Estimation of Soil Moisture Profile Using Wavelet Neural Networks

The main purpose of the presented study is to examine the usability of a Wavelet Neural Network (WNN) model for soil moisture estimation. In this study, the wavelet transformations and neural networks have been employed to estimate the daily soil moisture. Collected data have been decomposed into wavelet sub-time series using Discrete Wavelet Transformation (DWT) with Haar mother wavelets. The sub-time series have been selected as the inputs of neural network for estimation performance. Decomposition is done on different type of data. At the same time, those decomposed sub-time series data are used like inputs to the Time-Delay Neural Network (TDNN). The selection of sub-time series has effect on the output data also. Soil moisture values at different depths are estimated using inverse discrete wavelet transformation (IDWT). DWT and IDWT are applied with the quadrature mirror filters of decomposition and synthesis filters. Also, it is shown that selection of sub-time series has impact on the neural network model's performance. Consequently, the most appropriate wavelet-NN configuration is determined for each station which means of selecting the appropriate mother wavelet, number of scales and the neural network type. The main point, in WNN type configuration is the wavelet decomposition and usage of sub-time series as inputs of neural network. The results have been provided with the error metrics of the Root Mean Square Error (RMSE) and Coefficient of Efficiency (CE) by comparing the real and estimated values.