Demonstrating the use of remote sensing and machine-learning for agriculture
Real time information from a remote sensing satellite miles above earth helped identify crop stress due to rains in Bhadrak, Jajapur, Kendraparha and Puri along the eastern coast of India in Odisha state on 28 August. Remote sensing capabilities for such agricultural applications have increased manifold over the years and publicly available resources can be used for efficient farming and land management. This was demonstrated through ICRISAT’s work in combining remote sensing and machine learning for agriculture related applications during a recently held webinar for senior executives of the Asian Development Bank.
Using spatial and temporal maps resulting from ICRISAT and the work of its partners, Dr Murali Krishna Gumma, Head, Remote Sensing and GIS, ICRISAT, showed how remote sensing has helped identify irrigated and rainfed cropland in India. He also demonstrated crop type mapping in Jhansi district, replete with information on crop dominance. To demonstrate use of remote sensing for determining crop intensity, Dr Gumma used maps of Krishna Basin in south India.
He further said Google’s Earth Engine (GEE) provides satellite imagery from multiple open-source resources. Machine learning algorithms can then process images and classify information faster than traditional methods of processing and analysis.
Using specific datasets like Sentinel-1 imagery, Dr Gumma demonstrated how district-wise flood analysis was done in Ethiopia’s Afar region over a time. He also informed the webinar’s audience that his team is mapping rice-fallow lands in Asia using GEE for the year 2019-2020. Thus far in this exercise, the team has produced pan-Asia maps that reveal areas where rice was cultivated in one season and land left fallow next season and where a single season of rice was followed by rice/wheat or sugarcane in the next season.
For a few applications like crop type mapping, Dr Gumma demonstrated methods of using both satellite and ground data with the help of machine learning algorithms. These techniques were also used to estimate spatio-temporal distribution of drought in Myanmar between 2010 and 2017 and to witness changes in cropping patterns across India over the last two decades.
Remote sensing data can also help in yield estimation, Dr Gumma said. He showed how ICRISAT’s soil data, climate data from Indian states, remote sensing products and crop models help estimate yield. The use of remote sensing data for insurance purposes was also demonstrated with cases.