Evaluating four interpolation methods of temperature and vegetation indices obtained from satellite images in daily reference evapotranspiration modeling

Due to the requirement for extensive meteorological data, the standard FAO Penman-Monteith method for estimating reference evapotranspiration (ET0) is limited. Moreover, the lack of sufficient meteorological data in many regions has led to the utilization of remote sensing imagery as a valuable alternative. However, these images often have multi-day temporal resolutions. To obtain daily remote sensing data, in this study four mathematical functions: spline (S), cubic spline (CS), Bezier (B), and composite Bezier (CB) for interpolating 8-day land surface temperature (LST-D/N) and 16-day vegetation indices (NDVI and LAI) to daily values were compared. Subsequently, four remote sensing variables were used as inputs under 12 scenarios for two neural network models: Multi-Layer Perceptron (MLP) and Multi-Layer Perceptron combined with Stochastic Gradient Descent (MLP-SGD) to estimate ET0. This study was conducted at two stations, Urmia and Kerman, from 2001 to 2022. The determination coefficients of 0.89 in Urmia and 0.83 in Kerman demonstrated the superiority of spline-based interpolation methods in estimating ET0. Spline functions are recommended for interpolating remote sensing variables to estimate reference evapotranspiration.