|DRAJAT, DENA - Statistics Indonesia|
Submitted to: Remote Sensing of Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/25/2021
Publication Date: 9/2/2021
Citation: Thorp, K.R., Drajat, D. 2021. Deep machine learning with Sentinel satellite data to map paddy rice production stages across West Java, Indonesia. Remote Sensing of Environment. 265. Article 112679. https://doi.org/10.1016/j.rse.2021.112679.
Interpretive Summary: Agricultural production statistics are important for governments to promote food security, conduct planning for import and export of agricultural products, and assess impacts of crop production on environmental resources. Government agencies typically deploy field observers to collect crop production data at various locations, and these data are used to develop reports on crop production statistics. Satellite remote sensing methods have potential to make crop production statistics more comprehensive and accurate. In collaboration with Badan Pusat Statistik (BPS or Statistics Indonesia), this study aimed to combine Indonesia's novel, technology-driven method for collecting agricultural survey data with satellite remote sensing images to map paddy rice production across West Java, Indonesia. Results indicated that the remote sensing analysis method could identify and map paddy rice production stages with accuracy greater than 75%. The research partnership with Statistics Indonesia was initiated through a fellowship program within the USDA Foreign Agricultural Service, and the research supports USDA's efforts at the embassy in Jakarta, Indonesia.
Technical Abstract: Indonesia recently implemented a novel, technology-driven approach for conducting agricultural production surveys, which involves monthly observations at many thousands of strategic locations and automated data logging via a cellular phone application. Data from these comprehensive surveys offer immense value for advancing remote sensing technology to map crop production across Indonesia, particularly through the development of machine learning approaches to relate survey data with satellite imagery. The objective of this study was to develop a recurrent neural network method to classify paddy rice production stages across West Java, Indonesia using synthetic aperture radar (SAR) and optical remote sensing data from Sentinel-1 and Sentinel-2 satellites. Monthly paddy rice survey data at 21,696 locations across West Java from November 2018 through April 2019 was used for model training and testing. The recurrent neural network model was then applied to develop monthly maps of paddy rice production stages at 10-m spatial resolution across West Java. Probabilities were computed for the presence of five land use classifications, including rice at tillering, heading, and harvest stages, rice fields with little to no vegetation present, and non-rice areas. Among these five classifications, the neural network performed with classification accuracies of 79.6% and 75.9% for model training and testing, respectively. Cloud cover impacted Sentinel-2 optical data for up to two-thirds of the West Java land area, and inclusion of cloud-insensitive Sentinel-1 SAR data, particularly the VH polarization (cross-polarized), was required for comprehensive mapping. Temporal patterns of paddy rice production stages were consistent among the monthly ground-based agricultural survey data and the 10-m rice production maps across West Java. The results demonstrated the value of combining modern agricultural survey data, satellite remote sensing, and deep machine learning to develop multitemporal maps of paddy rice production stages.