Validation of the cotton crop coefficient estimation model based on Sentinel-2 imagery and eddy covariance measurements

Abstract

New technologies are needed in order to improve water use efficiency and aid irrigation management. To meet this challenge, crop coefficient (Kc) estimation models based on remotely sensed crop reflectance were developed using new public domain satellite imagery. Spectral modeling of Kc is possible due to the high correlations between Kc and the crop phenologic development and spectral reflectance. In this study, cotton evapotranspiration was measured in the field during two seasons using the eddy covariance method. Kc was estimated as the ratio between reference evapotranspiration and the measured cotton evapotranspiration. In addition, a time series of Sentinel-2 imagery was processed to produce 21 vegetation indices based on the sensor’s unique spectral bands. Empirical Kc – vegetation index models were derived and ranked according to their prediction error. This was performed for each season separately and in addition cross-validation between the seasons was performed. In accordance with previous findings, we found strong correlations between Kc and indices that are based on the red and red-edge bands (MTCI, REP, and S2REP). In addition, other spectral indices that are strongly correlated to Kc were identified. The Merris Terrestrial Chlorophyll Index (MTCI) had the closest relation to Kc (R2 > 0.91). It was also demonstrated that the model developed for the first season can be successfully applied to the second season data to predict Kc, and vice-versa with RMSE < 0.1. Therefore this study cross-validates the models for estimating cotton water consumption using satellite imagery that is available at no cost at a temporal resolution of five days and a spatial resolution of 10–20 m. The confidence inspired by this validation sets the scene for near-real-time irrigation decision support systems.