FAO-EOSTAT: EO data for Official Agricultural Statistics

Building capacity in Senegal, Uganda, Afghanistan, Lesotho.

The EOSTAT project kicked off in 2020 under the leadership of the  Office of the Chief Statistician  of FAO with the aim of building capacity in National Statistics Offices (NSO's) in developing countries in the operational use of EO data for the production of agricultural statistics.

Nowadays the conditions for the use of EO data for agricultural statistics are very favourable.

The generous availability of free and open high resolution EO data stemming from the Copernicus program which adds to the long-standing LANDSAT series at medium resolution offer an unprecedented possibility to detect crop phenology and spectral traits and derive accurate agricultural statistics.

Copernicus Space Component: Long term continuity with free and open access

The expansion of cloud storage and computing capabilities, and the rise of machine learning and artificial intelligence has opened the door to high flexibility and alternatives for deploying low-cost infrastructures and automation.

Commercial and open and free EO data and cloud services providers

Despite all, the actual uptake of EO data for operational use in NSO's is still very low globally, especially in developing countries, due to a series of technical, financial and administrative barriers.


Main barriers

A series of challenges are found on the innovative path of integrating EO data and statistics within NSO's with a view on operational use for official agricultural statistics.

Cloud masking

1) The complexity of image pre-processing (including image atmospheric correction and cloud masking), to more advanced temporal compositing and gap-filling which are required to derive analysis-ready data (ARD) also called data-cubes. Such operations are not trivial and require specialized expertise in Remote Sensing and big data handling.

Top figure, example of GPS traces that contain more than parcels boundaries. Bottom figure, challenge to localize the surveyed parcels in the statistical database due to the fact that parcels are localized by geo-points (i.e. a single GPS coordinate)

2) Low availability and quality of in-situ data due to I) the high cost of surveys and ii) the scarce integration of EO at the survey design level, and the inconsistent use of georeferencing methods in the field.

Nevertheless, crop classification models require a high volume of ground-truth data (in-situ data) for calibration and validation purposes. For instances, the recommendations from Jecam, SIGMA and Sen2Agri frameworks consider 100-150 samples should be collected per main crop types within each agro-ecological zone. Requirements for training data is even higher for deep learning models.

3) Models calibrated in one area and for a given agricultural season perform poorly if used to predict crops in different areas and over different periods. Research is still ongoing on transfer learning. Hence, new ground truth data needs to be collected for every new agricultural season fpr accurate results..

Google Earth Engine Interface, Code editor for Javascript.

4) Lack of user friendliness of existing EO platforms which provide access to data archives and to modelling environments. High level of programming skills are required

5) Data confidentiality!


EOSTAT: breaking through the barriers

EOSTAT aims at overcoming the identified barriers through technical innovation, training and change management.

EOSTAT operates in close collaboration with NSO's in countries and has established a partnership with key working groups (UCL and UN BIG Data group and ESA) and a dedicated country engagement programme.

International partners and NSO's currently collaborating with FAO under the EOSTAT programme

EOSTAT pilot Countries

Currently, EOSTAT collaborates with 6 Countries Senegal, Uganda, Afghanistan, Lesotho.


How does FAO overcome such barriers and fill existing capacity gaps through the EOSTAT project?

1) FAO delivers a full solutions for automatic preprocessing of EO images and analysis though graphic user interface such as Sen2Agri or through EOSTAT CropMapper app developed on top of Google Earth Engine.

Sen2-Agri system

The Sen2-Agri system is an operational standalone processing system generating agricultural products from Sentinel-2 (A&B) and Landsat 8 time series along the growing season. These different products consist of:

  • monthly cloud free composites
  • biophisical indicators (NDVI, LAI, fPAR)
  • Crop mask, several along the season
  • Crop type mask, seasonal

EOSTAT CropMapper

The EOSTAT CropMapper software, developed using Google stack, using GEE backend. Allows for supervised, and un-supervised classification. Currently being implemented in Afghanistan. Allows for the production of national and district level crop acreage statstics

2) FAO provides a secure and low cost cloud based solution, through the collaboration with the UN Global platform*

UN Global Platform solution diagram for the EO-STAT project in Senega

UN Global platform provides storage and computing power and ensures an optimal performance and low running costs. It is a scalable solution and ensures secure hosting of country data. The UN global platform serves also as sharing of trusted data, methods and algorythms and allows for an easy deployment of Sen2-Agri tool box.

3) FAO delivers customized training programs covering key aspects of the integration of EO data and statistics: best practices for georeferencing in-situ data and optimization of survery design using EO data; deployment and use of Sen2Agri and CropMapper; use of Google GEE and Google Colab for crop type mapping.

EOSTAT online Webinars

4) FAO develops novel methodologies for crop mapping and land cover mapping that cope with scarcity of ground truth data, and uses EO data to optimize survey designs and protocols.

5) FAO facilitatation of the access to ground truth data supported by the AGRIS programme (50 x 2030)

 

EOSTAT key activities

Activities are carried out in consultation and jointly with the NSO's in countries and-or with the line Ministries concerned with agricultural statistics and Earth Observations. Technical assistance activities are articulated through 3 main components: the in situ data, the EO data and the Training.

EOSTAT main areas of work


EOSTAT pilots in Countries


 

      Geospatial statistical strategies

          Copernicus Space Component: Long term continuity with free and open access

          Commercial and open and free EO data and cloud services providers

          Cloud masking

          Top figure, example of GPS traces that contain more than parcels boundaries. Bottom figure, challenge to localize the surveyed parcels in the statistical database due to the fact that parcels are localized by geo-points (i.e. a single GPS coordinate)

          Google Earth Engine Interface, Code editor for Javascript.

          International partners and NSO's currently collaborating with FAO under the EOSTAT programme

          Sen2-Agri system

          EOSTAT CropMapper

          UN Global Platform solution diagram for the EO-STAT project in Senega

          EOSTAT main areas of work