Google Leveraging Satellite Imagery To Build Agri Stack For India

With an eye on Indian farmers, tech giant Google is reportedly building a digital agri stack, which will leverage satellite imagery to help map the boundaries of individual farms, monitor crops, and improve yields.

As per news agency PTI, Google DeepMind’s senior director Manish Gupta said that the big tech major has already built a base model of the agri stack. 

The senior Google Deepmind executive said that the model will ascribe a unique ID to each farm and will act as a base layer of the “digital agri stack”. He cited the Indian government’s unique identification project Aadhaar as the inspiration behind the initiative. 

Speaking at the Mumbai Tech Week event on Saturday (March 1), Gupta said that the base model of the stack, besides identifying field boundaries via satellite imagery, can also screen which crops are being grown on the ground. 

Citing other use cases of the agri stack, he said that the analysis and data from the base layer can be used to build multiple applications such as “bettering the subsidy payments”, offering farm insurance or farm loans. Gupta also called on startups and other agritech players to leverage this model to build various solutions and tools. 

Explaining the need for such a platform, he noted that 40% of the Indian population works in the agriculture sector and the homegrown farm lending market alone is a $550 Bn opportunity, which is largely dominated by informal sector lenders.

Gupta also said that Google is working on a Matryoshka doll model to reduce the energy utilised by its data centres on the back of growing GenAI usage. Elaborating on this, he said that the model will club smaller models into a larger model for higher efficiency. 

“Our teams have been developing methods to make these large models efficient, one of which is called the Matryoshka models. The name was inspired by Matryoshka dolls, which have dolls nested one inside the other. We built the models in this nested manner where you have the largest model with progressively smaller models embedded inside it,” Gupta said, as per The Hindu businessline. 

He added that the model will assign a complex task to the biggest AI model, while a “moderate” and “simple” query can go to an intermediate-sized model and small model, respectively. This, he said, would lead to “tremendous savings” in energy and computational cost.

The Google Deepmind executive also rued the lack of Indic language data online, which poses a “challenge” in building multi-language AI. He noted that the big tech major is “quite determined” to close the gap, adding that his “dream” is to make Google’s AI model do better on Hindi than English.

It is pertinent to note that the big tech major has also launched the “Vaani” project, which aims to collect data in the form of voice recordings from across over 700 districts. Under the first phase of the project, Gupta noted that the company has collected 14,000 hours of audio data from 80 districts with people speaking in 59 languages.

That said, Google is building an agri stack at a time when the Indian government is also working on a similar project. It is pertinent to note that finance minister Nirmala Sitharaman proposed digital public infrastructure (DPI) for the agriculture sector in her Budget 2024-25 speech last year. Agri stack is one of the core components of DPI for agriculture. 

In her seventh budget speech, the FM said that the government, in partnership with the states, will advance the DPI initiative to cover farmers and their lands over the next three years. 

The Centre’s agri stack initiative is essentially a database of farmers, including details such as their identity, land records, coverage, income, insurance, loans and revenue history.