems (PASTIS-PAR) for continuous monitoring of Plant Area Index (PAI) and FAPAR have been installed over three sites:• Barrax site in Spain (March/August 2014). A subcontract was signed for the installation and maintenance of PASTIS-PAR sensors over this site. The data is currently under analysis at INRA.• Yanco site in Australia (November 2014 - now). PASTIS sensors are already installed, and first data was delivered in December. • Ottawa site in Canada (March/September 2014). Table 1 shows a detailed view of all the field campaigns, dates, Essential Sampling Units (ESU) characterized and up-scaled ground maps per site. The ground data was up-scaled using SPOT-5 or Lansat-7 or Landsat-8 images according to the CEOS LPV protocols. A total of 165 high resolution ground based maps (35 in 2013, 67 in 2014, 59 in 2015, 4 in 2016) have been generated from the ground measurements (see example on Figure 14). The complete database (ground measurements and reference maps) is available on the IMAGINES website (http://www.fp7-imagines.eu/pages/services-and-products/ground-data.php)Figure 14: Ground-based LAI (top-left), FAPAR (top-right), FCover (bottom-right) maps (20x20 km2) retrieved on the 25 de Mayo- La Pampa site (Argentina) – 9th February 2014. Quality Flag map (bottom-right): clear and dark blue correspond to the pixels belonging to the ‘good quality’ estimate; red corresponds to “poor quality” estimate.EOLAB has also contacted 450 sites where albedo ground data are collected: 196 of them have shared their data (91 sites with Albedo and Diffuse fraction). These sites belong to existing networks like AERONET, AMERIFLUX, ASIAFLUX, BSRN, CARBOEUROPE, CARBOAFRICA, ISIS, LTER, OZFLUX, TCOS-SIBERIA and the European Fluxes Data Cluster (EFDC). The spatial representativeness and suitability of these sites for the validation of CGLS Albedo has been analyzed. A total of 53 homogeneous sites were processed. The period covers from 2006 to 2013. However, only 7 sites shared recent data corresponding to 2013 or 2014.Potential Impact:The dissemination activities are performed through the project website, through participation in international conferences and workshops and through publications in peer-reviewed journals. The results are exploited in the Copernicus Global Land Service, for validation and evolution purposes mainly, and in the Copernicus Climate change Service. They are also exploited in other research projects and initiatives. 1.1.4.1 ImpactThe expected impacts of the call were:• “In the context of the already existing capabilities, projects will be expected to contribute to the integration of these data sources (Sentinel satellites) into service chain of the GMES services, particularly for global land applications”: IMAGINES has provided two operational processing chains to the Copernicus Global Land Service. These chains allow the introduction of the 300m biophysical variables into the CGLS portfolio and the evolution of the service operations towards medium resolution. Furthermore, these chains require no or limited adaptation to be able to ingest Sentinel-3 data. As such, the impact of IMAGINES is very positive. • “the projects are expected to establish a basis for the development of innovative new GMES products or applications based on operational space data availability from European Sentinels”: IMAGINES has opened the way to the integration of the LDAS in the operational processing lines of the national agro-meteorological services. Indeed, through national training sessions in France and Hungary, the potential of LDAS tool for agro-meteorology applications has been demonstrated. In particular, future collaborations are expected in Hungary with the Research Institute for Soil Science and Agricultural Chemistry (RISSAC) and the Ministry of Agriculture. Furthermore, IMAGINES has performed the first research about the fusion of multi-scale multi-sensor products. Results will benefit, in the coming months, the evolution of the CGLS towards the multi-sensor approach. As such, the impact of IMAGINES is positive.• “Apart from addressing specific knowledge generation enabling GMES service delivery at European level, projects could also help stimulate new commercial activities through innovative space applications, and thereby have a beneficial impact on SMEs active in the value-adding sectors.”: IMAGINES did not initiate any commercial activities: it was not possible since the Sentinel-2 data were not available during the project life. However, at short term, IMAGINES has stimulated the activities of HYGEOS and EOLAB, the 2 SMEs of the consortium. Both are involved in the consortium in charge of the production “Vegetation and Energy” of the CGLS. HYGEOS is also involved in the two other consortia in charge of the production “Cryosphere and Water” and of the “Dissemination”. As such, the impact of IMAGINES is more limited than expected but positive. • “Further insights into the uptake of products, possible models for operational supply, and the evolution and trends of value-added product delivery in light of future space sensors will be demonstrated”: IMAGINES did not make this demonstration. As such, the project has no impact. 1.1.4.2 Dissemination activitiesThe IMAGINES website is at the address: Figure 15. It is updated regularly to announce all projects events and achievements. The website gives generic information on the project, and on its progress. In particular, the “News” section reports the events where IMAGINES is promoted and the oral talks and posters are available. The “Documents” section contains the public documents of the project. The section “Services and Products” include one page per product type: High Resolution biophysical products, LDAS products, Ground measurements, Crops Maps and Global biophysical products. A simple registration procedure has been set-up to provide access information to users so that they can connect to a ftp site where the products (high resolution biophysical products, LDAS products, and crop maps) can be downloaded. The IMAGINES website is also a platform to disseminate the ground measurements database collected over the demonstration sites. The maps of biophysical variables, resulting from the up-scaling of the ground measurements using high resolution satellite images, are also made available to the community (without any registration) as they are very valuable material to assess the accuracy of satellite-derived biophysical products at kilometric or hectometric resolution.Figure 15: Home page of the IMAGINES websiteA flyer has also been prepared to describe the main characteristics of the project (Figure 16). Figure 16: Flyer of IMAGINES projectFigure 17: Front page of GV2M abstract book.From 3rd to 7th of February 2014, INRA organized The “Global Vegetation Monitoring and Modeling” International Conference (GV2M) in Avignon, France. The workshop webpage is https://colloque6.inra.fr/gv2m.The objective of GV2M was to bring researchers together to discuss the new developments in the use of remote sensing observations and Earth system modelling with emphasis on applications related to water, carbon and nitrogen cycles, climate processes and change, agriculture and forest monitoring over large spatial domains using possibly long time series of satellite observations. It was organized in the continuity of the previous Global Vegetation Monitoring workshops organized by the University of Montana and TERRABITES symposiums with an obvious synergy between these observational and modelling components. GV2M has also facilitated exchanges between the global land community and the agriculture community. It was an opportunity to shape the exploitation of the new observations that are or will be available in the very close future (VIIRS, PROBA-V, LDCM, Sentinels ...).IMAGINES achievements were presented through 5 talks and 3 posters. All can be downloaded on the GV2M web page, and on the “News and Events” page of IMAGINES website.IMAGINES was a sponsor of this event (Figure 17).All along the project life, IMAGINES and its achievements have been presented in more than 30 international events, inside and outside Europe. The details are given in Section 1.2 Table 3. Furthermore, IMAGINES activities have been disseminated through 9 scientific publications. The list is given in Table 2 of Section 1.2. In addition, 3 additional papers are in preparation. They are:• W. Li, M. Weiss, S. Buis, R. Lacaze, V. Demarez and F. Baret. “The DHF method to improve the temporal frequency of decametric products using daily hectometric products: the case of FAPAR as derived from the combination of LANDSAT8 and PROBA-V observations” to be submitted to Remote Sensing of Environment by INRA. • Camacho, F. et al. “Accuracy assessment of LAI and FAPAR GEOV1 and MODIS products over agricultural areas using the IMAGINES ground database”, to be submitted to Remote Sensing of Environment by EOLAB• Weiss et al. “Near Real Time LAI, FAPAR and FCOVER products at 300m resolution derived from PROBAV: principles and validation” to be submitted to Remote Sensing of Environment by INRA1.1.4.3 Exploitation of resultsExploitation in the Copernicus Global Land ServiceTo support the evolution of the CGLS in moving to a production at 333m resolution, IMAGINES has worked at two levels:1. Set-up innovative algorithms. INRA has defined the GEOV3 methodology to retrieve the LAI, FAPAR and FCover at 333m resolution which allows to better capture the spatial and temporal variations of the ecosystems compared to the existing versions of products (GEOV1 and GEOV2 currently in operations in the CGLS). Similarly, Meteo-France has improved the retrieval methodology of the surface albedo taking advantage of the finer spatial resolution of PROBA-V input data.2. Develop the associated processing lines. HYGEOS has implemented the GEOV3 algorithm and the albedo retrieval methodologies into two processing lines able to ingest the 333m PROBA-V data and to generate systematically the biophysical products according to the operational specifications and constraints of the CGLS. The Copernicus Global Land service has integrated the two processing chains developed in IMAGINES into its production unit to generate systematically, and in a timely manner, the biophysical products which are then disseminated to users via the data portal (http://land.copernicus.vgt.vito.be/PDF/portal/Application.html#Home) and via Eumetcast.Before delivery to the users, the uncertainty of satellite products needs to be estimated. In particular, the product accuracy should be estimated over a significant set of locations and time periods by comparison with reference in situ data sets. The availability of ground data becomes more important when products based on new satellite sensors are being developed, such as PROBA-V products of the Copernicus Global Land Service. However, currently, there is no specific in-situ network oriented to the validation of biophysical (LAI, FAPAR) satellite products. Therefore, the contribution of IMAGINES is crucial for the proper accuracy assessment of the Copernicus Global Land products. Two different datasets are addressed in IMAGINES:• Vegetation variables: Leaf Area Index, Fraction of Absorbed PAR, and vegetation cover variables have been collected over demonstration sites following CEOS LPV protocols for validation of coarse resolution satellite products. Moreover, additional ground data has been shared coming mostly from JECAM (www.jecam.org/) network of sites. At final, more than 1700 elementary sampling units have been characterized over 23 different sites during 64 field campaigns over the world resulting in 40 ground-based maps of each variable for the period 2013-2016.• Albedo: Blue-sky albedo and fraction of diffuse radiation (when available) data over homogeneous sites (up to 53 sites) of the European Fluxnet Database Cluster (www.europe-fluxdata.eu/) and SURFRAD stations (http://www.esrl.noaa.gov/gmd/grad/surfrad/) have been requested and formatted. A total of 53 homogeneous sites with albedo data have been processed. The period covers from 2006 to 2014 but most of the sites does not distribute the most recent years. This dataset is exploited for the validation of Copernicus Global Land products as follows:1. Accuracy assessment• Direct comparison with ground-based maps allows computing uncertainty metrics. The IMAGINES dataset allows updating accuracy of Copernicus Global Land SPOT/VGT GEOV1, GEOV2 products, and to assess the accuracy of PROBA-V at 1 km (GEOV2) and 333 m (GEOV3). Figure 18 (left) shows an example of the PROBA-V GEOV3 accuracy assessment with the IMAGINES ground dataset.• Direct comparison with ground data over the ESUs allows assessing the uncertainty of the IMAGINES decametric products fully exploiting the richness of the ground data base. The consolidation of this network of sites will allow validating in the near future the Sentinel-2 biophysical products.• For the Albedo products, the accuracy assessment is made by comparing directly with the albedo tower measurement. The IMAGINES data set allow validating Copernicus Global Land SPOT/VGT (Figure 18, right), and PROBA-V 1 km products, and the IMAGINES PROBA-V 333m products.Figure 18: Direct validation results for PROBA-V GEOV3 LAI (left) and for SPOT/VGT Surface Albedo (right)2. Temporal realism• The temporal realism of both vegetation variables and albedo products are investigated by qualitative comparison mainly with continuous LAI/FAPAR and Albedo acquisitions. The cross-correlation between satellite and ground temporal data is useful to quantify the goodness of the temporal variation of satellite products. The comparisons are performed over homogeneous sites (see an example for Albedo on Figure 19). Figure 19: Temporal variations of SPOT/VGT Surface Albedo and field data over the Puechabon site, France.• Moreover, the multi-temporal ground campaigns over few demonstration Imagines sites (e.g. Pshenichne, South-West or La Albufera) allows to further investigating the temporal variations at the resolution of satellite products (see example on Figure 20)The IMAGINES LDAS activities conducted by Meteo-France contribute to the cross-cutting quality monitoring of the Copernicus Global Land Service. Integrating observations into a land surface model using the LDAS is a way to assess and monitor the observation errors of five satellite-derived biophysical variable products of the CGLS: LAI (Leaf Area Index), FAPAR (Fraction of Absorbed fraction of the Photosynthetically Active Radiation), SA (Surface Albedo), SWI (Soil Water Index), LST (Land Surface temperature). One cross-cutting quality monitoring report is released every 6 months. During the exercise performed on operational products generated between January and June 2014, the main findings were:• Suspicious low values of LAI and FAPAR in June from products derived from PROBA-V sensor (till May 2014, they were derived from SPOT/VGT sensor)• A trend towards higher soil moisture values, already detected over 2013 compared to the 2007-2012 period (Figure 21)The former confirmed the outcome of the per-variable quality monitoring and led to stop the production of PROBA-V LAI and FAPAR for further investigations. The latter has revealed a consistency issue in the SWI time series. The origin has been identified and a re-processing action was planned to solve it. Note that such inconsistency was not detected by the regular SWI quality monitoring.IMAGINES is the framework in which the research needed to the validation and to the evolution of the LDAS is conducted.Figure 21: Monthly average values of surface soil moisture over France from 1st January 2007 to 30th June 2014 derived from satellite (green circles), simulated by the model without assimilation (blue line), and simulated by the model after assimilation of satellite-derived LAI and SWI (red line). After mid-2012, the satellite-derived values are systematically higher than the model estimates.Contribution to other Copernicus ServicesECMWF develops and uses world class models and data assimilation systems. In particular, the Integrated Forecasting System (IFS) is shared by ECMWF with national member states. “OpenIFS” is a recent initiative aiming at making the IFS more easily accessible to research institutes and universities. The IFS includes CTESSEL and the offline system. ECMWF has a real-time capability which was essential for MACCIII project and useful for other carbon users and downstream applications to the Copernicus Atmosphere Monitoring Service (http://atmosphere.copernicus.eu/). Interoperability across operational land-ocean-atmosphere systems is ensured given that the surface modelling, atmospheric transport, data assimilation, and seasonal predictions are under unified system configurations.The LDAS and the developments to improve surface characterizations obtained within IMAGINES are directly usable by MACCIII, via the IFS infrastructure in a fully coupled online real-time system with consistent energy/carbon/water budgets. Synergies were also established with the EC EARTH European climate modelling (http://ecearth.knmi.nl/) and the EuroSIP (seasonal prediction) (http://www.ecmwf.int/en/forecasts/tools-and-guidance/documentation-and-support/long-range) initiatives as well as with the seasonal and climate oriented research projects (e.g. Global Soil Wetness Project, FP7/SPECS (http://www.specs-fp7.eu/)). The link to the ERA-CLIM2 project (http://www.era-clim.eu/) has permitted to benefit from improvements in surface vegetation and carbon description obtained as spinoff of the IMAGINES research and development. This is expected to be operationally supported by the Copernicus Climate Change Service (C3S).Exploitation of LDAS at national levelRegional Land Data Assimilation System is a useful tool to produce new drought indicators effective for agriculture yield monitoring and this is investigated over Hungary at 8km horizontal resolution. Now, this regional LDAS produce water, energy, and CO2 fluxes. The model is able to simulate the vegetation biomass and it is investigated how to optimally integrate PROBA-V LAI (and, in the future, the Sentinel LAI) and SWI satellite data into the LDAS, in order to relate the vegetation biomass to the crop/fodder production, through relevant drought indicators. The products are created with offline version of SURFEX (version 7.3) model. Data can be downloaded from ftp.met.hu. Login and password are available from OMSZ team. OMSZ has organized a user workshop (http://met.hu/omsz/OMSZ_hirek/index.php?id=1557) on 4th of May 2016 to present the IMAGINES activities and the potential of LDAS to 9 Hungarian institutions (eg. Ministry of Agriculture, National Service for Hydrology, two universities, two research institutes, ...). Issues like “What kind of drought indicators could be useful for the experts?”, “How can the users utilize the biomass and crop/fodder productions?” have been discussed. Future collaborations with the Institute for Soil Sciences (RISSAC) and the Ministry of Agriculture are promising.Exploitation in other projects and initiativesIn IMAGINES, the activities on the crop classification were still much research oriented. They aimed at answering questions that will impact the use of satellite data during the Sentinel era such as how the large swath of Sentinel-2 will affect the classification accuracy and how to integrate satellite images from different sensors to ensure a timely, accurate and robust crop monitoring. The concepts and methods developed in the framework of IMAGINES have been integral in shaping the Sentinel-2 for Agriculture (Sen2-Agri) project (http://www.esa-sen2agri.org). Sen2-Agri has been launched by ESA, as a major contribution to the R&D component of the GEOGLAM initiative. The project will demonstrate the benefit of the Sentinel-2 mission for the agriculture domain across a range of crops and agricultural practices. The intention is to provide the international user community with validated algorithms to derive Earth Observation products relevant for crop monitoring. Both the pre-seasonal cropland detection algorithm and the crop type mapping along the season developed in IMAGINES are considered as candidate algorithms for the benchmarking phase in their original form or fine-tuned for high spatial resolution.List of Websites:1.1.5 Contact detailsInitially, the IMAGINES consortium comprises 9 entities, 7 of them with public background and 2 private companies (HYGEOS and EOLAB) which are SMEs. CNES participated to the IMAGINES project up to 17th March 2014 because of its major interest in Sentinel-3 and Sentinel-2 products. Unfortunately, due to the delay in the launch of the satellites, no Sentinels products have been practically generated during IMAGINES lifetime. Consequently, CNES withdraws from the IMAGINES consortium which is then made of 8 entities (6 public bodies and 2 SMEs). The following table gives an overview of the participants, showing their nationality, the main tasks attributed, and the main contact person.Table 2: Consortium members and tasks attributed. INT means “International” organization; * CNES participation terminated on 17th March 2014.A project public website has been set-up at the address: http://www.fp7-imagines.eu. See 1.1.4.2 Dissemination activities for more details on the website. final1-imagines-311766-finalreport-figure21.png final1-imagines-311766-finalreport-figure19.png final1-imagines-311766-finalreport-figure2.png final1-imagines-311766-finalreport-figure8.png final1-imagines-311766-finalreport-table2.png final1-imagines-311766-finalreport-figure16.png final1-imagines-311766-finalreport-figure11.png final1-imagines-311766-finalreport-figure20.png final1-imagines-311766-finalreport-figure18.png final1-imagines-311766-finalreport-figure3.png final1-imagines-311766-finalreport-figure10.png final1-imagines-311766-finalreport-figure17.png final1-imagines-311766-finalreport-figure13.png final1-imagines-311766-finalreport-figure12.png final1-imagines-311766-finalreport-figure15.png final1-imagines-311766-finalreport-table1.png final1-imagines-311766-finalreport-figure14.png final1-imagines-311766-finalreport-figure6.png final1-imagines-311766-finalreport-figure1.png final1-imagines-311766-finalreport-figure7.png final1-imagines-311766-finalreport-figure5.png final1-imagines-311766-finalreport-figure4.png final1-imagines-311766-finalreport-figure9.png
