Indicative Mapping of Rice Cropping Systems in Timor-Leste Using PlanetScope NDVI Time Series and DTW Clustering Pedro Junior Fernandes (a*), Masahiko Nagai (b)
Yamaguchi University
Abstract
Monitoring rice cropping systems and their seasonal dynamics is crucial for assessing production and planning, particularly in regions with limited field observations. We introduce a hexagonal grid-based time-series framework that segments rice-growing patterns and phenology in Timor-Leste using monthly 3 m PlanetScope imagery. The Normalized Difference Vegetation Index (NDVI) was extracted for approximately 6,000 hexagonal grid cells (50 m) from December 2018 to September 2022 and clustered using Dynamic Time Warping (DTW) to group similar growth trajectories. The optimal number of phenological regimes was determined by a majority decision across internal validity indices: Silhouette, Davies-Bouldin, and Calinski-Harabasz. These indices consistently favored two clusters, although the Elbow curve indicated a bend near three. For agronomic interpretation, we summarized each cluster^s NDVI profile into phenology metrics, including mean, max, min, amplitude, counts of low-cover months below 0.45, sustained greenness above 0.55, and peaks at or above 0.60, and assigned functional labels. The analysis identified two temporally stable regimes aligned with the national cropping calendar: (1) a single-crop system with one main-season green-up from December to May and predominantly low off-season NDVI below 0.45, and (2) an irrigated double-crop system maintaining elevated off-season greenness, with NDVI frequently at or above 0.60 for multiple months from June to November, indicating a second cycle. The results indicate that DTW clustering of satellite NDVI can effectively delineate dominant phenology and distinguish between single and double rice cropping in data-scarce settings. While the workflow is scalable and transferable, findings are indicative rather than definitive due to limited ground truth. Future work should incorporate plot-level observations, such as sowing and harvest dates, irrigation records, and GPS photo points, to calibrate labels and quantify accuracy
Keywords: ndvi, dtw clustering, time series analysis, rice crop system, Timor-Leste
