Though possibly the most backward district of Andhra Pradesh, Srikakulam is home to one of the more interesting experiments in recent years which, if successful, could transform the way India administers its agriculture sector, particularly its farm insurance policies. A team of 18 persons, headed by a former ISRO employee, are running a data analytics company—SatSure—that integrates data from satellite imagery and IoT devices to try and map farm output in real time and correlate this with inputs like seed quality, water and fertiliser usage. Right now, SatSure is able to correctly predict output and its fluctuations around 85-90% of the time but if this improves, as it should, with more data, it can be a very useful tool to predict farm distress and its causes—we tend to think of farm distress as only crop failure but as recent events show, even a bumper crop can lead to distress if it is not anticipated and marketing support not stepped up. So, if satellite imagery is able to predict a bumper crop early enough, the government could either step up procurement efforts or lower export duties quickly.
To get its algorithm right—this is critical since, by itself, the satellite imagery doesn’t give that much information—SatSure took a piece of farm land in Srikakulam and divided it into many parcels, each with different agricultural practices. So, there was one piece with too much fertiliser, one with too little, one with too much water, one with too little … Satellite imagery was captured from time to time, as were real-life conditions captured by measuring devices placed in the soil and connected to the internet.
Once the algorithm was finalised, crop output projections were made by the satellite team and this was measured against independent data from the ground team. This used the conventional crop-cutting experiments (CCE) that India uses—every year, India has 4 million CCEs in different parts of the country to determine crop yields and losses. When the results were matched, SatSure found it was getting it right 85-90% of the time (see table) for different types of agriculture practices. So, in the case of maize that was grown on over-fertilised land in Pathapatnam, the satellite team predicted an output of 1,248 bushels per hectare versus, when the actual crop came in, a yield of 1,361 bushels.
Imagine how this will revolutionise crop insurance if, based on interpretation of satellite imagery and on-ground IoT instruments, you are able to predict crop loss before it happens. Getting 4 million CCEs done each year by village patwaris is time-consuming and, even if you assume the sampling methodology to choose the farms for CCE is correct, there is no saying whether the CCEs were actually conducted or vouching for their accuracy. Such inefficiencies in the patwari system lead to huge delays in insurance reimbursement—a recent CAG report points out that, during 2011-16, five out of the nine selected states took more than the prescribed time of 45 days, with delays of up to 1,069 days, in processing farm insurance claims.
It is, of course, right to argue a 85-90% confidence limit is not good enough, so the SatSure pilot needs to be extended to larger areas and, in all likelihood, there will also be other groups working on similar solutions and possibly with better algorithms. In all of this, the richness of the data on each farming area is critical—SatSure itself is building an archive of satellite data going back the past 15 years and if, in addition to this, there can be data on soil-types, that will increase the system’s ability to predict outcomes correctly. Once mainstreamed, such interpretations based on satellite data can open up many more avenues for policy-makers.
If satellite and IoT data can help delineate excessively fertilised areas and show how this is lowering productivity, for instance, this can be built into government policy for the region or for fertilisers. SatSure, in fact, talks of how it has used satellite imagery to create algorithms that have helped it detect, in Brazil, which parts of various cities had the most plastic waste, to check on palm-tree cover in Dubai, to analyse where forest cover was reducing and even levels of oil inventories—the lids on oil containers rose as they were filled up, so the length of the shadow helped determine how much oil was being stocked! Images of prime ministers and chief ministers getting on to helicopters to do an aerial survey of crop and other damage look very attractive, but a fully functioning dashboard—with predictive facilities – on a district official’s desktop is a lot more useful.