Aarhus University Seal

Machine learning creates ways to better assess the suitability of agricultural land

A comparison of maps of land suitability for specialty crops produced using machine learning and a mechanistic crop model gives rise to rethinking the concept of “suitability” and perhaps even replacing it with the two concepts: ecological suitability and socioeconomic suitability.

[Translate to English:] Foto: Colourbox

Adverse weather, pests, diseases, laws and regulations, and fluctuating market prices are just some of the factors that affect farmers around the world every day. Therefore, it is very important for farmers to know if their land is suitable for the crops they plan to grow. There are a number of different methods for this, and the ever-increasing amount of available geographical data makes it both faster and easier to assess the suitability of the land. But which methods should be used? And what type of suitability do they actually assess? A group of researchers have now investigated this by comparing land suitability maps using two different methods: 

  1. Machine learning with the algorithm Maxent
  2. The mechanistic crop model ECOCROP 

“In practice, very different methods have been used for land evaluation. Some have focused on the climate, others socio-economic conditions, others again have focused on crop growth models, soil systems, and the environment. What they all have in common is that they are heavily dependent on expert knowledge and are very time-consuming,” explains postdoc Anders Bjørn Møller from the Department of Agroecology at Aarhus University.

Automated methods

There are not always experts or large amounts of time available for mapping, which is why there is a great deal of interest in automated methods.

“One of the most widely used mechanistic methods for assessing land suitability is the ECOCROP model (ed. Ecological Crop Requirements). Another very common automated method is machine learning based on land use,” explains Anders Bjørn Møller.

The ECOCROP model is based on the ECOCROP database, developed by the Food and Agriculture Organisation of the United Nations (FAO) and used to determine the suitability of a particular crop in a specific environment. Machine learning is also a widespread method of assessing the suitability of the land. The most common algorithm used is Maxent, which uses all available data on natural and socio-economic factors. This approach is based on the assumption that farmers grow crops where the best growing conditions exist. 

“Machine learning is very different from ECOCROP. Because while ECOCROP focuses on growing conditions, machine learning can potentially use any variable that we as researchers find relevant, including socio-economic conditions,” explains Anders Bjørn Møller.

Should be used for different purposes

Both models can be used to assess the land's suitability for cultivation. However, according to the researchers, the differences between the two methods are large, and their results shed light on different aspects of the land suitability, which is why, according to the researchers, a reassessment of the concept of land suitability will be necessary.

“We have compared maps of theland suitability for a total of 41 specialty crops such as potatoes, carrots, apples, and onions. The maps are made with the two methods, and just as we expected, there were big differences in the results with the two methods. Machine learning could often identify the areas where farmers typically grow the crops. The accuracy was especially high for potatoes and carrots. On the other hand, ECOCROP showed a low suitability for growing potatoes and carrots in the areas where they are most widespread,” says Anders Bjørn Møller.

The difference does not mean that machine learning is thus a better method than ECOCROP, but rather it shows that the two methods measure on different factors.

“ECOCROP simply finds the area that has the best growth conditions for e.g., potatoes or carrots. There are good growing conditions in the eastern part of the country, however, more potatoes and carrots are grown on the sandy soils in the western part of the country. This can be done due to a lot of technological and socio-economic factors, including irrigation and fertiliser, which machine learning automatically takes into account,” explains Anders Bjørn Møller 

The concept of land suitability for reassessment

Although the results for the two methods are vastly different and can be used for different purposes, they are both used to study the same concept, namely the suitability of the land, and according to the researchers this is problematic. But instead of highlighting one method as better or more accurate than the other, we should instead reconsider the concept of suitability, if you asks Anders Bjørn Møller.

“Using these two models for the same purpose is simply not possible, because the results are often completely different. Instead, as a researcher, you should make clear to yourself what you want to map, what it is to be used for, and choose the method that best suits these purposes. If you want to know if it is you will benefit financially from growing a certain crop somewhere, then machine learning is the way forward, but if you would rather know if the crop has good growth conditions, then ECOCROP is a safer bet. We believe that the concept of suitability should be divided into at least two concepts, namely ecological suitability and socio-economic suitability. It will help make the research results more transparent and useful in the future.”

In other words, according to the researchers here, it is important that researchers in the future consider the purpose of their research and what kind of suitability they aim for before they decide which methods to use.

Additional information
We strive to ensure that all our articles live up to the Danish universities' principles for good research communication (scroll down to find the English version on the web-site). Because of this the article will be supplemented with the following information:
Study type:Experimental and theoretical discussion
Collaborators:Wageningen University and Research, Holland - Close collaboration, has especially helped to develop the study, discuss the results, and write the article
Funding:The Innovation Fund Denmark funded this research as part of the project ProvenanceDK (grant number 6150.00035B). The Agricultural School of Nordsjælland Foundation facilitated the research through a travel grant. V.L. Mulder is member of the research consortium GLADSOILMAP supported by LE STUDIUM Loire Valley Institute for Advanced Studies through its LE STUDIUM Research Consortium Program
Conflict of interest:None
Read more:The publication ”Can We Use Machine Learning for Agricultural Land Suitability Assessment?” is published in the journal Agronomy. It is written by Anders Bjørn Møller, Vera Leatitia Mulder, Gerard B. M. Heuvelink, Niels Mark Jacobsen, and Mogens Humlekrog Greve
Contact:Post Doc Anders Bjørn Møller, Department of Agroecology, Aarhus University. Mail: anbm@agro.au.dk