The Center for Data Innovation spoke with Catherine Nakalembe, Africa program lead at NASA Harvest. NASA Harvest is NASA’s food security and agriculture program led and implemented by the University of Maryland. Nakalembe, who leads the program’s activities in Africa, is helping countries across the African continent build systems to monitor crops based on NASA and the European Space Agency’s free satellite data. Nakalembe discussed how she champions the use of satellite data to help countries make decisions related to food security more quickly and with a deeper evidence-base.
Hodan Omaar: Why is monitoring agriculture so important in Africa, how is NASA helping in this effort, and what role does machine learning play?
Catherine Nakalembe: Monitoring agriculture is the most important factor in understanding what kind of progress a farmer, region, or country is making and how they can improve, especially when agriculture is the livelihood of a large part of the population. Yet few individuals or institutions in Africa have invested in monitoring. You will find that many ministries of agriculture in sub-saharan Africa have very well developed research organizations that look at things like seed development and distribution, but they do not monitor the yield of these crops.
This is primarily because monitoring is very expensive and many institutions don’t have the funding to do it. It’s also extremely complicated; imagine monitoring individual fields every two weeks for a whole country with only traditional tools. It becomes even more difficult in complex landscapes where farmers are planting different crops, at different times, using different methods.
This is where remote sensing can really help. Using data from satellite images to monitor crops consistently and continuously, offers ways to characterize and quantify cropland and production which can then be presented to decision makers.
With the complexity in smallholder agricultural landscapes, machine learning is vital to extracting useful data on many different variables at an appropriate level of regional granularity. Which crop types have the highest yield in each season? Which fertilizer works best? What impact does irrigating the land have? NASA Harvest is really focused on developing the underlying algorithms, datasets, and satellite systems that can be used to monitor agriculture and inform food security decisions.
Omaar: Aside from satellite images, what types of field data do you need to get a fuller picture of regional crop production and how do you get this? Which of these data are fed to ML algorithms to predict yield and which of these data are used for contextual human decision making?
Nakalembe: This is a good question. One part of the framework for national agriculture monitoring is working with on-the-ground agents to collect data using, for example, the Open Data Kit. Any agent with an Android phone can install this software and use it to collect ground data on things such as crop type, fertilizer used, etc. This provides contextual information that helps us understand the variability between fields but is also really useful for training algorithms.
For instance, we have ground points of where locusts have been seen in East Africa. These locusts can destroy crops that could feed more than 30,000 people at a time and most normal crops can’t recover from such extensive damage. With remote sensing we should expect to see a signal that there is a sharp decline in crop conditions from the satellite images of these fields. But we haven’t seen that; the satellite data looks the same as every season over the past few years.
It could be that the on-the-ground control operations to mitigate crop damage are working, but this level of contextual understanding can only be obtained if we can get photos of where locusts have invaded and evidence of any damage to crops. If there is evidence of damage, we can use this as training data to find other areas that present the same features in satellite data.
The primary focus of my work is improving baseline datasets, including data on what type of crops are present and where they are, by harnessing ground data. It is my hope that we can work with more on-the-ground agents in a systemized way that leverages existing government structures. Working with existing government agents who might be based at districts, counties, or villages, and training them to provide georeferenced ground data will help at the reporting stage when we are trying to inform decision makers. When I created a drought report in 2015 using satellite data, the real magic came when I was able to show reference pictures of what the affected areas on the report actually looked like on the ground.
Omaar: If NASA technology shows that a small farmer needs additional minerals or pesticides, but the farmer doesn’t have the resources to implement the recommendations, how do you measure the value of informing agronomic practices?
Nakalembe: For now, there is no way to effectively measure the value to farmers, but that hasn’t been the primary focus of the NASA Harvest program; it requires a whole other level of thinking. For the longest time, the focus of my work in Africa has been trying to get these tools into the hands of people who work in ministries of agriculture, to integrate this earth data into their reporting and monitoring.
But once they have this information, packaging it and translating it in a way that informs farmers on what to grow, when to grow, and when to harvest is a different ball game. I think doing this requires engagement with smaller companies who would be able to leverage their relationships with farmers. These SMEs are more likely to be able to encourage the farmers to employ better practices because they are the market who is buying from them.
For example, if an SME who buys crops from a farmer tells them the conditions are better for planting sorghum rather than maize this month, the farmer will be more incentivized to plant higher yield seeds to get more value from them. Governments could then support this through subsidies or tax breaks for the SMEs, which I think might be more sustainable in the long term than fully funded state programs that just give the farmers the seeds they should plant.
I’d be really interested in doing something at this level and showing the value to farmers. It’s easier to do in higher resource countries such as the United States, where larger farms can use instruments like automated sensors that turn off the water once the soil moisture is perfect. But in countries where farmers don’t have these kinds of instruments and you can’t really communicate with them, it’s a lot harder. Although, some of the extension groups I work with have created co-operations to educate each other about what is working for them so there is potential there.
Omaar: Your work on strengthening decision making in African agriculture and food security is part of a larger international G20-endorsed program: the Group on Earth Observations Global Agricultural Monitoring Initiative (GEOGLAM). How do data and insights from different complex national and regional food systems scale up to global policies and why do we need global policies on food systems?
Nakalembe: The existence of GEOGLAM really started with the food price crisis in 2008. Price volatility between bigger producer countries caused major frictions in the global market. One of the initial ideas behind the GEOGLAM initiative was to establish crop monitors that have a global focus to show that if there is a crop failure in country A and a surplus in country B there really shouldn’t be huge price fluctuations that could put many people at significant risk of malnutrition.
Such failures can even happen within a country. Tanzania enforced an export ban a few years ago because there was a particular region affected by a drought, but this region does not contribute much to the national grain supply. In fact, Tanzania had a national grain surplus so the ban caused the country to lose a significant portion of grain that year because they couldn’t export it, which then impacted the entire market. This was all because there was no clear understanding of which region produces what, how much each region contributes to national supply, or how much can be safely exported. Such bans often negatively affect the incomes of farmers and traders by hindering their access to regional and international markets.
If we know what’s happening at a global scale we can make regional and national strategies. That’s what GEOGLAM is concerned with: facilitating cross coordination and influencing policy at a much higher level.
Omaar: Where are we with remote sensing today, and where do you see the field in the next 5-10 years?
Nakalembe: The level of maturity we see today in remote sensing has really increased compared to the last 5 years. There is greater access to data, better computing power, and more sophisticated algorithms. But some countries in Africa still lack the robust IT infrastructure needed to make use of the current capabilities in remote sensing.
In countries like the United States, farmers have access to the resources they need to report what they are planting and what methodologies they are employing. Remote sensing can therefore be effectively used to verify if they are doing what they said they did and ML algorithms can process much of this data to reveal valuable lessons. But some low resource countries are unfortunately still using paper surveys and lack basic computers.
If these countries could invest in IT and train their human resources, they could definitively answer very basic questions about how much food they have, the percentage of the country suffering from drought, and the level of assistance they do or do not need. Over the next 10 years we just need to work on getting these countries to be able to employ monitoring methods properly to truly propel these countries forward.