They basically build a training dataset by identifying cropping pixels that have a higher NDVI than neighbouring pixels during the dry season. Global dry season definitions are picked up from an FAO list. An ML model is then trained on this data. But how is this then different from saying that in India double cropping == irrigation?
The definition of irrigated areas has always been quite confusing. Should areas which get supplementary groundwater irrigation during kharif also be called irrigated? This paper https://www.nature.com/articles/sdata2016118 took such an approach by building NDVI profiles for crops and compared differences at various crop stages to distinguish between irrigated and non-irrigated crops. Very tough to calibrate though!
Adi
--
Aaditeshwar Seth Microsoft Chair Professor, Computer Science and Engineering, IIT Delhi