Optical trapezoid model (OPTRAM) as a method has been proposed to retrieve soil moisture from remote sensing data. It is based on the assumption that a trapezoidal shape would be derived from plotting of vegetation index (VI) versus shortwave-infrared transformed reflectance (STR) data assuming a linear relationship between VI and STR. A literature review and the present study indicate that the relationship between VI and STR under both dry and wet conditions, especially in the wide range of vegetation cover, is nonlinear. Therefore, we modified the OPTRAM model by introducing nonlinear edges to the VI-STR space and then employed the modified OPTRAM (denoted as MOPTRAM) using Sentinel-2 observations to predict surface (θsurf) and root zone (θrz) soil moistures. Soil moisture predicted by MOPTRAM and OPTRAM methods were compared with ground truth volumetric soil moisture data of a maize field. Accuracy of the predictions from the MOPTRAM increased as compared to that from the OPTRAM for both θsurf and θrz with a wide range of vegetation cover. The root-mean-square error and R2 for the θsurf estimates from MOPTRAM were 0.036 cm3 / cm3 and 0.748, respectively, with corresponding figures of 0.047 and 0.692 from OPTRAM, implying greater prediction accuracy for the modified model in the studied area.