The world population keeps increasing; where there were about 6 billion people on Earth in the year 2000, in 2020 that number will be nearly 8 billion. The Earth’s surface on the other hand, of course, remains the same. With this growing world population, less land remains available for agriculture. So how do we make sure that the world is provided with sufficient food in the future? One of the answers to this challenge is precision farming.
Van den Borne Aardappelen, a potato farmer in the Dutch village of Reusel, is one of the most progressive companies in the field of precision farming in The Netherlands. Early 2008, the company began using sensors to measure chlorophyll in the potato plant, on which the issue of the fertilizer was adjusted. This resulted in the decrease in the use of unnecessary fertilizer and potatoes being of better quality. In the subsequent years, new state of the art technology was acquired. Cameras, soil sensors, and drones made it possible to measure all sorts of variables that have an influence on the growth of the potato plants.
Large scale statistical analysis
This collection of data creates a rich dataset containing information on 167 different plots of Van den Borne Aardappelen. What is the conductivity of the soil? What is the composition of the bottom? How moist is the soil? How well do the plants grow? By using this information, the shortest route is always driven, and the farmers can very accurately determine where to plant, harvest, and fertilize the potato plants. This saves water, fertilizer, pesticides, a lot of diesel, and also ensures the maximum yield of potatoes.
Up to now Van den Borne Aardappelen has only been using fragmented and small sections of the dataset. In addition, the dataset has not yet been used as a single entity in order to carry out a large scale data analysis. This in fact was the major challenge that TU/e student Puck Mulders of the Department of Mathematics and Computer Science worked on during her internship at Van den Borne Aardappelen. Which variables cause the difference in yield between plots? What is the interaction between the variables? And what is the ideal combination of variables to create an optimal harvest?
Explaining differences between plots
The goal of the internship was making a model that explains the differences between the plots of Van den Borne Aardappelen. First, for each plot a function has been estimated for the haulm weight, root weight, and other features of the potato plant. Think for example of an estimation of the haulm weight during the growing season of the potato. Then clusters of plots with similar features were made. This clustering clearly showed a significantly higher yield per hectare for one group of plots compared to another group. On the basis of these clusters Puck Mulders examined whether properties of the lots, fertilizer or the way of planting could explain these differences in yield.
An example is shown in this graph, which displays the haulm weight of plots with a certain soil composition. What seemed to be the case? Cluster 1 with the highest haulm weight structurally delivers the highest yield. And another example, the model has shown that an increased planting distance combined with larger seed potatoes structurally ensures a higher yield.
With the assistance of this model, it will become possible to make predictions about the yield when changing certain variables. And consequently determine the optimal combination of variables to maximize the yield of a certain cluster of plots. By drawing such conclusions from the model, the yield map will not show as much red areas as in the image below (which indicates low yield) anymore in the future. The model ensures that the land and resources of Van den Borne Aardappelen are used optimally, achieving maximum efficiency with the smallest possible impact on the environment.
And does it work? Yes! Since the Van den Borne brothers took over the company from their father, the yield increased from 46 tons to 53 tons of potatoes. Based on the new models, they estimate that the optimal yield is 75 tons, meaning there is still room for growth.