The Frederick Forsyth classic called The Devil’s Alternative starts with a dramatic sequence of the Oval Office pouring over images beamed down by US satellites of large-scale crop failures in the (erstwhile) USSR. If the images from the eyes in space were correct, USSR was headed for major turmoil and geopolitical crisis. This was the stuff of legend during the Cold War. Today, startups in India are using similar technology to solve more real-life (though less dramatic) problems for the humblest of farmers across the country.
One such startup is RMSI Cropalytics, who are deploying a combination of cutting edge technology like satellite imaging, weather forecasting, geo-location, artificial intelligence (AI) and analytics on one hand with government land records, historical crop yield data and agribusiness metrics to provide vital information to every stakeholder in the value creation chain.
Let’s start with a simple question – who will be interested in predicting crop yields and why? The answer is not as simple, and neither is the sheer economic value at stake. The key stakeholders in the agri economy other than the farmers themselves are commodity traders (including the government), farm credit companies, agri input (seed & pesticides) providers, agro insurance companies and agri-data and analytics service provides. These stakeholders control represent a combined economic value of ~$600 billion – which is almost equal to the combined GDP of all SAARC nations excluding India.
It is in this context that their trademarked PInCER (Profile & Information of Crop Exposure & Risk) platform provides the government, crop lenders, insurers and traders with information and analytics to set up and manage consistent and efficient operations in the agriculture sector.
“Basically what we do is collect data from different sources, including remote sensing, historical yield and weather data, apply actuarial modelling and agronomy knowledge on it, and pull all of it into an AI/ML model which estimates crop yields accurately at very high resolutions. This impacts the smallest farmer in every aspect from the amount of insurance premium to crop process,” says Roli Jindal, Co-Founder of RMSI Cropalytics
“Imagine a situation where an insurance company is determining crop insurance premium basis the average yield of a district. By definition, there will be a large number of farmers who will be paying a high premium despite being above the average yield value. With farm-level yield predictions, every farmer will be paying the right premium which is applicable for their individual land only,” she added.
Eye in the sky
Today’s satellite imaging can provide resolutions of 10mx10m. That provides inputs for fairly high accuracy of predictions. But how do you correlate the image of a patch of land from space to its owner? One solution is matching the lat-long of the location with land records. With this crops up 2 impediments. Firstly, on-ground verification of location indicated by the satellite image. Secondly, digital mapping of land records, a large number of which are in physical paper form. The first problem is solved through a formatted SMS by the farm owner from the farm location, which is analyzed by Cropalytics to determine geo-location. For the second problem, advanced image recognition and image-to-text conversion solutions are required, which can cater to multiple vernacular languages.
Optical imaging satellites go blind for 3-4 months during monsoons. No clear sky imaging is available during this period, which is a major sowing season. Since tracking farm productivity during this period is crucial to yield predictions, Cropalytics has started using SAR (Synthetic Aperture Radar) satellites that are impervious to weather conditions. This will help build scalable, reliable and consistent models even in difficult conditions.
Even a formidable combination of technologies cannot ensure that satellite images effectively differentiate between small, adjacent farm plots in assessing yields. Cropalytics has been deploying an established (and distinctly low tech) technique called crop cutting to tackle this. Here, farms are segregated into 10mx10m plots (equivalent to satellite image resolutions) and the progress and yield of this pinpointed piece of land is measured over a crop cycle. 1,000+ such crop cutting experiments have been conducted over the last 4 years to improve the accuracy of farm-level yield predictions.
To augment satellite imaging to gauge crop health, mobile phone, IoT and drone image based solutions are also being developed which will deliver very granular recognition of crops, crop health and farm conditions. These inputs are collated and extrapolated to a much wider area to deliver village, district and state-level crop yield estimates and verified historical data.
Information in your hands
No matter how powerful your technology and how accurate your prediction models, its utility is determined by its usability. Cropalytics’ has provided an extremely detailed, flexible and intuitive interface which allows users to zoom in to individual farm levels and zoom out to district levels for historical, progressive and predictive information. This is available on computer as well as smartphone screens which help identify distress hotspots that are critical of insurers, governments and service providers to be aware of.
The mobile app also collects sizeable amounts of data which are used to further train the ML model. Images captured on mobile devices are especially useful in identifying crop varieties, pests and crop diseases. These information and analytics sets are invaluable in suggesting solutions to factors impeding farm yields which can avoid farm distress.
Timeliness and comprehensiveness of data are imperative for agri-service providers. Given the fragmented nature of multiple data sources, this becomes a herculean task. As a result, many service providers depend on outdated data which lead to suboptimal results. The Cropalytics solution clearly overcomes these problems and provides a single view to users which is suitable for their particular requirement.
The future is green
Crop indices measurement and prediction is still in a nascent stage from an analytics and adoption stage. New technological opportunities are being experimented with for increasingly varied applications across geographies and service segments. While it is difficult to quantify benefits at an aggregate level, the efficacy of these solutions is beyond doubt.
Governments and corporates are leading the charge in leveraging technology for higher farm productivity. This can only be good news and indicative of a truly lush green future for every farmer in India.
Source: indiaai.gov.in
