Skip to main content
European Commission logo
English English
CORDIS - EU research results
CORDIS
CORDIS Web 30th anniversary CORDIS Web 30th anniversary
Content archived on 2024-06-18

Information Service on Agricultural Change

Final Report Summary - ISAC (Information Service on Agricultural Change)


Executive Summary:
Current operational services for agro-environmental monitoring mostly rely on the use of satellite images with a high temporal frequency (daily) but low spatial resolution (> 250m). However, since the early 2000, new high frequency, high resolution (< 30m) wide swath sensors have become available.

In the frame of the ISAC project three services were developed and demonstrated:

1. High Resolution Biophysical Parameters (ABP-CMS)
2. Drought stress (DS-CIS)
3. Agricultural change (AC-CIS)

The ABP-CMS service developments included the set-up of a processing chain for high resolution DMC/Deimos-1 and Landsat-TM/ETM+ images (25m pixel size). To demonstrate the service, NDVI and biophysical products such as fAPAR, LAI and fCover were produced for Belgium (2009-2013) and Ethiopia (2008, 2012).

Based on the above high resolution biophysical parameters, products for crop monitoring and damage assessment were developed and demonstrated in 2012 in Belgium and Ethiopia, as part of the DS-CIS service. Products for pasture monitoring were prototyped and demonstrated in 2012 for two regions in Spain.

The ABP-CMS high resolution biophysical parameters and the derived DS-CIS prototype products were validated extensively with field data collected in 2012 in the demonstration countries.

AC-CIS focused on short and long term agricultural change detection and provided yield projections for Ethiopia, Spain and Belgium, accounting for different climate change and land use change scenarios, mainly in support of decision-making. Further, a framework was presented to study risk spreading techniques such as insurance versus risk reduction strategies, such as the use of fertilizers, in developing countries to lower the risk of falling below the minimum nutrition requirement level.

The ABP-CMS service for providing high resolution biophysical parameters is considered a potential candidate for future uptake in the Copernicus services, as this service potentially may form the basis and input for a variety of downstream services. It is also complementary to the GIO / Copernicus Global Land Service.

The proposed DS-CIS and AC-CIS services are well appreciated by the users, both in the public domain (authorities, calamity funds, emergency and early warning sector) as by the private sector (insurance sector, agro-industry, farmers). It is expected that the outputs of ISAC will form the basis for setting up paying downstream services with these users and in particular with the insurance sector.
Project Context and Objectives:
Expansion of the world population is one of the primary reasons for an increased demand on land resources. Exacerbated by climate change, this puts major constraints to sustainable land management. Occurrence of natural disasters significantly increased in the past 40 years, directly affecting the population through the events itself or indirectly through loss agricultural production. Countries try to deal with these global trends in their own way, pending their priorities, resources and level of development. In Europe, agriculture is regulated by the EC’s Common Agricultural Policy. Farmers are mostly protected against abnormal conditions and calamities and are supported through the government or specific agricultural insurances in case of exceptional conditions leading to harvest failure. In Africa crop and rangeland systems are less well understood and the resilience after disasters is low, directly affecting the livelihoods of vulnerable populations. Food aid and development programmes are needed in order to remedy the worst situations and formulate actions towards sustainable land management and development. Although challenges are different from one country to another there is one major commonality: the need for transparent and reliable information on agriculture and the agro-environment.

Current operational services for agro-environmental monitoring mostly rely on the use of satellite images with a high temporal frequency (daily) but low spatial resolution (> 250m). However, since the early 2000, new high frequency, high resolution (< 30m) wide swath sensors have become available.

ISAC aims to develop and demonstrate three services:
1. High Resolution Biophysical Parameters (ABP-CMS)
2. Drought stress (DS-CIS)
3. Agricultural change (AC-CIS)

ABP-CMS intends to expand the capabilities of Copernicus Land Monitoring Core Services (LMCS) for what concerns the provision of vegetation indices and biophysical parameters with a much higher spatial resolution compared to the products offered today. DS-CIS aims to provide information for monitoring the agro-environment, addressing the needs of users from the private sector (insurers, agro-industry, farmers) and the public sector (authorities, calamity funds, international organizations).

The objective of the ABP-CMS, in analogy to the geoland2-BIOPAR service of the LMCS, is to produce vegetation and yield indices (NDVI, DMP) and biophysical parameters (LAI, fAPAR,…) with high frequency and much greater spatial detail based on DMC-1/2 and Deimos-1 sensors (22-32m resolution). The biophysical parameters are generated using model inversion, similar to the methodology used in geoland2-BIOPAR for low resolution images.

DS-CIS aims to derive indices from the above high resolution biophysical parameters to detect drought phenomena and to assess crop damage and related risk. Particular attention is paid to pasture monitoring in drought sensitive areas.

AC-CIS focuses on short and long term agricultural change detection. AC-CIS aims to provide information on future risks by providing model based crop yield estimates and yield projections accounting for different climate changes scenarios, mainly in support of decision-making.

In summary the objectives of ISAC are to:
O1. generate a directory of potential users of the ISAC services
O2. develop ISAC services assuring scientific and technical quality
O3. demonstrate the validity, utility and usage of ISAC services in Belgium, Spain and Ethiopia
O4. benchmark the services against existing alternatives
O5. generate service specifications and a service deployment roadmap
O6. create awareness to the emergency response, land monitoring and the scientific community

Project Results:
1. Downstream user requirements and benchmarking plans
The ISAC project started with an inventory of potential users and a detailed analysis of their needs (WP 21, 22, 23). The users of the final ISAC results are situated in:
- the private insurance sector
- the food security sector, including organizations dealing with micro-insurance
- the emergency response sector

Information requirements from insurance companies, government agencies supporting these insurance initiatives, downstream service providers providing information to insurers, national and international organizations active in early warning for food security and humanitarian organizations were collected from numerous discussions at meetings and conferences, via e-mail or phone and from sending out questionnaires. The collected information requirements form the basis for the development of the ISAC downstream services and products.

The general background of agricultural risk insurance in different countries in Europe was explored with the help of CEA, the European Federation of Insurers and Re-insurers. In total 7 European countries have activities in the domain and expressed interest in the use of EO data for crop damage and risk assessment in general, and in the ISAC project in particular. For Spain and Belgium, the needs were assessed in detail. In addition, users in Russia, Brazil and Argentina also showed interest in the project.

When looking at the information requirements which were gathered, basically three types of users can be discerned:
- Insurance companies, supported by government agencies, which have already put in place EO based insurance (ISAC users in Spain). These users are merely looking to improve their existing products (e.g. pasture insurance) by passing to EO products with a higher spatial resolution though still with a high temporal frequency.
- Insurance companies, supported by government agencies, which are planning to set-up crop insurances (ISAC users in Belgium and Russia). In the first place these users need ‘basic’ information on calamities, yield statistics, etc. for risk assessment. Their EO data needs are less specific.
- Downstream service providers which are providing EO based services to insurance companies (ISAC users in Brazil and Argentina), also requiring EO data with a high spatial and temporal resolution.

It was found that in many cases EO data are not the only source of information for insurers. EO data can be used as a direct trigger for payments but more often they serve to guide field visits. When EO indices present anomalies experts are sent to the field to assess the damage.

Most users indicate that information on future risks related to climate change would be very useful.

With regard to data dissemination, a secured, interactive web application is requested in order to access ISAC products and documentation, to visualize and explore maps and graphs and to provide feedback. Eventually this web application could be combined with an e-mail and/or SMS service for sending out alerts.

Micro-insurance is a special type of insurance set-up in developing countries to protect low-income people, often it is coupled with micro-credits. A number of pilot projects on micro-insurance have been initiated in Ethiopia and other countries in the IGAD region by local insurers with support of the World Food Program, the Food and Agriculture Organization, the World Bank, Oxfam America, SwissRe, etc. Most pilots use a weather index (rainfall) although the availability of weather data is often limited and correlation with actual harvested yield is sometimes low. Investments are made to improve meteorological infrastructure rather than looking at the use of alternative indices such as EO based vegetation indices. Exception is the IBLI livestock insurance for Northern Kenya (and soon also Southern Ethiopia) which uses NDVI data from AVHRR (MODIS in the future) and the pasture insurance which is being developed by NISCO. Higher resolution EO data are believed to bring additional benefit here.

Current early warning systems for monitoring food security also make use of EO data. It basically concerns meteorological and vegetation indices derived from low (1-10 km) and medium resolution (250m) satellite data. At national level in Ethiopia (DRMFSS) the availability of EO data with a higher spatial resolution could improve the existing services as complicated topography and climate variability can be better be accounted for. At regional level (FAO-GIEWS, WFP-VAM) the high temporal frequency (daily overpasses) and especially the availability of long historical time series of the currently used low to medium resolution sensors (Meteosat, AVHRR and SPOT-VGT) in combination with their large geographical coverage is considered as critical for agricultural monitoring. It is feared that the ISAC high resolution service will not meet these criteria, though it is recognized that it offers new opportunities for early warning (Fewsnet), but also for improved crop mapping (FAO-GIEWS). In addition, Information on short and long term impacts of climate change on agriculture in Ethiopia, including estimates of projected yields, caloric availability and vulnerability indicators, would be appreciated.

Even though emergency management and food security may share some strong commonalities or relationships, as in the case of Ethiopia where drought-induced famine and floodings are the most prevalent disasters, the two concepts are clearly distinguishable. “Emergency” is typically a critical situation occurring in a relatively small area and causing needs for immediate intervention. Whereas, “food insecurity” concerns a critical situation affecting numerous communities, vast regions or even nations to produce or access sufficient food within their reach; it develops more slowly thus making time a less critical issue. Besides the need for high resolution information on natural resources conditions, pasture conditions, crop yield,… for risk management and vulnerability assessment, similar to what is described above for early warning, the users, in particular WFP-VAM, express a need for timely (immediate) information from high resolution sensors to support ground information in case of a disaster.

Finally, it would be appreciated if ISAC products for the food security and emergency response sector would be made available through existing data dissemination initiatives such as the Geonetwork nodes and the SAFER network.

The planned ISAC information services, DS-CIS and AC-CIS, only focus on the delivery of information related to drought (anomaly maps, potential damage maps, risk maps, biomass estimates,…) and agricultural change (simulations to assess the effect of climate change on agriculture). Some of the users expressed a need for high resolution crop maps or for maps indicating hail damage. Due to time and budget constraints this type of applications cannot be elaborated within ISAC. However, the high resolution biophysical parameters generated by the ISAC mapping service ABP-CMS can be used as input for such applications (to be elaborated by the users themselves). Other users require information on soil moisture, status of water bodies, area of cultivated land, weather data or yield statistics, but unfortunately the provision of this kind of information is out of the scope of the ISAC project.

2. Service R&D
2.1. High resolution biophysical parameters
Archived UK-DMC-2 and Deimos-1 images (L1T corrected, i.e. ortho-rectified) over Belgium for the period 2010-2011 were obtained from the GSCDA Data Warehouse. Additionally Deimos-1 images over Spain, Ethiopia and Belgium were made available by ISAC partner Deimos Imaging. These datasets served as input for prototyping the ABP-CMS service and products.

A processing chain for high resolution DMC/Deimos-1 images was developed and a first series of prototype products (version 1) was made available in December 2011. A second series of prototype products (version 2) was made available in May 2012, including improved atmospheric correction, LAI, fCover, albedo estimates and improved fAPAR estimates.

The processing chain comprises the following components:
- Data import: conversion of the external data, i.e. calibrated Top-of-Atmosphere (TOA) images, to common formats, accessible by the further routines, initialisation of log-files which will contain the processing history of each segment. Problems of lacking viewing geometry data had to be solved.
- Geometric correction*: model-based remapping generally removes the bulk of the geometrical deformations of the raw segment but most often a considerable shift remains, due to uncertainty/imprecision in the models and their inputs (satellite time, platform attitude, orbital elements, etc.). This translation is corrected by means of a polynomial regression, applied on control points which are automatically searched by means of chip matching. Geometric accuracy indicators are provided with the final images. Current geometric accuracy amounts to 2 pixels on average.
*For L1T corrected input images this step is skipped.
- Atmospheric correction: the calibrated image values are still perturbed by atmospheric (and angular) effects. To convert them into more relevant information, intrinsic to the scanned targets at the earth surface, the SMAC algorithm (Rahmann & Dedieu, 1994) is used. SMAC requires a set of sensor and band specific coefficients. For the first prototype products Landsat coefficients were used as Landsat and DMC/Deimos have a similar spectral behavior, for the second / final prototype products DMC/Deimos specific SMAC coefficients were derived and used.
- Cloud & cloud shadow detection: threshold setting is used to eliminate clouds and snow/ice at segment level. Cloud mask translation is used to remove cloud shadows.
- Calculation of vegetation and yield indices NDVI and DMP and biophysical parameters LAI, fAPAR, fCover and albedo:
o The first fAPAR prototype products were derived linearly from NDVI
(fAPAR = 1.346*NDVI - 0.234 with R²=0.96).
o The final prototype biophysical parameters are calculated following the Geoland2/GIO-BIOPAR approach by applying neural networks on directionally normalized DMC/Deimos-1 top of canopy red and near infrared reflectance values. The algorithm development was subcontracted to INRA-EMMAH, the institute which also developed the algorithms for Geoland2/GIO.

The products and service and the methodology used to generate the products is described in detail in Deliverable 31.2 (Service Prototyping Report for ABP-CMS). The processing flowchart is shown in Figure 1.

Figure 1: Flowchart ABP-CMS processing

New UK-DMC2 and Deimos-1 images for the 2011-2012 growing season in Belgium and Ethiopia were ordered from the GMES Data Warehouse and from Deimos Imaging. They were processed immediately after delivery and placed on a FTP server to make them accessible for the users. These data mainly served to demonstrate the ABP-CMS service. The Demonstration Products are presented in Deliverable 41.1 (Ethiopia), 42.1 (Spain) and 43.1 (Belgium).

To evaluate their quality and accuracy, the ABP-CMS intermediate and final products were extensively validated. For a detailed description of the validation methodology and results we refer to Deliverable 51.1 (Technical Validation and Benchmarking Report).

A basic sensitivity analysis was carried to check the effect of using climatology data (as an alternative for real time measurements) of water vapour, ozone and aerosols as input for SMAC. Variations were simulated by multiplying the climatology values with a factor ranging between 0.25 and 2.00. It was found that variations in water vapour, aerosols and ozone inputs indeed have an influence on TOC reflectances (RTOC) and also on the derived fAPAR, fCover and LAI values, but the errors introduced (in terms of MBE) are still acceptable. This is important to know when transferring the processing chain to an operational environment as the permitted use of climatology values simplifies operations.

The high resolution images used for generating the ISAC products were ortho-rectified by the EO data provider. In order to assess the geometric accuracy of these products, images from different providers were checked for pixel shifts by overlaying a vector file with the field boundaries as declared by the Flemish farmers. The geo-location of the Landsat and Deimos-1 sample images obtained from the USGS and from Deimos Imaging, respectively, is highly accurate. The precision of the UK-DMC2 images acquired from the GMES Data Warehouse seemed to be equal to or somewhat lower than the precision of the Landsat and Deimos-1 images. However, the accuracy may change within the scene and is generally lower at the edges of the scene.
Pixel shifts, even small ones, introduce errors in the time series of observations. Therefore, to remove the remaining shifts in the input EO data, an additional geometric correction algorithm should be applied.

The high resolution biophysical parameters and intermediate products were also validated with field data collected in 2012 in Belgium. NDVI values derived from high resolution DMC/Deimos-1 sensors were compared with close-range sensor measurements on a number of potato fields. LAI, fAPAR and fCover values were validated with DHP (Digital Hemispherical Photography) measurements on selected maize and potato fields. In addition, high resolution NDVI and fAPAR values were compared, after downsizing, with medium resolution MODIS observations.

DMC based fAPAR was found to be lower than DHP based fAPAR (differences of 0.2) whereas LAI and fCover differences between DMC and DHP were much smaller. From the comparison with MODIS it could also be concluded that the DMC derived fAPAR as well as the NDVI was underestimated (with 0.2 for NDVI and from 0 up to 0.2 in the higher values range for fAPAR in Belgium). In Ethiopia the differences were smaller (below 0.05 for NDVI). However, it has to be mentioned that the comparison was based on a limited number of images only. Close range sensor measurements of NDVI, finally, were lower than DMC derived NDVI values, though these sensor measurements need to be interpreted with caution.

In theory, biophysical parameters such as fAPAR, LAI or fCover, should be sensor independent. In practice, however, differences are observed between sensors. Depending on the algorithm used, the values may differ. Systematic differences between time series of biophysical parameters derived from different sensors can be corrected for by applying the appropriate equation.

2.2. Products for crop monitoring and damage assessment

For crop monitoring and damage assessment, actual images are often compared with a reference image, based on a long time series of observations and representing a “normal” situation. Disadvantage of using high resolution images is the lack of archive data. This has several reasons: the frequency of acquisitions is low (as for Landsat-TM/ETM+, overpass frequency of 16 days), the satellite is relatively new (as for Deimos-1, since end 2009) or the acquired images simply cannot be used because of the presence of clouds.

VITO developed a procedure within ISAC to derive high resolution (25-30m) reference images for DMC/Deimos-1 and Landsat-ETM+ images from time series of medium resolution (250m) images of the TERRA-MODIS sensor (operational since 2000), under the hypothesis that these sensors are comparable from a spectral point of view. The MODIS land bands have corresponding bandwidths to the Landsat7-ETM+ and DMC-1/2 and Deimos-1 sensors except their bandwidths are narrower (Gao et al., 2006). By means of orthogonal regression the relation between NDVI and fAPAR values of MODIS on the one hand and DMC/Deimos-1 or Landsat on the other hand, was determined.

For Flanders, crop type information at field level is available from farmers’ declarations from 2003 until 2011 (Mestbank-EPR dataset). This information was used to “unmix” the MODIS spectral values. First, weighted mean MODIS fAPAR values were computed for the main crops per municipality (Genovese et al., 2001) on a 10-daily basis and stored in a database. Secondly, the previously derived equation relating MODIS and DMC/Deimos-1 values was applied. The resulting DMC fAPAR values were then reconverted to images with a spatial resolution of 25m, compatible with the DMC images. Finally, the long term average fAPAR was computed from these images (per dekad).

For Ethiopia (West-Shewa province), a slightly different procedure had to be applied as no detailed crop type information is available for the country. Only those pixels which are labeled as “agricultural land” in the “global cropland map” of the JRC (Van Cutsem et al., 2011) were used. Long term average NDVI values were computed from 10-daily MODIS NDVI images from the period 2000-2011. In order to obtain Landsat7-ETM+ NDVI averages, the previously derived equation was applied. Finally, the images (250m pixel size) were resampled to a spatial resolution of 30m, compatible with the Landsat7-ETM+ images.

For Ethiopia, anomaly maps were generated showing the absolute differences between the actual daily NDVI derived from Landsat or DMC/Deimos-1 images and long term average NDVI values of the corresponding dekad.

For Flanders, Student T-scores and corresponding return periods were computed for the actual DMC/Deimos-1 derived fAPAR values. This way, information was provided on whether or not the fAPAR value is exceptional. Terms as “return period” or “probability” of an event are also more commonly used in the insurance world. Anomaly maps were created showing per pixel whether the return periods is more than 20 years, between 15 and 20 years, between 10 and 15 years, between 5 and 10 years, below 5 years or if normal/above average were observed. These “anomaly maps” allow the user to judge whether the situation (in terms of fAPAR) at a certain moment in time, for a certain pixel, is normal, abnormal, exceptional or very exceptional. The maps can be used, for instance, to guide field visits to problem areas.

Strongly negative fAPAR deviations (T-scores) are indicative for drought stress. If such deviations occur frequently during the critical period for crop growth and development, this may be an indication for drought related damage.

The products and service and the methodology used to generate the products is described in detail in Deliverable 32.3 (Service Prototyping Report for DS-CIS).

Anomaly maps with fAPAR return periods were produced for the 2011-2012 growing season in Flanders for winter wheat, maize and potatoes. NDVI based anomaly maps were provided for Ethiopia for the 2012 Meher season. The DMC/Deimos-1 derived biophysical parameters, produced by ABP-CMS, were used as input for generating these maps. These data mainly served to demonstrate the DS-CIS service. The Demonstration Products are presented in Deliverable 41.1 (Ethiopia) and 43.1 (Belgium).

An example for Belgium and Ethiopia is shown in Figure 2 and Figure 3, respectively.

Figure 2: DS-CIS anomaly maps (fAPAR return period) for a potato field in Retie (Flanders)

Figure 3: DS-CIS anomaly maps (NDVI absolute differences – ADVI) for key area 1 in the east of West-Shewa

To evaluate the quality and accuracy of the DS-CIS products field data were collected in the summer and autumn of 2012 in Flanders and Ethiopia.

To check if the DMC derived NDVI, fAPAR, LAI and fCover values for potatoes observed throughout the season can be used for monitoring potato growth and development or even for yield estimation, the index values were compared with sensor measurements of potato yield for about 80 fields located in the north of Flanders.

Analysis of NDVI, fAPAR, LAI and fCover profiles of fields with normal, high and very high yields showed that these indices were all indicative, at least to some extent, for potato yield. In fact, the four indices were very similar, following a nearly parallel seasonal course, except for LAI which fluctuated more than the other indices. Sensitivity of NDVI, however, was lower compared to fAPAR and fCover and also to CI green and WDRVI, two additional indices which had been tested for this validation exercise.
Correlations of DMC indices of individual dates with potato yields at harvest were rather low. The availability of longer time series of observations, covering different seasons with different growing conditions, as well as summing or averaging index values over the most critical part of the season should give better results. Splitting up the dataset by planting date and variety could probably also improve the yield model. Still, the availability of continuous time series is crucial. In practice, however, observations are often lacking during parts of the season because of the presence of clouds. If a solution could be found to solve this problem, the way is open to use these products for index insurance or for precision agriculture to monitor crops at intra-field level.

In order to detect anomalies, the actual index values (fAPAR for Flanders) were compared with average values. In this case, the DMC historical average was derived at municipality level from historical (2003-2011) 10-daily MODIS fAPAR values of pixels which contain at least 50% of a certain crop. Especially in Flanders where the agricultural fields are very small, the MODIS pixels often contain different crops or vegetation, which may flatten the fAPAR curve. In addition, seasonal advance / delay and differences in planting date between early-normal-late varieties can cause the curve to broaden. As a result, the differences with respect to the average may be more extreme. Therefore, ideally, the historical average should be based on a longer time series (15-20 years) of high resolution images and preferably it should be split up by variety (early-normal-late).

fAPAR based anomaly maps were studied, together with the actual and average fAPAR profiles, for four randomly selected potato fields, respectively with low, normal, high and very high yields. In general, it may be concluded that the anomaly maps provided a good indication of actual potato growth and development in 2012 with respect to average conditions at the different fields.

However, the fact that the field has already been harvested, seasonal advance or delay can also lead to below average index values and high return periods, with or without impacting crop yields. The resulting red values in the anomaly maps may be confusing in such cases. Also, when comparing anomaly maps of two consecutive days, differences may occur, often related to imperfect correction for atmospheric effects. In some cases, aberrant index values are noticed at the field borders due to the influence of surrounding vegetation or due to pixel shifts. Therefore, to enable a correct interpretation of the maps and to obtain information on the cause of anomalies, it is recommended to always check the complete time series of anomaly maps in combination with the temporal profile of the underlying index, comparing the actual and average index values for the concerned region / field.

Finally, the DMC derived anomaly maps were compared with similar maps derived from 250m-MODIS data. The maps were compared for different crops (winter wheat, maize and potatoes), for different seasons and different locations (winter wheat only). Large differences were found between DMC and MODIS derived anomaly maps. The main reason for this is the different spatial resolution of both sensors. Agricultural parcels in Flanders are very small, with average parcel sizes of 1.5 ha, corresponding to about a quarter of a MODIS pixel. As a consequence, the MODIS signal of a single field is often contaminated with signal from surrounding fields or trees, bushes, streets,… nearby. Indeed, when comparing anomaly maps and time profiles of more homogeneous MODIS pixels, the agreement between MODIS and DMC was much higher. This clearly states the need for time series of high resolution satellite data, to allow monitoring at field level.

Also in Ethiopia, the products were validated against field observations (Figure 4). In 8 key areas, detailed field surveys were conducted by GeoSAS in the beginning, the middle and the end of the growing season. Detailed interviews with the farmers were done based on questionnaires. These questionnaires focused on the timing of the different Meher crop stages in 2012 (planting/sowing, emergence, flowering, ripening and harvest) for a range of crops. Important questions also included the onset and distribution of rains. Road surveys were performed at five different moments during the crop season. During these surveys, the timing of the different crop growth stages was reported across the province. The advantage of this type of survey is that a much larger amount of fields can be surveyed, with less effort. The disadvantage is that there is no possibility of interviewing the farmer using detailed questionnaires.

Figure 4: Field and road surveys in West-Shewa in 2012

The results of the comparison of satellite observations with both field and road surveys showed that the DMC products were able to detect the late onset of the crop growth season, with important delays in crop emergence in the beginning of the season. This is related to a late onset of the rains, with below average rainfall in May-June 2012. The analyses over different key areas, different crops and different crop performance classes indicated that DMC NDVI and ADVI profiles corresponded to variations in timing of the crop stages as reported in the field questionnaires.

The generation of a validation dataset using field and road surveys can be evaluated positively. Crop type distribution of the sample was representative with the general crop type distribution in West-Shewa. The advantage of the road survey is that a much larger amount of fields can be surveyed with less effort. The disadvantage is that there is no possibility of performing interviews with the farmers using detailed questionnaires as for the field surveys. The combination of the two types of surveys therefore was constructive.

The validation of DMC products using MODIS time series showed that DMC and MODIS NDVI are highly correlated, both in absolute values as in temporal evolution. The correlation of DMC and MODIS Absolute Difference Vegetation Index (ADVI, vegetation status anomaly) was lower (overall R²=0.62) but for key areas where also DMC observations were available in the beginning of the season, the correlation coefficients were significantly higher (up to R²=0.92). This strengthens the hypothesis that with more valid DMC observations, thus lower cloud cover over the season, the results would improve, and that the DMC ADVI compares well with the MODIS ADVI.

From the validation work in Belgium and Ethiopia it can be concluded that further research could focus on the fusion of MODIS with DMC, trying to fill up the DMC series with MODIS imagery, without losing spatial detail. This fusion process could be based on phenological parameters. Also the relation between MODIS and DMC NDVI should be further monitored.

For a detailed description of the validation methodology and results we refer to Deliverable 51.1 (Technical Validation and Benchmarking Report).

2.3. Products for pasture monitoring

Deimos Imaging developed prototype products for pasture monitoring, based on Deimos-1 images of 2011 and MODIS-250m images from 2000-2011 for two important pasture areas in Spain, Trujillo and Pozoblanco. The products are:

- Pasture Mask: The product identifies the location of the pasture lands and those mixed land uses with a dominant occurrence of pastures. For each patch, the associated table indicates for each patch and ID, the area (both in m2 and ha) and the perimeter (in m).
To generate the pasture mask, 12 multi-temporal Deimos-1 images over Trujillo and 11 over Pozoblanco were used. A Principal Component Analysis was performed with the multi-temporal NDVI dataset for each of the locations. Using the first three components as input, a non-supervised classification was performed (ISODATA algorithm). Up to five classes were requested in the output. The classes were re-assigned to get a binary mask pasture-no pasture.

- Management Units: The product identifies the management units (plots) that will be considered as being operated by one farmer. Each unit is assigned with and ID and rivers, paths and natural vegetation patches not being used by farmers are left out. The product is generated by means of on-screen digitalization of a Deimos-1 image of the area of interest.

- Biomass: This product provides dry matter biomass values (kg dry matter/ha). As an approximation, typical mean biomass values in pasture lands were investigated in the literature. Then the mean Deimos-1 NDVI value for the pasture lands identified in the Pasture Mask product was related with this value, a direct relationship was assumed for the other NDVI values:
BIOMASS = (average dry matter availability / average NDVI value for pasture pixels in the spring image)*NDVI

- Drought Identification Test: A drought occurrence/intensity indicator is derived by comparing the current MODIS NDVI 10-daily observations with the historical reference (2000-). Four scenarios are defined depending on the current NDVI value (NDVICURRENT):
o “Better than the historical reference” if NDVICURRENT> NDVIHISTMEAN
o “Worse than the historical reference” if
NDVIHISTMEAN>NDVICURRENT> [NDVIHISTMEAN – a• NDVIHISTSD]
o “Low intensity drought” if
[NDVIHISTMEAN – a• NDVIHISTSD]> NDVICURRENT>[NDVIHISTMEAN – b• NDVIHISTSD]
o “High intensity drought” if [NDVIHISTMEAN – b• NDVIHISTSD]> NDVICURRENT
Values of a and b coefficients are defined by the insurance company, and reflects the historical risk and the price of the insurance policies. For this prototype the following values were used: a=0.5; b=1. The information is presented in the same way as the user currently request for other similar products.

- Maximum Livestock Stocking Rate: This product estimates the maximum number of cattle units that could be allocated per ha, based on the Deimos-1 derived dry matter availability (Biomass product) and assumptions from the literature. It is assumed that these units will be there until the end of the outdoor grassing period in the area.

- Temporal Availability of Grassing Feeding: Planning Supplementary Feeding: This product shows the number of weeks for which available biomass (from the Biomass product) could feed the cattle units without the inclusion of supplementary feeding. It is based on Maximum Livestock Stocking Rate and uses the applicable legal maximum cattle density (1.5 cattle units/ha in the Trujillo area and 1.2 cattle units/ha in the Pozoblanco area). In principle cattle density figures should be provided by the user.
PLANNING SUPPLEMENTARY FEEDING = BIOMASS/(cattle density*9.6Kg DM)
The outcomes are averaged for each Management Unit (plots). Only pasture pixels identified by the product Pasture mask are considered for this procedure.

- Overstocking Risk: Overstocking occurs when plants are exposed to intensive overgrazing for long periods, or without sufficient recovery periods. It can be caused by livestock in poorly managed farming, or by overpopulations of native or non-native wild animals. It reduces the usefulness, productivity and biodiversity of the land, and it is one of the sources of desertification and erosion.
The product is based on the previous products and shows the difference between demand of dry matter by cattle units and current available dry matter (Deimos-1 derived Biomass).
A worst case scenario is proposed whereby it is assumed that during the summer months no biomass is generated, and that the year has been dry since spring. If information about biomass production in summer is available, it could be included in the methodology.
Three maps are calculated, one for the 1st of July, second for the 1st of August and third for the 1st of September.
The maps show graphically the plots that may have overstocking problems by assigning risk classes to these plots (very low-low-medium-high-very high risk).

The products and service and the methodology used to generate the products is described in detail in Deliverable 32.3 (Service Prototyping Report for DS-CIS).

An example for the Trujillo region (2012) is provided in Figure 5.

Figure 5: DS-CIS products for pasture monitoring for Trujillo

The ISAC products for Pasture Monitoring were validated qualitatively against observations of pasture state, height and need of supplementary feeding made in 2012 during springtime in Trujillo and Pozoblanco by insurance adjusters from Agroseguro. In Trujillo the correlation with the field data appeared to be high with regard to pasture quantity and state. The Supplementary Feeding product, however, underestimates the needs for supplementary feeding during the winter and early spring, at least for the pastures with “high intensity drought” in spring. In Pozoblanco the amount of biomass available for grazing is also overestimated and the need for supplementary feeding underestimated in early spring. This could be related, however, to the exceptional lack of rain during autumn and winter.

Quantitative validation of the ISAC products with monthly biomass measurements for plots restricted to grazing and plots open to grazing, located in the wider area of Trujillo and Pozoblanco, showed high correlations for biomass, however, the ISAC biomass estimates were significantly lower than the measured biomass values.

Finally, Deimos-1 derived NDVI and biomass estimates were compared with NDVI and DMP derived from 10-daily MODIS products from JRC-MARS. The agreement between Deimos-1 and MODIS values was high though Deimos-1 NDVI and biomass estimates were lower than the corresponding MODIS based NDVI and DMP estimates, respectively. This may be due to the difference in spatial resolution, but also due to differences in spectral band characteristics. With regard to DMP, the differences can be explained by the fact that DMP estimates potential dry matter production without accounting explicitly for water limitation.
Further research is required to optimize the algorithms for biomass retrieval and estimation of supplementary feeding needs.

For a detailed description of the validation methodology and results we refer to Deliverable 51.1 (Technical Validation and Benchmarking Report).

2.4. Study of risk spreading (EO based insurance) versus risk reduction strategies

IIASA presented a new approach, developed within AC-CIS, explaining why and how satellite images could be used to create valuable insurance arrangements for small household farmers in developing countries. Regarding the how, it was shown that, given that one can find a satisfactory relationship between vegetation indices (such as the VHI) and crop yield responses (here done in the form of a simple regression modelling approach), it is possible to relate satellite triggering points for claim payments as in the typical weather station based insurance products, and at the same time being able to calculate premiums reflecting the specific risk situation of the household. Furthermore, IIASA investigated the possibilities to use, instead of risk spreading techniques, risk reduction strategies, here through fertilizers. Again, also here the VHI index could be a possible alternative to define the trigger event.

The methodology is summarized below. A more detailed description is provided in Deliverable 33.2 (Service Prototyping Report for AC-CIS).

- Establishment of a weather-loss relationship:
First of all an empirical relationship has to be established between NDVI, VCI (vegetation condition index), TCI (Temperature Condition Index) or VHI (Vegetation Health Index) with observed crop yields in the past.
As it is not clear which one of the indices would fit best, in principle all of them have to be tested. Different indices over a given cropping season may have to be used as well as possible transformations may have to be done. Other indicators/variables are beneficial to increase the robustness of the model. By using the most satisfactory relationship, a satellite index distribution can be constructed and via simulation a corresponding crop yield distribution can be generated. In more detail, when a given index is found, past data is used to estimate a distribution, e.g. extreme value distribution. With such a distribution and the corresponding yield-index relationship, simulations can be performed to calculate risk premiums for a possible insurance contract.
- Defining the level of risk:
it is important to determine first which kind of risk a farmer wants to address via the insurance contract. A common approach to define the trigger point for payout in case of a regional drought, is to set the VHI trigger at 35, for instance. However, it is not always obvious what this value means for a farmer as not all farmers will be affected in the same way. Additionally, with such an approach it is not possible to calculate premium payments. Therefore it is considered as more relevant to assess the risk that (for a given farmer) the minimum nutrition requirements are not met. This would provide information about the “protection” level that is needed to survive a disaster.
- Growth through the use of fertilizers:
Protection from risk is not enough for making insurance products in developing countries a success. Hence, an additional “promotion” level has to be introduced. For this analysis, it is defined as potential increase in growth through the use of fertilizers. A Mitscherlich-Baule crop response function is to used to determine the effect of varying inputs of fertilizers on crop yield. To simplify the methodology, only the N input production is considered (P is ignored).
- 3-dimensional mapping:
The crop response function is finally linked to crop and satellite indices to create a 3-dimensional map. The 3-dimensional map can be used to analyze if for given criteria (e.g. <1% probability of falling below the subsistence level) insurance and/or investments in fertilizer use is needed.

The methodology was demonstrated for Ethiopia, North Shewa (K4). This region was selected because of its importance from a poverty reduction perspective and because of the availability of sufficient input data for the modeling approach.

Based on assumptions on household size, area of land under cultivation (ha), minimum daily energy requirements (kJ) and energy of teff (kJ), average teff yield (tons/ha), nutrition supply and demand (kJ) was estimated. Teff yields (tons/ha) estimated from the satellite index model (VHI) were used to calculate the probability that one falls below the minimum nutrition level and the corresponding expected loss (tons). Using average prices of teff per ton (USD) the actuarial fair price of insurance could be estimated (no premium loadings are assumed). With this sort of insurance contract und the assumptions mentioned above, the risk of starvations would be reduced.

By decreasing the crop yields during a drought event according to the VHI regression function and doing the same for the Mitscherlich Baule crop response function, for different farmers with different inputs, such as fertilizer use and crop yield risks, the risk of falling below the minimum nutrition requirement level can be calculated.

With an optimization algorithm the required amount of N fertilizer (urea) per hectare to decrease the probability to the required level and the corresponding costs of the fertilizers was computed. These costs appeared to be much lower than the cost of insurance. Therefore, increases in yields are far more effective to get the downside risk out of the minimum threshold level. However, it is also clear that this strategy will not work for severe droughts, where water availability will be the limiting factor for growth. This would be the point where the risk reduction should be considered.

The approach for Ethiopia , North Shewa, is illustrated in Figure 6.

Figure 6: AC-CIS modelling results for North Shewa

To make this methodology operable various issues have to be kept in mind. First of all, the correlation between the VHI index and crop yield has to be established in a satisfactory manner. Hence, a future research agenda should focus on how to estimate and also accurately predict the most suitable indices. As said, for a rough analysis of drought situations, threshold approaches could be used as an alternative. Also, the relationship between fertilizer use and VHI data has to be analyzed. In this approach we adopted a version which did not include possible influences of fertilizer use and water availability. From previous studies on teff, it was found that the relationship between crop yield and water availability is more or less linear. Whether this is also the case for VHI and crop yields and fertilizers has to be investigated in the future. Additionally, as different farmers have different initial assets, such as farm size, fertilizer use and also savings etc., in a next step it is necessary to perform a detailed analysis from a livelihood perspective, i.e. calculating the opportunity costs of taking insurance or using fertilizers. This is important as small farms usually have very little money to afford such instruments and probably some subsidized products have to be established first. The analysis neglected the problems of non-functioning or the absence of markets. This was the case for insurance as well as fertilizers. For example, usually, fertilizers cannot be bought in specific amounts but just in 50 kg bags and therefore farmers might be unable to buy them. Additional research in this field has to be done in the future, also related to the price uncertainty of such products as well as to changes in crop prices during drought periods. We analyzed not only the possibility of using satellite images for prevention issues but also for promotion of development. This is especially important as farmers should have the possibility to get out of risk and poverty traps in the future by investing in growth. Without that additional dimension it is unlikely that such products will be beneficial/successfull in the long run.

2.5. Study on the comparison of NDVI and soil moisture products

In support of early warning activities and/or the development of index insurance products for drought prone areas, IIASA compared soil moisture and NDVI estimates in drought years in the Horn of Africa.

Satellite derived vegetation indices such as NDVI are commonly used for analysing vegetation health and crop production. However, another promising biophysical parameter that might have a high potential to serve as an early warning indicator is soil moisture. It is supposed to be a key parameter that links precipitation estimates and vegetation parameters. Further, soil moisture affects crop status and crop yield potential better that other climatological parameters like precipitation and temperature do. Nevertheless, none of the current early warning systems are using satellite derived soil moisture data to assess the drought situation of a region.

IIASA analysed the contemporaneous correlation between regional average NDVI (from NOAA-AVHRR) and soil moisture (from ASCAT) and compared the regional average soil moisture and NDVI estimates for the drought years 2004 and 2010/11. Correlations were found to be high. In addition, a clear time lag is seen between regional averages of NDVI and soil moisture for the respective year. This is indicating that the vegetation needs a certain time to react to increasing soil moisture.

Figure 7: Soil moisture and MODIS NDVI time series for the drought year 2010-2011

2.6. Modeled yields and yield projections for risk assessment

As described in detail in Deliverable 33.2 (Service Prototyping Report for AC-CIS), the EPIC modelling system was used for crop yield estimation in Belgium and Spain. To this aim, the system was extended with information available at the EU level and adjusted to a 1 km resolution. Globally available data were replaced by information provided by various European authorities, including EUROSTAT and the Joint Research Centre (JRC).

Crop production in Belgium was simulated and validated as a part of IIASA's Pan-European EPIC implementation, focusing on winter wheat, maize and winter rye. Belgium produces winter wheat and maize yields, which are among the highest in Europe. EPIC simulations underestimated both wheat and maize yields in the majority of the Belgian regions, but were quite successful for winter rye. Simulated yields contain significantly less variance compared to the reported yields in Belgium – which is exceptionally visible for maize – indicating that the EPIC setup reached the upper limit here and does not provide enough elasticity to reproduce variance along these high yields. This could be eliminated by improving the crop management options in the model.

Yields of winter wheat, winter rye and rainfed and irrigated maize were estimated in Spain from 1997 to 2007 in a similar way as described for the Belgian case study. In contrast to Belgium, Spain produces substantially lower rye and wheat yields. On the other hand, some regions in Spain produce maize yields, which are among the highest in Europe. EPIC’s performance significantly varies by crop and region: winter rye yields were reproduced successfully; however, the simulations were less efficient for winter wheat and maize. Nevertheless, we can conclude that EPIC reproduced crop yields in Spain better than in Belgium.

Unfortunately the resolution of the EU EPIC implementation does not provide sufficiently detailed information to derive index insurance products on the level appropriate enough to be used by insurance companies. The basic risk is considered too high.

The EPIC modeling system was also used to estimate maize yields in Ethiopia. The system performed reasonably well in the prediction of regional maize yields as well as of year-to-year variability. The modelling error can be especially attributed to insufficient information on maize management, including maize calendar, irrigation intensity and fertilization, and soil information – both in attribute and spatial domains. Some uncertainties are coming from up-scaling of the field-scale EPIC model to regional scales and data and process generalization.

In addition, a risk assessment was performed for Ethiopia (see Figure 8). The EPIC model was used to simulate maize yields from 1970 until 2009. The spatial distribution and probability of critically low yields was analyzed, in order to identify areas that would have a large benefit from index-based products. The farmers operating in these areas should be targeted for index-based insurance products. It was found that the highest probability of critically low maize yield (more than 50%) occurs in the dry region of Somali, which also belongs to the less productive regions in Ethiopia. The maize yields in the central and western part are more stable due to better soil moisture regimes and the probability of critically low yield is under 20%.

Figure 8: East-Africa (a) Long term annual maize yields (in ton/ha at 15% water content) simulated by EPIC for 1971-2009 and (b) probability of maize yield of less than 1 ton/ha

Nevertheless, even if the absolute values are uncertain, looking at relative projected changes observed under climate change is still useful. Projected yield changes of a country are potentially important for an insurance company, as these projections provide crucial information for the future decision making process. For example, if a country is projected to increase its yields based on higher inputs, it would become more attractive for higher agricultural investment. On the contrary, if the long-term yield projections are decreasing, the insurance sector might react by investing in more promising countries.

EPIC scenario runs were explored for the future crop yields and crop yield projections under SRES Scenarios A1B and B1 for Belgium and Spain. These results demonstrate effects of projected climate change on overall yield of major crops under static fertilization, irrigation and technological conditions. The picture may be altered by adaptive crop management, including crop selection (e.g. early or late cultivars) or various fertilization and irrigation strategies.

The projections with current crop and crop-management distribution indicate a slightly diiferent development in Flanders and Wallonia (as illustrated in Figure 9). While for the Flanders region only a small increase in crop yield is projected, future crop yields of the Wallonia region will increase with greater intensity for both, A1 and B1A, scenarios. In contrast, crop yields of Spain are most likely to decrease over the entire country under given crop scenario design.

Figure 9: Crop yield change for Belgium from 1991-2010 to (a) 2031-2050 and (b) 2081-2100 under A1B scenario.

Information on the long-term agricultural changes of a country is important for stakeholders and provides crucial information for the decision making process. Especially with respect to researchers, policy makers and development practitioners in developing countries working to improve food security, environmental health and rural livelihoods face many uncertainties when envisaging the future of food systems.

Information about long-term agricultural change is also important information for the insurance sector. Such information is needed to design robust insurance strategies. Future yield projections are usually used for setting up premium payments, also since recently, scenario based analysis are performed to estimate the effects of climate change with yield response. This is also closely connected with agricultural development as agricultural development is needed to estimate future exposure rates (different crop varieties etc.). Such information may be important to estimate future back-up capital as well.

Long term agricultural change was studied using the GLOBIOM model for all three ISAC focus areas, namely Belgium, Spain and Ethiopia. We applied the global version of GLOBIOM for the scenarios of Ethiopia. To analyze impacts of future land use scenarios on Belgium and Spain the European version of GLOBIOM (GLOBIOM-EU) was applied. Two scenarios were developed, namely an optimistic and pessimistic scenario. For Belgium and Spain, strong growth on the demand side until 2030 for agricultural products is expected in both scenarios (as illustrated in Figure 10 for Spain). With respect to the agricultural sector, Ethiopia is experiencing strong growth in agricultural production and productivity in 2030. However, this development is mainly oriented towards export markets, thus exposing local markets to difficulties.

Figure 10: Net trade of Spain in 2000 and 2030 for SSP1 (fast technological change) and SSP3 (slow technological change)
Potential Impact:
ISAC aims to deliver information on drought stress and agricultural change at the benefit of authorities in Europe and in developing countries, such as in Africa, and, of the insurance sector. The project covers a broad range of sites from mild risk (Belgium) over severe (Spain) to extreme risk (Ethiopia) drought conditions. These test sites are chosen to demonstrate the generic nature and potential multi-purpose usage of the services.

Technically, ISAC builds upon the developments of previous and existing GMES/Copernicus services and projects (such as geoland2) with a focus on developments and trends of future (high resolution) sensors. The ABP-CMS service for providing high resolution biophysical parameters is considered a potential candidate for future uptake in the Copernicus services, as this service potentially may form the basis and input for a variety of downstream services.

The DS-CIS and AC-CIS services are fundamentally different in nature as they target a well-known user community, namely the agricultural monitoring community both in developed as developing countries. Although the user communities addressed are different in terms of legal status (private, intergovernmental and governmental) the underlying need for managing agricultural risks is similar.

Existing contacts from previous experiences and projects served as starting point and but the user base gradually enlarged in the course of the project thanks to the continuous efforts of the ISAC partners to disseminate and promote the project results at workshops and conferences and by visiting the users.

Feedback and recommendations made by the consulted users are presented in Deliverable 51.1-2 (Reports of user meeting, feedback and recommendations).

It can be said that, in general, the proposed ISAC services are well appreciated by the users, both in the public (calamity funds, emergency and early warning) domain as by the private sector (insurance, agro-industry, farmers). This is demonstrated amongst others through the active involvement of the users, specifically in Spain. It is expected that the outputs of ISAC will form the basis for setting up downstream services with actual paying clients (insurance, re-insurance or the development sector).

The ISAC products offer clear benefits and may have an impact at different levels:
- The high temporal frequency along with the high level of detail provided is a strong aspect of the service, especially when crop monitoring at field level is envisaged. Though even at lower temporal frequency (eg. monthly), the availability of standardized high resolution products could still be interesting for agri-environmental monitoring, for instance as input for land cover mapping.
- Societal benefit:
o the operational availability of ISAC products (especially the ABP-CMS and DS-CIS products) would greatly enhance current systems for crop monitoring and early warning, resulting in a direct societal benefit.
o EO based information helps to bring more transparency into the current systems, for instance when used for control purposes (calamity fund – insurance claims – control of farmers declarations).
o Index insurance has emerged as a novel mechanism for risk management in developing countries. The proposed insurance service “Kujisaidia” (by Astrium-geo) intends to cover the risks of yield failures for small stakeholders and farmers just in the case of drought. It is designed as a tool to transfer individual risk onto the community and is driven by social work.
- Economical impact:
o it was noted by the insurance users that the DS-CIS products can be used as a proxy for the final estimation of the compensation to be paid, that should be assessed at the site by field sampling. The products allow the insurance user to greatly decrease the number of field visits and, thus, the associated costs. This is also valid for public authorities such as calamity funds.
o the generation of temporal series of these products will allow a new perspective of historical risk estimation for the insurance companies, that in turn would imply a better calculation of the insurance fees to be paid by the farmers.
o from various discussions with the agro-industry, farmers and farmers organisations, it became clear that they could also benefit from a high resolution monitoring service as a tool for decision-making or control, leading to a more efficient planning and use of resources (human and financial).
- Scientific impact:
o it is widely recognized, that specifically the ABP-CMS service provides a large amount of opportunities, beyond the scope of the ISAC project. For example the ABP-CMS could be used to improve land cover and crop mapping, combining strengths of temporal and spatial resolution.

The ISAC services and its ABP-CMS in particular, can also be considered as a proof of concept for future operational Copernicus Land services such as those based on Sentinel-2. As such ISAC can be of benefit for the EC for defining future high resolution services in terms of infrastructure requirements (e.g. data storage volumes, processing capacity,…), but also in terms of upstream user requirements (e.g. need for ortho-rectified and atmospherically corrected products as input for downstream services).

Deliverable 60.1 (Service Implementation handbook) analyzes the outputs of the service prospects, development, demonstration and benchmarking activities and formulates recommendations and guidelines for the operational implementation of the service. It further takes into account the recommendations made during the ISAC review meetings and the mid-term evaluation.

The following observations are made:
- the current services as set up by ISAC are pre-operational
- in the time frame of the ISAC project only a basic product validation could be performed. Before operational services can be rolled out more information is required on the exact error budget of the various products. Nevertheless, the developed products could already be interesting for the users at this pre-operational stage as currently they do not dispose of any ancillary information at all. Once deployed they can gradually be improved. Moreover, most downstream products are “relative” products in the sense that they compare actual estimates with the reference, which are in principle subject to the same errors. As such the individual product accuracy requirements are less stringent.
- high temporal resolution combined with high spatial resolution is crucial for the studied applications. The demonstration activities have shown that data availability at present is still an issue and that regular availability of the products could not be guaranteed. For example, data acquisitions could be very regular, but there could be, for several weeks, no suitable data acquisition due to cloudiness. This may be partially solved, when the service is migrated to use Sentinel-2, once this becomes available, however, given the characteristics of the sensors and cloudiness of some areas, it is recommended to expand the service to a true multi-sensor service integrating sensor data from different sensors such as Sentinel-2, LDCM, DEIMOS/DMC.
- the automatic screening of clouds, especially for what concerns haze, fog and flight lines, is an important challenge, which could only be partially solved during the ISAC project. Research should be dedicated to fully automate this.
- the products offered by the service are single scenes with geometrically and atmospherically corrected and cloud masked vegetation indices (NDVI) and biophysical parameters (fAPAR, LAI, fCover), however to increase utility and usefulness of the products, research is needed to assure:
o Sub-pixel geometric accuracy (especially in a multi-sensor context, with different input accuracies, this becomes important
o Automatic cloud detection (incl. haze, fog, flight lines,…)
o A suitable compositing method
- availability of SWIR or thermal bands on future high resolution sensors may improve atmospheric correction and cloud detection. This should be accounted for when planning research activities.
- in principle the APB-CMS service could be provided on a global scale, however, the quantity of data and data processing power needed, will be significant (order of petabytes of storage needed), and it can be questioned if this would be feasible in the short term using a centralized data management system; it may be more realistic to foresee this service on a local and regional scale (eg. Europe), which is most likely better manageable and which would already serve the basic needs of a range of users. In any case distributed storage and processing (cloud computing) should be studied and considered when upscaling this service.
- to enable correct calculation of the cost price of a service based on high resolution satellite images, data policies should become clearer, as data cost may be significant due to the quantity of satellite data needed for this service. If costs are too high, this may be a limitation for the (commercial) success of the service.

For the downstream services, DS-CIS and to a lesser extent AC-CIS (as it is different in nature: it does not have a regular delivery time cycle and provides services more on an ad-hoc basis), an exploitation plan has been prepared (Deliverable 70.6). The exploitation plan builds upon existing collaborations of the ISAC partners but it also includes a range of new opportunities resulting from numerous discussions with potential users throughout the project.

In view of future exploitation, the services have been regrouped thematically and according to the type of users they address:
- Pasture monitoring services (DS-CIS)
- Crop monitoring services (DS-CIS and AC-CIS)
- Insurance services for Developing Countries (DS-CIS and AC-CIS)

The proposed services are no exact copies of the DS-CIS or AC-CIS services developed by ISAC. They rather build upon these services. In addition, they propose customized services to meet specific information needs or provide solutions for the shortcomings identified from user consultations.

The main goals of the exploitation plan are:
- To identify the types of users and customers for the ISAC products
- To identify the target markets
- To establish mechanisms for an effective exploitation of the products

It provides specific information on the products, actions required for implementation, costs and a business plan for the operational ISAC services.

List of Websites:

http://www.gmes-isac.info

final1-isac-final-report-draft-v0-2.pdf