Home Articles The future lies in machine learning

The future lies in machine learning



Rogerio Bonifacio, Head, Geospatial Analysis Unit, UN World Food Programme, explains how the agency is utilizing geospatial data and information

Could you elaborate on the mission of the World Food Programme and how it is utilizing geospatial technologies?

The World Food Programme is involved in humanitarian systems. We have wide presence in more than 75 countries across the world. And we assist around 75-85 million beneficiaries across the world. Our operations range from very small to very large in scale, like in South Sudan and Ethiopia. The usage of earth observation data has grown manifold in recent years, with more focus on medium- and low-resolution data streams. This is mainly because these are able to provide wide spatial coverage, and their high temporal frequency allow us to keep track of hazards, such as, droughts that tend to have large impact on our operations — both in terms of new beneficiaries that may arise from such events and keeping up with the expected rise in beneficiaries at the end of the season.

What is the UN mandate for use of spatial technologies, especially in agriculture and food?

We work in close collaboration with the Food and Agriculture Organization (FAO), the International Fund for Agricultural Development (IFAD) and other agencies. Each agency has a separate, but coordinated, usage of earth observation data. So, despite having related mandates, we have different requirements and operational aspects. As a result, we tend to make complementary, but slightly different use of the information. The World Food Programme is an operational agency. It is not in the business of providing guaranteed quality global datasets. We provide information that is operationally relevant and useful. And we make it available on a ‘best effort’ basis. We share the data we generate if it serves the purposes of other users downstream, but we don’t have a mandate to produce specific types of information. Our information is freely available on our websites in terms of reports that summarize the development of growing seasonal crops. But we also provide agri information from our data resolution platforms.

How is earth observation data being used in agriculture and land use activities by the UN?

One of the major things we do is in terms of early warnings to have some before time estimates of the likely impact on agricultural production. We also use long-term data to support our climate services related activities. And for the planning of our interventions, we also have to factor in hazard frequencies, land degradation status and a variety of other information streams that are generated from earth observation data.

What are the benefits of EO data compared to other information sources?

There are times when anything else would not allow us to obtain the same pattern of information in a timely manner. No other data would allow us to monitor an entire continent in a near-real-time basis. Occasionally, ground station data can be used. But, then you face accessibility issues and experience difficulty of organizing such data in consistent datasets.

How is the UN collaborating with private sector companies for data collection?

The private sector has an important contribution to make to the UN, which is not limited only to earth observation data. But, if we talk specifically about EO, we have now stepped into a stage where we are trying to prove the usefulness of the information we have. So, a lot of new activity is going on in this regard. However, since the sector is in a state of flux, it is a bit difficult to identify which types of systems would be more relevant three years from now. Nonetheless, once we have a more established industry, we will have to make the case for the use of earth observation data.

Each agency has a separate, but coordinated usage of earth observation data. So, despite having related mandates, we have different requirements and operational aspects.

Cost is a crucial factor and hopefully it can be cut down. But there are also unresolved questions that deal with the extraction of information from huge datasets. We, as an institution, do not have the capacity to deal with these large volumes of data. We don’t have the modern analytics to extract the relevant information, so we need to work together both on the economic access, and on the storage and processing of the information, and ultimately, the digestion of all the raw data into usable end-user products. That’s going to be the challenge and it is not a simple one.

Are the available datasets enough or are we lacking on the analysis part?

If you talk about the amount and variety of data, it is something that we need to resolve. We have to review our use cases; we have to review different constellations because they do differ, even if so slightly. This is a considerable amount of work involved. First, we need to identify which constellation and services are more suitable to each of the use cases. Because we have very diverse use cases — from very small geographically restricted applications to very wide geographical scope requirements — it is not even clear which of the services will be best suited.

I believe that the future is in the machine learning side of the analysis because of the sheer volume of data. We need automated system that can translate vast quantities of data into easily digestible information. The ground is quite fertile, both on the research and on the operational front. I foresee exciting times ahead.