Modeling plant density and ponding water effects on flooded rice evapotranspiration and crop coefficients: critical discussion about the concepts used in current methods

Abstract

The aim of the study is to propose new modeling approaches for daily estimations of crop coefficient K c for flooded rice (Oryza sativa L., ssp. indica) under various plant densities. Non-linear regression (NLR) and artificial neural networks (ANN) were used to predict K c based on leaf area index LAI, crop height, wind speed, water albedo, and ponding water depth. Two years of evapotranspiration ET c measurements from lysimeters located in a Mediterranean environment were used in this study. The NLR approach combines bootstrapping and Bayesian sensitivity analysis based on a semi-empirical formula. This approach provided significant information about the hidden role of the same predictor variables in the Levenberg-Marquardt ANN approach, which improved K c predictions. Relationships of production versus ET c were also built and verified by data obtained from Australia. The results of the study showed that the daily K c values, under extremely high plant densities (e.g., for LAI max > 10), can reach extremely high values (K c > 3) during the reproductive stage. Justifications given in the discussion question both the K c values given by FAO and the energy budget approaches, which assume that ET c cannot exceed a specific threshold defined by the net radiation. These approaches can no longer explain the continuous increase of global rice yields (currently are more than double in comparison to the 1960s) due to the improvement of cultivars and agriculture intensification. The study suggests that the safest method to verify predefined or modeled K c values is through preconstructed relationships of production versus ET c using field measurements.