Accuracies of support vector machine and random forest in rice mapping with Sentinel-1A, Landsat-8 and Sentinel-2A datasets

Abstract

SVM and RF are widely used in rice mapping. However, their performance with single and different combinations of satellite datasets is rarely reported. Hence we report their rice mapping accuracies for two seasons using Sentinel-1A, Landsat-8 and Sentinel-2A images. The VH and VV polarizations of Sentinel-1A, and two spectral indices (SIs) of Landsat-8 and Sentine1-2A were used to obtain seven datasets (VH, VV, SI, VHVV, VHSI, VVSI and VHVVSI), and on which SVM and RF were applied and accuracies were assessed. VHSI showed the highest overall accuracy for both algorithms in both years. SVM with VHSI had a slightly higher accuracy (90.8%) than RF with VHSI (89.2%) in 2015 while in 2016 RF with VHSI showed a slightly higher accuracy (95.2%) than SVM with VHSI (93.4%). Although they produced equivalent accuracies within years, RF is more sensitive to additional data, given a 6.0% increase from 2015 to 2016 with VHSI.

Keywords:

• Sentinel-1A
• Landsat-8
• Sentinel-2A
• rice mapping

Acknowledgments

The authors express their gratitude to students of the Key Laboratory of Agricultural Remote Sensing and Information Systems at Zhejiang University for their assistance during field work.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This study was funded by the National Key R&D Programme of China under the thematic areas “Monitoring and prediction methods of paddy rice and winter wheat in the middle and lower reaches of the Yangtze River (grant no. 2016YFD0300603-5)” and “Monitoring methods of paddy rice agro-meteorological disasters in the middle and lower reaches of the Yangtze River (grant no. 2017YFD0300402-3)”.