Philippe Stoop, Director of Research & Innovation at itk, took part in the AgroTIC webinar “Measuring and estimating yield: what contributions and uses of digital tools? ». During the webinar, several topics were discussed.

Decision support tools for agriculture and breeding: itk’s core business

Itk is an SME based in Montpellier with a hundred people whose job is to develop agronomic decision support tools for agriculture and breeding.

Our goal is to help farmers optimize their technical itineraries from the point of view of productivity. Moreover, it is to minimize their environmental impact in particular with an optimal use of inputs (irrigation, fertilization, and pesticide). Our core business is the development of mechanistic crop models.

Our platforms are intended for farmers but we do not address them directly. The services we develop are distributed by cooperatives. We work a lot with the Winfield group in the United States.

As far as crop production is concerned, we have two service platforms:

  • Vintel for perennial agriculture (viticulture, fruit growing). In the case of these crops, we are not yet on the yield forecast even though we are working on it. Vintel is more downstream. We advise crops where the yield potential is determined by the way the crop grown, its size, its management… What we bring to the wine grower and the fruit grower is the assurance that the water supply of his crop and its fertilization will allow him to reach quality yields. We enable him to be sure of his objectives but Vintel does not predict them as we can do with field crops.
  • Cropwin for field crops (wheat, corn, soybean). In this case, it is indeed the modeling and prediction of yield and maturity date that will be the very driving force of the tool. Based on our crop models, we will first be able to predict the development curve of the crop and its final yield. This will also allow us to know day after day what its photosynthesis potential will be. Depending on the climate, the soil in which it grows and its biomass prediction potential, we will be able to determine its water requirements (if we need irrigation advice) and its fertilizer requirements (for fertilization). If we add to this epidemiological models, it will also allow us to determine its needs in pesticides, for diseases and pests.


Example of a dashboard provided on the CropWin tool

Webinar screenshot – CropWin® platform demonstration


With the Cropwin platform, farmers can check the yield potential of their crop at any time. They can also have a presentation of the factors for which they may need to make an intervention in the next few days, whether for irrigation or fertilization.


The mechanistic model and its advantages in extreme weather conditions

Concerning the technologies used, the main input is modelling and more precisely mechanistic models. That is to say models based on research work on crop ecophysiology and crop behaviour.

The interest in this type of approach is that the models have been developed from field data but also from research in controlled conditions where crops have been tested outside their usual comfort zone. As a result, these models can be used to predict crop behaviour in extreme weather events with a high degree of reliability. This is really an advantage that is becoming more and more common now with the problems that climate change is causing. Farmers are increasingly facing conditions that they have never encountered in previous years, which statistical models based on historical records take less account of.
For the same reasons, these models are very robust and can easily be adapted to new contexts, especially if they are to be used in a new country. We recently had requests from Ukraine for our wheat model, which had been validated so far only in Western Europe. With this type of model, which deals with the physiological constants of the crop, adaptation to a new geographical context is easier. In general, all that is needed is a calibration on some historical data provided by our partners to easily readjust the model to this new context. They are also models that lend themselves easily to multi-scale work. Maximum precision is obtained on services for farmers where the latter can provide us with a lot of information on their varieties, the sowing dates… The model alone will bring the most precision. For precision work at the scale of a catchment area or a country, or even a continent, we will have to work with input data. Mechanistic models can do this by compensating for the lack of input data by assimilation with remotely sensed data.

Also, an important element is the use of realistic weather scenarios over the entire campaign. This is a growing field, which has made great progress in recent years and can now be used reliably, although not for all types of models and variables. As far as yields are concerned, we are now entering a phase where seasonal climate prediction models allow for fairly good accuracy. We saw an example during our collaboration with the Winfield cooperative in the United States last year. Indeed, as soon as the corn was semi-sown, we were able to alert farmers to the fact that, given the “El Nino” phenomenon predicted in the climate scenarios, yield difficulties were to be expected. In fact, it didn’t show up in the USDA yield forecast until several months later, when the phenomenon appeared in the field.

With these phenomena of complementary data to modeling, weather scenarios or remote sensing, our goal is to work as an integrator. We must be able to identify the best players to run our models reliably, according to the business issues that our customers submit to us. In the same way, for remote sensing, we work with several different players to adapt to our customers. Indeed, as was the case for our client Winfield, we have performance forecasting requests with people who already have access to remote sensing data. Our goal is then to be able to integrate the data already used by our clients into our models. That’s what we do at Winfield, in partnership with Geosys. Geosys works in the remote sensing part and we, at itk, in the forecasting part of the R7 offer of the cooperative.

On other subjects we can also work with biophysical indicators calculated by operators such as Airbus from satellite images. The objective is to be able to follow the progress in all these areas while enriching our own models.


Paths of progress

In terms of yield forecasting the major lock is being able to produce early in the season on reliable crops without having a lot of crop information. Techniques for automatic remote sensing of crops, thanks to the improved availability of high temporal frequency remote sensing supplies, are currently major paths of progress.

Another important player may be to lift the availability of satellite images depending on weather conditions. Currently, all techniques based on the exploitation of satellite images in the optical and infrared range are limited by cloud cover. CIRAD has opened up some interesting new possibilities.

Concerning the modelling track, what is interesting is to lift the locks on the valuation of yield maps. Currently, yield maps produced by sensors are under-exploited because of the difficulty in interpreting them. This is where the models seem very promising, when used in “reverse” mode. Based on a history of yield maps provided by the sensors of the harvesters crossed with our models, the idea is to map the soil. This will make it possible to search, by model version, for other quantities of soil that may stick, and to better explain the heterogeneities found in the field and find out how to correct these heterogeneities.

Also, another important element is to add value by coupling this yield prediction work with the optimization of collection and logistics.