3.1. Current State of IoT for Agriculture
For this overview of IoT for agriculture, sensor and communication technology applications are classified into five categories of (a) climate, (b) livestock, (c) plant, (d) soil, and (e) water. Within each category, there are many common, measurable parameters that can influence the performance of the agricultural system (Figure 4
). This categorization focuses on “ground-based” measurements, while other methods exist including aerial- and space-based earth observation or remote sensing, for example in [26
]. In some cases, remote sensing and ground-based measurements are combined to provide temporally (ground-based sensors) and spatially (remote sensing) dense crop measurements [29
summarizes common electronic sensors, their applications in agriculture, and articles that describe those in more detail. Wherever available, articles that report on applications in LMICs are included, although relevant precision agriculture research from across the globe is also included. Some parameters can use different technologies to estimate the same output. Methods for measuring basic agricultural parameters, including soil and atmospheric conditions, are well-established, and commercial products are available for IoT applications. New applications for optical sensors, in particular, are evolving as the cost of semiconductor technology and data storage and transmission decreases. A particular challenge with using low-cost sensors in agriculture is the need to calibrate the sensor for the specific implementation conditions. For applications where artificial intelligence (AI) is applied to classify events based on measurement patterns, complex training datasets are necessary to teach the AI algorithm.
In addition to the measurement device, careful consideration should be given to the path and process by which measurements are transformed into insights. Data transmission can represent a major challenge in smallholder agriculture applications. Edge computing is becoming more common in commercial IoT products, in which onboard device memory and processing capacity is utilized to carry out some data reduction and analysis [80
]. However, for insights that require a large quantity of measurements or computational power, transmitting data to the cloud for computing will be required. Table 2
summarizes common IoT data transmission protocols and their associated advantages and disadvantages.
The IoT for smallholder agriculture represents a challenge for data transmission due to remote locations, with devices distributed over large areas or multiple farms that have potentially limited access to electricity and cellular networks. Therefore, the range, data rate, and power consumption are important design considerations and are compared for the common communication protocols.
In Figure 5
, communication protocols are grouped based on data rate and range of transmission, which is divided into “Long range”, including low-power wide area network (LPWAN), which includes LoRa and SigFox, “Short range”, including Bluetooth Low Energy (BLE), Zigbee, and Z-Wave, and cellular communication including GSM 2G, 3G, 4G, and 5G. With respect to power consumption, solutions that are tailored to IoT applications offer superior performance to more general protocols (Figure 6
). While the range and power consumption of protocols like LoRa and SigFox are well suited for IoT applications, their device compatibility is more limited compared to generic wireless protocols like Bluetooth and WiFi. Additionally, cellular and satellite communication offer the advantage of providing a direct link to a web server instead of passing through an intermediate gateway.
Even though many organizations in LMICs have developed IoT systems for agriculture, few studies have reported on the implementation in a smallholder context and associated outcomes. Research in Embu County, Kenya tested different proximal soil sensors to estimate soil properties and composition on smallholder plots and provided management recommendations to farmers [57
]. The measurements showed considerable variation, both within plots and regionally, indicating a need for management recommendations based on both local measurements and regional soil maps. Another application implemented a tracing system for smallholders who raised live poultry that were traded in Vietnam to prevent the spread of avian influenza [75
]. Low-cost radio-frequency identification (RFID) tags on transport cages and electronic tag readers at markets tracked birds from farm-to-market, revealing that birds passed through up to six intermediate traders, creating a high risk of virus transmission. Many commercial entities serving smallholder farmers use the IoT to track assets and interface digitally with customers; for example, tractors and implements for hire by Hello Tractor in West Africa and EM3 AgriServices in India [83
], and irrigation equipment by Agriworks, Futurepump, and SunCulture in East Africa [84
]. Enrollment in these services is enabling more productive growing; however, low digital literacy is a barrier that requires a mix of traditional and technology-enabled engagement with farmers. A notable case is the use of sensors to monitor borehole pump reliability for potable and irrigation water supply in East Africa, which is discussed later in this article.
3.2. Implementation Cases
In addition to reviewing the literature and interviewing experts, we found it useful to study on-the-ground implementations of the IoT for smallholder agriculture. In particular, this research aimed to understand what helps to create an enabling environment at the country level, and the keys to success and the risks associated with active IoT in smallholder agriculture projects in India and Kenya.
India is home to over 130 million smallholder farms [85
] and is a common testbed and incubator for digital services for farmers. Site visits and discussions in India focused on identifying the factors that enable innovations in the agricultural sector in order to promote those in GFSS countries. The following is a summary of those factors.
● Mobile network connectivity and cost: Relatively cheap monthly mobile data plans of 0.21 USD/GB [38
] have helped IoT companies explore opportunities to work with sensors using cellular services for data transmission.
● Market opportunity: The large population size and density, and increasing incomes of the Indian middle class, make India a lucrative market for IoT providers [41
● Policies to support farmers: Agriculture accounts for 17.32% of India’s GDP and employs over 50% of the population, and some state governments are providing subsidies for new farm equipment that could be leveraged towards precision agriculture purchases [40
● Academic institutions: Some of the highest ranking academic institutions in India are performing research that benefits Indian farmers and are raising the awareness of farming challenges to students through hackathons.
Kenya represents a very different landscape compared to India. The population and total number of farm holdings are vastly smaller, and many of the enabling factors discussed above are yet to emerge. However, Kenya was selected because, among the different GFSS countries, the authors’ investigations and discussions revealed several active IoT for agriculture cases across different applications and locations in Kenya. These cases are summarized in this section along with some of the keys to the success and the associated risks with project sustainability.
As part of the Kenya Resilient Arid Lands Partnership for Integrated Development (RAPID) project to manage the recently discovered Lodwar Basin Aquifer in northern Kenya, electrical current sensors were installed on solar-electric borehole pumps to monitor “water system functionality, the approximate number of pumping hours and volume extracted per day, and the last report date for the sensor” [86
]. Data from the current sensors are transmitted via cellular or satellite network to a web server and dashboard where county government staff can monitor borehole use. While the pumped water serves a variety of community needs, some is used for irrigating small farm plots on municipal land that would otherwise be infertile (Figure 7
). One of the keys to the success of this IoT implementation was the clear value and utility to the government officials responsible for the pumps, who indicated that access to the dashboard significantly reduced maintenance costs and pump downtime. The Kenya RAPID project also benefits from having a large, distributed, and well-coordinated team where each organization plays a clear role, including the supplier of the IoT technology. The sensor measurements have revealed detailed pump usage patterns in relation to rainfall [74
], and this, coupled with a machine learning algorithm to predict failures and reduce detection time, has resulted in an increase in system-wide pump uptime from 70% to > 99% [87
]. A risk for the project is the current reliance on grant funding, although the IoT component has now been incorporated into the county government budget.
Borehole pump monitoring for small plot community farming in Turkana
The IoT solution for the aquaponics system at Kikaboni Farm was developed by Upande, and it monitors water conditions (temperature and pH) in the fish tanks and environmental conditions (ambient temperature and relative humidity) in the hydroponic vegetable growing area (Figure 8
). The sensors are battery-powered, charged by 3–10 W solar PV panels, and transmit measurements over a LoRa network to a gateway with a mobile data connection. Data are stored and processed on Upande servers and regularly fed back to the farm’s horticulture manager who can perform necessary adjustments. For example, the cover material for the vegetable structure was changed after measurements showed that temperatures were far above the recommended growing temperatures for leafy green vegetables (Figure 8
). One of the keys to success for this application is the ability of the horticulture manager to interpret the data, which has allowed them to make significant improvements to their product yield and quality. Kikaboni Farm has also collaborated on product development by providing a testbed for improvements to Upande Vipimo IoT products.
A risk for this project is the reliance on the expertise and willingness of the horticulture manager.
Water and greenhouse monitoring for aquaponics in Olooloitikosh
With little vegetation in the Mara River watershed, rainy season precipitation causes rapid river level rise and destructive flooding (Figure 9
). The IoT solution developed by Upande consists of solar-battery powered sound navigation and ranging (SONAR) level sensors at several points along the river, connected by LoRa to several grid-connected gateways with cellular access. In the event that the level sensors detect a rapid river level rise, an SMS-based system is activated, which alerts downstream farmers to pump water out of the river in order to open capacity to receive the upstream surge. One of the keys to the success of this system is the dedication of volunteers to coordinate activities, maintain the IoT system, and host workshops to engage the local communities. Some of the risks to the sustainability of this project are a lack of financial support, the rugged and remote conditions that the IoT must survive in, and the inconsistent support from the county government to allow the system to operate.
Flood detection and alert on the Mara River.
Greenhouses offer the opportunity for small farmers to grow high-value vegetables throughout the year in a controlled environment. However, uncontrolled greenhouse conditions can be harsh and damaging to plants. Researchers at the Dedan Kimathi University of Technology developed an IoT temperature, relative humidity, and soil moisture sensor coupled to an internet-connected gateway to assist farmers on their research farm (Figure 10
]. In this case, the system relies on the expertise of the farmers to interpret data and make the proper adjustments to the greenhouse. The farmers showed a high level of satisfaction with the system and reported that it had greatly improved productivity of tomatoes in their greenhouses. A key to the success of this project is the close connection and proximity between the IoT developers and the farmers, and the ability of the farmers to interpret the sensor measurements. This project is also equipping engineering students with the skills and experience needed to provide commercial IoT for agriculture solutions in Kenya [88
]. A potential risk to the sustainability of this project is that steady funding is needed to build a pipeline of work given the expertise, location, and access to research agriculture facilities.
Greenhouse monitoring in Nyeri
To supplement the limited information available to smallholder farmers, Arable has developed a multi-parameter IoT device with a suite of sensors for measurements including ambient temperature, humidity, precipitation, Normalized Difference Vegetation Index (NDVI), and photosynthetic active radiation (Figure 11
]. During current pilots in central Kenya, the devices are installed on smallholder farms and data is transmitted through cellular network to cloud servers where it is stored and analyzed. Reduced data is fed back to Kenyan partner researchers and agriculture extension agents who offer advice to area farmers. Agriculture extension agents reported that the local-scale information is a valuable supplement to regional forecasts provided by the Kenyan government and help them to provide better advice to farmers. A risk to this project is the challenge of determining an economical pathway to sustain and expand the IoT technology and staff in the ecosystem.
Monitoring smallholder maize plots in Nanyuki
3.3. Discussion of Challenges and Recommendations
Based on our literature review, expert interviews, surveys, and site visits, the team has synthesized a list of the challenges in IoT for smallholder agriculture in GFSS countries (summarized in Figure 12
), and proposed recommendations for some of the relevant players involved. The following section is a summary of challenges grouped into five categories, which correspond to the IoT architecture in Figure 1
: (i) measurement device, (ii) data transmission, (iii) data storage and analytics, (iv) feedback and implementation, and (v) project structure and support. A detailed discussion of the challenges, opportunities, and recommendations for the IoT for smallholder agriculture can be found in the full project report [25
]. We believe that this section will be appealing to audiences beyond the academic and research community, and specific recommendations are segmented towards Technologists, Project managers, and Funders.
3.3.1. Measurement Device Challenges
Access to components: Off-the-shelf IoT products are often not available, suitable, or affordable for commercial technologies in developing countries. Therefore, custom made solutions are designed in-house or by local IoT companies. During our site visit in Kenya, a number of IoT implementation teams that we spoke to indicated that procuring electronic and hardware components during the product development phase often delayed the project and increased costs.
Technologists: A good starting point for prototype circuit components is the local electronics and scrap market (e.g., CBD in Nairobi, Kisenyi in Kampala, Suame Magazine in Kumasi). These markets usually stock basic circuit and prototyping components that can be used as a “good enough” solution to reach a proof-of-concept prototype.
Device design: Smallholder farms are often in rugged and remote locations and require special consideration when designing a connected, electronic device for long-term monitoring.
Technologists: As soon as possible in the design process, test your device at a pilot site that is representative of the actual implementation site.
Project managers: Involve the farmers, agriculture extension agents, and farmer co-operatives in the product design phase to help your team identify non-obvious challenges and improve the likelihood that farmers will accept the idea.
Sensor calibration: Correlating the raw sensor measurements to actual physical values requires performing controlled calibration tests that can be expensive and time consuming.
Technologists: Aim to eventually provide calibration documentation for your product so that it can be benchmarked against other products and measurement methods. Perform some simple tests to check factory-calibrated sensors in conditions as close to the implementation conditions as possible in case there is a need to apply a correction factor.
Access to expertise: Many teams we visited mentioned a lack of access to expertise and resources for technical and business challenges. For example, several engineers reported spending significant amounts of time combing through websites to find technical solutions during their product development.
: Participate in online communities focused on IoT for agriculture hardware, especially for idea exchange, technical support, and recruiting. For example, we found the Gathering of Open Ag Tech to be a good example of such a resource in the agriculture sector (forum.goatech.org
3.3.2. Data Transmission Challenges:
Poor connectivity: Due to the remote location of some farming communities, poor mobile network connectivity and reliability is a common challenge.
Technologists: As a last resort, data can be collected manually, i.e., from a central hub connected to individual devices by a local wireless network.
Project managers: Check mobile network coverage in your implementation area using GSMA maps and cross-check with non-industry sources, for example, a compilation of user contributed Nperf data.
Transmission cost: While data costs have decreased significantly, the recurring cost of providing IoT services was frequently identified as a challenge for commercial applications.
Technologists: For some applications, satellite and LPWAN-based service providers are increasingly cost-competitive with conventional mobile data.
3.3.3. Data Storage and Analytics
Measurement to feedback: Raw sensor measurements can be difficult to reduce into actionable recommendations for farmers.
Project managers: While measurements are relatively easy to display on a dashboard, correlating them with crop growth and other effects requires input from a topical expert. Accessing the right expertise can be a challenge in itself, but resources at universities and agricultural extension agencies can often give some direction.
Technologists: Take a human centered approach by having farmers and agriculture experts from the area provide input on (a) what kind of recommendations would be useful and actionable and (b) the best means for the farmer to receive the recommendation.
Equity of access: The ownership of data collected is often overlooked in IoT projects and can lead to disagreements among stakeholders.
Project managers: Negotiate data access with project partners and funders. While it is important to maintain community access to data, service providers and funders may have requirements for data access.
Funders: Communicate early with the project implementers and have clear guidelines on data access, privacy, ownership, and sharing mechanisms of the data with the community.
3.3.4. Feedback and Implementation
: Many agriculture advisory mobile apps are designed for use on a smartphones; however, smartphone penetration is low among rural populations in GFSS countries [89
: Make sure the output is reaching the intended audience in a format that is accessible to them. If smartphone penetration is low, then using another medium such as radio, television, local print media, or extension agents can be appropriate to provide recommendations to farming communities [17
]. Early discussions with farmers, farmer co-operatives, and agriculture extension agents could help to make sure that the output is reaching the audience through the right communication channel.
Remote location: Many smallholder farms are in remote, difficult-to-access locations, which can add significant cost to IoT services if technicians need to regularly visit farms.
Technologists: Incorporate onboard diagnostics for your device (e.g., battery voltage, microcontroller temperature, accelerometer) in order to minimize maintenance visits.
: Identify a local farmer or extension agent who can assist with basic sensor maintenance and troubleshooting, for example, the horticulture manager at Kikaboni Farm described in Section 3.2
IoT revenue generation: Identifying the best customer for the information collected is important. In many cases, there is an opportunity to provide a service along with the hardware, which can provide recurring revenue.
Project managers: Many IoT for agriculture projects are initially grant funded, and determining approaches to monetize data and analytics services will help ensure project sustainability. Consider this aspect when agreeing on the terms of the funding.
3.3.5. Project Structure and Support
Complicated ecosystem: Successfully implementing an agriculture-IoT project involves coordination of a large, and sometimes complicated, ecosystem of actors over a long timescale.
Project managers: Try to organize a team of collaborators that includes data scientists, sensor experts, and agriculture experts so that each is contributing to solving a specific part of the problem in which he/she is an expert.
IoT business model: Most people we connected with during this project agreed that data from on-farm sensors is valuable. However, identifying a specific customer, or buyer, of the device or service is less straightforward, especially when the beneficiary is unlikely to have the means to afford the technology.
: Identifying and understanding whom the end customer is and who will be paying for the data is important. A good example is the Kenya RAPID project, described in Section 3.2
, in which the county government includes the IoT services in their annual budget.
Short funding timelines: Two- and three-year grants from donor agencies are short timelines; when it comes to farming seasons, this is a relatively short time period because you get data from only one or two cropping cycles, so collecting quality data and making actionable recommendations is a challenge.
Funders: Increase funding timelines to five to seven years to allow for data collection over multiple cropping periods. A larger dataset also enables better pattern recognition and predictive analytics, leading to suitable actionable recommendations.
Vertical integration: Many organizations are providing an end-to-end solution in which a single organization takes on the responsibility of farmer recruitment, IoT technology development, implementation, data analysis and recommendations, and monitoring and evaluation.
Project managers: Some of the key stakeholders the team spoke with indicated that, while they were experts in a few aspects of the IoT and agriculture, trying to work on the entire end-to-end process of an agriculture IoT ecosystem was stretching them thin; thus, incorporating a horizontal structure in which each organization offers specific expertise to solve the overall problem in a piecemeal fashion is favorable for project success.
Funders: Funding agencies can play a facilitation role to connect diverse groups with specific expertise to solve the overall problem. Organizations forming new partnerships could particularly benefit from guidance in negotiating agreements and contracts.