SCALABLE DISTRIBUTED RANDOM FOREST CLASSIFICATION FOR PADDY RICE MAPPING

The present work[1] deals with the satellite based monitoring of agriculture, and specifically using Sentinel data, for the purposes of food security monitoring in South Korea. South Korea’s food security concerns have to do with the overproduction of rice and the low self-sufficiency in the production of other major grains. For this reason the systematic and large scale monitoring of the paddy rice extent has been identified as key knowledge for the high-level decision-making in regard to food security. This work addresses the Big Data implications, derived from a large scale and high-resolution paddy rice mapping application.  In this regard, a distributed Random Forest classifier has been implemented, using the cluster-computing framework Apache Spark in a High Performance Data Analytics environment. The input data to the classifier comprise of long time-series of Sentinel-1 and Sentinel-2 images, but also pertinent vegetation indices. The proposed paddy rice classification method achieves an accuracy of more than 85%, for a study site in Northwestern South Korea.

[1] This work was supported by EOPEN project, partially funded by the European Commission, under the contract number H2020-776019.