Mapping Croplands of South Asia using Landsat 8 Imagery and Google Earth Engine

At the 2017 CSI Meeting in Cali, Murali Gumma from ICRISAT presented an impressive progress on the mapping of super high resolution (30 meter) croplands in South Asia using Landsat 8 imagery (16-day, 8-band). More than 2,000 field samples were collected to train a random forest classification algorithm and used to generate five types of classified data products:

  1. Cropland vs. non-croplands
  2. Irrigated vs, rainfed (including water bodies)
  3. Cropping intensities: single, double, or continuous
  4. Major crop types
  5. Cropland change
Map of cropland extent in South Asia
Map of rainfed vs. irrigated croplands in South Asia

This method was successfully applied in more than 446 million ha area spanning across five countries in South Asia with the accuracy levels ranging from 80% (for rainfed vs. irrigated classification) to 90% (for cropland vs non-cropland classification). This approach is now being used to generate global-scale datasets.

Another welcoming update from Murali’s presentation was on the development of iCrops mobile app, being tested for streamlining groundtruthing data collection. We discussed the exciting potential of CGIAR-wide adoption of this tool to generate a large groundtruthing database in collaboration, which will be an invaluable global public good.

Screenshots of iCrops mobile app, streamlining groundtruthing data collection

Learn more about this study and ICRISAT’s geospatial analysis work from Murali’s presentation: Murali Gumma (ICRISAT) 2017 – Center Update – CGIAR-CSI in Cali (Sep 22)


1 comment

Comments are closed.