Soil moisture values derived from remote sensing platforms only accounts for the near surface soil layers, generally the top 5cm. Passive microwave data at L-band (1.4 GHz, 21cm wavelength) measurements are shown to be a very effective observation for surface soil moisture retrieval. An optimization model is developed for the Backpropagation Neural Network model. This optimization model utilizes the combination of the mean and standard deviation of the soil moisture values, together with the prediction process at different pre-determined, equal size regions to cope with the spatial and temporal variation of soil moisture values. This optimized model coupled with an ANN of optimum architecture, in terms of inputs and the number of neurons in the hidden layers, is developed to predict scale-to-scale and downscaling of soil moisture values. The dependency on the accuracy of the mean and standard deviation values of soil moisture data is also studied in this research by simulating the soil moisture values using a multiple regression model. This model obtains very encouraging results for these research problems.