Automated near-real-time mapping and monitoring of rice growth extent and stages in Selangor Malaysia

Abstract

High-resolution mapping of paddy rice is crucial to inform the potential yield and future rice production. Although remote sensing studies have been widely used to map paddy rice extent, mapping rice fields in tropical regions are still difficult as rice is grown throughout the year with varying planting dates and cropping frequency. This study addressed this challenge using continuously updated Sentinel-1 data and machine learning algorithms for near-real-time monitoring of paddy rice extent and growth stages. The study was conducted in IADA Barat Laut Selangor, one of Malaysia's highest paddy yield producers, covering an area of about 18,000 ha. The study used a phenology-based approach with Sentinel-1 monthly time series data from January to December 2020 as inputs. The results show that the developed model can accurately predict and produce a 10-m resolution map of rice growth stages by using current data and previous 5 consecutive months Sentinel-1 VH data as inputs with the SVM classifier. Validated using field survey data conducted in 2022, the model had an overall accuracy of 83.33% with kappa coefficient of 0.76. The predicted growth stages also agreed well with the published ground cropping calendar. The algorithms were implemented in the Earth Engine App, which automatically maps paddy rice growth stages as new data are available. The developed app is the first online platform that provides near real-time paddy growth stages at 10-m resolution. This information will help measure the achievement of self-sufficiency in rice production.