Technology intervention in agriculture evolves continuously. Adoption varies. Small landholders can find technology overwhelming. TCS Digital Farming Initiative (DFI) personalizes its solution to the level of each farmer and field by bringing the earth to the cloud through the unique intersection of advanced remote sensing and IoT technologies. This enables scalable, affordable, accurate and high tech/ low touch solutions. It works with groups on new business models. DFI’s mKRISHI® platform uses intuitive interfaces with voice and local language, making for easier adoption. The backbone of the DFI platform is a unique protocol-centric, marketdriven production approach that leverages cyclic optimization of extensive field data collected across the agro value chain and predictive analytics of the data collected, which is fed back to guide on-field operations.
Farmer Perumal used to look at the sky to tell time; catch a whiff of the breeze to predict the weather. His son Nagaraj prefers his mobile phone and TCS’ Digital Farming app. Change has swept not just Sirukaranai village in Tamil Nadu, but all of the agricultural hinterland of India.
The picture is still grim: soil depletion due to overuse of fertilizers and pesticides, changes in the water table, an increase in temperature, vast change in market demand for agricultural produce, among other things, has the Indian farmer reeling. In an environment where social and cultural changes make white-collar professions appealing, Indian agriculture has lost both the validity of traditional knowledge and its passionate practitioners. However, there is a realization that if current, relevant information and technology are provided to the farmer in local languages and consumable form, a revival can happen.
Many governmental, for-profit, and non-profit groups have introduced farmer support initiatives. TCS’ Digital Farming Initiatives (DFI) group uses easy-to-use technology to optimize agricultural practices in India. From crop management and health assessment to marketrelated analytics, the DFI group is changing the landscape by implementing high-value, userconscious digital farming offerings on the innovative, patented TCS mKRISHI® platform. We present some technology and business model breakthroughs here.
Prominent in the DFI group’s portfolio is the Progressive Rural Integrated Digital Enterprise (PRIDE™) model. PRIDE™ guides users progressively through four phases—crop planning, just-intime input procurement, crop cycle management, and harvest planning—to help them improve crop production efficiently and sustainably, by creating effective linkages across partner ecosystems. The backbone of the platform is a unique protocol-centric, marketdriven production approach that leverages cyclic optimization of extensive field data collected across the agro value chain and predictive analytics of the data collected, which is fed back to guide onfield operations across the various phases.
PRIDE™ is coupled with an extensive agricultural knowledge base, an e-commerce engine for efficient market and agricultural input linkages, and an efficient delivery channel for distributing personalized cultivation practices to all farmers. It is based on extensive research in rural participatory sensing, wireless sensor networks, agricultural activity detection, hyperspectral image processing, low-cost drones, disease prediction and dispersal, and predictive analytics. It uses an integrated ‘from lab to land’ model involving hypothesis creation and validation through rapid prototyping and field data collection, with the validated hypotheses converted to use cases for productization in Agile development mode. PRIDE™ therefore ensures that the research consistently addresses real-world problems.
Cognitive remote sensing is a highly specialized area of study that involves using artificial intelligence in remote sensing, and cognitive remote sensing services in agriculture (CRSSA) are an advanced feature of the PRIDE™ platform
Improving and optimizing the practice of agriculture requires accurate and timely information based on spatial data provided at acceptable resolutions. This information is grouped into categories, as shown in Figure 1. For CRSSA, the core data for this comes from a constellation of commercial and government satellites.
The type of imaging used determines the nature of the agricultural problem that can be addressed because the agricultural knowledge held in an image (for example, phenology of the crop of interest; crop growth measures such as leaf area index, stomatal activity, and chlorophyll content; and soil-health metrics such as soil moisture, salinity, and pH) depends on the wavelength chosen for the imaging.
For instance, the normalized difference vegetation index (NDVI), which maps the density of green cover in a geographical area, needs imaging in the red and near-infrared wavelengths to identify crop health. The knowledge acquired from this ground truthing is then upscaled for use with satellite imagery.
For crop identification, satellite imagery of crop-growth characteristics is studied. For example, to identify rice across large watersheds, images of the ponded conditions in which rice grows, are examined, while machine learning algorithms are run on images of paddy leafing patterns, which are the crop’s unique signatures.
A point to note here is that not all imaging is satellite-based. Drones (see Box 1) have also been used effectively. Imagery using visible, near-visible, and radio wavelengths forms the basis of most current CRSSA. A combination of spectrum choices and plant phenology-based markers can help in developing newer CRSSA.
Disease and pest prediction
Spatial weather forecasting models can temporally predict the vulnerability of crops to fungal, bacterial, and viral diseases and pests based on the location of a farm, as such diseases require certain temperature, humidity, and wind conditions for emergence and propagation. TCS’ DFI group therefore uses spatial data on farms, along with forecasting data from global climate models obtained from the India Meteorological Department, the US National Oceanographic and Atmospheric Administration, and the European Centre for MediumRange Weather Forecasts, to make farm-level predictions. The group has created forecasting models covering 80 diseases and pests for 30 crops.
Crop-stress detection and drought monitoring
Evapotranspiration (ET)—the movement of water from the earth’s surface into the atmosphere through evaporation from the soil and transpiration from plants— is an important component of the hydrologic cycle. Accurate information on crop ET is critical in water management and hydrologic modelling, as it quantifies field- or pixel-scale water requirements, water stress, and yield prediction. ET anomalies can also be used for drought monitoring and forecasting.
Traditionally, ET has been estimated through human observation and mass balance calculation. However, remote sensing, which uses the surface energy balance approach, is a far superior technology today. The DFI group has therefore deployed an adapted “mapping evapotranspiration at high resolution with internalized calibration” algorithm for ET estimation, successfully estimating crop-stress throughout the crop season and using it as an input for soil-water balance to gauge realtime irrigation requirements. The group is also using pixel-level ET anomalies for drought monitoring.
Crop insurance is an ideal solution for mitigating the risks faced by a farmer during the cropping season. However, despite the fact that many crop insurance schemes have been launched, their implementation is hampered by the inability of the insurer to ascertain the reasons for crop failure. TCS’ DFI group has developed a methodology which, using spatial and economic data, calculates the efficiency of agricultural activity in farmlands and can evaluate the reasons behind crop failure or yield reduction.
Revolutionary Potential of Drones in Agricultural Insurance
In February 2016, the Indian government mandated that every insurance company should provide agricultural insurance to farmers. Via a notification, it would identify certain crops for insurance coverage. Insurers would then be assigned a given number of revenue blocks—which are different from administrative blocks—per state and, for low annual premiums (between 1.5 and 5%), reimburse farmers for losses in those crops caused by failed (or prevented) sowing, mid-season and localized calamities, and poor or no yield.
Loss due to failed sowing is caused by natural calamities and becomes reimbursable—to the tune of 25% of the sum assured—when more than 75% of arable land is not sown. The reimbursement rules are similar for mid-season calamities, localized calamities (hailstorms, localized floods, thunderstorms, etc.), and loss of yield (as compared with expected yield), although the insurer’s reimbursement goes up to 40% for mid-season losses, and 50% for yield losses.
Since vast sums of money are involved, working with accurate data is of essence. However, the collection of this data, which is done by each state government, presently involves laborious, time-consuming, and inaccurate methods. As a result, the insurance companies are not only unable to carry out the needed financial planning and forecasting, but must also factor in much bigger risks and costs, which would be commercially unviable.
To address this issue, the DFI group is working on gathering crop data from satellite- and dronebased remote sensing imaging, processed using artificial intelligence algorithms. The group has already tested these techniques successfully in the field through its mKRISHI® platform.
Remote sensing is done chiefly using two complementary technologies—satellites and unmanned aerial vehicles (UAVs or drones). Satellites image large areas at low resolutions,
while drones image much smaller tracts at high temporal and spatial resolutions. Judiciously combining the two gives us the best of both.
TCS’ DFI group has introduced innovation in three areas of drone-based solutions for agro insurance:
1. Payload customization:
Having customized COTS drones to fly for 45 minutes instead of 20, it is tweaking payload by adding multispectral and thermal cameras, which can image crops at various resolutions and cover 400 acres in a single flight.
2. Drone-health monitoring: A drone comes packed with complex electronic circuitry that governs a slew of sensors—a barometer, a compass, a GPS, accelerometers and other speed controllers, volt integrators, a flight controller, and more.
However, the drone has in-built systems to monitor those. However, what needs incorporating are monitoring mechanisms to ensure the drone’s structural health (for instance, reducing vibrations in the motor mount caused by wear and tear in the bearings or stabilizing and balancing a flight when one or more propellers are damaged), as these factors can otherwise adversely impact navigation and control. The result can be sometimes drastic, with a drone crashing or even lost beyond the line of sight in an area that’s hundreds of thousands of hectares in size. TCS’ customized drones have been field-tested for a 13-plus km range.
3. Crop classification algorithm development: The DFI group is working on a Proof of Concept (PoC) for drone-based crop imaging that uses machine learning and deep learning algorithms to classify crops. The group has to start work when the crop is anywhere between three and 15 days old, about 4-5 inches tall, with 3-4 tiny leaves per seedling. The leaves, at this stage, have practically no resemblance to those of a mature plant. The early-season imaging must, therefore, be of very high resolution so that the patterns in the image data from each stage of the plant’s development are distinctive. What’s more, the algorithms processing the image data must fix on a technique that succeeds in classifying crop growth stage as accurately for one acre as for the millions of acres that a drone flies over.
Drones fly “under” the weather and, armed with multispectral cameras, can tell how well a plant is growing. The cameras’ high-resolution capabilities track key indices and parameters, such as NDVI, EVI, and crop height, among others.
Drone images will be valuable at the yield stage too. At this stage, the government currently engages in the crop-cutting experiment—randomly identifying, for a given notified crop, four 5 m x 5 m plots in every revenue village; manually harvesting the crop from each; averaging the harvest across the plots; and extrapolating it to a per-acre number. Since the country has three million revenue villages, the exercise is conducted three million times. This translates to a lot of person-hours of effort (read money and time).
However, aerial images from drones flying at 50 m and 30 m deliver 1.5 cm and 1 cm resolutions, respectively. This is better than the 30-cm images that satellites can provide. The high-resolution pictures from a drone would enable monitoring a range of crop growth parameters, making it possible to estimate yield with unprecedented accuracy and speed.
Crop identification and monitoring through remote sensing
Crop identification through remote sensing takes plant phenology and agronomic practices into consideration. Several indices calculated from multispectral images—NDVI, normalized difference water index, enhanced vegetation index (EVI), and others—can monitor crop health during the growing season. The DFI group therefore uses a collection of satellite images for weekly updates on crop status and growth areas and, by analyzing time series data on crop health and factoring in weather forecasts, makes crop-yield predictions.
Cognitive IoT in agriculture
IoT for open farms
Knowing the instantaneous value of a parameter, such as field soil moisture, helps just-in-time irrigation, especially when water is scarce. Some precision monitoring scenarios may require measuring not only soil moisture at varying depths but also instantaneous pH values of ponds in aquaculture where, by continuous pH monitoring, one can estimate the quality of fish. All such applications require the internet of things (IoT). High-value crops such as grapes, which often have camouflage cover, need precise monitoring of microclimatic parameters such as temperature and humidity for the assessment of specific conditions such as fruit ripening (which affects quality) and precise prediction of specific diseases.
TCS’ DFI group has extensively monitored the potato crop in India’s northern state of Uttar Pradesh using a low-power method— Zigbee-based sensor nodes on the ground, integrated with TCS’ cloud platform. The group has validated the JHULSACAST model for the late blight of potato, which causes widespread damage to the crop. Long-range, low-data rate radio technologies, such as LoRa, can significantly upscale monitoring today. The group has also developed highly efficacious portable, mobilebased soil moisture sensing for irrigation planning that also maintains a complete digital trail of measurements.
IoT for closed farms
From greenhouses and cropprocessing units to indoor soilless cultivation methods, such as hydroponics and aeroponics, closed-farming setups benefit immensely from IoT, which enables precision control and monitoring. (To get an idea of plant or crop sensitivities in such settings, imagine the withering of leaves on a factory floor, which uses non-digital, heuristic methods and necessitates close monitoring of ambient temperature and other parameters.) Digital intervention with IoT, however, can help precisely regulate the usage of the machinery required for such a process, and save on electricity.
TCS’ DFI group has created a lowcost IoT solution framework that uses multiple sensors for the indoor monitoring of plants.
Wearables in farming
The DFI group has invested considerable effort in creating wearables systems—using sensors for agricultural activity recognition and training. To start with, creating a wearables-powered shirt has helped categorize agricultural activities. Accelerometer signals tracked from both hands of the human wearer have been used to classify activities, such as digging, bed-making, and harvesting. Based on signal strength and electronic pulse rate, a given activity can be scored for activity quality and, along with expert videos, used to train workers.
This research area has promise with crops such as tea, where the primary activity, tea-picking, involves carefully plucking two young leaves and a bud. Through extensive pilots conducted in the tea gardens of West Bengal, various sub-activities associated with tea-picking, along with their duration, have been clearly identified, which has led to important findings in characterizing the activity sequence and the resultant quality of leaves picked.
IoT in the agriculture supply chain
IoT is not restricted to on-farm interventions but also plays a pivotal role in monitoring activities at various touch points during the journey of the crop from farm to consumer. Whether in tracking vehicles, as part of harvest planning, or in regulating the conditions of cold storage along the way, IoT can transform the digital experience of the entire supply chain. IoT could be integrated at various levels within sugar PRIDEs in which mKRISHI® has been deployed for harvest management to transform and digitize the operations of the enterprises concerned.
The 4 ‘A’s Decision Support Model
TCS’ DFI group’s services and products have helped develop an application for managing large-scale farms. The goal was to define a decision support system for the whole farm so that field-level management and resources could be optimized, while factoring in farmer-level constraints.
The platform supports aggregation of structured and unstructured data and has a complete data management module, predictive and big data analytics modules, smart push notifications, and support for intelligent information gateways. The application provides the following key services:
• Acquisition based on IoT, UAVs, remote sensing, images, and voice
• Analysis through machine learning algorithms, data analytics, and real-time data processing
• Advisory via direct in-field and supply chain operations, based on intelligence gathered through data analytics
• Actuation of various types of in-field machinery, based on data analytics intelligence
A cognitive IoT framework
To address the varied IoT application requirements within the DFI group, we have developed a comprehensive solution stack that extends from the cloud to silicon-inthe-field. The group has deployed a flexible IoT adaptor on the cloud, as part of a precision agriculture solution that can acquire data from a variety of sensing sources. For the Edge, the Group has developed a configurable gateway middleware that runs on COTS-embedded boards to collect and process data locally from sensor clusters and receive commands from the cloud for actuation.
The gateway talks to devices on the local network through Zigbee, Wi-Fi, and other protocols. The gateway layer also includes the smartphone, which acts as a mobile gateway for low-cost portable sensing at scale. As evident from the applications discussed earlier, many contextual aspects come into the picture when we bring an IoT intervention to meet a given challenge in agriculture. These analytical solutions are translated into cognitive elements that empower our IoT stack, both on the cloud and the Edge, to improve the quality of our offerings.