Neural network with wavelet preprocessing for wheat growth stage estimation

The main purpose of this study is to demonstrate performance of Wavelet Preprocessed Neural Network (WPNN) for the estimation of winter wheat phenological stage estimation. Agrometeorological indices in correlation with the phenological stages are used as the input vector of an artificial neural network based estimator. Acquired data have been decomposed into wavelet sub-time series, using the Discrete Wavelet Transformation (DWT) with Haar mother wavelets. This procedure has been applied on different types of data and for different scales of decomposition. Decomposed sub-time series data are selected as the inputs of Neural Network (NN) for estimation performance. Phenological development, from seeding to maturity, is related to the accumulation of the temperature units above the base temperature which is commencement of changes in growth. Weekly GDD, VPD, ET0, PTU, T_min and Precipitation data were preprocessed using DWT and selected as the input to the NN to estimate winter wheat growth stages. Likewise, data segmented with respect to the phenological stage are used for validating the network. Two models were compared and the results have been provided with the error metrics through comparing the real and the estimated values.