How to Turn the Promise of Digital Agriculture into Reality
While there is clear evidence of the technology’s benefits, its full promise has yet to be realized. What is holding us back? In addition to policy challenges such as data governance (ownership, privacy, security, etc.) and business model challenges, there are at least three key technical hurdles we must overcome:
1. Sensing in agriculture is still too expensive. The industry has made enormous advances in instrumenting agricultural machines with sophisticated sense and control solutions. However, the lack of a corresponding movement towards standardization and interoperability between these devices has prevented them from driving more cost efficiencies. End users often wind up with an assortment of incompatible systems, gateways, power sources, and software packages. This increases costs and makes these systems more expensive to maintain. We must implement effective standards that exploit the synergies between different sensing and control solutions, thereby reducing their cost.
2. Agricultural data is vast, complex and difficult to exploit. The sheer size and complexity of agricultural data makes it stand out among industries that leverage data-driven insights. This poses a unique challenge. Take, for example, the growth of critical weather-related information. Today, hundreds of terabytes of weather data are generated daily. Remote observations from drones or satellites can provide information about vegetative growth, irrigation, pests, and more. Governments and private entities are launching more and higher resolution satellite constellations. Despite the richness of all this information, which is growing by the hour, it is highly underutilized for many reasons. First, it is distributed widely among the many public, private, and non-profit entities that generate it. Second, it comes in a variety of formats: projections, resolutions, and reference systems, to name just a few. Third, and perhaps most importantly, the data is often too large to be downloaded in time to be acted upon. We must develop scalable data and analytics platforms that can process petabytes of this data (one petabyte is equal to one million gigabytes). These platforms must also enable seamless, real-time integration of different data sources (weather, elevation, soil, sensors) and support computation in the data without having to move or download it.
New technologies pose challenges, yet we move forward with confidence.
3. Agricultural analytics today are narrowly focused on specific domains and therefore not scalable. Because a variety of factors determine farming outcomes (crop type, genotype, phenotype, cultivar, farm practices, etc.), researchers have developed numerous domain-specific analytical models for individual applications such as irrigation, growth, yield, and pests. For example, there are well over a dozen models for potato late blight disease and there are more than thirty different irrigation models. These models utilize highly specific data and information from multiple sources for any given application. Our challenge is to tap into the knowledge underlying these domain-specific agronomic models and integrate it with universal, big data machine-learning based techniques. Only in this way can we generate scalable, actionable insights for farmers across the globe.
There is no question that rapid advances in digital agriculture are taking place, and with a clear sense of the challenges ahead, we can move forward confidently. As one example, IBM Research has been working with Gallo Winery to co-develop a precision irrigation method and prototype system that has led to a 20 percent increase in water use efficiency -- meaning that Gallo uses 20 percent less water for each pound of grapes it produces.
My experience has taught me that there will always be significant hurdles to overcome with any new technology. But we can’t allow them to deter us from achieving our vision of providing a safe and abundant food supply for future generations.