We will create an end-to-end digital platform that measures agricultural insect pest pressure in the field and transmit this information to mobile devices in real-time, allowing the smallholder to make the appropriate interventions (e.g., pesticides, attract-and-kill formulations or traps, release of parasitoids/predators, neem-seed extract, etc.) in a timely fashion.
Our platform will leverage proprietary low-cost sensors and state-of-the-art machine learning algorithms to instantaneously and accurately count and classify live flying insects, down to the sex/species level. The proposed system will be designed to be generic and adaptable, working with the majority of crops and associated flying pests. Furthermore, we will optimize our system for use in the developing world.
In the last year we have demonstrated the feasibility of using our digital surveillance platform in large-scale successful USA-based field trials monitoring the Corn Earworm, a sibling species to the Fall Armyworm (FAW), a pest that first appeared in Africa in 2016, and which is already causing devastating crop losses.
As USAID notes, “Even if farmers select an appropriate pesticide, they may not spray at the right stage in a FAW growth cycle to make a difference.” The temporal resolution of most lepidoptera pest surveillance traps is at best, several days. In contrast, the temporal resolution our platform is one second. Thus, we can provide real-time, unassisted, 24/7, accurate monitoring of insect pest density in the field, together with short-term forecasts, both at a high spatial and temporal resolution.
Many parts of Africa have dozens of local languages and a low literacy rate (78% overall, but much lower in rural communities, and for women). These issues offer significant challenges. However, cell phone ownership rates are extremely high. Our platform exploits this cell phone penetration. We will treat insect encounters as digital objects, which can be “sliced-and-diced” at multiple levels of aggregation and presented in multiple formats. For example, where bandwidth is very limited, insect counts can be communicated by a text message (effectively or actually free on many African networks).
How this information is used depends on the intervention options available to the smallholder. Intervention options include pesticides, attract-and-kill formulations, mass trapping, release of parasitoids or predators, natural pesticides such as neem-seed extracts, etc.
Note that no matter what intervention is used, knowing exactly where and when a pest first enters a field provides the user the ability to intervene before populations cause damage, which is critical to have the maximum effect, and to make the best use of scarce resources and human labor. The grower will know almost immediately if the intervention is working or not.
At a higher level, our platform will aggregate information from numerous growers across regions, allowing government agencies (and/or NGOs) to determine patterns of pest distribution, the effectiveness of treatments, correlating it to the different biotic and abiotic conditions found in different areas and even conduct A/B tests to compare the effectiveness of different interventions. This is because our platform records insect density before, during and after intervention.
Note that many attempts to measure pest density are indirect; they actually measure proxies for the insects, such as leaf damage or metrological conditions. In contrast, our philosophy is “if you want to know FAW prevalence, then measure the prevalence of FAW.”
Our system will fill the “information-gap”. Smallholders cannot intervene efficiently without knowledge of which insects, in what densities, are at what locations in their farm. This information need is well understood in the developed world, even though first-world farmers have a much greater arsenal of interventions available to them, including blanket spraying.
FAW is especially difficult to control because gravid females migrate into the field and lay eggs when corn or sorghum is being sown, and the tiny larvae that hatch start the infestation as soon as the first leaves of the crop emerge from the ground. Timing of pesticide application is critical for effective pest suppression: after their second larval stage the FAW larvae escape control. Farmsense will provide the needed alarm for growers fighting FAW. With this prompt from Farmsense, growers will be able to deliver the timely applications needed to make the current pest control tools work.
Our team has identified this information need with two decades of “boots on the ground” experience of lepidoptera pest management in the developing world, including Ethiopia, Tunisia, Nigeria, Tanzania, Kenya, South Africa, Morocco, Egypt, and a dozen Indian states.
Our philosophy in developing this system is inspired by our experience dealing with another class of harmful insects found in Africa: mosquitoes. Many African health agencies are tired of well-meaning but impractical solutions being suggested by first world research groups. Thus, there is the unofficial malaria mantra, “make it cheap and easy to use, or don’t bother.”
Given this, we will obsessively optimize our system to be as cheap as possible. This means cheap to manufacture, transport, operate, and crucially, cheap to use in terms of human labor. We also understand from first hand experience that low levels of literacy and diversity of languages is an issue in the developing world; thus, we will produce wordless instruction sheets to explain how to use our sensors. Here we take inspiration from companies such as Lego, Ikea, and Dyson, which communicate complex instructions with simple graphics.
Versions of our sensors used in the developed world have a handful of switches and controls that the grower needs to interact with. However, the sensors we will build for this project have zero buttons. As soon as the protective cover over the solar panel is removed, our sensors start to work. We have found from experience that users like to have some direct feedback as they install sensors, to ensure that everything is working correctly. We will allow them to wave a feather or small twig in front of the sensor (emulating an insect), and they will see a small light flash to reassure them that the sensor is working correctly (note, our machine learning algorithms are not fooled by these test runs, and report only “NonInsectEvent detected” with the appropriate time stamp).
It is well-understood in IPM that the earlier you detect a pest, and the more accurately you can locate them, then the less drastic an intervention you need. Our ideas allow the farmer to go from current remedial “kill-all” blanket sprays using conventional broad-spectrum pesticides, to a more preventative approach, in which the grower applies pesticides only where the invasive adult pest is, before damage can occur.
In the developing world this precise information, aside from reducing costs, has the desirable side effect of potentially allowing the farmer to use less pesticides, reducing the possible harm to humans, or in some cases allow her to forego pesticides and use only biological controls.
We plan to adapt the “one-for-one” business model made famous by Toms Shoes. When Toms sells a pair of shoes, a new pair of shoes is given to an impoverished child. Similarly, we will sell our system to capitalized growers—in the first and developed world—and use the profits to give away the sensors to smallholders, especially the poorest of the poor in the developed world.
Comments