Mapping & Classifying Paddy Fields Applying Machine Learning Algorithms with Multi-temporal Sentinel-1A in Ampara district

Abstract

In Sri Lanka, Seasonal paddy field area mapping is still doing based on the traditional methods with poor technologies. Therefore this research focuses on the machine approach of mapping paddy fields area accurately on remote sensing data taken from the satellite. Multi-temporal Sentinel1A Synthetic Aperture Radar(SAR) data was used to map the spatial distribution of the secretary’s divisions paddy area in the Ampara district during the period from April 2019 to September 2019. The classifying algorithms were mainly used under the multi-temporal spectral filter classification with 11 dual-polarization(VH/VV) SAR using SNAP, QGIS, ENVI tools. The Time series model was used for each VH and VV bands separately. According to minimum and maximum value of both VH and VV bands, paddy field area was classified using deference of min and max value respectively The overall precision of paddy fields is shown to be 0.92 Also use random forest classification method to processed images with ENVI and It shows 0.86 accuracy rate. Each divisional secretary area showed accurate paddy classification according to non-remote sensing data provided by the district agriculture office of Ampara. This method can easily be used to classify paddy cultivation areas than its traditional methods. Also, it is low cost and very fast method. As further development, Rice prediction model is proposed using the same classified area with vegetation indexes of Sentinel 2 imagery.