Agriculture will remain the most important economic activity in Africa for at least the next 30 years.
The sector contributes more than 30 per cent of the GDP in most African countries. It also employs more than 70 per cent of the workforce in most developing countries.
The sector is, however, under intense pressure from climate change and cultural practices of land use (sub-divisions) that have consistently reduced farming acreage. Unless we apply scientific methods to improve productivity, we are at risk of food insecurity.
Gro Intelligence, an agricultural research and data company located in New York and Nairobi, has released a US corn yield model.
With it, those who have an interest in anticipating the size of the US corn crop can do so more effectively.
Gro will publish the results of the model for free on their website. This summary grossly oversimplifies the complex process of human learning and artificial intelligence (AI) training that produced the Gro model.
The Gro modeling team was composed of members of the company’s geospatial, data, and software development teams.
The company decided to model US corn first, because it is the best-understood crop in the world and therefore has the best available data coverage.
They used advanced computer-assisted statistical techniques to choose the following sources from among the broad array of data in libraries:
Normalised difference vegetation index (NDVI) — a measure of relative greenness, in this case collected remotely by satellite, used to assess the health of plants in a given area; land surface temperature (LST) - a measure of temperature, a key variable for determining crop progress, again from satellites.
Others are Gridded soil survey geographic data (gSSURGO) — information from the US Department of Agriculture (USDA) on each parcel of land’s soil type, which allows the modeling team to evaluate the significance of greenness and temperature; USDA crop conditions - survey-based subjective appraisals of the corn crop’s status that come out weekly during the growing season; Cropland data layer (CDL) - USDA’s appraisal of the crop usage for each piece of land in the US.
They took this mass of data and fed it into machine learning algorithms. The machines compared their results with the known values for the final yields between 2001 and 2015, and iterated toward the best possible use of the data to make an accurate estimate.
Then Gro’s agronomic experts properly weighted the machines’ results over the course of the growing season.
This created a hybrid process/statistical method that Gro ran live in 2016. It generated results that outperformed the models that are currently widely used.
Then they added the data for 2016 to the machines’ input and improved the prospective performance of the model.
All this has great importance for agriculture globally owing to various reasons. First, it is available for free, whereas previously models of this nature were generally expensive.
Second, the open approach used by Gro creates a new standard for community involvement and will result in continuous improvement.
Third, the methods being used here will generalise beyond the US; the methods will generalise beyond corn to crops that, while critical to their consumers, have largely been ignored due to their small financial size.
Gro presents Africa with a great opportunity to access knowledge that has been hitherto in the hands of few who could afford the high price of accessing it.
For this to happen, governments must deploy extension officers to make the models appetising to the farming community that may not interpret its usefulness.
I have no doubt that countries like Kenya could leverage the emerging models to predict crop yield and make farming more attractive to the youth.
It will also provide a mechanism for planning to enhance food security.