RICE CROP MAPPING USING SENTINEL-1 TIME SERIES IMAGES (CASE STUDY: MAZANDARAN, IRAN)
- 1School of Surveying and Geospatial Engineering, College of Engineering, University of Tehran, Iran
- 2Centre Eau Terre Environnement, INRS-Quebec, Canada
Keywords: Rice, Sentinel-1, Random forest, Iran, Time series images, Classification
Abstract. Policymaking and planning agricultural improvement require accurate and timely information and statistics. In Iran, collecting and acquiring agricultural statistics is often done in the traditional methods. Related studies have proved that these methods mostly contain some mistakes. Multi-temporal acquisition strategies of remotely sensed data provide an opportunity to improve rice monitoring and mapping. Studying and monitoring rice paddies in vast areas is limited by the presence of cloud cover, the spatial and temporal resolution of optical sensors, and the lack of open access or systematic Radar data. Sentinel-1 satellite data, which are free to access and has a high quality of spatial and temporal resolution, can provide a great opportunity for monitoring crop products, especially rice. In this study, Sigma Nought, Gamma Nought and Beta Nought time series of Sentinel-1 data in VV, VH and VV+VH polarizations were employed for extracting areas under rice cultivation in the region of Mazandaran province, Iran. These satellite data are taken regularly every 12 days, according to the season of the region, from March 21st to September 22nd of 2018. In this study, in order to specify the rice paddies area, several fieldworks were randomly carried out for two weeks, and field data were collected as well. Field data including rice paddies areas and non-rice areas were collected as ‘Test and Train data set’ and then the Random Forrest (RF) algorithm was carried out to determine the rice paddies area. The classification result was validated using test samples. The accuracy of all classifications results are over 80% and the best result is related to Sigma Nought and gamma Nought of VH polarization, with an accuracy of 91.37%. The results showed a high capability to evaluate and monitor rice production at moderate levels in a vast area which is regularly exposed to the cloud cover.