Interpretability of machine learning models
21 January 2021
The development of machine learning models that process large amounts of data greatly improves the performance of predictions.
Unlocking the value of data in outdoor agriculture is as much about lowering technology costs as it is about finding sound go-to-market strategies
Throughout their lives, farmers benefit from a limited number of crop cycles and seasons (~60 years for farmers spending their life farming) to understand the various agronomic dynamics of their parcels, acquire experience and adapt their practices over time. Against this limited number of trials stands an extensive range of parameters to leverage in order to increase yields: irrigation level; fertilizers’ rate; phytosanitary products’ rate; soil compaction; sun exposure; synergies between crops; etc. The infinite number of combinations makes decision making in farming a very complex task. On top of this complexity, farmers face pressure to produce more to satisfy increasing food demand (e.g. 8.5bn of individuals to feed by 2030, Food and Agricultural Organization) while dealing with an increasing number of challenges (e.g. soil depletion; water stress; legal environnemental constaints; intrants rising prices; etc.).
This complexity is also compounded by the unpredictable and uncontrollable nature of most parameters with which farmers must contend, especially the weather, implying that the same crop and management choices do not always lead to the same yields. This unpredictability can been removed in controlled environment farming (e.g. vertical farming; greenhouse farming) but remains a major issue for outdoor crops which still represent the overwhelming majority of cultivated surfaces across the globe.
This complexity, historically tackled through transgenerational experience and agronomic know-how, can now also be understood thanks to multiple outdoor agriculture technological innovations. These innovations enable farmers to collect data on their immediate agricultural environment (e.g. fields; crops; weather; etc.) and extract precise insights on their crops’ needs in order to increase yields while optimizing resource allocation at the plant level. The combined emergence of these technologies is generally refered to as ‘precision agriculture’, currently carried by a wide variety of players and technologies.
These technologies carry the promise to support farmers in their daily decision making process and operations by helping them chose the best combinations among all the parameters they have to leverage (e.g. how much inputs should I spread? how many times should I plough? etc.). One broadly adopted use case of precision agriculture is for instance the automation of section cutting and spraying modulation. Based on geolocation data, this technique connects one GPS location system with crop protection sprayers or seed drillers to avoid zone overdosing during field works. According to Arvalis, a French agricultural institute, this technique can represent up to 10-23€/ha savings in phytosanitary products depending on considered crop and farm. Many other technologies and decision support tools have been available for several years now to help farmers better manage their operations. However, despite their visibility and long-time presence on farm shows, these technologies are still unevenly adopted by farmers today. If local weather stations and basic GPS systems seem to have found their ways towards the majority of farms, more sophisticated technologies and techniques such as parcels modulation (e.g. modify inputs and works at the plant level based on parcels’ heterogeneity maps) still remain the exclusivity of a limited number of modern farms.
Several factors are holding back the broader adoption of agricultural and precision AG technologies, which are as much related to financial stakes as to go-to-market issues.
The first innovation hindering factor is parcels’ heterogeneity and ardous comparability. Unlike indoor agriculture where growing conditions are controlled and replicable regardless of considered geography, the agronomic diversity of outdoor parcels is infinite and complicates the production of universal return on investment demonstrations for technology providers. The ‘burden of proof’ must be demonstrated multiple times and on large scales to prove the value of a technological innovation to farmers. For the moment, a lot of AgTech companies demonstrate their value on a use-case basis, limiting the conviction impact to local specific conditions.
Beyond this sometimes challenging value demonstration, AgTech technologies still represent high investment costs for most farmers. Against these costs, return on investment can sometimes prove uncertain for outdoor farmers mostly because of the uncontrollable and unpredictable nature of major growing factors such as the weather, reducing the ability to reproduce with certainty the same result with the same parameters from one year to another year, despite extensive technological equipment.
Eventually, the lack of support and training for farmers can also hinder technology adoption. While farmers are a particularly connected and innovation-friendly public (e.g. 71% of farmers own a smartphone, 72% of them have at least one agricultural app), they also have a complex and time-constrained job. Against this limited amount of time available for technological watch, the AgTech landscape and its multitude of offers is still too fragmented to be easily understood, slowing down technologies adoption.
However, above-mentionned difficulties can all be addressed and partially overcome, thanks to both technical and commercial levers. Technological value demonstrations must first be run on a large scale and multiplied at local levels in order to get over parcel-level experiments’ limited trust generation potential. These large-scale demonstrations close to farmers can be carried out through partnerships with local relays in order to generate trust. Such demonstrations were for instance carried in Britany (France), where the Terrena farming cooperative ran an AB testing of its modulation technology over 4000 parcels, in partnerships with the Brittany Ille Armor’s CUMA federation.
Beyond this trust and value demonstration issue, several ways can be considered to moderate technologies’ cost. Greater sobriety can be integrated by-design in innovations to build technologically nimble offers. Several use cases could for instance be deployed by leveraging the data and computing power of hardware already widely deployed within farms, such as smartphones or tractors (provided that some partnerships are built with OEMs). Likewise, Kuhn developped a free smartphone app (‘Kuhn Easy Maps’) to help farmers navigate through their field and avoid overlap during spraying work.
The other way around, the automation of some technologies can also democratize their use by reducing costs associated to services. This is the promise of autonomous drones (such as those developed by American Robotics), which cut drones’ driving costs by removing the human from the loop.
Besides, non-technical approaches can also help to drive innovation costs down, thanks to investment cost sharing among farmers (e.g. through local collective institutions such as coops, agricultural chambers or technical institutes) or between farmers and other third party players. Costs could for instance be shared with insurance companies (since low risk exposure induced by innovation could imply lower premiums) or local authorities (since private investments can have positive externalities for local communities). Eventually, innovation costs can also be partially passed on to end-consumers through higher selling prices if investments have an established impact on the ecological value of produced crop and strong traceability is ensured. In this respect, current changing consumption habits of people regarding food and sustainability are a positive sign for innovation in agriculture, recognized by investors as funding amounts raised by the AgTech sectors reached record levels this year.
Technological value demonstrations must first be run on a large scale and multiplied at local levels in order to get over parcel-level experiments limited trust generation
Finally, AgTech players can improve their offers’ perceived value and adoption through simple and easy to use designs as well as through better user guidance and training. Such simplicity can be achieved by promoting sober integration into farmers’ existing digital ecosystems (e.g. avoid additional apps to access the service), limited calibration and maintenance requirements but also effortless interoperability with other farmers’ technological ecosystem components. One way to ensure easiness-of-use by-design is to involve partner farmers in the product development and innovation process (e.g. as promoted by the start-up Karnott).
Regarding user guidance and trainings, several best practices can help to foster adoption, from affiliated farmers enrollment in marketing (e.g. as promoted by the start-up Farmitoo) to enhanced presence on farmers’ preferred online communication channels. Indeed farmers are increasingly using social networks on a global scale (e.g. private facebook groups, youtube channels, specialized forums, etc.) to both reach out for advices among peers and engage on a personnal level with their end-customers (e.g. livestream platforms such as Taobao were massively used by Chinese farmers during the Covid-19 crisis), creating unique online spaces for AgTech players to engage with their potential users.
Although it has not reached its full potential yet due to its slow democratization, agricultural data analysis and precision farming represent part of the answer to the challenges of tomorrow’s agriculture as it can help – through technology – to increase yields while rationalizing costs and reduce pressure on ecosystems. However, precision agriculture will have to be coupled with other innovations – not always technological, ranging from biotechnologies to new cultural practices (e.g. vertical farming) as well as fairer and optimized distribution chains (e.g. less waste; shorter circuits) – to take on the challenge of ensuring food security globally while building sustainable farming and food systems.
Sur le même sujet