Spatio-temporal soil moisture estimation using neural network with wavelet preprocessing

Soil moisture is an important indicator that defines the land surface-atmosphere interactions by contributing precisely to the surface energy and water balance. In this study, we examine the utility of the Neural Network (NN) model using Discrete Wavelet Transform (DWT) as preprocessing mechanism for soil moisture estimation. The decomposed wavelet sub-time series data was used as input to the 3-layered NN. Recognition of, as well as understanding the changes and spatial distributions of soil moisture are crucial in order to determine water usage, droughts, floods and surface runoffs. This study aims to use remote sensing data together with ground-based agro-meteorological data. The soil moisture data from sites were composed of the 15- A nd 45-cm, measured at intervals of 8 days during the period of crop growth season (October to June) between 2014 and 2015. Simultaneously, soil moisture data were selected as remotely sensed images were acquired. Utilizing remotely sensed data (Landsat 7 and Landsat 8) Vegetation Indices (VI): Landsat Surface Temperature (LST), Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI) and Modified Soil Adjusted Vegetation Index (MSAVI) were obtained. Temperature Vegetation Dryness Index (TVDI) was computed with respect to LST. The results of this study showed that the proposed model, using ground-based and remotely sensed data, should serve as an enhanced method to obtain highly reliable soil moisture values.