Assessing Spatial Variability of Boro Rice (Oryza Sativa) Phenology Across Different Climatic Zones in Bangladesh Using Sentinel-1 and -2 Time Series Data

Abstract Accurate estimation of rice phenology is crucial for effective crop management and food security, particularly in data-scarce regions like Bangladesh. This study investigates the application of Sentinel-1 (S1) and Sentinel-2 (S2) remote sensing data to monitor the key phenological stages of Boro rice across seven diverse climatic zones. In our analysis, we first employed a k-means clustering algorithm to identify and extract Boro rice parcels from S2 optical imagery, with the results validated through field visits. We then utilized the complementary strengths of VH/VV (cross-ratio, CR) polarization from S1 and the Enhanced Vegetation Index (EVI) from S2 time series, employing local minima and breakpoints methods to estimate two critical stages of rice growth: transplanting and harvesting. The results indicated a better agreement between CR and EVI in detecting harvest dates (mean absolute error, MAE = 10.4; mean error, ME = 9.7) than transplanting dates (MAE = 17.7; ME = 17.3), with CR consistently showing a systematic delay. Although EVI generally exhibited lower Root Mean Square Error (RMSE) values than CR for most stages, the superiority of EVI over CR could not be conclusively determined due to the generic nature of the reference data, which did not fully account for climatic variations. Notably, hilly regions in the southeast and northeastern parts of the country exhibited earlier transplanting (day of the year, DOY, 9 to 16) and harvesting (DOY 109 to 129) dates for both EVI and CR, likely due to higher rainfall and moderate temperatures during the growing period. Additionally, our study found that integrating S1 and S2 data provided a more comprehensive understanding of rice phenology, highlighting the importance of considering local environmental factors for accurate phenological monitoring.