Phenology-based classifier for mapping rainfed rice in Ebonyi State, Nigeria, using dual-polarized Sentinel-1A SAR images

Maps of rice systems are integral to programs that require reliable data to improve the productivity of smallholder farmers and ensure food security. Conventional data collection techniques and complex classification models are expensive, data-intensive, and computationally intensive. The need for effective, simple algorithms is particularly acute in Nigeria, where diverse rice production systems and crop management practices prevail. Phenology-based classifiers, integrated with microwave satellite data, have proven effective for detecting cultivated rice fields by leveraging temporal backscatter characteristics of rice throughout its growth cycle. Multi-temporal, dual-polarized Sentinel-1A SAR images at 20m resolution and field-observed data were obtained in the wet rice season of Ebonyi State, Nigeria, from June to November 2023. Center and boundary coordinates of the sample rice plots and other land cover classes were measured in the fields. Extracted backscatter values at critical rice growth stages were used to compute statistics for parametrizing rulesets. Assessment of the confusion matrix showed a user accuracy of (92%, 90%) and kappa coefficients (0.88, 0.86) for the VH and VV polarized rice maps, respectively. Comparisons with government rice data showed high accuracy, with VH outperforming VV, with lower percentage errors of -2% and -12%, respectively. A key finding showed that VH and VV polarizations were sensitive to detecting rice sown at different periods during the wet rice season. The study demonstrates the effectiveness of combining SAR data with a phenology-based classifier for detecting multiple seasons and mapping rice fields in complex agricultural systems.