Monitoring rice crop and yield estimation with Sentinel-2 data

Abstract

The future success of rice farming will lie in developing productive, sustainable, and resilient farming systems in relation to coexistent ecosystems. Thus, accurate information on agricultural practices and grain yield at optimum temporal and spatial scales is crucial. This study evaluates the potential application of Sentinel-2 (S2) to monitor the dynamics of rice fields in two consecutive seasons (2018 and 2019) in the Ebro Delta growing area. For this purpose, time series of four different spectral indices (NDVI, NDWI MF , NDWI GAO , and BSI), derived from smoothed S2 data at 20 m spatial resolution, were generated. Then, a combination of the first and second derivative analysis on the temporal profiles of spectral indices was used to automatically identify key phenology and management features from regional to field scale; and for estimating crop yield at fields. Features extracted from NDVI and NDWI GAO were used for identifying significant phenological stage dates ( i.e. Tillering, Heading Date, and Maturity), and field status ( i.e. hydroperiod), although the performance of the proposed method at field-scale was limited by S2 data gaps. The absolute minimum of NDWI MF showed great potential for estimating rice yield, including different cultivars ( r = - 0.8), and less sensibility to the number of valid images. Sentinel-2 alone cannot assure a consistent phenology monitoring at all fields but demonstrated strong capabilities for studying the performance of rice fields, thus must be considered in the development of new strategies for the management of rice-growing areas. • Clouds and shadows masking in Sentinel-2 imagery is still challenging under certain scenarios. • Sentinel-2 is a great tool for monitoring agricultural practices at small rice fields. • Rice yield is better estimated from NDWI than NDVI at the heading date. • Fast changes on spectral indices are related to variations in phenology and hydroperiod.